the quarterly journal of economics...the quarterly journal of economics vol. 131 november 2016 issue...

44
THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J. DAVIS We develop a new index of economic policy uncertainty (EPU) based on newspaper coverage frequency. Several types of evidence—including human readings of 12,000 newspaper articles—indicate that our index proxies for move- ments in policy-related economic uncertainty. Our U.S. index spikes near tight presidential elections, Gulf Wars I and II, the 9/11 attacks, the failure of Lehman Brothers, the 2011 debt ceiling dispute, and other major battles over fiscal policy. Using firm-level data, we find that policy uncertainty is associated with greater stock price volatility and reduced investment and employment in policy-sensitive sectors like defense, health care, finance, and infrastructure con- struction. At the macro level, innovations in policy uncertainty foreshadow de- clines in investment, output, and employment in the United States and, in a panel vector autoregressive setting, for 12 major economies. Extending our U.S. index back to 1900, EPU rose dramatically in the 1930s (from late 1931) and has drifted upward since the 1960s. JEL Codes: D80, E22, E66, G18, L50. We thank Adam Jorring, Kyle Kost, Abdulla Al-Kuwari, Sophie Biffar, Jo ¨rn Boehnke, Vladimir Dashkeyev, Olga Deriy, Eddie Dinh, Yuto Ezure, Robin Gong, Sonam Jindal, Ruben Kim, Sylvia Klosin, Jessica Koh, Peter Lajewski, David Nebiyu, Rebecca Sachs, Ippei Shibata, Corinne Stephenson, Naoko Takeda, Melissa Tan, Sophie Wang, and Peter Xu for research assistance and the National Science Foundation, MacArthur Foundation, Sloan Foundation, Becker Friedman Institute, Initiative on Global Markets, and Stigler Center at the University of Chicago for financial support. We thank Ruedi Bachmann, Sanjai Bhagat, Vincent Bignon, Yongsung Chang, Vladimir Dashkeyev, Jesus Fernandez-Villaverde, Laurent Ferrara, Luis Garicano, Matt Gentzkow, Yuriy Gorodnichenko, Kevin Hassett, Takeo Hoshi, Greg Ip, Anil Kashyap, Patrick Kehoe, John Makin, Johannes Pfeifer, Meijun Qian, Itay Saporta, John Shoven, Sam Schulhofer-Wohl, Jesse Shapiro, Erik Sims, Stephen Terry, Cynthia Wu, and many seminar and conference audiences for comments. We also thank the referees and editors, Robert Barro and Larry Katz, for comments and suggestions. ! The Author(s) 2016. Published by Oxford University Press, on behalf of President and Fellows of Harvard College. All rights reserved. For Permissions, please email: [email protected] The Quarterly Journal of Economics (2016), 1593–1636. doi:10.1093/qje/qjw024. Advance Access publication on July 11, 2016. 1593 by guest on November 3, 2016 http://qje.oxfordjournals.org/ Downloaded from

Upload: others

Post on 18-Mar-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

THE

QUARTERLY JOURNALOF ECONOMICS

Vol 131 November 2016 Issue 4

MEASURING ECONOMIC POLICY UNCERTAINTY

SCOTT R BAKER

NICHOLAS BLOOM

STEVEN J DAVIS

We develop a new index of economic policy uncertainty (EPU) based onnewspaper coverage frequency Several types of evidencemdashincluding humanreadings of 12000 newspaper articlesmdashindicate that our index proxies for move-ments in policy-related economic uncertainty Our US index spikes near tightpresidential elections Gulf Wars I and II the 911 attacks the failure ofLehman Brothers the 2011 debt ceiling dispute and other major battles overfiscal policy Using firm-level data we find that policy uncertainty is associatedwith greater stock price volatility and reduced investment and employment inpolicy-sensitive sectors like defense health care finance and infrastructure con-struction At the macro level innovations in policy uncertainty foreshadow de-clines in investment output and employment in the United States and in apanel vector autoregressive setting for 12 major economies Extending our USindex back to 1900 EPU rose dramatically in the 1930s (from late 1931) and hasdrifted upward since the 1960s JEL Codes D80 E22 E66 G18 L50

We thank Adam Jorring Kyle Kost Abdulla Al-Kuwari Sophie Biffar JornBoehnke Vladimir Dashkeyev Olga Deriy Eddie Dinh Yuto Ezure Robin GongSonam Jindal Ruben Kim Sylvia Klosin Jessica Koh Peter Lajewski DavidNebiyu Rebecca Sachs Ippei Shibata Corinne Stephenson Naoko TakedaMelissa Tan Sophie Wang and Peter Xu for research assistance and theNational Science Foundation MacArthur Foundation Sloan Foundation BeckerFriedman Institute Initiative on Global Markets and Stigler Center at theUniversity of Chicago for financial support We thank Ruedi Bachmann SanjaiBhagat Vincent Bignon Yongsung Chang Vladimir Dashkeyev JesusFernandez-Villaverde Laurent Ferrara Luis Garicano Matt Gentzkow YuriyGorodnichenko Kevin Hassett Takeo Hoshi Greg Ip Anil Kashyap PatrickKehoe John Makin Johannes Pfeifer Meijun Qian Itay Saporta John ShovenSam Schulhofer-Wohl Jesse Shapiro Erik Sims Stephen Terry Cynthia Wu andmany seminar and conference audiences for comments We also thank the refereesand editors Robert Barro and Larry Katz for comments and suggestions

The Author(s) 2016 Published by Oxford University Press on behalf of Presidentand Fellows of Harvard College All rights reserved For Permissions please emailjournalspermissionsoupcomThe Quarterly Journal of Economics (2016) 1593ndash1636 doi101093qjeqjw024Advance Access publication on July 11 2016

1593

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

I Introduction

Concerns about policy uncertainty have intensified in thewake of the global financial crisis serial crises in the Eurozoneand partisan policy disputes in the United States For examplethe Federal Open Market Committee (2009) and theInternational Monetary Fund (IMF) (2012 2013) suggest thatuncertainty about US and European fiscal regulatory and mon-etary policies contributed to a steep economic decline in 2008ndash2009 and slow recoveries afterward1

To investigate the role of policy uncertainty we first developan index of economic policy uncertainty (EPU) for the UnitedStates and examine its evolution since 19852 Our index reflectsthe frequency of articles in 10 leading US newspapers that con-tain the following trio of terms lsquolsquoeconomicrsquorsquo or lsquolsquoeconomyrsquorsquo lsquolsquouncer-tainrsquorsquo or lsquolsquouncertaintyrsquorsquo and one or more of lsquolsquoCongressrsquorsquo lsquolsquodeficitrsquorsquolsquolsquoFederal Reserversquorsquo lsquolsquolegislationrsquorsquo lsquolsquoregulationrsquorsquo or lsquolsquoWhite HousersquorsquoThe index spikes near tight presidential elections Gulf Wars I andII the 911 attacks the 2011 debt ceiling dispute and other majorbattles over fiscal policy We extend our newspaper-based ap-proach to measuring policy uncertainty along three dimensionsback in time across countries and to specific policy categories

To push back to 1900 we rely on archives for six major USnewspapers published throughout the past century Thislong-span EPU index highlights prendashWorld War II political de-velopments and shocks like the Gold Standard Act of 1900 theoutbreak of World War I the Versailles conference in 1919 and asustained surge in policy uncertainty from late 1931 whenPresident Herbert Hoover and then President FranklinRoosevelt introduced a rash of major new policies The indexalso shows an upward drift since the 1960s perhaps due torising political polarization or the growing economic role for gov-ernment (Baker et al 2014)Using similar methods we constructEPU indexes for 11 other countries including all G10 economiesThese indexes are particularly helpful in countries with feweralternative uncertainty measures We develop category-specific

1 lsquolsquoWidespread reports from business contacts noted that uncertainties abouthealth-care tax and environmental policies were adding to businessesrsquo reluctanceto commit to higher capital spendingrsquorsquo (Federal Open Market Committee 2009) Seealso IMF (2012 pp xvndashxvi and 49ndash53 and 2013 pp 70ndash76)

2 Our data are available at monthly and daily frequencies at httpwwwpolicyuncertaintycom and are carried by Bloomberg Haver FRED and Reuters

QUARTERLY JOURNAL OF ECONOMICS1594

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty indexes for the United States by specifyingmore restrictive criteria for those articles that contain termsabout the economy policy and uncertainty For example wedevelop indexes of health care policy uncertainty and nationalsecurity policy uncertainty based on the presence of additionalterms like lsquolsquohealth carersquorsquo lsquolsquohospitalrsquorsquo or lsquolsquohealth insurancersquorsquo andlsquolsquowarrsquorsquo lsquolsquoterrorismrsquorsquo or lsquolsquodepartment of defensersquorsquo respectivelyCategory-specific shocks and policy initiatives are clearly visible

Our approach to measuring policy uncertainty raises potentialconcerns about newspaper reliability accuracy bias and consis-tency To address these concerns we evaluate our EPU index inseveral ways First we show a strong relationship between ourmeasure of EPU and other measures of economic uncertainty forexample implied stock market volatility Second we also show astrong relationship between our index and other measures of policyuncertainty for example the frequency with which the FederalReserve Systemrsquos Beige Books mention policy uncertainty Thirdwe find very similar movements in EPU indexes based on right-leaning and left-leaning newspapers suggesting that politicalslant does not seriously distort our overall EPU index

Fourth we conducted an extensive audit study of 12000 ran-domly selected articles drawn from major US newspapers Workingunder close supervision teams of University of Chicago studentsunderwent a training process and then carefully read overlappingsets of randomly selected articles guided by a 65-page referencemanual and weekly team meetings The auditors assessed whethera given article discusses economic policy uncertainty based on ourcriteria We use the audit results to select our policy term set eval-uate the performance of our computer-automated methods and con-struct additional data There is a high correlation between ourhuman- and computer-generated indexes (086 in quarterly datafrom 1985 to 2012 and 093 in annual data from 1900 to 2010) Thediscrepancy between the human and computer-generated indexes isuncorrelated with GDP growth rates and with the level of EPU

Finally our indexes have a market use validation commer-cial data providers that include Bloomberg FRED Haver andReuters carry our indexes to meet demands from banks hedgefunds corporations and policy makers This pattern of marketadoption suggests that our indexes contain useful information fora range of decision makers

In Section IV we provide evidence of how firm-level and ag-gregate outcomes evolve in the wake of policy uncertainty

ECONOMIC POLICY UNCERTAINTY 1595

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

movements Causal inference is challenging because policy re-sponds to economic conditions and is likely to be forward lookingTo make progress we follow a micro and a macro estimation ap-proach First the micro approach exploits firm-level differences inexposure to certain aspects of policy mainly government pur-chases of goods and services We use micro data from the FederalRegistry of Contracts and data on government health care spend-ing to calculate the share of firm and industry revenues derivedfrom sales to the government Next in firm-level regressions thatinclude time and firm fixed effects and other controls we show thatfirms with greater exposure to government purchases experiencegreater stock price volatility when policy uncertainty is high andreduced investment rates and employment growth when policyuncertainty rises Adding the VIX as an explanatory variable (in-teracted with firm-level exposure to government purchases) westill find greater stock price volatility and falls in investment andemployment with heightened policy uncertainty which points to apolicy uncertainty channel at work rather than a broader uncer-tainty effect We also find that firms in the defense health careand financial sectors are especially responsive to their own cate-gory-specific EPU measures confirming their information value

These firm-level results are suggestive of a causal impact ofpolicy uncertainty on investment and employment in sectors thatrely heavily on government spending and in sectors like healthcare and finance with strong exposure to major shifts in regula-tory policy However the firm-level results offer limited guidanceabout the magnitude of aggregate effects in part because theycapture only a limited set of potential policy uncertainty channels

Our second approach fits vector autoregressive (VAR) modelsto US data and to an international panel VAR that exploits ourEPU indexes for 12 countries The US VAR results indicate thata policy uncertainty innovation equivalent to the actual EPU in-crease from 2005ndash2006 to 2011ndash2012 foreshadows declines ofabout 6 in gross investment 11 in industrial productionand 035 in employment The 12-country panel VAR yields sim-ilar results3 Although our results are not necessarily causal oneplausible interpretation of our micro and macro evidence is that

3 Stock and Watson (2012) use our EPU index to investigate the factorsbehind the 2007ndash2009 recession and slow recovery and come to a similar conclu-sionmdashnamely that policy uncertainty is a strong candidate to partly explain thepoor economic performance but causal identification is hard

QUARTERLY JOURNAL OF ECONOMICS1596

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty retards investment hiring and growth in pol-icy-sensitive sectors like defense finance healthcare and con-struction and these sectors are important enough for policyuncertainty to matter at the aggregate level

This article relates to at least three strands of literature Thefirst is research on the impact of uncertainty on growth and in-vestment Theoretical work on this topic dates at least toBernanke (1983) who points out that high uncertainty givesfirms an incentive to delay investment and hiring wheninvestment projects are costly to undo or workers are costly tohire and fire4 Of course once uncertainty recedes firms increasehiring and investment to meet pent-up demand Other reasons fora depressive effect of uncertainty include precautionary spendingcutbacks by households upward pressure on the cost of finance(eg Pastor and Veronesi 2013 Gilchrist Sim and Zakrajsek2014) managerial risk aversion (eg Panousi and Papanikolaou2012) and interactions between nominal rigidities and searchfrictions (Basu and Bundick 2012 Leduc and Liu 2015)

Second there is a literature focused explicitly on policy un-certainty Friedman (1968) Rodrik (1991) Higgs (1997) andHassett and Metcalf (1999) among others consider the detrimen-tal economic effects of monetary fiscal and regulatory policy un-certainty More recently Born and Pfeifer (2014) and Fernandez-Villaverde at al (2015) study policy uncertainty in DSGE modelsfinding moderately negative effects while Pastor and Veronesi(2012 2013) model the theoretical links among fluctuationspolicy uncertainty and stock market volatility5

4 Dixit and Pindyck (1994) offer a review of the early theoretical literatureincluding papers by Oi (1961) Hartman (1972) and Abel (1983) that highlightpotentially positive effects of uncertainty Recent empirical papers include Bloom(2009) Bachman Elstener and Sims (2013) Bloom et al (2014) and Scotti (2016)with a review in Bloom (2014)

5 In other related work Julio and Yook (2012) find that investment fallsaround national elections Durnev (2010) finds that corporate investment becomesless responsive to stock prices in election years Brogaard and Detzel (2015) findthat policy uncertainty reduces asset returns Handley and Limao (2015) find thattrade policy uncertainty delays firm entry Gulen and Ion (2016) find negative re-sponses of corporate investment to our EPU index Koijen Philipson and Uhlig(2016) develop evidence that government-induced uncertainty about profitabilitygenerates a large equity risk premium for firms in the health care sector and redu-ces their medical RampD and Giavazzi and McMahon (2012) find that policy uncer-tainty led German households to increase savings in the run-up to the close andconsequential general elections in 1998

ECONOMIC POLICY UNCERTAINTY 1597

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Finally there is a rapidly growing literature on text searchmethodsmdashusing newspaper archives in particularmdashto measure avariety of outcomes Examples include Gentzkow and Shapiro(2010) Hoberg and Phillips (2010) Boudoukh et al (2013) andAlexopoulos and Cohen (2015) Our work suggests that newspa-per text search can yield useful proxies for economic and policyconditions stretching back several decades which could be espe-cially valuable in earlier eras and in countries with fewer datasources

Section II describes the data we use to construct our policyuncertainty indexes Section III evaluates our EPU measures inseveral ways and develops additional evidence about movementsin policy-related uncertainty over time Section IV investigateshow firm-level outcomes covary with policy uncertainty and thedynamic responses of aggregate outcomes to policy uncertaintyinnovations Section V concludes and offers some thoughts aboutdirections for future research

II Measuring EPU

We build indexes of policy-related economic uncertaintybased on newspaper coverage frequency6 We aim to capture un-certainty about who will make economic policy decisions whateconomic policy actions will be undertaken and when and theeconomic effects of policy actions (or inaction)mdashincluding uncer-tainties related to the economic ramifications of lsquolsquononeconomicrsquorsquopolicy matters for example military actions Our measures cap-ture both near-term concerns (eg when will the Fed adjust itspolicy rate) and longer term concerns (eg how to fund entitle-ment programs) as reflected in newspaper articles We first de-scribe the construction of our monthly and daily EPU indexes forthe United States from 1985 onward and then turn to indexes forspecific policy categories indexes for other countries and histor-ical indexes for the United States and United Kingdom

6 Earlier drafts of this article include index components based on (i) the pre-sent value of future scheduled tax code expirations and (ii) disagreement amongprofessional forecasters over future government purchases and consumer pricesHowever to extend our EPU measures over time and across countries we focushere on the newspaper approach while continuing to report the other componentsat httpwwwpolicyuncertaintycom

QUARTERLY JOURNAL OF ECONOMICS1598

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

IIA US Economic Policy Uncertainty Indexes from 1985

Our modern monthly EPU index for the United States relieson 10 leading newspapers USA Today Miami Herald ChicagoTribune Washington Post Los Angeles Times Boston Globe SanFrancisco Chronicle Dallas Morning News New York Timesand Wall Street Journal We search the digital archives of eachpaper from January 1985 to obtain a monthly count of articlesthat contain the following trio of terms lsquolsquouncertaintyrsquorsquo or lsquolsquouncer-tainrsquorsquo lsquolsquoeconomicrsquorsquo or lsquolsquoeconomyrsquorsquo and one of the following policyterms lsquolsquoCongressrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederal Reserversquorsquo lsquolsquolegislationrsquorsquolsquolsquoregulationrsquorsquo or lsquolsquoWhite Housersquorsquo (including variants likelsquolsquouncertaintiesrsquorsquo lsquolsquoregulatoryrsquorsquo or lsquolsquothe Fedrsquorsquo) In other words tomeet our criteria an article must contain terms in all three cat-egories pertaining to uncertainty the economy and policy Weuse our audit study to select the policy terms as explained inSection IIIA

An obvious difficulty with these raw counts is that the over-all volume of articles varies across newspapers and time Thuswe scale the raw counts by the total number of articles in thesame newspaper and month We standardize each monthlynewspaper-level series to unit standard deviation from 1985 to2009 and then average across the 10 papers by month Finallywe normalize the 10-paper series to a mean of 100 from 1985 to2009 To be precise let Xit denote the scaled EPU frequencycounts for newspaper i = 1 2 10 in month t and let T1 andT2 denote the time intervals used in the standardization andnormalization calculations We proceed in the following steps(i) Compute the times-series variance 2

i in the interval T1 foreach paper i (ii) Standardize Xit by dividing through by thestandard deviation i for all t This operation yields for eachpaper a series Yit with unit standard deviation in the intervalT1 (iii) Compute the mean over newspapers of Yit in each monthto obtain the series Zt (iv) Compute M the mean value of Zt inthe interval T2 (v) Multiply Zt by (100M) for all t to obtain thenormalized EPU time-series index We use the same approachfor other countries and indexes

Figure I plots the resulting index which shows clear spikesaround the Gulf Wars close presidential elections the 911 ter-rorist attack the stimulus debate in early 2008 the LehmanBrothers bankruptcy and TARP legislation in late 2008 thesummer 2011 debt ceiling dispute and the battle over the lsquolsquofiscal

ECONOMIC POLICY UNCERTAINTY 1599

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

cliffrsquorsquo in late 2012 among other events and developments Somenotable political events do not generate high EPU according toour index For instance our EPU index shows no large spike inconnection with the partial federal government shutdowns fromNovember 1995 to January 1996 although those shutdowns re-ceived quite a lot of press coverage7

In addition to our monthly index we produce a daily EPUindex using the Newsbank news aggregator which coversaround 1500 US newspapers Newsbankrsquos extensive coverageyields enough articles to generate a meaningful daily countTaking monthly averages of our daily index it correlates at 085with our 10-paper monthly index indicating a high degree of sim-ilarity Because papers enter and leave the Newsbank archive andits count of newspapers expands greatly over time compositionalshifts potentially distort the longer term behavior of the daily EPU

FIGURE I

EPU Index for the United States

7 We find more than 8000 articles about these shutdowns in Newsbank ar-chives but less than 25 also mention the economy less than 2 mention uncer-tainty and only 1 mentions both Thus politically tumultuous episodes do notnecessarily raise EPU by our measure

QUARTERLY JOURNAL OF ECONOMICS1600

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

index Hence we focus on our 10-paper monthly EPU index butthe daily index provides a useful high-frequency alternative8

IIB EPU Indexes for Policy Categories

To create indexes for policy categories we apply additionalcriteria to those articles that contain our trio of terms about theeconomy policy and uncertainty The additional criteria involvethe presence of one or more category-relevant terms lsquolsquothe Fedrsquorsquolsquolsquocentral bankrsquorsquo lsquolsquointerest ratersquorsquo lsquolsquoinflationrsquorsquo and so on for the mon-etary policy category for example Online Appendix B reports thefull set of terms that define our 11 policy categories and subcat-egories We use Newsbank for the category indexes because itshigh text density facilitates measurement by time period andpolicy category As seen in Figure II the national security EPUindex spiked sharply in connection with the 911 attacks GulfWar I and the onset of Gulf War II The health care EPU index

FIGURE II

National Security and Health Care EPU Indexes

8 We update the daily EPU index at approximately 9 am EST each day andpost it at httpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1601

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

rose sharply during the Clinton health care reform initiative in1993ndash1994 and has fluctuated at high levels from 2009 to 2014

Table I reports all 11 category-specific EPU indexes9 It alsoreports an overall economic uncertainty (EU) index that drops thepolicy requirement in the EPU index The first two rows reportaverage EU and EPU values for the indicated periods expressedrelative to the average EPU value from 1985 to 2014 For exam-ple the EU value of 2182 says the (scaled) frequency of EU arti-cles from 19851 to 19906 is somewhat more than twice theaverage frequency of EPU articles from 1985 to 2014 The next11 rows report relative frequency values for specific policy cate-gories and time periods For example the 548 value for lsquolsquoNationalSecurityrsquorsquo says the frequency of EPU articles during 20019 to200212 that mention national security matters is 548 of the1985ndash2014 average EPU frequency and 43 548

1285

of the EPU

frequency from 20019 to 200212Fiscal matters especially tax policy stand out in Table I as the

largest source of policy uncertainty especially in recent years Thefiscal policy EPU index rose from values near 33 in the precrisisyears to 615 in 20089 to 200912 and 783 from 2010 to 2013Health care policy is the second largest source of elevated EPU inrecent years Policy uncertainty related to financial regulations andentitlement programs also rose sharply after 2008 but from initiallylower levels Concerns related to sovereign debt and currency crisesare up by an order of magnitude during 2010 to 2013 but from sucha low base as to have little impact on the overall EPU index EPUconcerns related to monetary policy are important throughout the1985ndash2014 period but perhaps surprisingly they are not elevatedin recent years by our measure We interpret this result as a reflec-tion of low and stable inflation rates in recent years which appar-ently drive newspaper coverage more than disputes amongprofessional economists about unconventional monetary policies10

Several other researchers develop measures related to uncer-tainty about government behavior Marina Azzimonti (2015) con-structs a newspaper index of partisan conflict at the federal levelthat shows similarities to our EPU index but also notable

9 In contrast to Figure III which normalizes each category-specific EPUseries to 100 Table I expresses each category-specific EPU series as a percentageof the overall EPU frequency from 1985 to 2014

10 Other evidence also points to subdued levels of inflation uncertainty inrecent years See Nalewaik (2015) for a presentation and discussion of evidencebased on time-series models surveys and financial markets data

QUARTERLY JOURNAL OF ECONOMICS1602

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EI

EC

ON

OM

ICP

OL

ICY

UN

CE

RT

AIN

TY

BY

PO

LIC

YC

AT

EG

OR

YA

ND

TIM

EP

ER

IOD

1985ndash2014

Tim

ep

erio

d19851

ndash19906

19907

ndash19911

219921

ndash20018

20019

ndash20021

220031

ndash20076

20077

ndash20088

20089

ndash20091

220101

ndash20131

019851

ndash20141

2

Mid

-80s

toG

ulf

War

IG

ulf

War

I1990s

boo

mto

91

191

1att

ack

s2000s

boo

m

Earl

ycr

edit

cru

nch

Leh

man

coll

ap

seamp

rece

ssio

n

Fis

cal

pol

icy

batt

les

Over

all

aver

age

Over

all

econ

omic

un

cert

ain

ty2182

3498

1859

3269

1598

1848

3709

2521

2193

Eco

nom

icp

olic

yu

nce

rtain

ty1096

1419

881

1285

714

834

1321

1275

1000

Fis

cal

pol

icy

496

596

359

554

323

331

615

783

461

Taxes

399

484

319

512

302

314

569

681

403

Gov

ern

men

tsp

end

ing

ampot

her

227

268

121

173

85

66

171

332

171

Mon

etary

pol

icy

327

418

261

452

222

316

278

261

281

Hea

lth

care

70

154

149

184

131

134

293

393

173

Nati

onal

secu

rity

250

536

180

548

254

159

213

198

238

Reg

ula

tion

157

230

145

196

112

155

292

281

174

Fin

an

cial

regu

lati

on33

70

13

53

17

36

102

61

33

Sov

erei

gn

deb

tamp

curr

ency

cris

es14

06

23

05

04

03

04

39

16

En

titl

emen

tp

rogra

ms

73

126

115

187

88

82

153

247

124

Tra

de

pol

icy

38

40

63

26

17

20

14

21

38

Su

mof

pol

icy

cate

gor

ies

1425

2107

1295

2151

1152

1200

1863

2222

1506

Rati

oof

EP

Uto

over

all

EU

05

004

104

703

904

504

503

605

104

7

Not

es

Qu

erie

sru

nF

ebru

ary

12

2015

onU

S

new

spap

ers

inA

cces

sW

orld

New

sN

ewsb

an

k

usi

ng

the

cate

gor

y-s

pec

ific

pol

icy

term

sets

list

edin

On

lin

eA

pp

end

ixB

E

xce

pt

for

the

last

row

all

entr

ies

are

exp

ress

edre

lati

ve

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

lsquolsquoOver

all

econ

omic

un

cert

ain

tyrsquorsquo

qu

an

tifi

esth

efr

equ

ency

ofart

icle

sth

at

mee

tou

rlsquolsquoe

con

omyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

requ

irem

ents

(ie

d

rop

pin

gth

elsquolsquop

olic

yrsquorsquo

requ

irem

ent)

an

dis

als

oex

pre

ssed

rela

tive

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

Th

eca

tegor

y-

spec

ific

ind

exvalu

essu

mto

mor

eth

an

100

for

two

reaso

ns

firs

tw

eu

sea

few

pol

icy

term

sin

mor

eth

an

one

pol

icy

cate

gor

y

For

exam

ple

lsquolsquoM

edic

aid

rsquorsquoap

pea

rsin

the

term

sets

for

bot

hh

ealt

hca

rean

den

titl

emen

tp

rogra

ms

Sec

ond

a

new

spap

erart

icle

that

mee

tsth

elsquolsquoe

con

omyrsquorsquo

lsquolsquopol

icyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

crit

eria

can

refe

rto

mor

eth

an

one

pol

icy

cate

gor

y

ECONOMIC POLICY UNCERTAINTY 1603

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 2: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

I Introduction

Concerns about policy uncertainty have intensified in thewake of the global financial crisis serial crises in the Eurozoneand partisan policy disputes in the United States For examplethe Federal Open Market Committee (2009) and theInternational Monetary Fund (IMF) (2012 2013) suggest thatuncertainty about US and European fiscal regulatory and mon-etary policies contributed to a steep economic decline in 2008ndash2009 and slow recoveries afterward1

To investigate the role of policy uncertainty we first developan index of economic policy uncertainty (EPU) for the UnitedStates and examine its evolution since 19852 Our index reflectsthe frequency of articles in 10 leading US newspapers that con-tain the following trio of terms lsquolsquoeconomicrsquorsquo or lsquolsquoeconomyrsquorsquo lsquolsquouncer-tainrsquorsquo or lsquolsquouncertaintyrsquorsquo and one or more of lsquolsquoCongressrsquorsquo lsquolsquodeficitrsquorsquolsquolsquoFederal Reserversquorsquo lsquolsquolegislationrsquorsquo lsquolsquoregulationrsquorsquo or lsquolsquoWhite HousersquorsquoThe index spikes near tight presidential elections Gulf Wars I andII the 911 attacks the 2011 debt ceiling dispute and other majorbattles over fiscal policy We extend our newspaper-based ap-proach to measuring policy uncertainty along three dimensionsback in time across countries and to specific policy categories

To push back to 1900 we rely on archives for six major USnewspapers published throughout the past century Thislong-span EPU index highlights prendashWorld War II political de-velopments and shocks like the Gold Standard Act of 1900 theoutbreak of World War I the Versailles conference in 1919 and asustained surge in policy uncertainty from late 1931 whenPresident Herbert Hoover and then President FranklinRoosevelt introduced a rash of major new policies The indexalso shows an upward drift since the 1960s perhaps due torising political polarization or the growing economic role for gov-ernment (Baker et al 2014)Using similar methods we constructEPU indexes for 11 other countries including all G10 economiesThese indexes are particularly helpful in countries with feweralternative uncertainty measures We develop category-specific

1 lsquolsquoWidespread reports from business contacts noted that uncertainties abouthealth-care tax and environmental policies were adding to businessesrsquo reluctanceto commit to higher capital spendingrsquorsquo (Federal Open Market Committee 2009) Seealso IMF (2012 pp xvndashxvi and 49ndash53 and 2013 pp 70ndash76)

2 Our data are available at monthly and daily frequencies at httpwwwpolicyuncertaintycom and are carried by Bloomberg Haver FRED and Reuters

QUARTERLY JOURNAL OF ECONOMICS1594

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty indexes for the United States by specifyingmore restrictive criteria for those articles that contain termsabout the economy policy and uncertainty For example wedevelop indexes of health care policy uncertainty and nationalsecurity policy uncertainty based on the presence of additionalterms like lsquolsquohealth carersquorsquo lsquolsquohospitalrsquorsquo or lsquolsquohealth insurancersquorsquo andlsquolsquowarrsquorsquo lsquolsquoterrorismrsquorsquo or lsquolsquodepartment of defensersquorsquo respectivelyCategory-specific shocks and policy initiatives are clearly visible

Our approach to measuring policy uncertainty raises potentialconcerns about newspaper reliability accuracy bias and consis-tency To address these concerns we evaluate our EPU index inseveral ways First we show a strong relationship between ourmeasure of EPU and other measures of economic uncertainty forexample implied stock market volatility Second we also show astrong relationship between our index and other measures of policyuncertainty for example the frequency with which the FederalReserve Systemrsquos Beige Books mention policy uncertainty Thirdwe find very similar movements in EPU indexes based on right-leaning and left-leaning newspapers suggesting that politicalslant does not seriously distort our overall EPU index

Fourth we conducted an extensive audit study of 12000 ran-domly selected articles drawn from major US newspapers Workingunder close supervision teams of University of Chicago studentsunderwent a training process and then carefully read overlappingsets of randomly selected articles guided by a 65-page referencemanual and weekly team meetings The auditors assessed whethera given article discusses economic policy uncertainty based on ourcriteria We use the audit results to select our policy term set eval-uate the performance of our computer-automated methods and con-struct additional data There is a high correlation between ourhuman- and computer-generated indexes (086 in quarterly datafrom 1985 to 2012 and 093 in annual data from 1900 to 2010) Thediscrepancy between the human and computer-generated indexes isuncorrelated with GDP growth rates and with the level of EPU

Finally our indexes have a market use validation commer-cial data providers that include Bloomberg FRED Haver andReuters carry our indexes to meet demands from banks hedgefunds corporations and policy makers This pattern of marketadoption suggests that our indexes contain useful information fora range of decision makers

In Section IV we provide evidence of how firm-level and ag-gregate outcomes evolve in the wake of policy uncertainty

ECONOMIC POLICY UNCERTAINTY 1595

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

movements Causal inference is challenging because policy re-sponds to economic conditions and is likely to be forward lookingTo make progress we follow a micro and a macro estimation ap-proach First the micro approach exploits firm-level differences inexposure to certain aspects of policy mainly government pur-chases of goods and services We use micro data from the FederalRegistry of Contracts and data on government health care spend-ing to calculate the share of firm and industry revenues derivedfrom sales to the government Next in firm-level regressions thatinclude time and firm fixed effects and other controls we show thatfirms with greater exposure to government purchases experiencegreater stock price volatility when policy uncertainty is high andreduced investment rates and employment growth when policyuncertainty rises Adding the VIX as an explanatory variable (in-teracted with firm-level exposure to government purchases) westill find greater stock price volatility and falls in investment andemployment with heightened policy uncertainty which points to apolicy uncertainty channel at work rather than a broader uncer-tainty effect We also find that firms in the defense health careand financial sectors are especially responsive to their own cate-gory-specific EPU measures confirming their information value

These firm-level results are suggestive of a causal impact ofpolicy uncertainty on investment and employment in sectors thatrely heavily on government spending and in sectors like healthcare and finance with strong exposure to major shifts in regula-tory policy However the firm-level results offer limited guidanceabout the magnitude of aggregate effects in part because theycapture only a limited set of potential policy uncertainty channels

Our second approach fits vector autoregressive (VAR) modelsto US data and to an international panel VAR that exploits ourEPU indexes for 12 countries The US VAR results indicate thata policy uncertainty innovation equivalent to the actual EPU in-crease from 2005ndash2006 to 2011ndash2012 foreshadows declines ofabout 6 in gross investment 11 in industrial productionand 035 in employment The 12-country panel VAR yields sim-ilar results3 Although our results are not necessarily causal oneplausible interpretation of our micro and macro evidence is that

3 Stock and Watson (2012) use our EPU index to investigate the factorsbehind the 2007ndash2009 recession and slow recovery and come to a similar conclu-sionmdashnamely that policy uncertainty is a strong candidate to partly explain thepoor economic performance but causal identification is hard

QUARTERLY JOURNAL OF ECONOMICS1596

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty retards investment hiring and growth in pol-icy-sensitive sectors like defense finance healthcare and con-struction and these sectors are important enough for policyuncertainty to matter at the aggregate level

This article relates to at least three strands of literature Thefirst is research on the impact of uncertainty on growth and in-vestment Theoretical work on this topic dates at least toBernanke (1983) who points out that high uncertainty givesfirms an incentive to delay investment and hiring wheninvestment projects are costly to undo or workers are costly tohire and fire4 Of course once uncertainty recedes firms increasehiring and investment to meet pent-up demand Other reasons fora depressive effect of uncertainty include precautionary spendingcutbacks by households upward pressure on the cost of finance(eg Pastor and Veronesi 2013 Gilchrist Sim and Zakrajsek2014) managerial risk aversion (eg Panousi and Papanikolaou2012) and interactions between nominal rigidities and searchfrictions (Basu and Bundick 2012 Leduc and Liu 2015)

Second there is a literature focused explicitly on policy un-certainty Friedman (1968) Rodrik (1991) Higgs (1997) andHassett and Metcalf (1999) among others consider the detrimen-tal economic effects of monetary fiscal and regulatory policy un-certainty More recently Born and Pfeifer (2014) and Fernandez-Villaverde at al (2015) study policy uncertainty in DSGE modelsfinding moderately negative effects while Pastor and Veronesi(2012 2013) model the theoretical links among fluctuationspolicy uncertainty and stock market volatility5

4 Dixit and Pindyck (1994) offer a review of the early theoretical literatureincluding papers by Oi (1961) Hartman (1972) and Abel (1983) that highlightpotentially positive effects of uncertainty Recent empirical papers include Bloom(2009) Bachman Elstener and Sims (2013) Bloom et al (2014) and Scotti (2016)with a review in Bloom (2014)

5 In other related work Julio and Yook (2012) find that investment fallsaround national elections Durnev (2010) finds that corporate investment becomesless responsive to stock prices in election years Brogaard and Detzel (2015) findthat policy uncertainty reduces asset returns Handley and Limao (2015) find thattrade policy uncertainty delays firm entry Gulen and Ion (2016) find negative re-sponses of corporate investment to our EPU index Koijen Philipson and Uhlig(2016) develop evidence that government-induced uncertainty about profitabilitygenerates a large equity risk premium for firms in the health care sector and redu-ces their medical RampD and Giavazzi and McMahon (2012) find that policy uncer-tainty led German households to increase savings in the run-up to the close andconsequential general elections in 1998

ECONOMIC POLICY UNCERTAINTY 1597

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Finally there is a rapidly growing literature on text searchmethodsmdashusing newspaper archives in particularmdashto measure avariety of outcomes Examples include Gentzkow and Shapiro(2010) Hoberg and Phillips (2010) Boudoukh et al (2013) andAlexopoulos and Cohen (2015) Our work suggests that newspa-per text search can yield useful proxies for economic and policyconditions stretching back several decades which could be espe-cially valuable in earlier eras and in countries with fewer datasources

Section II describes the data we use to construct our policyuncertainty indexes Section III evaluates our EPU measures inseveral ways and develops additional evidence about movementsin policy-related uncertainty over time Section IV investigateshow firm-level outcomes covary with policy uncertainty and thedynamic responses of aggregate outcomes to policy uncertaintyinnovations Section V concludes and offers some thoughts aboutdirections for future research

II Measuring EPU

We build indexes of policy-related economic uncertaintybased on newspaper coverage frequency6 We aim to capture un-certainty about who will make economic policy decisions whateconomic policy actions will be undertaken and when and theeconomic effects of policy actions (or inaction)mdashincluding uncer-tainties related to the economic ramifications of lsquolsquononeconomicrsquorsquopolicy matters for example military actions Our measures cap-ture both near-term concerns (eg when will the Fed adjust itspolicy rate) and longer term concerns (eg how to fund entitle-ment programs) as reflected in newspaper articles We first de-scribe the construction of our monthly and daily EPU indexes forthe United States from 1985 onward and then turn to indexes forspecific policy categories indexes for other countries and histor-ical indexes for the United States and United Kingdom

6 Earlier drafts of this article include index components based on (i) the pre-sent value of future scheduled tax code expirations and (ii) disagreement amongprofessional forecasters over future government purchases and consumer pricesHowever to extend our EPU measures over time and across countries we focushere on the newspaper approach while continuing to report the other componentsat httpwwwpolicyuncertaintycom

QUARTERLY JOURNAL OF ECONOMICS1598

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

IIA US Economic Policy Uncertainty Indexes from 1985

Our modern monthly EPU index for the United States relieson 10 leading newspapers USA Today Miami Herald ChicagoTribune Washington Post Los Angeles Times Boston Globe SanFrancisco Chronicle Dallas Morning News New York Timesand Wall Street Journal We search the digital archives of eachpaper from January 1985 to obtain a monthly count of articlesthat contain the following trio of terms lsquolsquouncertaintyrsquorsquo or lsquolsquouncer-tainrsquorsquo lsquolsquoeconomicrsquorsquo or lsquolsquoeconomyrsquorsquo and one of the following policyterms lsquolsquoCongressrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederal Reserversquorsquo lsquolsquolegislationrsquorsquolsquolsquoregulationrsquorsquo or lsquolsquoWhite Housersquorsquo (including variants likelsquolsquouncertaintiesrsquorsquo lsquolsquoregulatoryrsquorsquo or lsquolsquothe Fedrsquorsquo) In other words tomeet our criteria an article must contain terms in all three cat-egories pertaining to uncertainty the economy and policy Weuse our audit study to select the policy terms as explained inSection IIIA

An obvious difficulty with these raw counts is that the over-all volume of articles varies across newspapers and time Thuswe scale the raw counts by the total number of articles in thesame newspaper and month We standardize each monthlynewspaper-level series to unit standard deviation from 1985 to2009 and then average across the 10 papers by month Finallywe normalize the 10-paper series to a mean of 100 from 1985 to2009 To be precise let Xit denote the scaled EPU frequencycounts for newspaper i = 1 2 10 in month t and let T1 andT2 denote the time intervals used in the standardization andnormalization calculations We proceed in the following steps(i) Compute the times-series variance 2

i in the interval T1 foreach paper i (ii) Standardize Xit by dividing through by thestandard deviation i for all t This operation yields for eachpaper a series Yit with unit standard deviation in the intervalT1 (iii) Compute the mean over newspapers of Yit in each monthto obtain the series Zt (iv) Compute M the mean value of Zt inthe interval T2 (v) Multiply Zt by (100M) for all t to obtain thenormalized EPU time-series index We use the same approachfor other countries and indexes

Figure I plots the resulting index which shows clear spikesaround the Gulf Wars close presidential elections the 911 ter-rorist attack the stimulus debate in early 2008 the LehmanBrothers bankruptcy and TARP legislation in late 2008 thesummer 2011 debt ceiling dispute and the battle over the lsquolsquofiscal

ECONOMIC POLICY UNCERTAINTY 1599

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

cliffrsquorsquo in late 2012 among other events and developments Somenotable political events do not generate high EPU according toour index For instance our EPU index shows no large spike inconnection with the partial federal government shutdowns fromNovember 1995 to January 1996 although those shutdowns re-ceived quite a lot of press coverage7

In addition to our monthly index we produce a daily EPUindex using the Newsbank news aggregator which coversaround 1500 US newspapers Newsbankrsquos extensive coverageyields enough articles to generate a meaningful daily countTaking monthly averages of our daily index it correlates at 085with our 10-paper monthly index indicating a high degree of sim-ilarity Because papers enter and leave the Newsbank archive andits count of newspapers expands greatly over time compositionalshifts potentially distort the longer term behavior of the daily EPU

FIGURE I

EPU Index for the United States

7 We find more than 8000 articles about these shutdowns in Newsbank ar-chives but less than 25 also mention the economy less than 2 mention uncer-tainty and only 1 mentions both Thus politically tumultuous episodes do notnecessarily raise EPU by our measure

QUARTERLY JOURNAL OF ECONOMICS1600

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

index Hence we focus on our 10-paper monthly EPU index butthe daily index provides a useful high-frequency alternative8

IIB EPU Indexes for Policy Categories

To create indexes for policy categories we apply additionalcriteria to those articles that contain our trio of terms about theeconomy policy and uncertainty The additional criteria involvethe presence of one or more category-relevant terms lsquolsquothe Fedrsquorsquolsquolsquocentral bankrsquorsquo lsquolsquointerest ratersquorsquo lsquolsquoinflationrsquorsquo and so on for the mon-etary policy category for example Online Appendix B reports thefull set of terms that define our 11 policy categories and subcat-egories We use Newsbank for the category indexes because itshigh text density facilitates measurement by time period andpolicy category As seen in Figure II the national security EPUindex spiked sharply in connection with the 911 attacks GulfWar I and the onset of Gulf War II The health care EPU index

FIGURE II

National Security and Health Care EPU Indexes

8 We update the daily EPU index at approximately 9 am EST each day andpost it at httpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1601

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

rose sharply during the Clinton health care reform initiative in1993ndash1994 and has fluctuated at high levels from 2009 to 2014

Table I reports all 11 category-specific EPU indexes9 It alsoreports an overall economic uncertainty (EU) index that drops thepolicy requirement in the EPU index The first two rows reportaverage EU and EPU values for the indicated periods expressedrelative to the average EPU value from 1985 to 2014 For exam-ple the EU value of 2182 says the (scaled) frequency of EU arti-cles from 19851 to 19906 is somewhat more than twice theaverage frequency of EPU articles from 1985 to 2014 The next11 rows report relative frequency values for specific policy cate-gories and time periods For example the 548 value for lsquolsquoNationalSecurityrsquorsquo says the frequency of EPU articles during 20019 to200212 that mention national security matters is 548 of the1985ndash2014 average EPU frequency and 43 548

1285

of the EPU

frequency from 20019 to 200212Fiscal matters especially tax policy stand out in Table I as the

largest source of policy uncertainty especially in recent years Thefiscal policy EPU index rose from values near 33 in the precrisisyears to 615 in 20089 to 200912 and 783 from 2010 to 2013Health care policy is the second largest source of elevated EPU inrecent years Policy uncertainty related to financial regulations andentitlement programs also rose sharply after 2008 but from initiallylower levels Concerns related to sovereign debt and currency crisesare up by an order of magnitude during 2010 to 2013 but from sucha low base as to have little impact on the overall EPU index EPUconcerns related to monetary policy are important throughout the1985ndash2014 period but perhaps surprisingly they are not elevatedin recent years by our measure We interpret this result as a reflec-tion of low and stable inflation rates in recent years which appar-ently drive newspaper coverage more than disputes amongprofessional economists about unconventional monetary policies10

Several other researchers develop measures related to uncer-tainty about government behavior Marina Azzimonti (2015) con-structs a newspaper index of partisan conflict at the federal levelthat shows similarities to our EPU index but also notable

9 In contrast to Figure III which normalizes each category-specific EPUseries to 100 Table I expresses each category-specific EPU series as a percentageof the overall EPU frequency from 1985 to 2014

10 Other evidence also points to subdued levels of inflation uncertainty inrecent years See Nalewaik (2015) for a presentation and discussion of evidencebased on time-series models surveys and financial markets data

QUARTERLY JOURNAL OF ECONOMICS1602

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EI

EC

ON

OM

ICP

OL

ICY

UN

CE

RT

AIN

TY

BY

PO

LIC

YC

AT

EG

OR

YA

ND

TIM

EP

ER

IOD

1985ndash2014

Tim

ep

erio

d19851

ndash19906

19907

ndash19911

219921

ndash20018

20019

ndash20021

220031

ndash20076

20077

ndash20088

20089

ndash20091

220101

ndash20131

019851

ndash20141

2

Mid

-80s

toG

ulf

War

IG

ulf

War

I1990s

boo

mto

91

191

1att

ack

s2000s

boo

m

Earl

ycr

edit

cru

nch

Leh

man

coll

ap

seamp

rece

ssio

n

Fis

cal

pol

icy

batt

les

Over

all

aver

age

Over

all

econ

omic

un

cert

ain

ty2182

3498

1859

3269

1598

1848

3709

2521

2193

Eco

nom

icp

olic

yu

nce

rtain

ty1096

1419

881

1285

714

834

1321

1275

1000

Fis

cal

pol

icy

496

596

359

554

323

331

615

783

461

Taxes

399

484

319

512

302

314

569

681

403

Gov

ern

men

tsp

end

ing

ampot

her

227

268

121

173

85

66

171

332

171

Mon

etary

pol

icy

327

418

261

452

222

316

278

261

281

Hea

lth

care

70

154

149

184

131

134

293

393

173

Nati

onal

secu

rity

250

536

180

548

254

159

213

198

238

Reg

ula

tion

157

230

145

196

112

155

292

281

174

Fin

an

cial

regu

lati

on33

70

13

53

17

36

102

61

33

Sov

erei

gn

deb

tamp

curr

ency

cris

es14

06

23

05

04

03

04

39

16

En

titl

emen

tp

rogra

ms

73

126

115

187

88

82

153

247

124

Tra

de

pol

icy

38

40

63

26

17

20

14

21

38

Su

mof

pol

icy

cate

gor

ies

1425

2107

1295

2151

1152

1200

1863

2222

1506

Rati

oof

EP

Uto

over

all

EU

05

004

104

703

904

504

503

605

104

7

Not

es

Qu

erie

sru

nF

ebru

ary

12

2015

onU

S

new

spap

ers

inA

cces

sW

orld

New

sN

ewsb

an

k

usi

ng

the

cate

gor

y-s

pec

ific

pol

icy

term

sets

list

edin

On

lin

eA

pp

end

ixB

E

xce

pt

for

the

last

row

all

entr

ies

are

exp

ress

edre

lati

ve

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

lsquolsquoOver

all

econ

omic

un

cert

ain

tyrsquorsquo

qu

an

tifi

esth

efr

equ

ency

ofart

icle

sth

at

mee

tou

rlsquolsquoe

con

omyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

requ

irem

ents

(ie

d

rop

pin

gth

elsquolsquop

olic

yrsquorsquo

requ

irem

ent)

an

dis

als

oex

pre

ssed

rela

tive

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

Th

eca

tegor

y-

spec

ific

ind

exvalu

essu

mto

mor

eth

an

100

for

two

reaso

ns

firs

tw

eu

sea

few

pol

icy

term

sin

mor

eth

an

one

pol

icy

cate

gor

y

For

exam

ple

lsquolsquoM

edic

aid

rsquorsquoap

pea

rsin

the

term

sets

for

bot

hh

ealt

hca

rean

den

titl

emen

tp

rogra

ms

Sec

ond

a

new

spap

erart

icle

that

mee

tsth

elsquolsquoe

con

omyrsquorsquo

lsquolsquopol

icyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

crit

eria

can

refe

rto

mor

eth

an

one

pol

icy

cate

gor

y

ECONOMIC POLICY UNCERTAINTY 1603

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 3: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

policy uncertainty indexes for the United States by specifyingmore restrictive criteria for those articles that contain termsabout the economy policy and uncertainty For example wedevelop indexes of health care policy uncertainty and nationalsecurity policy uncertainty based on the presence of additionalterms like lsquolsquohealth carersquorsquo lsquolsquohospitalrsquorsquo or lsquolsquohealth insurancersquorsquo andlsquolsquowarrsquorsquo lsquolsquoterrorismrsquorsquo or lsquolsquodepartment of defensersquorsquo respectivelyCategory-specific shocks and policy initiatives are clearly visible

Our approach to measuring policy uncertainty raises potentialconcerns about newspaper reliability accuracy bias and consis-tency To address these concerns we evaluate our EPU index inseveral ways First we show a strong relationship between ourmeasure of EPU and other measures of economic uncertainty forexample implied stock market volatility Second we also show astrong relationship between our index and other measures of policyuncertainty for example the frequency with which the FederalReserve Systemrsquos Beige Books mention policy uncertainty Thirdwe find very similar movements in EPU indexes based on right-leaning and left-leaning newspapers suggesting that politicalslant does not seriously distort our overall EPU index

Fourth we conducted an extensive audit study of 12000 ran-domly selected articles drawn from major US newspapers Workingunder close supervision teams of University of Chicago studentsunderwent a training process and then carefully read overlappingsets of randomly selected articles guided by a 65-page referencemanual and weekly team meetings The auditors assessed whethera given article discusses economic policy uncertainty based on ourcriteria We use the audit results to select our policy term set eval-uate the performance of our computer-automated methods and con-struct additional data There is a high correlation between ourhuman- and computer-generated indexes (086 in quarterly datafrom 1985 to 2012 and 093 in annual data from 1900 to 2010) Thediscrepancy between the human and computer-generated indexes isuncorrelated with GDP growth rates and with the level of EPU

Finally our indexes have a market use validation commer-cial data providers that include Bloomberg FRED Haver andReuters carry our indexes to meet demands from banks hedgefunds corporations and policy makers This pattern of marketadoption suggests that our indexes contain useful information fora range of decision makers

In Section IV we provide evidence of how firm-level and ag-gregate outcomes evolve in the wake of policy uncertainty

ECONOMIC POLICY UNCERTAINTY 1595

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

movements Causal inference is challenging because policy re-sponds to economic conditions and is likely to be forward lookingTo make progress we follow a micro and a macro estimation ap-proach First the micro approach exploits firm-level differences inexposure to certain aspects of policy mainly government pur-chases of goods and services We use micro data from the FederalRegistry of Contracts and data on government health care spend-ing to calculate the share of firm and industry revenues derivedfrom sales to the government Next in firm-level regressions thatinclude time and firm fixed effects and other controls we show thatfirms with greater exposure to government purchases experiencegreater stock price volatility when policy uncertainty is high andreduced investment rates and employment growth when policyuncertainty rises Adding the VIX as an explanatory variable (in-teracted with firm-level exposure to government purchases) westill find greater stock price volatility and falls in investment andemployment with heightened policy uncertainty which points to apolicy uncertainty channel at work rather than a broader uncer-tainty effect We also find that firms in the defense health careand financial sectors are especially responsive to their own cate-gory-specific EPU measures confirming their information value

These firm-level results are suggestive of a causal impact ofpolicy uncertainty on investment and employment in sectors thatrely heavily on government spending and in sectors like healthcare and finance with strong exposure to major shifts in regula-tory policy However the firm-level results offer limited guidanceabout the magnitude of aggregate effects in part because theycapture only a limited set of potential policy uncertainty channels

Our second approach fits vector autoregressive (VAR) modelsto US data and to an international panel VAR that exploits ourEPU indexes for 12 countries The US VAR results indicate thata policy uncertainty innovation equivalent to the actual EPU in-crease from 2005ndash2006 to 2011ndash2012 foreshadows declines ofabout 6 in gross investment 11 in industrial productionand 035 in employment The 12-country panel VAR yields sim-ilar results3 Although our results are not necessarily causal oneplausible interpretation of our micro and macro evidence is that

3 Stock and Watson (2012) use our EPU index to investigate the factorsbehind the 2007ndash2009 recession and slow recovery and come to a similar conclu-sionmdashnamely that policy uncertainty is a strong candidate to partly explain thepoor economic performance but causal identification is hard

QUARTERLY JOURNAL OF ECONOMICS1596

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty retards investment hiring and growth in pol-icy-sensitive sectors like defense finance healthcare and con-struction and these sectors are important enough for policyuncertainty to matter at the aggregate level

This article relates to at least three strands of literature Thefirst is research on the impact of uncertainty on growth and in-vestment Theoretical work on this topic dates at least toBernanke (1983) who points out that high uncertainty givesfirms an incentive to delay investment and hiring wheninvestment projects are costly to undo or workers are costly tohire and fire4 Of course once uncertainty recedes firms increasehiring and investment to meet pent-up demand Other reasons fora depressive effect of uncertainty include precautionary spendingcutbacks by households upward pressure on the cost of finance(eg Pastor and Veronesi 2013 Gilchrist Sim and Zakrajsek2014) managerial risk aversion (eg Panousi and Papanikolaou2012) and interactions between nominal rigidities and searchfrictions (Basu and Bundick 2012 Leduc and Liu 2015)

Second there is a literature focused explicitly on policy un-certainty Friedman (1968) Rodrik (1991) Higgs (1997) andHassett and Metcalf (1999) among others consider the detrimen-tal economic effects of monetary fiscal and regulatory policy un-certainty More recently Born and Pfeifer (2014) and Fernandez-Villaverde at al (2015) study policy uncertainty in DSGE modelsfinding moderately negative effects while Pastor and Veronesi(2012 2013) model the theoretical links among fluctuationspolicy uncertainty and stock market volatility5

4 Dixit and Pindyck (1994) offer a review of the early theoretical literatureincluding papers by Oi (1961) Hartman (1972) and Abel (1983) that highlightpotentially positive effects of uncertainty Recent empirical papers include Bloom(2009) Bachman Elstener and Sims (2013) Bloom et al (2014) and Scotti (2016)with a review in Bloom (2014)

5 In other related work Julio and Yook (2012) find that investment fallsaround national elections Durnev (2010) finds that corporate investment becomesless responsive to stock prices in election years Brogaard and Detzel (2015) findthat policy uncertainty reduces asset returns Handley and Limao (2015) find thattrade policy uncertainty delays firm entry Gulen and Ion (2016) find negative re-sponses of corporate investment to our EPU index Koijen Philipson and Uhlig(2016) develop evidence that government-induced uncertainty about profitabilitygenerates a large equity risk premium for firms in the health care sector and redu-ces their medical RampD and Giavazzi and McMahon (2012) find that policy uncer-tainty led German households to increase savings in the run-up to the close andconsequential general elections in 1998

ECONOMIC POLICY UNCERTAINTY 1597

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Finally there is a rapidly growing literature on text searchmethodsmdashusing newspaper archives in particularmdashto measure avariety of outcomes Examples include Gentzkow and Shapiro(2010) Hoberg and Phillips (2010) Boudoukh et al (2013) andAlexopoulos and Cohen (2015) Our work suggests that newspa-per text search can yield useful proxies for economic and policyconditions stretching back several decades which could be espe-cially valuable in earlier eras and in countries with fewer datasources

Section II describes the data we use to construct our policyuncertainty indexes Section III evaluates our EPU measures inseveral ways and develops additional evidence about movementsin policy-related uncertainty over time Section IV investigateshow firm-level outcomes covary with policy uncertainty and thedynamic responses of aggregate outcomes to policy uncertaintyinnovations Section V concludes and offers some thoughts aboutdirections for future research

II Measuring EPU

We build indexes of policy-related economic uncertaintybased on newspaper coverage frequency6 We aim to capture un-certainty about who will make economic policy decisions whateconomic policy actions will be undertaken and when and theeconomic effects of policy actions (or inaction)mdashincluding uncer-tainties related to the economic ramifications of lsquolsquononeconomicrsquorsquopolicy matters for example military actions Our measures cap-ture both near-term concerns (eg when will the Fed adjust itspolicy rate) and longer term concerns (eg how to fund entitle-ment programs) as reflected in newspaper articles We first de-scribe the construction of our monthly and daily EPU indexes forthe United States from 1985 onward and then turn to indexes forspecific policy categories indexes for other countries and histor-ical indexes for the United States and United Kingdom

6 Earlier drafts of this article include index components based on (i) the pre-sent value of future scheduled tax code expirations and (ii) disagreement amongprofessional forecasters over future government purchases and consumer pricesHowever to extend our EPU measures over time and across countries we focushere on the newspaper approach while continuing to report the other componentsat httpwwwpolicyuncertaintycom

QUARTERLY JOURNAL OF ECONOMICS1598

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

IIA US Economic Policy Uncertainty Indexes from 1985

Our modern monthly EPU index for the United States relieson 10 leading newspapers USA Today Miami Herald ChicagoTribune Washington Post Los Angeles Times Boston Globe SanFrancisco Chronicle Dallas Morning News New York Timesand Wall Street Journal We search the digital archives of eachpaper from January 1985 to obtain a monthly count of articlesthat contain the following trio of terms lsquolsquouncertaintyrsquorsquo or lsquolsquouncer-tainrsquorsquo lsquolsquoeconomicrsquorsquo or lsquolsquoeconomyrsquorsquo and one of the following policyterms lsquolsquoCongressrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederal Reserversquorsquo lsquolsquolegislationrsquorsquolsquolsquoregulationrsquorsquo or lsquolsquoWhite Housersquorsquo (including variants likelsquolsquouncertaintiesrsquorsquo lsquolsquoregulatoryrsquorsquo or lsquolsquothe Fedrsquorsquo) In other words tomeet our criteria an article must contain terms in all three cat-egories pertaining to uncertainty the economy and policy Weuse our audit study to select the policy terms as explained inSection IIIA

An obvious difficulty with these raw counts is that the over-all volume of articles varies across newspapers and time Thuswe scale the raw counts by the total number of articles in thesame newspaper and month We standardize each monthlynewspaper-level series to unit standard deviation from 1985 to2009 and then average across the 10 papers by month Finallywe normalize the 10-paper series to a mean of 100 from 1985 to2009 To be precise let Xit denote the scaled EPU frequencycounts for newspaper i = 1 2 10 in month t and let T1 andT2 denote the time intervals used in the standardization andnormalization calculations We proceed in the following steps(i) Compute the times-series variance 2

i in the interval T1 foreach paper i (ii) Standardize Xit by dividing through by thestandard deviation i for all t This operation yields for eachpaper a series Yit with unit standard deviation in the intervalT1 (iii) Compute the mean over newspapers of Yit in each monthto obtain the series Zt (iv) Compute M the mean value of Zt inthe interval T2 (v) Multiply Zt by (100M) for all t to obtain thenormalized EPU time-series index We use the same approachfor other countries and indexes

Figure I plots the resulting index which shows clear spikesaround the Gulf Wars close presidential elections the 911 ter-rorist attack the stimulus debate in early 2008 the LehmanBrothers bankruptcy and TARP legislation in late 2008 thesummer 2011 debt ceiling dispute and the battle over the lsquolsquofiscal

ECONOMIC POLICY UNCERTAINTY 1599

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

cliffrsquorsquo in late 2012 among other events and developments Somenotable political events do not generate high EPU according toour index For instance our EPU index shows no large spike inconnection with the partial federal government shutdowns fromNovember 1995 to January 1996 although those shutdowns re-ceived quite a lot of press coverage7

In addition to our monthly index we produce a daily EPUindex using the Newsbank news aggregator which coversaround 1500 US newspapers Newsbankrsquos extensive coverageyields enough articles to generate a meaningful daily countTaking monthly averages of our daily index it correlates at 085with our 10-paper monthly index indicating a high degree of sim-ilarity Because papers enter and leave the Newsbank archive andits count of newspapers expands greatly over time compositionalshifts potentially distort the longer term behavior of the daily EPU

FIGURE I

EPU Index for the United States

7 We find more than 8000 articles about these shutdowns in Newsbank ar-chives but less than 25 also mention the economy less than 2 mention uncer-tainty and only 1 mentions both Thus politically tumultuous episodes do notnecessarily raise EPU by our measure

QUARTERLY JOURNAL OF ECONOMICS1600

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

index Hence we focus on our 10-paper monthly EPU index butthe daily index provides a useful high-frequency alternative8

IIB EPU Indexes for Policy Categories

To create indexes for policy categories we apply additionalcriteria to those articles that contain our trio of terms about theeconomy policy and uncertainty The additional criteria involvethe presence of one or more category-relevant terms lsquolsquothe Fedrsquorsquolsquolsquocentral bankrsquorsquo lsquolsquointerest ratersquorsquo lsquolsquoinflationrsquorsquo and so on for the mon-etary policy category for example Online Appendix B reports thefull set of terms that define our 11 policy categories and subcat-egories We use Newsbank for the category indexes because itshigh text density facilitates measurement by time period andpolicy category As seen in Figure II the national security EPUindex spiked sharply in connection with the 911 attacks GulfWar I and the onset of Gulf War II The health care EPU index

FIGURE II

National Security and Health Care EPU Indexes

8 We update the daily EPU index at approximately 9 am EST each day andpost it at httpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1601

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

rose sharply during the Clinton health care reform initiative in1993ndash1994 and has fluctuated at high levels from 2009 to 2014

Table I reports all 11 category-specific EPU indexes9 It alsoreports an overall economic uncertainty (EU) index that drops thepolicy requirement in the EPU index The first two rows reportaverage EU and EPU values for the indicated periods expressedrelative to the average EPU value from 1985 to 2014 For exam-ple the EU value of 2182 says the (scaled) frequency of EU arti-cles from 19851 to 19906 is somewhat more than twice theaverage frequency of EPU articles from 1985 to 2014 The next11 rows report relative frequency values for specific policy cate-gories and time periods For example the 548 value for lsquolsquoNationalSecurityrsquorsquo says the frequency of EPU articles during 20019 to200212 that mention national security matters is 548 of the1985ndash2014 average EPU frequency and 43 548

1285

of the EPU

frequency from 20019 to 200212Fiscal matters especially tax policy stand out in Table I as the

largest source of policy uncertainty especially in recent years Thefiscal policy EPU index rose from values near 33 in the precrisisyears to 615 in 20089 to 200912 and 783 from 2010 to 2013Health care policy is the second largest source of elevated EPU inrecent years Policy uncertainty related to financial regulations andentitlement programs also rose sharply after 2008 but from initiallylower levels Concerns related to sovereign debt and currency crisesare up by an order of magnitude during 2010 to 2013 but from sucha low base as to have little impact on the overall EPU index EPUconcerns related to monetary policy are important throughout the1985ndash2014 period but perhaps surprisingly they are not elevatedin recent years by our measure We interpret this result as a reflec-tion of low and stable inflation rates in recent years which appar-ently drive newspaper coverage more than disputes amongprofessional economists about unconventional monetary policies10

Several other researchers develop measures related to uncer-tainty about government behavior Marina Azzimonti (2015) con-structs a newspaper index of partisan conflict at the federal levelthat shows similarities to our EPU index but also notable

9 In contrast to Figure III which normalizes each category-specific EPUseries to 100 Table I expresses each category-specific EPU series as a percentageof the overall EPU frequency from 1985 to 2014

10 Other evidence also points to subdued levels of inflation uncertainty inrecent years See Nalewaik (2015) for a presentation and discussion of evidencebased on time-series models surveys and financial markets data

QUARTERLY JOURNAL OF ECONOMICS1602

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EI

EC

ON

OM

ICP

OL

ICY

UN

CE

RT

AIN

TY

BY

PO

LIC

YC

AT

EG

OR

YA

ND

TIM

EP

ER

IOD

1985ndash2014

Tim

ep

erio

d19851

ndash19906

19907

ndash19911

219921

ndash20018

20019

ndash20021

220031

ndash20076

20077

ndash20088

20089

ndash20091

220101

ndash20131

019851

ndash20141

2

Mid

-80s

toG

ulf

War

IG

ulf

War

I1990s

boo

mto

91

191

1att

ack

s2000s

boo

m

Earl

ycr

edit

cru

nch

Leh

man

coll

ap

seamp

rece

ssio

n

Fis

cal

pol

icy

batt

les

Over

all

aver

age

Over

all

econ

omic

un

cert

ain

ty2182

3498

1859

3269

1598

1848

3709

2521

2193

Eco

nom

icp

olic

yu

nce

rtain

ty1096

1419

881

1285

714

834

1321

1275

1000

Fis

cal

pol

icy

496

596

359

554

323

331

615

783

461

Taxes

399

484

319

512

302

314

569

681

403

Gov

ern

men

tsp

end

ing

ampot

her

227

268

121

173

85

66

171

332

171

Mon

etary

pol

icy

327

418

261

452

222

316

278

261

281

Hea

lth

care

70

154

149

184

131

134

293

393

173

Nati

onal

secu

rity

250

536

180

548

254

159

213

198

238

Reg

ula

tion

157

230

145

196

112

155

292

281

174

Fin

an

cial

regu

lati

on33

70

13

53

17

36

102

61

33

Sov

erei

gn

deb

tamp

curr

ency

cris

es14

06

23

05

04

03

04

39

16

En

titl

emen

tp

rogra

ms

73

126

115

187

88

82

153

247

124

Tra

de

pol

icy

38

40

63

26

17

20

14

21

38

Su

mof

pol

icy

cate

gor

ies

1425

2107

1295

2151

1152

1200

1863

2222

1506

Rati

oof

EP

Uto

over

all

EU

05

004

104

703

904

504

503

605

104

7

Not

es

Qu

erie

sru

nF

ebru

ary

12

2015

onU

S

new

spap

ers

inA

cces

sW

orld

New

sN

ewsb

an

k

usi

ng

the

cate

gor

y-s

pec

ific

pol

icy

term

sets

list

edin

On

lin

eA

pp

end

ixB

E

xce

pt

for

the

last

row

all

entr

ies

are

exp

ress

edre

lati

ve

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

lsquolsquoOver

all

econ

omic

un

cert

ain

tyrsquorsquo

qu

an

tifi

esth

efr

equ

ency

ofart

icle

sth

at

mee

tou

rlsquolsquoe

con

omyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

requ

irem

ents

(ie

d

rop

pin

gth

elsquolsquop

olic

yrsquorsquo

requ

irem

ent)

an

dis

als

oex

pre

ssed

rela

tive

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

Th

eca

tegor

y-

spec

ific

ind

exvalu

essu

mto

mor

eth

an

100

for

two

reaso

ns

firs

tw

eu

sea

few

pol

icy

term

sin

mor

eth

an

one

pol

icy

cate

gor

y

For

exam

ple

lsquolsquoM

edic

aid

rsquorsquoap

pea

rsin

the

term

sets

for

bot

hh

ealt

hca

rean

den

titl

emen

tp

rogra

ms

Sec

ond

a

new

spap

erart

icle

that

mee

tsth

elsquolsquoe

con

omyrsquorsquo

lsquolsquopol

icyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

crit

eria

can

refe

rto

mor

eth

an

one

pol

icy

cate

gor

y

ECONOMIC POLICY UNCERTAINTY 1603

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 4: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

movements Causal inference is challenging because policy re-sponds to economic conditions and is likely to be forward lookingTo make progress we follow a micro and a macro estimation ap-proach First the micro approach exploits firm-level differences inexposure to certain aspects of policy mainly government pur-chases of goods and services We use micro data from the FederalRegistry of Contracts and data on government health care spend-ing to calculate the share of firm and industry revenues derivedfrom sales to the government Next in firm-level regressions thatinclude time and firm fixed effects and other controls we show thatfirms with greater exposure to government purchases experiencegreater stock price volatility when policy uncertainty is high andreduced investment rates and employment growth when policyuncertainty rises Adding the VIX as an explanatory variable (in-teracted with firm-level exposure to government purchases) westill find greater stock price volatility and falls in investment andemployment with heightened policy uncertainty which points to apolicy uncertainty channel at work rather than a broader uncer-tainty effect We also find that firms in the defense health careand financial sectors are especially responsive to their own cate-gory-specific EPU measures confirming their information value

These firm-level results are suggestive of a causal impact ofpolicy uncertainty on investment and employment in sectors thatrely heavily on government spending and in sectors like healthcare and finance with strong exposure to major shifts in regula-tory policy However the firm-level results offer limited guidanceabout the magnitude of aggregate effects in part because theycapture only a limited set of potential policy uncertainty channels

Our second approach fits vector autoregressive (VAR) modelsto US data and to an international panel VAR that exploits ourEPU indexes for 12 countries The US VAR results indicate thata policy uncertainty innovation equivalent to the actual EPU in-crease from 2005ndash2006 to 2011ndash2012 foreshadows declines ofabout 6 in gross investment 11 in industrial productionand 035 in employment The 12-country panel VAR yields sim-ilar results3 Although our results are not necessarily causal oneplausible interpretation of our micro and macro evidence is that

3 Stock and Watson (2012) use our EPU index to investigate the factorsbehind the 2007ndash2009 recession and slow recovery and come to a similar conclu-sionmdashnamely that policy uncertainty is a strong candidate to partly explain thepoor economic performance but causal identification is hard

QUARTERLY JOURNAL OF ECONOMICS1596

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty retards investment hiring and growth in pol-icy-sensitive sectors like defense finance healthcare and con-struction and these sectors are important enough for policyuncertainty to matter at the aggregate level

This article relates to at least three strands of literature Thefirst is research on the impact of uncertainty on growth and in-vestment Theoretical work on this topic dates at least toBernanke (1983) who points out that high uncertainty givesfirms an incentive to delay investment and hiring wheninvestment projects are costly to undo or workers are costly tohire and fire4 Of course once uncertainty recedes firms increasehiring and investment to meet pent-up demand Other reasons fora depressive effect of uncertainty include precautionary spendingcutbacks by households upward pressure on the cost of finance(eg Pastor and Veronesi 2013 Gilchrist Sim and Zakrajsek2014) managerial risk aversion (eg Panousi and Papanikolaou2012) and interactions between nominal rigidities and searchfrictions (Basu and Bundick 2012 Leduc and Liu 2015)

Second there is a literature focused explicitly on policy un-certainty Friedman (1968) Rodrik (1991) Higgs (1997) andHassett and Metcalf (1999) among others consider the detrimen-tal economic effects of monetary fiscal and regulatory policy un-certainty More recently Born and Pfeifer (2014) and Fernandez-Villaverde at al (2015) study policy uncertainty in DSGE modelsfinding moderately negative effects while Pastor and Veronesi(2012 2013) model the theoretical links among fluctuationspolicy uncertainty and stock market volatility5

4 Dixit and Pindyck (1994) offer a review of the early theoretical literatureincluding papers by Oi (1961) Hartman (1972) and Abel (1983) that highlightpotentially positive effects of uncertainty Recent empirical papers include Bloom(2009) Bachman Elstener and Sims (2013) Bloom et al (2014) and Scotti (2016)with a review in Bloom (2014)

5 In other related work Julio and Yook (2012) find that investment fallsaround national elections Durnev (2010) finds that corporate investment becomesless responsive to stock prices in election years Brogaard and Detzel (2015) findthat policy uncertainty reduces asset returns Handley and Limao (2015) find thattrade policy uncertainty delays firm entry Gulen and Ion (2016) find negative re-sponses of corporate investment to our EPU index Koijen Philipson and Uhlig(2016) develop evidence that government-induced uncertainty about profitabilitygenerates a large equity risk premium for firms in the health care sector and redu-ces their medical RampD and Giavazzi and McMahon (2012) find that policy uncer-tainty led German households to increase savings in the run-up to the close andconsequential general elections in 1998

ECONOMIC POLICY UNCERTAINTY 1597

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Finally there is a rapidly growing literature on text searchmethodsmdashusing newspaper archives in particularmdashto measure avariety of outcomes Examples include Gentzkow and Shapiro(2010) Hoberg and Phillips (2010) Boudoukh et al (2013) andAlexopoulos and Cohen (2015) Our work suggests that newspa-per text search can yield useful proxies for economic and policyconditions stretching back several decades which could be espe-cially valuable in earlier eras and in countries with fewer datasources

Section II describes the data we use to construct our policyuncertainty indexes Section III evaluates our EPU measures inseveral ways and develops additional evidence about movementsin policy-related uncertainty over time Section IV investigateshow firm-level outcomes covary with policy uncertainty and thedynamic responses of aggregate outcomes to policy uncertaintyinnovations Section V concludes and offers some thoughts aboutdirections for future research

II Measuring EPU

We build indexes of policy-related economic uncertaintybased on newspaper coverage frequency6 We aim to capture un-certainty about who will make economic policy decisions whateconomic policy actions will be undertaken and when and theeconomic effects of policy actions (or inaction)mdashincluding uncer-tainties related to the economic ramifications of lsquolsquononeconomicrsquorsquopolicy matters for example military actions Our measures cap-ture both near-term concerns (eg when will the Fed adjust itspolicy rate) and longer term concerns (eg how to fund entitle-ment programs) as reflected in newspaper articles We first de-scribe the construction of our monthly and daily EPU indexes forthe United States from 1985 onward and then turn to indexes forspecific policy categories indexes for other countries and histor-ical indexes for the United States and United Kingdom

6 Earlier drafts of this article include index components based on (i) the pre-sent value of future scheduled tax code expirations and (ii) disagreement amongprofessional forecasters over future government purchases and consumer pricesHowever to extend our EPU measures over time and across countries we focushere on the newspaper approach while continuing to report the other componentsat httpwwwpolicyuncertaintycom

QUARTERLY JOURNAL OF ECONOMICS1598

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

IIA US Economic Policy Uncertainty Indexes from 1985

Our modern monthly EPU index for the United States relieson 10 leading newspapers USA Today Miami Herald ChicagoTribune Washington Post Los Angeles Times Boston Globe SanFrancisco Chronicle Dallas Morning News New York Timesand Wall Street Journal We search the digital archives of eachpaper from January 1985 to obtain a monthly count of articlesthat contain the following trio of terms lsquolsquouncertaintyrsquorsquo or lsquolsquouncer-tainrsquorsquo lsquolsquoeconomicrsquorsquo or lsquolsquoeconomyrsquorsquo and one of the following policyterms lsquolsquoCongressrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederal Reserversquorsquo lsquolsquolegislationrsquorsquolsquolsquoregulationrsquorsquo or lsquolsquoWhite Housersquorsquo (including variants likelsquolsquouncertaintiesrsquorsquo lsquolsquoregulatoryrsquorsquo or lsquolsquothe Fedrsquorsquo) In other words tomeet our criteria an article must contain terms in all three cat-egories pertaining to uncertainty the economy and policy Weuse our audit study to select the policy terms as explained inSection IIIA

An obvious difficulty with these raw counts is that the over-all volume of articles varies across newspapers and time Thuswe scale the raw counts by the total number of articles in thesame newspaper and month We standardize each monthlynewspaper-level series to unit standard deviation from 1985 to2009 and then average across the 10 papers by month Finallywe normalize the 10-paper series to a mean of 100 from 1985 to2009 To be precise let Xit denote the scaled EPU frequencycounts for newspaper i = 1 2 10 in month t and let T1 andT2 denote the time intervals used in the standardization andnormalization calculations We proceed in the following steps(i) Compute the times-series variance 2

i in the interval T1 foreach paper i (ii) Standardize Xit by dividing through by thestandard deviation i for all t This operation yields for eachpaper a series Yit with unit standard deviation in the intervalT1 (iii) Compute the mean over newspapers of Yit in each monthto obtain the series Zt (iv) Compute M the mean value of Zt inthe interval T2 (v) Multiply Zt by (100M) for all t to obtain thenormalized EPU time-series index We use the same approachfor other countries and indexes

Figure I plots the resulting index which shows clear spikesaround the Gulf Wars close presidential elections the 911 ter-rorist attack the stimulus debate in early 2008 the LehmanBrothers bankruptcy and TARP legislation in late 2008 thesummer 2011 debt ceiling dispute and the battle over the lsquolsquofiscal

ECONOMIC POLICY UNCERTAINTY 1599

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

cliffrsquorsquo in late 2012 among other events and developments Somenotable political events do not generate high EPU according toour index For instance our EPU index shows no large spike inconnection with the partial federal government shutdowns fromNovember 1995 to January 1996 although those shutdowns re-ceived quite a lot of press coverage7

In addition to our monthly index we produce a daily EPUindex using the Newsbank news aggregator which coversaround 1500 US newspapers Newsbankrsquos extensive coverageyields enough articles to generate a meaningful daily countTaking monthly averages of our daily index it correlates at 085with our 10-paper monthly index indicating a high degree of sim-ilarity Because papers enter and leave the Newsbank archive andits count of newspapers expands greatly over time compositionalshifts potentially distort the longer term behavior of the daily EPU

FIGURE I

EPU Index for the United States

7 We find more than 8000 articles about these shutdowns in Newsbank ar-chives but less than 25 also mention the economy less than 2 mention uncer-tainty and only 1 mentions both Thus politically tumultuous episodes do notnecessarily raise EPU by our measure

QUARTERLY JOURNAL OF ECONOMICS1600

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

index Hence we focus on our 10-paper monthly EPU index butthe daily index provides a useful high-frequency alternative8

IIB EPU Indexes for Policy Categories

To create indexes for policy categories we apply additionalcriteria to those articles that contain our trio of terms about theeconomy policy and uncertainty The additional criteria involvethe presence of one or more category-relevant terms lsquolsquothe Fedrsquorsquolsquolsquocentral bankrsquorsquo lsquolsquointerest ratersquorsquo lsquolsquoinflationrsquorsquo and so on for the mon-etary policy category for example Online Appendix B reports thefull set of terms that define our 11 policy categories and subcat-egories We use Newsbank for the category indexes because itshigh text density facilitates measurement by time period andpolicy category As seen in Figure II the national security EPUindex spiked sharply in connection with the 911 attacks GulfWar I and the onset of Gulf War II The health care EPU index

FIGURE II

National Security and Health Care EPU Indexes

8 We update the daily EPU index at approximately 9 am EST each day andpost it at httpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1601

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

rose sharply during the Clinton health care reform initiative in1993ndash1994 and has fluctuated at high levels from 2009 to 2014

Table I reports all 11 category-specific EPU indexes9 It alsoreports an overall economic uncertainty (EU) index that drops thepolicy requirement in the EPU index The first two rows reportaverage EU and EPU values for the indicated periods expressedrelative to the average EPU value from 1985 to 2014 For exam-ple the EU value of 2182 says the (scaled) frequency of EU arti-cles from 19851 to 19906 is somewhat more than twice theaverage frequency of EPU articles from 1985 to 2014 The next11 rows report relative frequency values for specific policy cate-gories and time periods For example the 548 value for lsquolsquoNationalSecurityrsquorsquo says the frequency of EPU articles during 20019 to200212 that mention national security matters is 548 of the1985ndash2014 average EPU frequency and 43 548

1285

of the EPU

frequency from 20019 to 200212Fiscal matters especially tax policy stand out in Table I as the

largest source of policy uncertainty especially in recent years Thefiscal policy EPU index rose from values near 33 in the precrisisyears to 615 in 20089 to 200912 and 783 from 2010 to 2013Health care policy is the second largest source of elevated EPU inrecent years Policy uncertainty related to financial regulations andentitlement programs also rose sharply after 2008 but from initiallylower levels Concerns related to sovereign debt and currency crisesare up by an order of magnitude during 2010 to 2013 but from sucha low base as to have little impact on the overall EPU index EPUconcerns related to monetary policy are important throughout the1985ndash2014 period but perhaps surprisingly they are not elevatedin recent years by our measure We interpret this result as a reflec-tion of low and stable inflation rates in recent years which appar-ently drive newspaper coverage more than disputes amongprofessional economists about unconventional monetary policies10

Several other researchers develop measures related to uncer-tainty about government behavior Marina Azzimonti (2015) con-structs a newspaper index of partisan conflict at the federal levelthat shows similarities to our EPU index but also notable

9 In contrast to Figure III which normalizes each category-specific EPUseries to 100 Table I expresses each category-specific EPU series as a percentageof the overall EPU frequency from 1985 to 2014

10 Other evidence also points to subdued levels of inflation uncertainty inrecent years See Nalewaik (2015) for a presentation and discussion of evidencebased on time-series models surveys and financial markets data

QUARTERLY JOURNAL OF ECONOMICS1602

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EI

EC

ON

OM

ICP

OL

ICY

UN

CE

RT

AIN

TY

BY

PO

LIC

YC

AT

EG

OR

YA

ND

TIM

EP

ER

IOD

1985ndash2014

Tim

ep

erio

d19851

ndash19906

19907

ndash19911

219921

ndash20018

20019

ndash20021

220031

ndash20076

20077

ndash20088

20089

ndash20091

220101

ndash20131

019851

ndash20141

2

Mid

-80s

toG

ulf

War

IG

ulf

War

I1990s

boo

mto

91

191

1att

ack

s2000s

boo

m

Earl

ycr

edit

cru

nch

Leh

man

coll

ap

seamp

rece

ssio

n

Fis

cal

pol

icy

batt

les

Over

all

aver

age

Over

all

econ

omic

un

cert

ain

ty2182

3498

1859

3269

1598

1848

3709

2521

2193

Eco

nom

icp

olic

yu

nce

rtain

ty1096

1419

881

1285

714

834

1321

1275

1000

Fis

cal

pol

icy

496

596

359

554

323

331

615

783

461

Taxes

399

484

319

512

302

314

569

681

403

Gov

ern

men

tsp

end

ing

ampot

her

227

268

121

173

85

66

171

332

171

Mon

etary

pol

icy

327

418

261

452

222

316

278

261

281

Hea

lth

care

70

154

149

184

131

134

293

393

173

Nati

onal

secu

rity

250

536

180

548

254

159

213

198

238

Reg

ula

tion

157

230

145

196

112

155

292

281

174

Fin

an

cial

regu

lati

on33

70

13

53

17

36

102

61

33

Sov

erei

gn

deb

tamp

curr

ency

cris

es14

06

23

05

04

03

04

39

16

En

titl

emen

tp

rogra

ms

73

126

115

187

88

82

153

247

124

Tra

de

pol

icy

38

40

63

26

17

20

14

21

38

Su

mof

pol

icy

cate

gor

ies

1425

2107

1295

2151

1152

1200

1863

2222

1506

Rati

oof

EP

Uto

over

all

EU

05

004

104

703

904

504

503

605

104

7

Not

es

Qu

erie

sru

nF

ebru

ary

12

2015

onU

S

new

spap

ers

inA

cces

sW

orld

New

sN

ewsb

an

k

usi

ng

the

cate

gor

y-s

pec

ific

pol

icy

term

sets

list

edin

On

lin

eA

pp

end

ixB

E

xce

pt

for

the

last

row

all

entr

ies

are

exp

ress

edre

lati

ve

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

lsquolsquoOver

all

econ

omic

un

cert

ain

tyrsquorsquo

qu

an

tifi

esth

efr

equ

ency

ofart

icle

sth

at

mee

tou

rlsquolsquoe

con

omyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

requ

irem

ents

(ie

d

rop

pin

gth

elsquolsquop

olic

yrsquorsquo

requ

irem

ent)

an

dis

als

oex

pre

ssed

rela

tive

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

Th

eca

tegor

y-

spec

ific

ind

exvalu

essu

mto

mor

eth

an

100

for

two

reaso

ns

firs

tw

eu

sea

few

pol

icy

term

sin

mor

eth

an

one

pol

icy

cate

gor

y

For

exam

ple

lsquolsquoM

edic

aid

rsquorsquoap

pea

rsin

the

term

sets

for

bot

hh

ealt

hca

rean

den

titl

emen

tp

rogra

ms

Sec

ond

a

new

spap

erart

icle

that

mee

tsth

elsquolsquoe

con

omyrsquorsquo

lsquolsquopol

icyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

crit

eria

can

refe

rto

mor

eth

an

one

pol

icy

cate

gor

y

ECONOMIC POLICY UNCERTAINTY 1603

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 5: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

policy uncertainty retards investment hiring and growth in pol-icy-sensitive sectors like defense finance healthcare and con-struction and these sectors are important enough for policyuncertainty to matter at the aggregate level

This article relates to at least three strands of literature Thefirst is research on the impact of uncertainty on growth and in-vestment Theoretical work on this topic dates at least toBernanke (1983) who points out that high uncertainty givesfirms an incentive to delay investment and hiring wheninvestment projects are costly to undo or workers are costly tohire and fire4 Of course once uncertainty recedes firms increasehiring and investment to meet pent-up demand Other reasons fora depressive effect of uncertainty include precautionary spendingcutbacks by households upward pressure on the cost of finance(eg Pastor and Veronesi 2013 Gilchrist Sim and Zakrajsek2014) managerial risk aversion (eg Panousi and Papanikolaou2012) and interactions between nominal rigidities and searchfrictions (Basu and Bundick 2012 Leduc and Liu 2015)

Second there is a literature focused explicitly on policy un-certainty Friedman (1968) Rodrik (1991) Higgs (1997) andHassett and Metcalf (1999) among others consider the detrimen-tal economic effects of monetary fiscal and regulatory policy un-certainty More recently Born and Pfeifer (2014) and Fernandez-Villaverde at al (2015) study policy uncertainty in DSGE modelsfinding moderately negative effects while Pastor and Veronesi(2012 2013) model the theoretical links among fluctuationspolicy uncertainty and stock market volatility5

4 Dixit and Pindyck (1994) offer a review of the early theoretical literatureincluding papers by Oi (1961) Hartman (1972) and Abel (1983) that highlightpotentially positive effects of uncertainty Recent empirical papers include Bloom(2009) Bachman Elstener and Sims (2013) Bloom et al (2014) and Scotti (2016)with a review in Bloom (2014)

5 In other related work Julio and Yook (2012) find that investment fallsaround national elections Durnev (2010) finds that corporate investment becomesless responsive to stock prices in election years Brogaard and Detzel (2015) findthat policy uncertainty reduces asset returns Handley and Limao (2015) find thattrade policy uncertainty delays firm entry Gulen and Ion (2016) find negative re-sponses of corporate investment to our EPU index Koijen Philipson and Uhlig(2016) develop evidence that government-induced uncertainty about profitabilitygenerates a large equity risk premium for firms in the health care sector and redu-ces their medical RampD and Giavazzi and McMahon (2012) find that policy uncer-tainty led German households to increase savings in the run-up to the close andconsequential general elections in 1998

ECONOMIC POLICY UNCERTAINTY 1597

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Finally there is a rapidly growing literature on text searchmethodsmdashusing newspaper archives in particularmdashto measure avariety of outcomes Examples include Gentzkow and Shapiro(2010) Hoberg and Phillips (2010) Boudoukh et al (2013) andAlexopoulos and Cohen (2015) Our work suggests that newspa-per text search can yield useful proxies for economic and policyconditions stretching back several decades which could be espe-cially valuable in earlier eras and in countries with fewer datasources

Section II describes the data we use to construct our policyuncertainty indexes Section III evaluates our EPU measures inseveral ways and develops additional evidence about movementsin policy-related uncertainty over time Section IV investigateshow firm-level outcomes covary with policy uncertainty and thedynamic responses of aggregate outcomes to policy uncertaintyinnovations Section V concludes and offers some thoughts aboutdirections for future research

II Measuring EPU

We build indexes of policy-related economic uncertaintybased on newspaper coverage frequency6 We aim to capture un-certainty about who will make economic policy decisions whateconomic policy actions will be undertaken and when and theeconomic effects of policy actions (or inaction)mdashincluding uncer-tainties related to the economic ramifications of lsquolsquononeconomicrsquorsquopolicy matters for example military actions Our measures cap-ture both near-term concerns (eg when will the Fed adjust itspolicy rate) and longer term concerns (eg how to fund entitle-ment programs) as reflected in newspaper articles We first de-scribe the construction of our monthly and daily EPU indexes forthe United States from 1985 onward and then turn to indexes forspecific policy categories indexes for other countries and histor-ical indexes for the United States and United Kingdom

6 Earlier drafts of this article include index components based on (i) the pre-sent value of future scheduled tax code expirations and (ii) disagreement amongprofessional forecasters over future government purchases and consumer pricesHowever to extend our EPU measures over time and across countries we focushere on the newspaper approach while continuing to report the other componentsat httpwwwpolicyuncertaintycom

QUARTERLY JOURNAL OF ECONOMICS1598

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

IIA US Economic Policy Uncertainty Indexes from 1985

Our modern monthly EPU index for the United States relieson 10 leading newspapers USA Today Miami Herald ChicagoTribune Washington Post Los Angeles Times Boston Globe SanFrancisco Chronicle Dallas Morning News New York Timesand Wall Street Journal We search the digital archives of eachpaper from January 1985 to obtain a monthly count of articlesthat contain the following trio of terms lsquolsquouncertaintyrsquorsquo or lsquolsquouncer-tainrsquorsquo lsquolsquoeconomicrsquorsquo or lsquolsquoeconomyrsquorsquo and one of the following policyterms lsquolsquoCongressrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederal Reserversquorsquo lsquolsquolegislationrsquorsquolsquolsquoregulationrsquorsquo or lsquolsquoWhite Housersquorsquo (including variants likelsquolsquouncertaintiesrsquorsquo lsquolsquoregulatoryrsquorsquo or lsquolsquothe Fedrsquorsquo) In other words tomeet our criteria an article must contain terms in all three cat-egories pertaining to uncertainty the economy and policy Weuse our audit study to select the policy terms as explained inSection IIIA

An obvious difficulty with these raw counts is that the over-all volume of articles varies across newspapers and time Thuswe scale the raw counts by the total number of articles in thesame newspaper and month We standardize each monthlynewspaper-level series to unit standard deviation from 1985 to2009 and then average across the 10 papers by month Finallywe normalize the 10-paper series to a mean of 100 from 1985 to2009 To be precise let Xit denote the scaled EPU frequencycounts for newspaper i = 1 2 10 in month t and let T1 andT2 denote the time intervals used in the standardization andnormalization calculations We proceed in the following steps(i) Compute the times-series variance 2

i in the interval T1 foreach paper i (ii) Standardize Xit by dividing through by thestandard deviation i for all t This operation yields for eachpaper a series Yit with unit standard deviation in the intervalT1 (iii) Compute the mean over newspapers of Yit in each monthto obtain the series Zt (iv) Compute M the mean value of Zt inthe interval T2 (v) Multiply Zt by (100M) for all t to obtain thenormalized EPU time-series index We use the same approachfor other countries and indexes

Figure I plots the resulting index which shows clear spikesaround the Gulf Wars close presidential elections the 911 ter-rorist attack the stimulus debate in early 2008 the LehmanBrothers bankruptcy and TARP legislation in late 2008 thesummer 2011 debt ceiling dispute and the battle over the lsquolsquofiscal

ECONOMIC POLICY UNCERTAINTY 1599

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

cliffrsquorsquo in late 2012 among other events and developments Somenotable political events do not generate high EPU according toour index For instance our EPU index shows no large spike inconnection with the partial federal government shutdowns fromNovember 1995 to January 1996 although those shutdowns re-ceived quite a lot of press coverage7

In addition to our monthly index we produce a daily EPUindex using the Newsbank news aggregator which coversaround 1500 US newspapers Newsbankrsquos extensive coverageyields enough articles to generate a meaningful daily countTaking monthly averages of our daily index it correlates at 085with our 10-paper monthly index indicating a high degree of sim-ilarity Because papers enter and leave the Newsbank archive andits count of newspapers expands greatly over time compositionalshifts potentially distort the longer term behavior of the daily EPU

FIGURE I

EPU Index for the United States

7 We find more than 8000 articles about these shutdowns in Newsbank ar-chives but less than 25 also mention the economy less than 2 mention uncer-tainty and only 1 mentions both Thus politically tumultuous episodes do notnecessarily raise EPU by our measure

QUARTERLY JOURNAL OF ECONOMICS1600

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

index Hence we focus on our 10-paper monthly EPU index butthe daily index provides a useful high-frequency alternative8

IIB EPU Indexes for Policy Categories

To create indexes for policy categories we apply additionalcriteria to those articles that contain our trio of terms about theeconomy policy and uncertainty The additional criteria involvethe presence of one or more category-relevant terms lsquolsquothe Fedrsquorsquolsquolsquocentral bankrsquorsquo lsquolsquointerest ratersquorsquo lsquolsquoinflationrsquorsquo and so on for the mon-etary policy category for example Online Appendix B reports thefull set of terms that define our 11 policy categories and subcat-egories We use Newsbank for the category indexes because itshigh text density facilitates measurement by time period andpolicy category As seen in Figure II the national security EPUindex spiked sharply in connection with the 911 attacks GulfWar I and the onset of Gulf War II The health care EPU index

FIGURE II

National Security and Health Care EPU Indexes

8 We update the daily EPU index at approximately 9 am EST each day andpost it at httpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1601

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

rose sharply during the Clinton health care reform initiative in1993ndash1994 and has fluctuated at high levels from 2009 to 2014

Table I reports all 11 category-specific EPU indexes9 It alsoreports an overall economic uncertainty (EU) index that drops thepolicy requirement in the EPU index The first two rows reportaverage EU and EPU values for the indicated periods expressedrelative to the average EPU value from 1985 to 2014 For exam-ple the EU value of 2182 says the (scaled) frequency of EU arti-cles from 19851 to 19906 is somewhat more than twice theaverage frequency of EPU articles from 1985 to 2014 The next11 rows report relative frequency values for specific policy cate-gories and time periods For example the 548 value for lsquolsquoNationalSecurityrsquorsquo says the frequency of EPU articles during 20019 to200212 that mention national security matters is 548 of the1985ndash2014 average EPU frequency and 43 548

1285

of the EPU

frequency from 20019 to 200212Fiscal matters especially tax policy stand out in Table I as the

largest source of policy uncertainty especially in recent years Thefiscal policy EPU index rose from values near 33 in the precrisisyears to 615 in 20089 to 200912 and 783 from 2010 to 2013Health care policy is the second largest source of elevated EPU inrecent years Policy uncertainty related to financial regulations andentitlement programs also rose sharply after 2008 but from initiallylower levels Concerns related to sovereign debt and currency crisesare up by an order of magnitude during 2010 to 2013 but from sucha low base as to have little impact on the overall EPU index EPUconcerns related to monetary policy are important throughout the1985ndash2014 period but perhaps surprisingly they are not elevatedin recent years by our measure We interpret this result as a reflec-tion of low and stable inflation rates in recent years which appar-ently drive newspaper coverage more than disputes amongprofessional economists about unconventional monetary policies10

Several other researchers develop measures related to uncer-tainty about government behavior Marina Azzimonti (2015) con-structs a newspaper index of partisan conflict at the federal levelthat shows similarities to our EPU index but also notable

9 In contrast to Figure III which normalizes each category-specific EPUseries to 100 Table I expresses each category-specific EPU series as a percentageof the overall EPU frequency from 1985 to 2014

10 Other evidence also points to subdued levels of inflation uncertainty inrecent years See Nalewaik (2015) for a presentation and discussion of evidencebased on time-series models surveys and financial markets data

QUARTERLY JOURNAL OF ECONOMICS1602

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EI

EC

ON

OM

ICP

OL

ICY

UN

CE

RT

AIN

TY

BY

PO

LIC

YC

AT

EG

OR

YA

ND

TIM

EP

ER

IOD

1985ndash2014

Tim

ep

erio

d19851

ndash19906

19907

ndash19911

219921

ndash20018

20019

ndash20021

220031

ndash20076

20077

ndash20088

20089

ndash20091

220101

ndash20131

019851

ndash20141

2

Mid

-80s

toG

ulf

War

IG

ulf

War

I1990s

boo

mto

91

191

1att

ack

s2000s

boo

m

Earl

ycr

edit

cru

nch

Leh

man

coll

ap

seamp

rece

ssio

n

Fis

cal

pol

icy

batt

les

Over

all

aver

age

Over

all

econ

omic

un

cert

ain

ty2182

3498

1859

3269

1598

1848

3709

2521

2193

Eco

nom

icp

olic

yu

nce

rtain

ty1096

1419

881

1285

714

834

1321

1275

1000

Fis

cal

pol

icy

496

596

359

554

323

331

615

783

461

Taxes

399

484

319

512

302

314

569

681

403

Gov

ern

men

tsp

end

ing

ampot

her

227

268

121

173

85

66

171

332

171

Mon

etary

pol

icy

327

418

261

452

222

316

278

261

281

Hea

lth

care

70

154

149

184

131

134

293

393

173

Nati

onal

secu

rity

250

536

180

548

254

159

213

198

238

Reg

ula

tion

157

230

145

196

112

155

292

281

174

Fin

an

cial

regu

lati

on33

70

13

53

17

36

102

61

33

Sov

erei

gn

deb

tamp

curr

ency

cris

es14

06

23

05

04

03

04

39

16

En

titl

emen

tp

rogra

ms

73

126

115

187

88

82

153

247

124

Tra

de

pol

icy

38

40

63

26

17

20

14

21

38

Su

mof

pol

icy

cate

gor

ies

1425

2107

1295

2151

1152

1200

1863

2222

1506

Rati

oof

EP

Uto

over

all

EU

05

004

104

703

904

504

503

605

104

7

Not

es

Qu

erie

sru

nF

ebru

ary

12

2015

onU

S

new

spap

ers

inA

cces

sW

orld

New

sN

ewsb

an

k

usi

ng

the

cate

gor

y-s

pec

ific

pol

icy

term

sets

list

edin

On

lin

eA

pp

end

ixB

E

xce

pt

for

the

last

row

all

entr

ies

are

exp

ress

edre

lati

ve

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

lsquolsquoOver

all

econ

omic

un

cert

ain

tyrsquorsquo

qu

an

tifi

esth

efr

equ

ency

ofart

icle

sth

at

mee

tou

rlsquolsquoe

con

omyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

requ

irem

ents

(ie

d

rop

pin

gth

elsquolsquop

olic

yrsquorsquo

requ

irem

ent)

an

dis

als

oex

pre

ssed

rela

tive

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

Th

eca

tegor

y-

spec

ific

ind

exvalu

essu

mto

mor

eth

an

100

for

two

reaso

ns

firs

tw

eu

sea

few

pol

icy

term

sin

mor

eth

an

one

pol

icy

cate

gor

y

For

exam

ple

lsquolsquoM

edic

aid

rsquorsquoap

pea

rsin

the

term

sets

for

bot

hh

ealt

hca

rean

den

titl

emen

tp

rogra

ms

Sec

ond

a

new

spap

erart

icle

that

mee

tsth

elsquolsquoe

con

omyrsquorsquo

lsquolsquopol

icyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

crit

eria

can

refe

rto

mor

eth

an

one

pol

icy

cate

gor

y

ECONOMIC POLICY UNCERTAINTY 1603

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 6: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

Finally there is a rapidly growing literature on text searchmethodsmdashusing newspaper archives in particularmdashto measure avariety of outcomes Examples include Gentzkow and Shapiro(2010) Hoberg and Phillips (2010) Boudoukh et al (2013) andAlexopoulos and Cohen (2015) Our work suggests that newspa-per text search can yield useful proxies for economic and policyconditions stretching back several decades which could be espe-cially valuable in earlier eras and in countries with fewer datasources

Section II describes the data we use to construct our policyuncertainty indexes Section III evaluates our EPU measures inseveral ways and develops additional evidence about movementsin policy-related uncertainty over time Section IV investigateshow firm-level outcomes covary with policy uncertainty and thedynamic responses of aggregate outcomes to policy uncertaintyinnovations Section V concludes and offers some thoughts aboutdirections for future research

II Measuring EPU

We build indexes of policy-related economic uncertaintybased on newspaper coverage frequency6 We aim to capture un-certainty about who will make economic policy decisions whateconomic policy actions will be undertaken and when and theeconomic effects of policy actions (or inaction)mdashincluding uncer-tainties related to the economic ramifications of lsquolsquononeconomicrsquorsquopolicy matters for example military actions Our measures cap-ture both near-term concerns (eg when will the Fed adjust itspolicy rate) and longer term concerns (eg how to fund entitle-ment programs) as reflected in newspaper articles We first de-scribe the construction of our monthly and daily EPU indexes forthe United States from 1985 onward and then turn to indexes forspecific policy categories indexes for other countries and histor-ical indexes for the United States and United Kingdom

6 Earlier drafts of this article include index components based on (i) the pre-sent value of future scheduled tax code expirations and (ii) disagreement amongprofessional forecasters over future government purchases and consumer pricesHowever to extend our EPU measures over time and across countries we focushere on the newspaper approach while continuing to report the other componentsat httpwwwpolicyuncertaintycom

QUARTERLY JOURNAL OF ECONOMICS1598

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

IIA US Economic Policy Uncertainty Indexes from 1985

Our modern monthly EPU index for the United States relieson 10 leading newspapers USA Today Miami Herald ChicagoTribune Washington Post Los Angeles Times Boston Globe SanFrancisco Chronicle Dallas Morning News New York Timesand Wall Street Journal We search the digital archives of eachpaper from January 1985 to obtain a monthly count of articlesthat contain the following trio of terms lsquolsquouncertaintyrsquorsquo or lsquolsquouncer-tainrsquorsquo lsquolsquoeconomicrsquorsquo or lsquolsquoeconomyrsquorsquo and one of the following policyterms lsquolsquoCongressrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederal Reserversquorsquo lsquolsquolegislationrsquorsquolsquolsquoregulationrsquorsquo or lsquolsquoWhite Housersquorsquo (including variants likelsquolsquouncertaintiesrsquorsquo lsquolsquoregulatoryrsquorsquo or lsquolsquothe Fedrsquorsquo) In other words tomeet our criteria an article must contain terms in all three cat-egories pertaining to uncertainty the economy and policy Weuse our audit study to select the policy terms as explained inSection IIIA

An obvious difficulty with these raw counts is that the over-all volume of articles varies across newspapers and time Thuswe scale the raw counts by the total number of articles in thesame newspaper and month We standardize each monthlynewspaper-level series to unit standard deviation from 1985 to2009 and then average across the 10 papers by month Finallywe normalize the 10-paper series to a mean of 100 from 1985 to2009 To be precise let Xit denote the scaled EPU frequencycounts for newspaper i = 1 2 10 in month t and let T1 andT2 denote the time intervals used in the standardization andnormalization calculations We proceed in the following steps(i) Compute the times-series variance 2

i in the interval T1 foreach paper i (ii) Standardize Xit by dividing through by thestandard deviation i for all t This operation yields for eachpaper a series Yit with unit standard deviation in the intervalT1 (iii) Compute the mean over newspapers of Yit in each monthto obtain the series Zt (iv) Compute M the mean value of Zt inthe interval T2 (v) Multiply Zt by (100M) for all t to obtain thenormalized EPU time-series index We use the same approachfor other countries and indexes

Figure I plots the resulting index which shows clear spikesaround the Gulf Wars close presidential elections the 911 ter-rorist attack the stimulus debate in early 2008 the LehmanBrothers bankruptcy and TARP legislation in late 2008 thesummer 2011 debt ceiling dispute and the battle over the lsquolsquofiscal

ECONOMIC POLICY UNCERTAINTY 1599

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

cliffrsquorsquo in late 2012 among other events and developments Somenotable political events do not generate high EPU according toour index For instance our EPU index shows no large spike inconnection with the partial federal government shutdowns fromNovember 1995 to January 1996 although those shutdowns re-ceived quite a lot of press coverage7

In addition to our monthly index we produce a daily EPUindex using the Newsbank news aggregator which coversaround 1500 US newspapers Newsbankrsquos extensive coverageyields enough articles to generate a meaningful daily countTaking monthly averages of our daily index it correlates at 085with our 10-paper monthly index indicating a high degree of sim-ilarity Because papers enter and leave the Newsbank archive andits count of newspapers expands greatly over time compositionalshifts potentially distort the longer term behavior of the daily EPU

FIGURE I

EPU Index for the United States

7 We find more than 8000 articles about these shutdowns in Newsbank ar-chives but less than 25 also mention the economy less than 2 mention uncer-tainty and only 1 mentions both Thus politically tumultuous episodes do notnecessarily raise EPU by our measure

QUARTERLY JOURNAL OF ECONOMICS1600

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

index Hence we focus on our 10-paper monthly EPU index butthe daily index provides a useful high-frequency alternative8

IIB EPU Indexes for Policy Categories

To create indexes for policy categories we apply additionalcriteria to those articles that contain our trio of terms about theeconomy policy and uncertainty The additional criteria involvethe presence of one or more category-relevant terms lsquolsquothe Fedrsquorsquolsquolsquocentral bankrsquorsquo lsquolsquointerest ratersquorsquo lsquolsquoinflationrsquorsquo and so on for the mon-etary policy category for example Online Appendix B reports thefull set of terms that define our 11 policy categories and subcat-egories We use Newsbank for the category indexes because itshigh text density facilitates measurement by time period andpolicy category As seen in Figure II the national security EPUindex spiked sharply in connection with the 911 attacks GulfWar I and the onset of Gulf War II The health care EPU index

FIGURE II

National Security and Health Care EPU Indexes

8 We update the daily EPU index at approximately 9 am EST each day andpost it at httpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1601

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

rose sharply during the Clinton health care reform initiative in1993ndash1994 and has fluctuated at high levels from 2009 to 2014

Table I reports all 11 category-specific EPU indexes9 It alsoreports an overall economic uncertainty (EU) index that drops thepolicy requirement in the EPU index The first two rows reportaverage EU and EPU values for the indicated periods expressedrelative to the average EPU value from 1985 to 2014 For exam-ple the EU value of 2182 says the (scaled) frequency of EU arti-cles from 19851 to 19906 is somewhat more than twice theaverage frequency of EPU articles from 1985 to 2014 The next11 rows report relative frequency values for specific policy cate-gories and time periods For example the 548 value for lsquolsquoNationalSecurityrsquorsquo says the frequency of EPU articles during 20019 to200212 that mention national security matters is 548 of the1985ndash2014 average EPU frequency and 43 548

1285

of the EPU

frequency from 20019 to 200212Fiscal matters especially tax policy stand out in Table I as the

largest source of policy uncertainty especially in recent years Thefiscal policy EPU index rose from values near 33 in the precrisisyears to 615 in 20089 to 200912 and 783 from 2010 to 2013Health care policy is the second largest source of elevated EPU inrecent years Policy uncertainty related to financial regulations andentitlement programs also rose sharply after 2008 but from initiallylower levels Concerns related to sovereign debt and currency crisesare up by an order of magnitude during 2010 to 2013 but from sucha low base as to have little impact on the overall EPU index EPUconcerns related to monetary policy are important throughout the1985ndash2014 period but perhaps surprisingly they are not elevatedin recent years by our measure We interpret this result as a reflec-tion of low and stable inflation rates in recent years which appar-ently drive newspaper coverage more than disputes amongprofessional economists about unconventional monetary policies10

Several other researchers develop measures related to uncer-tainty about government behavior Marina Azzimonti (2015) con-structs a newspaper index of partisan conflict at the federal levelthat shows similarities to our EPU index but also notable

9 In contrast to Figure III which normalizes each category-specific EPUseries to 100 Table I expresses each category-specific EPU series as a percentageof the overall EPU frequency from 1985 to 2014

10 Other evidence also points to subdued levels of inflation uncertainty inrecent years See Nalewaik (2015) for a presentation and discussion of evidencebased on time-series models surveys and financial markets data

QUARTERLY JOURNAL OF ECONOMICS1602

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EI

EC

ON

OM

ICP

OL

ICY

UN

CE

RT

AIN

TY

BY

PO

LIC

YC

AT

EG

OR

YA

ND

TIM

EP

ER

IOD

1985ndash2014

Tim

ep

erio

d19851

ndash19906

19907

ndash19911

219921

ndash20018

20019

ndash20021

220031

ndash20076

20077

ndash20088

20089

ndash20091

220101

ndash20131

019851

ndash20141

2

Mid

-80s

toG

ulf

War

IG

ulf

War

I1990s

boo

mto

91

191

1att

ack

s2000s

boo

m

Earl

ycr

edit

cru

nch

Leh

man

coll

ap

seamp

rece

ssio

n

Fis

cal

pol

icy

batt

les

Over

all

aver

age

Over

all

econ

omic

un

cert

ain

ty2182

3498

1859

3269

1598

1848

3709

2521

2193

Eco

nom

icp

olic

yu

nce

rtain

ty1096

1419

881

1285

714

834

1321

1275

1000

Fis

cal

pol

icy

496

596

359

554

323

331

615

783

461

Taxes

399

484

319

512

302

314

569

681

403

Gov

ern

men

tsp

end

ing

ampot

her

227

268

121

173

85

66

171

332

171

Mon

etary

pol

icy

327

418

261

452

222

316

278

261

281

Hea

lth

care

70

154

149

184

131

134

293

393

173

Nati

onal

secu

rity

250

536

180

548

254

159

213

198

238

Reg

ula

tion

157

230

145

196

112

155

292

281

174

Fin

an

cial

regu

lati

on33

70

13

53

17

36

102

61

33

Sov

erei

gn

deb

tamp

curr

ency

cris

es14

06

23

05

04

03

04

39

16

En

titl

emen

tp

rogra

ms

73

126

115

187

88

82

153

247

124

Tra

de

pol

icy

38

40

63

26

17

20

14

21

38

Su

mof

pol

icy

cate

gor

ies

1425

2107

1295

2151

1152

1200

1863

2222

1506

Rati

oof

EP

Uto

over

all

EU

05

004

104

703

904

504

503

605

104

7

Not

es

Qu

erie

sru

nF

ebru

ary

12

2015

onU

S

new

spap

ers

inA

cces

sW

orld

New

sN

ewsb

an

k

usi

ng

the

cate

gor

y-s

pec

ific

pol

icy

term

sets

list

edin

On

lin

eA

pp

end

ixB

E

xce

pt

for

the

last

row

all

entr

ies

are

exp

ress

edre

lati

ve

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

lsquolsquoOver

all

econ

omic

un

cert

ain

tyrsquorsquo

qu

an

tifi

esth

efr

equ

ency

ofart

icle

sth

at

mee

tou

rlsquolsquoe

con

omyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

requ

irem

ents

(ie

d

rop

pin

gth

elsquolsquop

olic

yrsquorsquo

requ

irem

ent)

an

dis

als

oex

pre

ssed

rela

tive

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

Th

eca

tegor

y-

spec

ific

ind

exvalu

essu

mto

mor

eth

an

100

for

two

reaso

ns

firs

tw

eu

sea

few

pol

icy

term

sin

mor

eth

an

one

pol

icy

cate

gor

y

For

exam

ple

lsquolsquoM

edic

aid

rsquorsquoap

pea

rsin

the

term

sets

for

bot

hh

ealt

hca

rean

den

titl

emen

tp

rogra

ms

Sec

ond

a

new

spap

erart

icle

that

mee

tsth

elsquolsquoe

con

omyrsquorsquo

lsquolsquopol

icyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

crit

eria

can

refe

rto

mor

eth

an

one

pol

icy

cate

gor

y

ECONOMIC POLICY UNCERTAINTY 1603

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 7: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

IIA US Economic Policy Uncertainty Indexes from 1985

Our modern monthly EPU index for the United States relieson 10 leading newspapers USA Today Miami Herald ChicagoTribune Washington Post Los Angeles Times Boston Globe SanFrancisco Chronicle Dallas Morning News New York Timesand Wall Street Journal We search the digital archives of eachpaper from January 1985 to obtain a monthly count of articlesthat contain the following trio of terms lsquolsquouncertaintyrsquorsquo or lsquolsquouncer-tainrsquorsquo lsquolsquoeconomicrsquorsquo or lsquolsquoeconomyrsquorsquo and one of the following policyterms lsquolsquoCongressrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederal Reserversquorsquo lsquolsquolegislationrsquorsquolsquolsquoregulationrsquorsquo or lsquolsquoWhite Housersquorsquo (including variants likelsquolsquouncertaintiesrsquorsquo lsquolsquoregulatoryrsquorsquo or lsquolsquothe Fedrsquorsquo) In other words tomeet our criteria an article must contain terms in all three cat-egories pertaining to uncertainty the economy and policy Weuse our audit study to select the policy terms as explained inSection IIIA

An obvious difficulty with these raw counts is that the over-all volume of articles varies across newspapers and time Thuswe scale the raw counts by the total number of articles in thesame newspaper and month We standardize each monthlynewspaper-level series to unit standard deviation from 1985 to2009 and then average across the 10 papers by month Finallywe normalize the 10-paper series to a mean of 100 from 1985 to2009 To be precise let Xit denote the scaled EPU frequencycounts for newspaper i = 1 2 10 in month t and let T1 andT2 denote the time intervals used in the standardization andnormalization calculations We proceed in the following steps(i) Compute the times-series variance 2

i in the interval T1 foreach paper i (ii) Standardize Xit by dividing through by thestandard deviation i for all t This operation yields for eachpaper a series Yit with unit standard deviation in the intervalT1 (iii) Compute the mean over newspapers of Yit in each monthto obtain the series Zt (iv) Compute M the mean value of Zt inthe interval T2 (v) Multiply Zt by (100M) for all t to obtain thenormalized EPU time-series index We use the same approachfor other countries and indexes

Figure I plots the resulting index which shows clear spikesaround the Gulf Wars close presidential elections the 911 ter-rorist attack the stimulus debate in early 2008 the LehmanBrothers bankruptcy and TARP legislation in late 2008 thesummer 2011 debt ceiling dispute and the battle over the lsquolsquofiscal

ECONOMIC POLICY UNCERTAINTY 1599

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

cliffrsquorsquo in late 2012 among other events and developments Somenotable political events do not generate high EPU according toour index For instance our EPU index shows no large spike inconnection with the partial federal government shutdowns fromNovember 1995 to January 1996 although those shutdowns re-ceived quite a lot of press coverage7

In addition to our monthly index we produce a daily EPUindex using the Newsbank news aggregator which coversaround 1500 US newspapers Newsbankrsquos extensive coverageyields enough articles to generate a meaningful daily countTaking monthly averages of our daily index it correlates at 085with our 10-paper monthly index indicating a high degree of sim-ilarity Because papers enter and leave the Newsbank archive andits count of newspapers expands greatly over time compositionalshifts potentially distort the longer term behavior of the daily EPU

FIGURE I

EPU Index for the United States

7 We find more than 8000 articles about these shutdowns in Newsbank ar-chives but less than 25 also mention the economy less than 2 mention uncer-tainty and only 1 mentions both Thus politically tumultuous episodes do notnecessarily raise EPU by our measure

QUARTERLY JOURNAL OF ECONOMICS1600

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

index Hence we focus on our 10-paper monthly EPU index butthe daily index provides a useful high-frequency alternative8

IIB EPU Indexes for Policy Categories

To create indexes for policy categories we apply additionalcriteria to those articles that contain our trio of terms about theeconomy policy and uncertainty The additional criteria involvethe presence of one or more category-relevant terms lsquolsquothe Fedrsquorsquolsquolsquocentral bankrsquorsquo lsquolsquointerest ratersquorsquo lsquolsquoinflationrsquorsquo and so on for the mon-etary policy category for example Online Appendix B reports thefull set of terms that define our 11 policy categories and subcat-egories We use Newsbank for the category indexes because itshigh text density facilitates measurement by time period andpolicy category As seen in Figure II the national security EPUindex spiked sharply in connection with the 911 attacks GulfWar I and the onset of Gulf War II The health care EPU index

FIGURE II

National Security and Health Care EPU Indexes

8 We update the daily EPU index at approximately 9 am EST each day andpost it at httpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1601

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

rose sharply during the Clinton health care reform initiative in1993ndash1994 and has fluctuated at high levels from 2009 to 2014

Table I reports all 11 category-specific EPU indexes9 It alsoreports an overall economic uncertainty (EU) index that drops thepolicy requirement in the EPU index The first two rows reportaverage EU and EPU values for the indicated periods expressedrelative to the average EPU value from 1985 to 2014 For exam-ple the EU value of 2182 says the (scaled) frequency of EU arti-cles from 19851 to 19906 is somewhat more than twice theaverage frequency of EPU articles from 1985 to 2014 The next11 rows report relative frequency values for specific policy cate-gories and time periods For example the 548 value for lsquolsquoNationalSecurityrsquorsquo says the frequency of EPU articles during 20019 to200212 that mention national security matters is 548 of the1985ndash2014 average EPU frequency and 43 548

1285

of the EPU

frequency from 20019 to 200212Fiscal matters especially tax policy stand out in Table I as the

largest source of policy uncertainty especially in recent years Thefiscal policy EPU index rose from values near 33 in the precrisisyears to 615 in 20089 to 200912 and 783 from 2010 to 2013Health care policy is the second largest source of elevated EPU inrecent years Policy uncertainty related to financial regulations andentitlement programs also rose sharply after 2008 but from initiallylower levels Concerns related to sovereign debt and currency crisesare up by an order of magnitude during 2010 to 2013 but from sucha low base as to have little impact on the overall EPU index EPUconcerns related to monetary policy are important throughout the1985ndash2014 period but perhaps surprisingly they are not elevatedin recent years by our measure We interpret this result as a reflec-tion of low and stable inflation rates in recent years which appar-ently drive newspaper coverage more than disputes amongprofessional economists about unconventional monetary policies10

Several other researchers develop measures related to uncer-tainty about government behavior Marina Azzimonti (2015) con-structs a newspaper index of partisan conflict at the federal levelthat shows similarities to our EPU index but also notable

9 In contrast to Figure III which normalizes each category-specific EPUseries to 100 Table I expresses each category-specific EPU series as a percentageof the overall EPU frequency from 1985 to 2014

10 Other evidence also points to subdued levels of inflation uncertainty inrecent years See Nalewaik (2015) for a presentation and discussion of evidencebased on time-series models surveys and financial markets data

QUARTERLY JOURNAL OF ECONOMICS1602

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EI

EC

ON

OM

ICP

OL

ICY

UN

CE

RT

AIN

TY

BY

PO

LIC

YC

AT

EG

OR

YA

ND

TIM

EP

ER

IOD

1985ndash2014

Tim

ep

erio

d19851

ndash19906

19907

ndash19911

219921

ndash20018

20019

ndash20021

220031

ndash20076

20077

ndash20088

20089

ndash20091

220101

ndash20131

019851

ndash20141

2

Mid

-80s

toG

ulf

War

IG

ulf

War

I1990s

boo

mto

91

191

1att

ack

s2000s

boo

m

Earl

ycr

edit

cru

nch

Leh

man

coll

ap

seamp

rece

ssio

n

Fis

cal

pol

icy

batt

les

Over

all

aver

age

Over

all

econ

omic

un

cert

ain

ty2182

3498

1859

3269

1598

1848

3709

2521

2193

Eco

nom

icp

olic

yu

nce

rtain

ty1096

1419

881

1285

714

834

1321

1275

1000

Fis

cal

pol

icy

496

596

359

554

323

331

615

783

461

Taxes

399

484

319

512

302

314

569

681

403

Gov

ern

men

tsp

end

ing

ampot

her

227

268

121

173

85

66

171

332

171

Mon

etary

pol

icy

327

418

261

452

222

316

278

261

281

Hea

lth

care

70

154

149

184

131

134

293

393

173

Nati

onal

secu

rity

250

536

180

548

254

159

213

198

238

Reg

ula

tion

157

230

145

196

112

155

292

281

174

Fin

an

cial

regu

lati

on33

70

13

53

17

36

102

61

33

Sov

erei

gn

deb

tamp

curr

ency

cris

es14

06

23

05

04

03

04

39

16

En

titl

emen

tp

rogra

ms

73

126

115

187

88

82

153

247

124

Tra

de

pol

icy

38

40

63

26

17

20

14

21

38

Su

mof

pol

icy

cate

gor

ies

1425

2107

1295

2151

1152

1200

1863

2222

1506

Rati

oof

EP

Uto

over

all

EU

05

004

104

703

904

504

503

605

104

7

Not

es

Qu

erie

sru

nF

ebru

ary

12

2015

onU

S

new

spap

ers

inA

cces

sW

orld

New

sN

ewsb

an

k

usi

ng

the

cate

gor

y-s

pec

ific

pol

icy

term

sets

list

edin

On

lin

eA

pp

end

ixB

E

xce

pt

for

the

last

row

all

entr

ies

are

exp

ress

edre

lati

ve

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

lsquolsquoOver

all

econ

omic

un

cert

ain

tyrsquorsquo

qu

an

tifi

esth

efr

equ

ency

ofart

icle

sth

at

mee

tou

rlsquolsquoe

con

omyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

requ

irem

ents

(ie

d

rop

pin

gth

elsquolsquop

olic

yrsquorsquo

requ

irem

ent)

an

dis

als

oex

pre

ssed

rela

tive

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

Th

eca

tegor

y-

spec

ific

ind

exvalu

essu

mto

mor

eth

an

100

for

two

reaso

ns

firs

tw

eu

sea

few

pol

icy

term

sin

mor

eth

an

one

pol

icy

cate

gor

y

For

exam

ple

lsquolsquoM

edic

aid

rsquorsquoap

pea

rsin

the

term

sets

for

bot

hh

ealt

hca

rean

den

titl

emen

tp

rogra

ms

Sec

ond

a

new

spap

erart

icle

that

mee

tsth

elsquolsquoe

con

omyrsquorsquo

lsquolsquopol

icyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

crit

eria

can

refe

rto

mor

eth

an

one

pol

icy

cate

gor

y

ECONOMIC POLICY UNCERTAINTY 1603

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 8: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

cliffrsquorsquo in late 2012 among other events and developments Somenotable political events do not generate high EPU according toour index For instance our EPU index shows no large spike inconnection with the partial federal government shutdowns fromNovember 1995 to January 1996 although those shutdowns re-ceived quite a lot of press coverage7

In addition to our monthly index we produce a daily EPUindex using the Newsbank news aggregator which coversaround 1500 US newspapers Newsbankrsquos extensive coverageyields enough articles to generate a meaningful daily countTaking monthly averages of our daily index it correlates at 085with our 10-paper monthly index indicating a high degree of sim-ilarity Because papers enter and leave the Newsbank archive andits count of newspapers expands greatly over time compositionalshifts potentially distort the longer term behavior of the daily EPU

FIGURE I

EPU Index for the United States

7 We find more than 8000 articles about these shutdowns in Newsbank ar-chives but less than 25 also mention the economy less than 2 mention uncer-tainty and only 1 mentions both Thus politically tumultuous episodes do notnecessarily raise EPU by our measure

QUARTERLY JOURNAL OF ECONOMICS1600

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

index Hence we focus on our 10-paper monthly EPU index butthe daily index provides a useful high-frequency alternative8

IIB EPU Indexes for Policy Categories

To create indexes for policy categories we apply additionalcriteria to those articles that contain our trio of terms about theeconomy policy and uncertainty The additional criteria involvethe presence of one or more category-relevant terms lsquolsquothe Fedrsquorsquolsquolsquocentral bankrsquorsquo lsquolsquointerest ratersquorsquo lsquolsquoinflationrsquorsquo and so on for the mon-etary policy category for example Online Appendix B reports thefull set of terms that define our 11 policy categories and subcat-egories We use Newsbank for the category indexes because itshigh text density facilitates measurement by time period andpolicy category As seen in Figure II the national security EPUindex spiked sharply in connection with the 911 attacks GulfWar I and the onset of Gulf War II The health care EPU index

FIGURE II

National Security and Health Care EPU Indexes

8 We update the daily EPU index at approximately 9 am EST each day andpost it at httpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1601

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

rose sharply during the Clinton health care reform initiative in1993ndash1994 and has fluctuated at high levels from 2009 to 2014

Table I reports all 11 category-specific EPU indexes9 It alsoreports an overall economic uncertainty (EU) index that drops thepolicy requirement in the EPU index The first two rows reportaverage EU and EPU values for the indicated periods expressedrelative to the average EPU value from 1985 to 2014 For exam-ple the EU value of 2182 says the (scaled) frequency of EU arti-cles from 19851 to 19906 is somewhat more than twice theaverage frequency of EPU articles from 1985 to 2014 The next11 rows report relative frequency values for specific policy cate-gories and time periods For example the 548 value for lsquolsquoNationalSecurityrsquorsquo says the frequency of EPU articles during 20019 to200212 that mention national security matters is 548 of the1985ndash2014 average EPU frequency and 43 548

1285

of the EPU

frequency from 20019 to 200212Fiscal matters especially tax policy stand out in Table I as the

largest source of policy uncertainty especially in recent years Thefiscal policy EPU index rose from values near 33 in the precrisisyears to 615 in 20089 to 200912 and 783 from 2010 to 2013Health care policy is the second largest source of elevated EPU inrecent years Policy uncertainty related to financial regulations andentitlement programs also rose sharply after 2008 but from initiallylower levels Concerns related to sovereign debt and currency crisesare up by an order of magnitude during 2010 to 2013 but from sucha low base as to have little impact on the overall EPU index EPUconcerns related to monetary policy are important throughout the1985ndash2014 period but perhaps surprisingly they are not elevatedin recent years by our measure We interpret this result as a reflec-tion of low and stable inflation rates in recent years which appar-ently drive newspaper coverage more than disputes amongprofessional economists about unconventional monetary policies10

Several other researchers develop measures related to uncer-tainty about government behavior Marina Azzimonti (2015) con-structs a newspaper index of partisan conflict at the federal levelthat shows similarities to our EPU index but also notable

9 In contrast to Figure III which normalizes each category-specific EPUseries to 100 Table I expresses each category-specific EPU series as a percentageof the overall EPU frequency from 1985 to 2014

10 Other evidence also points to subdued levels of inflation uncertainty inrecent years See Nalewaik (2015) for a presentation and discussion of evidencebased on time-series models surveys and financial markets data

QUARTERLY JOURNAL OF ECONOMICS1602

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EI

EC

ON

OM

ICP

OL

ICY

UN

CE

RT

AIN

TY

BY

PO

LIC

YC

AT

EG

OR

YA

ND

TIM

EP

ER

IOD

1985ndash2014

Tim

ep

erio

d19851

ndash19906

19907

ndash19911

219921

ndash20018

20019

ndash20021

220031

ndash20076

20077

ndash20088

20089

ndash20091

220101

ndash20131

019851

ndash20141

2

Mid

-80s

toG

ulf

War

IG

ulf

War

I1990s

boo

mto

91

191

1att

ack

s2000s

boo

m

Earl

ycr

edit

cru

nch

Leh

man

coll

ap

seamp

rece

ssio

n

Fis

cal

pol

icy

batt

les

Over

all

aver

age

Over

all

econ

omic

un

cert

ain

ty2182

3498

1859

3269

1598

1848

3709

2521

2193

Eco

nom

icp

olic

yu

nce

rtain

ty1096

1419

881

1285

714

834

1321

1275

1000

Fis

cal

pol

icy

496

596

359

554

323

331

615

783

461

Taxes

399

484

319

512

302

314

569

681

403

Gov

ern

men

tsp

end

ing

ampot

her

227

268

121

173

85

66

171

332

171

Mon

etary

pol

icy

327

418

261

452

222

316

278

261

281

Hea

lth

care

70

154

149

184

131

134

293

393

173

Nati

onal

secu

rity

250

536

180

548

254

159

213

198

238

Reg

ula

tion

157

230

145

196

112

155

292

281

174

Fin

an

cial

regu

lati

on33

70

13

53

17

36

102

61

33

Sov

erei

gn

deb

tamp

curr

ency

cris

es14

06

23

05

04

03

04

39

16

En

titl

emen

tp

rogra

ms

73

126

115

187

88

82

153

247

124

Tra

de

pol

icy

38

40

63

26

17

20

14

21

38

Su

mof

pol

icy

cate

gor

ies

1425

2107

1295

2151

1152

1200

1863

2222

1506

Rati

oof

EP

Uto

over

all

EU

05

004

104

703

904

504

503

605

104

7

Not

es

Qu

erie

sru

nF

ebru

ary

12

2015

onU

S

new

spap

ers

inA

cces

sW

orld

New

sN

ewsb

an

k

usi

ng

the

cate

gor

y-s

pec

ific

pol

icy

term

sets

list

edin

On

lin

eA

pp

end

ixB

E

xce

pt

for

the

last

row

all

entr

ies

are

exp

ress

edre

lati

ve

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

lsquolsquoOver

all

econ

omic

un

cert

ain

tyrsquorsquo

qu

an

tifi

esth

efr

equ

ency

ofart

icle

sth

at

mee

tou

rlsquolsquoe

con

omyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

requ

irem

ents

(ie

d

rop

pin

gth

elsquolsquop

olic

yrsquorsquo

requ

irem

ent)

an

dis

als

oex

pre

ssed

rela

tive

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

Th

eca

tegor

y-

spec

ific

ind

exvalu

essu

mto

mor

eth

an

100

for

two

reaso

ns

firs

tw

eu

sea

few

pol

icy

term

sin

mor

eth

an

one

pol

icy

cate

gor

y

For

exam

ple

lsquolsquoM

edic

aid

rsquorsquoap

pea

rsin

the

term

sets

for

bot

hh

ealt

hca

rean

den

titl

emen

tp

rogra

ms

Sec

ond

a

new

spap

erart

icle

that

mee

tsth

elsquolsquoe

con

omyrsquorsquo

lsquolsquopol

icyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

crit

eria

can

refe

rto

mor

eth

an

one

pol

icy

cate

gor

y

ECONOMIC POLICY UNCERTAINTY 1603

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 9: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

index Hence we focus on our 10-paper monthly EPU index butthe daily index provides a useful high-frequency alternative8

IIB EPU Indexes for Policy Categories

To create indexes for policy categories we apply additionalcriteria to those articles that contain our trio of terms about theeconomy policy and uncertainty The additional criteria involvethe presence of one or more category-relevant terms lsquolsquothe Fedrsquorsquolsquolsquocentral bankrsquorsquo lsquolsquointerest ratersquorsquo lsquolsquoinflationrsquorsquo and so on for the mon-etary policy category for example Online Appendix B reports thefull set of terms that define our 11 policy categories and subcat-egories We use Newsbank for the category indexes because itshigh text density facilitates measurement by time period andpolicy category As seen in Figure II the national security EPUindex spiked sharply in connection with the 911 attacks GulfWar I and the onset of Gulf War II The health care EPU index

FIGURE II

National Security and Health Care EPU Indexes

8 We update the daily EPU index at approximately 9 am EST each day andpost it at httpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1601

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

rose sharply during the Clinton health care reform initiative in1993ndash1994 and has fluctuated at high levels from 2009 to 2014

Table I reports all 11 category-specific EPU indexes9 It alsoreports an overall economic uncertainty (EU) index that drops thepolicy requirement in the EPU index The first two rows reportaverage EU and EPU values for the indicated periods expressedrelative to the average EPU value from 1985 to 2014 For exam-ple the EU value of 2182 says the (scaled) frequency of EU arti-cles from 19851 to 19906 is somewhat more than twice theaverage frequency of EPU articles from 1985 to 2014 The next11 rows report relative frequency values for specific policy cate-gories and time periods For example the 548 value for lsquolsquoNationalSecurityrsquorsquo says the frequency of EPU articles during 20019 to200212 that mention national security matters is 548 of the1985ndash2014 average EPU frequency and 43 548

1285

of the EPU

frequency from 20019 to 200212Fiscal matters especially tax policy stand out in Table I as the

largest source of policy uncertainty especially in recent years Thefiscal policy EPU index rose from values near 33 in the precrisisyears to 615 in 20089 to 200912 and 783 from 2010 to 2013Health care policy is the second largest source of elevated EPU inrecent years Policy uncertainty related to financial regulations andentitlement programs also rose sharply after 2008 but from initiallylower levels Concerns related to sovereign debt and currency crisesare up by an order of magnitude during 2010 to 2013 but from sucha low base as to have little impact on the overall EPU index EPUconcerns related to monetary policy are important throughout the1985ndash2014 period but perhaps surprisingly they are not elevatedin recent years by our measure We interpret this result as a reflec-tion of low and stable inflation rates in recent years which appar-ently drive newspaper coverage more than disputes amongprofessional economists about unconventional monetary policies10

Several other researchers develop measures related to uncer-tainty about government behavior Marina Azzimonti (2015) con-structs a newspaper index of partisan conflict at the federal levelthat shows similarities to our EPU index but also notable

9 In contrast to Figure III which normalizes each category-specific EPUseries to 100 Table I expresses each category-specific EPU series as a percentageof the overall EPU frequency from 1985 to 2014

10 Other evidence also points to subdued levels of inflation uncertainty inrecent years See Nalewaik (2015) for a presentation and discussion of evidencebased on time-series models surveys and financial markets data

QUARTERLY JOURNAL OF ECONOMICS1602

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EI

EC

ON

OM

ICP

OL

ICY

UN

CE

RT

AIN

TY

BY

PO

LIC

YC

AT

EG

OR

YA

ND

TIM

EP

ER

IOD

1985ndash2014

Tim

ep

erio

d19851

ndash19906

19907

ndash19911

219921

ndash20018

20019

ndash20021

220031

ndash20076

20077

ndash20088

20089

ndash20091

220101

ndash20131

019851

ndash20141

2

Mid

-80s

toG

ulf

War

IG

ulf

War

I1990s

boo

mto

91

191

1att

ack

s2000s

boo

m

Earl

ycr

edit

cru

nch

Leh

man

coll

ap

seamp

rece

ssio

n

Fis

cal

pol

icy

batt

les

Over

all

aver

age

Over

all

econ

omic

un

cert

ain

ty2182

3498

1859

3269

1598

1848

3709

2521

2193

Eco

nom

icp

olic

yu

nce

rtain

ty1096

1419

881

1285

714

834

1321

1275

1000

Fis

cal

pol

icy

496

596

359

554

323

331

615

783

461

Taxes

399

484

319

512

302

314

569

681

403

Gov

ern

men

tsp

end

ing

ampot

her

227

268

121

173

85

66

171

332

171

Mon

etary

pol

icy

327

418

261

452

222

316

278

261

281

Hea

lth

care

70

154

149

184

131

134

293

393

173

Nati

onal

secu

rity

250

536

180

548

254

159

213

198

238

Reg

ula

tion

157

230

145

196

112

155

292

281

174

Fin

an

cial

regu

lati

on33

70

13

53

17

36

102

61

33

Sov

erei

gn

deb

tamp

curr

ency

cris

es14

06

23

05

04

03

04

39

16

En

titl

emen

tp

rogra

ms

73

126

115

187

88

82

153

247

124

Tra

de

pol

icy

38

40

63

26

17

20

14

21

38

Su

mof

pol

icy

cate

gor

ies

1425

2107

1295

2151

1152

1200

1863

2222

1506

Rati

oof

EP

Uto

over

all

EU

05

004

104

703

904

504

503

605

104

7

Not

es

Qu

erie

sru

nF

ebru

ary

12

2015

onU

S

new

spap

ers

inA

cces

sW

orld

New

sN

ewsb

an

k

usi

ng

the

cate

gor

y-s

pec

ific

pol

icy

term

sets

list

edin

On

lin

eA

pp

end

ixB

E

xce

pt

for

the

last

row

all

entr

ies

are

exp

ress

edre

lati

ve

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

lsquolsquoOver

all

econ

omic

un

cert

ain

tyrsquorsquo

qu

an

tifi

esth

efr

equ

ency

ofart

icle

sth

at

mee

tou

rlsquolsquoe

con

omyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

requ

irem

ents

(ie

d

rop

pin

gth

elsquolsquop

olic

yrsquorsquo

requ

irem

ent)

an

dis

als

oex

pre

ssed

rela

tive

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

Th

eca

tegor

y-

spec

ific

ind

exvalu

essu

mto

mor

eth

an

100

for

two

reaso

ns

firs

tw

eu

sea

few

pol

icy

term

sin

mor

eth

an

one

pol

icy

cate

gor

y

For

exam

ple

lsquolsquoM

edic

aid

rsquorsquoap

pea

rsin

the

term

sets

for

bot

hh

ealt

hca

rean

den

titl

emen

tp

rogra

ms

Sec

ond

a

new

spap

erart

icle

that

mee

tsth

elsquolsquoe

con

omyrsquorsquo

lsquolsquopol

icyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

crit

eria

can

refe

rto

mor

eth

an

one

pol

icy

cate

gor

y

ECONOMIC POLICY UNCERTAINTY 1603

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 10: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

rose sharply during the Clinton health care reform initiative in1993ndash1994 and has fluctuated at high levels from 2009 to 2014

Table I reports all 11 category-specific EPU indexes9 It alsoreports an overall economic uncertainty (EU) index that drops thepolicy requirement in the EPU index The first two rows reportaverage EU and EPU values for the indicated periods expressedrelative to the average EPU value from 1985 to 2014 For exam-ple the EU value of 2182 says the (scaled) frequency of EU arti-cles from 19851 to 19906 is somewhat more than twice theaverage frequency of EPU articles from 1985 to 2014 The next11 rows report relative frequency values for specific policy cate-gories and time periods For example the 548 value for lsquolsquoNationalSecurityrsquorsquo says the frequency of EPU articles during 20019 to200212 that mention national security matters is 548 of the1985ndash2014 average EPU frequency and 43 548

1285

of the EPU

frequency from 20019 to 200212Fiscal matters especially tax policy stand out in Table I as the

largest source of policy uncertainty especially in recent years Thefiscal policy EPU index rose from values near 33 in the precrisisyears to 615 in 20089 to 200912 and 783 from 2010 to 2013Health care policy is the second largest source of elevated EPU inrecent years Policy uncertainty related to financial regulations andentitlement programs also rose sharply after 2008 but from initiallylower levels Concerns related to sovereign debt and currency crisesare up by an order of magnitude during 2010 to 2013 but from sucha low base as to have little impact on the overall EPU index EPUconcerns related to monetary policy are important throughout the1985ndash2014 period but perhaps surprisingly they are not elevatedin recent years by our measure We interpret this result as a reflec-tion of low and stable inflation rates in recent years which appar-ently drive newspaper coverage more than disputes amongprofessional economists about unconventional monetary policies10

Several other researchers develop measures related to uncer-tainty about government behavior Marina Azzimonti (2015) con-structs a newspaper index of partisan conflict at the federal levelthat shows similarities to our EPU index but also notable

9 In contrast to Figure III which normalizes each category-specific EPUseries to 100 Table I expresses each category-specific EPU series as a percentageof the overall EPU frequency from 1985 to 2014

10 Other evidence also points to subdued levels of inflation uncertainty inrecent years See Nalewaik (2015) for a presentation and discussion of evidencebased on time-series models surveys and financial markets data

QUARTERLY JOURNAL OF ECONOMICS1602

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EI

EC

ON

OM

ICP

OL

ICY

UN

CE

RT

AIN

TY

BY

PO

LIC

YC

AT

EG

OR

YA

ND

TIM

EP

ER

IOD

1985ndash2014

Tim

ep

erio

d19851

ndash19906

19907

ndash19911

219921

ndash20018

20019

ndash20021

220031

ndash20076

20077

ndash20088

20089

ndash20091

220101

ndash20131

019851

ndash20141

2

Mid

-80s

toG

ulf

War

IG

ulf

War

I1990s

boo

mto

91

191

1att

ack

s2000s

boo

m

Earl

ycr

edit

cru

nch

Leh

man

coll

ap

seamp

rece

ssio

n

Fis

cal

pol

icy

batt

les

Over

all

aver

age

Over

all

econ

omic

un

cert

ain

ty2182

3498

1859

3269

1598

1848

3709

2521

2193

Eco

nom

icp

olic

yu

nce

rtain

ty1096

1419

881

1285

714

834

1321

1275

1000

Fis

cal

pol

icy

496

596

359

554

323

331

615

783

461

Taxes

399

484

319

512

302

314

569

681

403

Gov

ern

men

tsp

end

ing

ampot

her

227

268

121

173

85

66

171

332

171

Mon

etary

pol

icy

327

418

261

452

222

316

278

261

281

Hea

lth

care

70

154

149

184

131

134

293

393

173

Nati

onal

secu

rity

250

536

180

548

254

159

213

198

238

Reg

ula

tion

157

230

145

196

112

155

292

281

174

Fin

an

cial

regu

lati

on33

70

13

53

17

36

102

61

33

Sov

erei

gn

deb

tamp

curr

ency

cris

es14

06

23

05

04

03

04

39

16

En

titl

emen

tp

rogra

ms

73

126

115

187

88

82

153

247

124

Tra

de

pol

icy

38

40

63

26

17

20

14

21

38

Su

mof

pol

icy

cate

gor

ies

1425

2107

1295

2151

1152

1200

1863

2222

1506

Rati

oof

EP

Uto

over

all

EU

05

004

104

703

904

504

503

605

104

7

Not

es

Qu

erie

sru

nF

ebru

ary

12

2015

onU

S

new

spap

ers

inA

cces

sW

orld

New

sN

ewsb

an

k

usi

ng

the

cate

gor

y-s

pec

ific

pol

icy

term

sets

list

edin

On

lin

eA

pp

end

ixB

E

xce

pt

for

the

last

row

all

entr

ies

are

exp

ress

edre

lati

ve

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

lsquolsquoOver

all

econ

omic

un

cert

ain

tyrsquorsquo

qu

an

tifi

esth

efr

equ

ency

ofart

icle

sth

at

mee

tou

rlsquolsquoe

con

omyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

requ

irem

ents

(ie

d

rop

pin

gth

elsquolsquop

olic

yrsquorsquo

requ

irem

ent)

an

dis

als

oex

pre

ssed

rela

tive

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

Th

eca

tegor

y-

spec

ific

ind

exvalu

essu

mto

mor

eth

an

100

for

two

reaso

ns

firs

tw

eu

sea

few

pol

icy

term

sin

mor

eth

an

one

pol

icy

cate

gor

y

For

exam

ple

lsquolsquoM

edic

aid

rsquorsquoap

pea

rsin

the

term

sets

for

bot

hh

ealt

hca

rean

den

titl

emen

tp

rogra

ms

Sec

ond

a

new

spap

erart

icle

that

mee

tsth

elsquolsquoe

con

omyrsquorsquo

lsquolsquopol

icyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

crit

eria

can

refe

rto

mor

eth

an

one

pol

icy

cate

gor

y

ECONOMIC POLICY UNCERTAINTY 1603

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 11: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

TA

BL

EI

EC

ON

OM

ICP

OL

ICY

UN

CE

RT

AIN

TY

BY

PO

LIC

YC

AT

EG

OR

YA

ND

TIM

EP

ER

IOD

1985ndash2014

Tim

ep

erio

d19851

ndash19906

19907

ndash19911

219921

ndash20018

20019

ndash20021

220031

ndash20076

20077

ndash20088

20089

ndash20091

220101

ndash20131

019851

ndash20141

2

Mid

-80s

toG

ulf

War

IG

ulf

War

I1990s

boo

mto

91

191

1att

ack

s2000s

boo

m

Earl

ycr

edit

cru

nch

Leh

man

coll

ap

seamp

rece

ssio

n

Fis

cal

pol

icy

batt

les

Over

all

aver

age

Over

all

econ

omic

un

cert

ain

ty2182

3498

1859

3269

1598

1848

3709

2521

2193

Eco

nom

icp

olic

yu

nce

rtain

ty1096

1419

881

1285

714

834

1321

1275

1000

Fis

cal

pol

icy

496

596

359

554

323

331

615

783

461

Taxes

399

484

319

512

302

314

569

681

403

Gov

ern

men

tsp

end

ing

ampot

her

227

268

121

173

85

66

171

332

171

Mon

etary

pol

icy

327

418

261

452

222

316

278

261

281

Hea

lth

care

70

154

149

184

131

134

293

393

173

Nati

onal

secu

rity

250

536

180

548

254

159

213

198

238

Reg

ula

tion

157

230

145

196

112

155

292

281

174

Fin

an

cial

regu

lati

on33

70

13

53

17

36

102

61

33

Sov

erei

gn

deb

tamp

curr

ency

cris

es14

06

23

05

04

03

04

39

16

En

titl

emen

tp

rogra

ms

73

126

115

187

88

82

153

247

124

Tra

de

pol

icy

38

40

63

26

17

20

14

21

38

Su

mof

pol

icy

cate

gor

ies

1425

2107

1295

2151

1152

1200

1863

2222

1506

Rati

oof

EP

Uto

over

all

EU

05

004

104

703

904

504

503

605

104

7

Not

es

Qu

erie

sru

nF

ebru

ary

12

2015

onU

S

new

spap

ers

inA

cces

sW

orld

New

sN

ewsb

an

k

usi

ng

the

cate

gor

y-s

pec

ific

pol

icy

term

sets

list

edin

On

lin

eA

pp

end

ixB

E

xce

pt

for

the

last

row

all

entr

ies

are

exp

ress

edre

lati

ve

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

lsquolsquoOver

all

econ

omic

un

cert

ain

tyrsquorsquo

qu

an

tifi

esth

efr

equ

ency

ofart

icle

sth

at

mee

tou

rlsquolsquoe

con

omyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

requ

irem

ents

(ie

d

rop

pin

gth

elsquolsquop

olic

yrsquorsquo

requ

irem

ent)

an

dis

als

oex

pre

ssed

rela

tive

toth

eaver

age

EP

Ufr

equ

ency

from

1985

to2014

Th

eca

tegor

y-

spec

ific

ind

exvalu

essu

mto

mor

eth

an

100

for

two

reaso

ns

firs

tw

eu

sea

few

pol

icy

term

sin

mor

eth

an

one

pol

icy

cate

gor

y

For

exam

ple

lsquolsquoM

edic

aid

rsquorsquoap

pea

rsin

the

term

sets

for

bot

hh

ealt

hca

rean

den

titl

emen

tp

rogra

ms

Sec

ond

a

new

spap

erart

icle

that

mee

tsth

elsquolsquoe

con

omyrsquorsquo

lsquolsquopol

icyrsquorsquo

an

dlsquolsquou

nce

rtain

tyrsquorsquo

crit

eria

can

refe

rto

mor

eth

an

one

pol

icy

cate

gor

y

ECONOMIC POLICY UNCERTAINTY 1603

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 12: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

departuresmdashfor example war and national security threats pro-duce declines in partisan conflict but increases in policy uncer-tainty Shoag and Veuger (2015) develop policy uncertaintyindexes for US states based on newspapers and other local indi-cators finding a strong negative link to state-level economic per-formance Fernandez-Villaverde et al (2015) estimate stochasticvolatility processes for US capital taxes labor taxes and govern-ment expenditures in a DSGE model finding correlations with ourEPU index of 044 031 and 067 respectively Jurado Ludvigsonand Ng (2015) derive uncertainty measures from common variationin the unforecastable components of macroeconomic indicatorswith their main measure correlating at 042 with our EPU index

IIC EPU Indexes for Other Countries

We also construct EPU indexes for 11 other major econo-mies11 As with our US index we first obtain a monthly countof articles that contain a trio of terms about the economy (E) policy(P) and uncertainty (U) We then scale the raw counts standard-ize each newspaperrsquos variation average across papers in a countryby month and normalize12 To help develop suitable E P and Uterm sets we consulted persons with native-level fluency and eco-nomics expertise in the relevant language and country Our P termset differs across countries for reasons both obvious (eg usinglsquolsquoBOJrsquorsquo for Japan) and idiosyncratic (eg inclusion of lsquolsquocustomsdutiesrsquorsquo for India) Online Appendix A lists the term sets and news-papers for each country-level EPU index We perform all searchesin the native language of the newspaper drawing on archives forseven newspapers in India six each in Canada and South Koreatwo each in France Germany Italy Japan Spain and the UnitedKingdom and one each in China and Russia13

Figure III displays the EPU index for Russia and OnlineAppendix Figures A1ndashA10 display the other country-level

11 We have recently developed additional EPU indexes for Australia andBrazil and assisted other researchers in developing EPU indexes for Holland andIreland We are open to proposals to developing indexes for other countries

12 For certain papers outside the United States search platform limitationspreclude us from scaling by the count of all articles In these cases we instead scaleby the count of articles containing the common and neutral term lsquolsquotodayrsquorsquo

13 Censorship and state control of the media present special challenges for Russiaand China For China we use the South China Morning Post the leading English-language newspaper in Hong Kong For Russia we rely on Kommersant which focuseson financial matters and is reportedly fairly free of government pressures

QUARTERLY JOURNAL OF ECONOMICS1604

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 13: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

indexes14 The Russian index responds to Russian military con-flicts major political developments in Ukraine the Russian finan-cial crisis in 1998 the Lehman Brothers failure in 2008 the 2013lsquolsquotaper tantrumrsquorsquo triggered by a perceived shift in US monetarypolicy and other developments While the Russian index is noisyreflecting our reliance on a single paper it shows that our approachyields useful information even for countries with strong restric-tions on press freedoms Looking at EPU indexes across 12 coun-tries we see that a wide variety of global and domestic factors drivemovements in our newspaper-based measures of policyuncertainty

IID Long-Span EPU Indexes for the United States and UnitedKingdom

We also construct long-span monthly EPU indexes back to 1900for the United States (drawing on digital archives for the Wall StreetJournal New York Times Los Angeles Times Boston Globe Chicago

FIGURE III

Index of EPU for Russia

14 We provide regular monthly updates of the country-level EPU indexes athttpwwwpolicyuncertaintycom

ECONOMIC POLICY UNCERTAINTY 1605

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 14: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

Tribune and Washington Post) and the United Kingdom (Times ofLondon and the Guardian) Based on informal audits and our reviewof word usage patterns in newspapers and other text sources weexpanded the E term set for the historical indexes to include lsquolsquobusi-nessrsquorsquo lsquolsquoindustryrsquorsquo lsquolsquocommercersquorsquo and lsquolsquocommercialrsquorsquo The expanded andnarrower E term sets yield very similar results in recent decades butthe expanded set seems to perform better in the early decades of thetwentieth century Based on results of the audit analysis describedlater we also expanded the P term set for the historical indexes toinclude lsquolsquotariff rsquorsquo and lsquolsquowarrsquorsquo

Figure IV and Online Appendix Figure A1 display the histor-ical EPU indexes for the United States and United KingdomIndexes for these two countries exhibit similarities and notabledifferences For example the elevation of EPU levels in the1930s is dramatic in the United States but modest in the UnitedKingdom which experienced a less severe output fall during theGreat Depression World Wars I and II are more prominent in theUnited Kingdom EPU series Gulf Wars I and II are associatedwith sharp EPU spikes in both countries The mid-1970s stands

FIGURE IV

US Historical Index of EPU

QUARTERLY JOURNAL OF ECONOMICS1606

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 15: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

out as a period of unusually high EPU in the United Kingdom(which suffered severe economic turmoil over this period includingthe IMF bailout and resignation of Prime Minister Harold Wilson)but not in the United States The post-1960s upward drift of EPUevident for the United States is absent for the United KingdomThis long-term US-UK comparison reinforces our earlier infer-ence that a broad mix of domestic and international developmentsinfluences the extent of policy uncertainty in any given country

III Evaluating Our Policy Uncertainty Measures

As remarked in Section I using newspaper-based measuresof EPU raises several issues about accuracy and potential biasThis section explains how we sought to address those issues Westart with a discussion of our audit study which relies on humanreadings of newspaper articles We use the audit study to selectour P term set compare the time-series behavior of human andcomputer-generated EPU indexes and collect other informationabout the nature of policy uncertainty Next we consider the roleof political slant in our EPU index Last we compare our news-paper-based index to other measures of uncertainty stock marketvolatility the frequency of uncertainty and policy uncertaintydiscussions in the Beige Books the share of the lsquolsquoRisk Factorsrsquorsquosection in firmsrsquo 10-K filings devoted to government policies andregulations and the frequency of large daily stock market movestriggered by news about government policy

IIIA Audit Study Based on Human Readings

We spent six months developing an audit process designed toevaluate and refine our US EPU indexes and another 18 monthsrunning a large-scale human audit study During the latterphase student teams working under our close supervision readand coded articles drawn from eight newspapers from 1900 to201215 We now describe the audit process and results

1 Audit Process We began by reading a few hundred news-paper articles typically in batches of 50 and comparing notes to

15 To construct our EPU index it suffices to recover counts of articles thatcontain certain terms In contrast we need full-text articles (machine-readablefiles or images) to carry out the audit study We could not access full-text articlesfor the Boston Globe or USA Today but we did so for the other eight newspapers

ECONOMIC POLICY UNCERTAINTY 1607

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 16: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

develop classification criteria an audit template in the form of anExcel file and the first draft of a guidebook for auditors Early onwe concluded that the largest payoff to an audit study involvedselecting and evaluating the lsquolsquopolicyrsquorsquo or P term set Accordinglythe formal audit study described below samples from the universeof articles that meet our lsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteriawhich concentrates our (expensive) human resources on samplesthat are highly germane for our purposes16

Next we conducted a pilot audit Working with a team ofstudent research assistants we read and coded 2000 randomlyselected newspaper articles To identify coding difficulties andweaknesses in our training materials we held weekly review ses-sions with the auditors and assigned about 20 of articles tomultiple auditors We used the pilot study to develop a trainingprocess and refine our audit guide The resulting 65-page guideserves as a training tool and reference manual in our full-scaleaudit It explains how to assess whether an article meets ourcriteria for economic uncertainty and economic policy uncertaintyand how to code each field in the audit template17 The pilot studyalso led to improvements in the audit process For example toensure that auditor-learning effects are not confounded with dif-ferences across papers or over time the full-scale audit studypresents articles to auditors in a randomized order

To conduct the full-scale audit we recruited and trained newteams of research assistants Each new auditor underwent atraining process that included a review of the audit guide andtemplate trial codings of at least 100 articles (not included inthe audit sample) a one-on-one meeting to review the trial cod-ings and additional trial codings and feedback when needed Wemet with the audit teams on a weekly basis to address questionsreview lsquolsquohard callsrsquorsquo and coding differences and maintain esprit de

16 Only 05 of the articles in our 10 leading newspapers satisfy both thelsquolsquoeconomyrsquorsquo and lsquolsquouncertaintyrsquorsquo criteria Thus the vast majority of all articles readby our auditors would be useless for selecting and evaluating our P term set if wewere to sample randomly from all newspaper articles

17 The guide includes coding instructions numerous examples and FAQs Forexample one of the FAQs asks lsquolsquoAre remarks about uncertain tax revenues groundsfor EPU=1rsquorsquo and answers lsquolsquoYes if the article attributes uncertainty about tax rev-enues partly or entirely to uncertainty about policy choices No if the articleattributes uncertainty about tax revenues entirely to uncertainty about economicconditions rsquorsquo The audit guide is available at httpwwwpolicyuncertaintycomAudit_Guidepptx

QUARTERLY JOURNAL OF ECONOMICS1608

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 17: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

corps The auditors reviewed 12009 articles from 1900 to 2012that we selected using a two-stage approach18 First we specifieda target sample size (higher in 1985ndash2011 and certain key earlieryears) and then we randomly sampled a number of articles foreach newspaper and month To monitor audit quality and shar-pen incentives for careful work we randomly assigned about onequarter of the articles to multiple auditors

2 Selecting a P Term Set When an auditor codes an article asEPU = 1 he or she also records the policy terms contained in thepassages about EPU Using these records we identified 15 termsthat appear often in newspaper discussions of EPU from 1985 to2012 lsquolsquoregulationrsquorsquo lsquolsquobudgetrsquorsquo lsquolsquospendingrsquorsquo lsquolsquopolicyrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquotaxrsquorsquolsquolsquofederal reserversquorsquo lsquolsquowarrsquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoHouse ofRepresentativesrsquorsquo lsquolsquogovernmentrsquorsquo lsquolsquoCongressrsquorsquo lsquolsquoSenatersquorsquo lsquolsquopresi-dentrsquorsquo and lsquolsquolegislationrsquorsquo (and variants like lsquolsquoregulatoryrsquorsquo lsquolsquotaxationrsquorsquoetc) We then considered the approximately 32000 term set per-mutations with four or more of these policy terms For each per-mutation we generated computer assignments of EPUC = 0 or 1for each article in the sample By comparing these computer as-signments to the human codings we obtain sets of false negatives(EPUC = 0 EPUH = 1) and false positives (EPUC = 1 EPUH = 0) foreach permutation We chose the P term set that minimizes thegross error ratemdashthat is the sum of false positive and false nega-tive error rates This process yields our baseline policy term set forthe EPU index in Figure I lsquolsquoregulationrsquorsquo lsquolsquodeficitrsquorsquo lsquolsquoFederalReserversquorsquo lsquolsquoWhite Housersquorsquo lsquolsquoCongressrsquorsquo and lsquolsquolegislationrsquorsquo

Online Appendix Figures B1 to B6 display alternative EPUindexes constructed by dropping the six baseline terms one at atime Inspecting these figures it is apparent that the time-seriesbehavior of our EPU index is not particularly sensitive to anysingle policy term We also experimented with compound textfilters for example adding government AND tax to the baselineterm set Somewhat to our surprise we were unable to developsimple compound text filters that achieved a materially lowergross error rate than our baseline term set19

18 We reviewed more than 15000 articles across the preaudit phase pilotaudit auditor training exercises and full-scale audit but we draw only on the12009 articles in the full-scale audit for our analysis here

19 Our consideration of compound text filters focused on terms that materiallylowered the false negative rate when added to the baseline term setmdashat the cost of

ECONOMIC POLICY UNCERTAINTY 1609

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 18: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

We repeated this process to obtain the P term set for thehistorical EPU index in Figure IV which makes use of all sixterms in the P set for the modern index plus lsquolsquotariffrsquorsquo and lsquolsquowarrsquorsquoAdding these two policy terms accords well with the prominentrole of tariffs and tariff revenues in the first half of the twentiethcentury and with US participation in World Wars I and II theKorean War and the Vietnam War all of which involved muchgreater per capita rates of US military deployments and casual-ties than more recent military conflicts

3 Time-Series Comparison We chose the P term set for ourcomputer-automated EPU index to minimize the gross error raterelative to the human benchmark provided by our audit study Toassess the time-series performance implied by our automatedclassifications we now compare movements over time in humanand computer-generated EPU indexes To do so we compute thefraction of audit sample articles with EPUH = 1 in each quarterfrom 1985 to 2012 multiply by the EU rate for our 10 newspapersand normalize the resulting human EPU index to 100 over theperiod To obtain the corresponding computer EPU index we in-stead use the fraction of audit-sample articles with EPUC = 1Figure V compares these human and computer EPU indexesThere are differences between the two seriesmdashfor example alarger spike for the summer 2011 debt ceiling dispute in thehuman EPU indexmdashbut they are quite similar with a correlationof 086 Repeating the same type of comparison using annual datafrom 1900 to 2010 in Online Appendix Figure C1 we find a cor-relation of 093 between the human and computer EPU indexes

Figures V and C1 provide some assurance that our computer-automated EPU classifications track the actual time-series vari-ation in the intensity of concerns about EPU as judged by intel-ligent humans In this regard itrsquos worth stressing that our term-set selection criterion makes no use of time-series variation SoFigures V and C1 offer something of an independent check on the

even greater increases in the false positive rate Otherwise the term in questionwould be part of the baseline set lsquolsquoTaxrsquorsquo is the leading example in this regard As anexample of how adding lsquolsquotaxrsquorsquo to the policy term set yields a false positive see lsquolsquoCreditMarkets Little Change in Treasury Pricesrsquorsquo by Kenneth N Gilpin New York TimesFebruary 14 1991 The article discusses economic uncertainty and includes re-marks about taxable and tax-exempt securities but it contains no discussion ofpolicy matters

QUARTERLY JOURNAL OF ECONOMICS1610

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 19: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

performance of our automated classification criteria Howeveritrsquos important to understand the limitations of these comparisonsThey incorporate our computer-automated EU assignments andmore fundamentally they rely on the content of newspaper arti-cles We use other methods as discussed later to assess the reli-ability of newspaper content for the purposes of constructing anEPU index

For downstream econometric applications we also care aboutthe time-series properties of the net error rate given by the dif-ference between the computer and human EPU index valuesCalculating this net error rate from the series in Figure V wefind that it is essentially uncorrelated with quarterly real GDPgrowth rates (correlation of002) and with the lsquolsquotruersquorsquo (iehuman) EPU rate in the audit sample (correlation of 0004)

4 Other Audit Results Our audit study also speaks to severalother questions related to our EPU index First only 5 of audit-sample articles with EPUH = 1 mainly discuss actual or prospec-tive declines in policy uncertainty Apparently reporters and

FIGURE V

Human and Computer EPU Indexes

ECONOMIC POLICY UNCERTAINTY 1611

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 20: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

editors do not regard falling uncertainty as particularly newswor-thy Second 10 of EPUH = 1 articles discuss uncertainty aboutwho will make future economic policy decisions 68 discuss un-certainty about what economic policies will be undertaken (orwhen) and 47 discuss uncertainty about the economic effectsof past present or future policy actions Third the share of EPUH

= 1 articles that discuss who will make future economic policydecisions triples in presidential election years compared withother years indicating that the nature of policy uncertaintyshifts substantially over the election cycle20 Fourth 32 ofEPUH = 1 articles mention policy matters in other countriesoften alongside domestic policy concerns

IIIB Political Slant in Newspaper Coverage of EPU

Our audit study does not address the potential for politicalslant to skew newspaper coverage of EPU If right-leaning (left-leaning) newspapers seriously overplay EPU when Democrats(Republicans) are in power political slant could distort measuredchanges in our index To investigate this issue we split our 10newspapers into the 5 most Republican and 5 most Democraticpapers using the media slant index of Gentzkow and Shapiro(2010) They assign slant values based on how frequently news-papers use words preferred by one party or the other in congres-sional speech For example a newspaper that frequently useslsquolsquodeath taxrsquorsquo lsquolsquopersonal accountsrsquorsquo and lsquolsquowar on terrorrsquorsquo (terms pre-ferred by Republicans) falls on the right side of their slant indexand a newspaper that frequently uses lsquolsquoestate taxrsquorsquo lsquolsquoprivate ac-countsrsquorsquo and lsquolsquowar in Iraqrsquorsquo (terms preferred by Democrats) falls onthe left side Online Appendix Figure C3 plots the lsquolsquoleftrsquorsquo andlsquolsquorightrsquorsquo versions of our EPU index They move together closely

20 We also find electoral cycle effects on the level of policy uncertainty in amulticountry setting In particular we merge our country-level EPU indexeswith data on the timing and closeness of democratic national elections from Julioand Yook (2012 2016) updating their data to cover recent elections This effortyields an unbalanced panel with 12 countries 62 national elections (none forChina) and 3263 monthly observations Using country fixed effects and an electiontiming indicator as explanatory variables EPU is on average 16 log points higherduring the month of national elections (t-statistic of 53 clustering errors at thecountry level) Including ln(1 + jpercentage voting gap between first- and second-place finishersj) as an additional regressor we find statistically significant evidencethat close elections yield a further elevation of policy uncertaintymdashbut the close-ness effect is small

QUARTERLY JOURNAL OF ECONOMICS1612

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 21: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

with a correlation of 092 This finding suggests that politicalslant does not seriously distort variation over time in newspapercoverage of EPU and is not a major concern for our index

IIIC Comparisons to Other Measures of Uncertainty and PolicyUncertainty

Another way to evaluate our EPU index is by comparisonwith other measures of uncertainty and policy uncertainty Themost obvious comparator is the VIX an index of 30-day option-implied volatility in the SampP500 index available since 1990 Asseen in Figure VI the VIX and the EPU index often move to-gether (correlation of 058) but they also show distinct variationFor example the VIX reacts more strongly to the Asian financialcrisis the WorldCom fraud and the Lehman Brothers collapsemdashevents with strong financial and stock market connections Incontrast the EPU index shows stronger responses to war in theGulf region the election of a new president and political battlesover taxes and government spendingmdashevents that clearly involvemajor policy concerns but also affect stock market volatility

FIGURE VI

US EPU Compared to 30-Day VIX

ECONOMIC POLICY UNCERTAINTY 1613

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 22: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

Of course the two measures differ conceptually in severalrespects While the VIX reflects implied volatility over a 30-daylook-ahead period our EPU index involves no explicit horizonThe VIX pertains to uncertainty about equity returns while theEPU index reflects policy uncertainty and not just for equity re-turns The VIX covers publicly traded firms only which accountfor about one third of private employment (Davis et al 2007) Tothrow some light on the role of these differences we create anewspaper-based index of equity market uncertaintySpecifically we retain our E and U term sets but replace the Pterm set with lsquolsquostock pricersquorsquo lsquolsquoequity pricersquorsquo or lsquolsquostock marketrsquorsquo Theresulting index shown in Online Appendix Figure C2 correlateswith the VIX at 073 considerably higher than the EPU-VIXcorrelation21

This result tells us two things First it demonstrates that wecan construct a reasonable proxy for an important type of eco-nomic uncertainty using frequency counts of newspaper arti-clesmdasha proof-of-concept for our basic approach Second thestronger correlation of the newspaper-based equity index withthe VIX confirms that differences in topical scope between theVIX and the EPU index are an important source of distinct var-iation in the two measures

1 Other Text Sources We also consider uncertainty indicatorsbased on the Beige Book releases before each regularly scheduledmeeting of the Federal Open Market Committee (FOMC) TheBeige Book published eight times a year summarizes in roughly15000 words the views and concerns expressed by business andother contacts to the 12 regional Federal Reserve Banks Wecount the frequency of lsquolsquouncertainrsquorsquo in each Beige Book normal-ized to account for variation in word count22 We also read eachpassage that contains lsquolsquouncertainrsquorsquo to judge whether it pertains topolicy matters and if so we record the policy category

21 We make no effort here to develop an optimal term set for the news index ofequity market uncertainty something we are currently pursuing in other workInstead Online Appendix Figure C2 reflects our first attempt and can surely beimproved

22 That is we divide the raw frequency count by the number of words in theBeige Book and rescale to preserve the average frequency count per Beige Book overthe sample period

QUARTERLY JOURNAL OF ECONOMICS1614

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 23: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

Figure VII shows the resulting quarterly frequency countsper Beige Book (BB) It highlights many of the same shocks andpolicy developments as the EPU index in Figure I The quarterlytime-series correlation between the EPU index and the BB policyuncertainty indicator is 054 The BB policy uncertainty indicatorshows little immediate response to the financial crisis but beginsto rise in the second half of 2009 and is at highly elevatedlevels from 2010 to 2013 In a categorical breakdown analogousto Table I (not shown) the BBs also point to fiscal policy as themost important source by far of elevated policy uncertainty inrecent years Financial regulation and sovereign debt concernsfigure more prominently in the BBs than in newspapers In con-trast to newspapers (but rather unsurprisingly) the BBs almostnever mention monetary policy uncertainty

Figure VII also shows a policy uncertainty indicator based ontextual analysis of 10-K filings For each 10-K filing we countsentences in the Risk Factors section (mandatory since fiscalyear 2005) that contain one or more of the policy terms listed inOnline Appendix E We then divide by the total number of sen-tences in the Risk Factors section and average over firms by year

FIGURE VII

Policy Uncertainty Measures Based on Textual Analysis of the Fedrsquos BeigeBooks and Part 1A (Risk Factors) of Firmsrsquo 10-K Filings

ECONOMIC POLICY UNCERTAINTY 1615

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 24: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

to obtain the series in Figure VII23 Although the temporalcoarseness of the 10-K filings precludes fine-grained compari-sons our analysis reveals a strong upward drift after 2009 inthe degree to which firms express concerns about their exposureto policy-related risk factors24

2 Daily Stock Market Jumps Finally following BakerBloom and Davis (2015) we characterize all large daily moves(greater than j25j) in the SampP stock index from 1900 to 2012 Ineach instance we locate and read the next-day New York Timesand Wall Street Journal articles that cover the stock move Werecord the explanation(s) according to the article and classify itas policy-related or not The idea is that higher policy uncertaintyleads to a greater frequency of large equity market moves trig-gered by policy-related news As seen in Online Appendix FigureC6 we find precisely that The correlation of the annual fre-quency count of daily stock market jumps triggered by policynews and the annual version of the EPU index in Figure IV is078 The 1930s and the period during and after the GreatRecession stand out in both series

IIID Summary

In summary our audit study and comparison to other textsources and types of data indicate that our newspaper-based EPUindexes contain useful information about the extent and nature ofeconomic policy uncertainty Compared to other policy uncer-tainty measures newspaper-based indexes offer distinct advan-tages They can be extended to many countries and backward intime sometimes by a century or more For large countries like theUnited States it is feasible to construct useful newspaper-based

23 The average length of the Risk Factors section of 10-K filings has grownsteadily over time perhaps because firms are providing increasingly detailed dis-cussions in this regard For this reason we prefer to scale by the total number ofsentences so as not to overstate the rising importance of policy-related risk factors

24 Online Appendix Figure C5 reports another 10-K policy uncertainty indi-cator based on the fact that firms generally discuss risk factors in order of theirimportance to the firm Thus for each 10-K filing we calculate the percent of theRisk Factors section one must read before encountering a discussion of policy-re-lated risks Averaging across firms by year the mean value of this measure fallsfrom 252 for fiscal year 2005 to 170 for 2013 and the median falls from 152 to87 In other words the average firm perceives policy risks as increasingly impor-tant from 2005 to 2013 relative to other risks

QUARTERLY JOURNAL OF ECONOMICS1616

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 25: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

indexes at a daily frequency and by region Newspaper-based in-dexes are readily disaggregated and parsed to develop category-specific indexes

IV Policy Uncertainty and Economic Activity

To investigate whether policy uncertainty matters for eco-nomic outcomes we take two complementary approaches Thefirst uses firm-level data yielding better causal identificationbut capturing only a limited set of impact channelsmdashgovernmentpurchases of goods and services and certain aspects of regulatorypolicy The second uses macro data in VAR analyses potentiallycapturing many channels but offering little assurance about theidentification of causal effects

IVA Firm-Level Outcomes and Policy Uncertainty

Our firm-level analysis considers option-implied stock pricevolatility as a proxy for firm-level uncertainty and investmentrates and employment growth as real activity measures Weuse US panel data on publicly listed firms and an identificationstrategy that differentiates firms by exposure to uncertaintyabout government purchases of goods and services To measurethis exposure we draw on two sources of information For firms inHealth Services (SIC 80) we use the government share of UShealth care expenditures in 2010 which we calculate as 438 inOnline Appendix F For all other industries we exploit micro datain the Federal Registry of Contracts from 2000 to 2013 as follows

As a first step we match the federal contracts database toCompustat firms using DUNS numbers and the names of theparent firm and their US subsidiaries25 This match yields theparent firmrsquos revenue derived from federal contracts which weallocate to three-digit SIC industries using industry codes andline-of-business data in Compustat We then aggregate revenuesand contract awards to obtain the ratio of federal purchases torevenues in each three-digit industry by year To smooth outhigh-frequency variation from lumpy contract awards we

25 We do so using Dunn amp Bradstreetrsquos US database of all public and privatefirms which includes a firm name DUNS number industry and ownership infor-mation In this way we capture federal contracts of the publicly listed parent firm(eg lsquolsquoGeneral Electricrsquorsquo) and contracts with subsidiaries of the parent firm (eglsquolsquoGeneral Electric Capital Servicesrsquorsquo and lsquolsquoUSA Instrumentsrsquorsquo)

ECONOMIC POLICY UNCERTAINTY 1617

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 26: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

average these ratios from 2000 to 2013 to obtain our exposuremeasure for each three-digit SIC At the top end firms operatingin the guided missiles and space vehicles and parts industry (SIC376) derive 78 of their revenues from sales to the federal gov-ernment The corresponding figure for selected other industrieswith high exposures to federal purchases is 39 for ordnance andaccessories (SIC 348) 27 for search detection navigation guid-ance amp aeronautical systems (SIC 381) 21 for engineering ser-vices (SIC 871) 20 for aircrafts and parts (SIC 372) 15 forship and boat building and repairing (SIC 373) 11 for blankbooks loose leaf binders and bookbinding (SIC 278) and 9 forheavy construction (SIC 160) Direct sales to the federal govern-ment are comparatively small in most other industries

In a second step we measure each firmrsquos exposure to govern-ment purchases as its revenue-weighted mean (across its lines ofbusiness) of the industry-level exposure measures calculated inthe first step If the firm operates in a single three-digit SIC thenits exposure measure equals the corresponding industry exposuremeasure We prefer this two-step approach because it may lessenthe scope for reverse causality and because industry-levelmeasures may better proxy for the firmrsquos ex ante exposure touncertainty about government purchases Our robustness inves-tigations below consider several other firm-level policy exposuremeasures

IVB Implied Stock Price Volatility

Table II displays results from regressing firmsrsquo 30-day im-plied stock price volatility on economic policy uncertainty Weobtain the implied volatility measure from Options Metricswhich calculates the 30-day volatility implied by firm-levelequity options These options have been traded since the mid-1990s on the Chicago Board of Options and Exchange (CBOE2014) and our data begin in 1996 We use this volatility measurein quarterly regressions to match the quarterly company ac-counts averaging implied volatility over all trading days in thequarter We run regressions on a sample that extends from 1996to 2012 and weight by firm sales giving more weight to the largerfirms that also tend to have more actively traded equity options

Column (1) reports a very basic specification that regresseslogged 30-day implied volatility on our EPU index and the ratio offederal government purchases to GDP a control for the first

QUARTERLY JOURNAL OF ECONOMICS1618

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 27: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

TA

BL

EII

OP

TIO

N-I

MP

LIE

DS

TO

CK

PR

ICE

VO

LA

TIL

ITY

AN

DP

OL

ICY

UN

CE

RT

AIN

TY

Dep

var

log(3

0-d

ay

imp

lied

vol

)(1

)(2

)(3

)(4

)(5

)(6

)(7

)

Log

(EP

U)

04

32

00

44

07

52

(00

10)

(00

13)

(00

27)

Log

(EP

U)

inte

nsi

ty02

15

02

28

05

45

00

82

(00

69)

(01

00)

(02

02)

(01

17)

Log

(VIX

)07

34

(00

16)

Log

(VIX

)

inte

nsi

ty

00

20

(01

17)

Log

(EU

)10

80

(00

27)

Log

(EU

)

inte

nsi

ty

03

01

(01

77)

Fed

eral

pu

rch

ase

sG

DP

193

0

77

5

174

0

(15

0)

(14

9)

(14

9)

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

294

5

297

0

299

3

310

8(1

27

2)

(123

6)

(126

6)

(132

4)

Nati

onal

secu

rity

EP

U

def

ense

00

48

(00

12)

Hea

lth

care

EP

U

hea

lth

00

71

(00

43)

Fin

an

cial

regu

lati

onE

PU

fin

an

ce01

44

(00

30)

Fir

man

dti

me

effe

cts

No

Yes

No

Yes

No

Yes

Yes

Not

es

Th

esa

mp

leco

nta

ins

1365

78

obse

rvati

ons

on54

60

firm

sfr

om1996

to2012

Th

ed

epen

den

tvari

able

isth

en

atu

ral

log

ofth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

Inte

nsi

tyis

the

firm

rsquosex

pos

ure

tofe

der

al

pu

rch

ase

sof

goo

ds

an

dse

rvic

esco

mp

ute

dby

the

two-

step

met

hod

des

crib

edin

Sec

tion

IV

Fed

eral

pu

rch

ase

sG

DP

isfr

omN

IPA

table

sL

og(E

U)

isth

elo

gof

the

new

spap

er-b

ase

dec

onom

icu

nce

rtain

tyin

dex

N

ati

onal

secu

rity

EP

U

def

ense

isth

en

ati

onal

secu

rity

EP

Uin

dex

from

Table

Im

ult

ipli

edby

1fo

rfi

rms

ind

efen

sein

du

stri

es(S

ICs

348

372

376

379

381

871)

an

d0

oth

erw

ise

an

dan

alo

gou

sly

for

hea

lth

care

EP

U

hea

lth

(SIC

s800

to809)

an

dfi

nan

cial

regu

lati

onE

PU

fin

an

ce(S

ICs

600ndash699)

All

regre

ssio

ns

wei

gh

ted

by

the

firm

rsquosaver

age

sale

sin

the

sam

ple

per

iod

S

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1619

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 28: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

moment of policy Log(EPU) is highly statistically significantwith the coefficient of 0432 indicating that a 1 EPU increaseis associated with a roughly 043 increase in firm-level impliedvolatility To put this magnitude in perspective our EPU indexrose by 856 log points (135) from 2006 to 2012 which implies anestimated upward shift of 37 log points (45) in average firm-level implied volatility The negative coefficient on the controlvariable in column (1) says that conditional on log(EPU) averagefirm-level implied volatility is lower when the ratio of federalpurchases to GDP is higher

Column (2) contains the key result We add a full set of firmand time fixed effects to control for unobserved factors that differacross firms and unobserved common factors that vary over timeThe log(EPU) and federal purchasesGDP terms drop out as theyare collinear with the time effects But we now interact thesemeasures with our firm-level measures of exposure to govern-ment purchases This specification tests whether implied volatil-ity at firms with greater exposure to government purchasescovaries more strongly with policy uncertainty We find verystrong evidence for this The coefficient of 0215 on thelog(EPU) intensity measure suggests that for every 1 increasein our policy uncertainty index a firm with say a 50 govern-ment revenue share would see its stock volatility rise by 01126

Column (3) evaluates to what extent our EPU measure tellsus anything different from the VIX index the most commonlyused proxy for overall economic uncertainty As noted inSection IIIC our EPU index and the VIX have a correlation co-efficient of 058 Adding the VIX in a specification without firm ortime effects reverses the sign of the EPU term while the coeffi-cient on the VIX is large (at 0734) and highly significant Thisresult is unsurprising since the VIX is the 30-day implied volatil-ity on the SampP500 index and it should be highly correlated withthe average 30-day implied volatility for publicly listed USfirms

Column (4) again adds time and firm fixed effects and wenow interact the EPU federal purchasesGDP and VIX measureswith the intensity of the firmrsquos exposure to government pur-chases Strikingly we now find that the EPU index has a large

26 Using a quite different empirical design and source of variation KellyPastor and Veronesi (2016) find evidence that policy uncertainty related to electionoutcomes also raises option-implied stock market volatility

QUARTERLY JOURNAL OF ECONOMICS1620

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 29: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

and significant coefficient while the VIX drops out entirelyCombining columns (3) and (4) reveals that the 30-day impliedvolatility is best explained by the VIX index for the average firmbut the EPU index provides additional explanatory power for theimplied volatility of firms in sectors with high government expo-suremdashlike defense health care engineering services and heavyconstruction

Columns (5) and (6) run a similar evaluation for the EUindex yielding similar results In column (5) we run a regressionwith the EPU EU and federal purchasesGDP measures but notime or firm fixed effects The EU index dominates with a largeand highly significant coefficient Again this result is not surpris-ingmdashthe EU index reflects the overall frequency of newspaperarticles about economic uncertainty without any stipulationthat these articles also discuss policy Column (6) adds time andfirm fixed effects and we again interact the key measures witheach firmrsquos exposure to government purchases As before theEPU measure dominates the general uncertainty measure inthe interacted specification with controls for firm and time effectsIndeed the EU measure now takes on the opposite sign In sum-mary while the EU index is more closely related to the averagefirm-level implied volatility in the specification (5) that excludesfirm and time effects the EPU index outperforms the EU index inexplaining firm-specific movements in option-implied volatility

Finally in column (7) we add category-specific EPU mea-sures from Section IIB for firms in the defense finance andhealth care sectors These category-specific measures potentiallycapture a broad range of impact channels including ones thatinvolve regulatory policy Reassuringly all three measuresyield positive statistically significant coefficients at the 1ndash10level For example implied volatility for defense firms respondsto the national security EPU index which jumped up in GulfWars I and II and after the 911 terrorist attacks (Figure II)Similarly implied volatility for firms in the health care sectorresponds to the health care EPU index which rose during theClinton health care reform initiative and in response to uncer-tainties surrounding the Affordable Care Act The large highlysignificant coefficient on the financial regulation EPU index isespecially noteworthy because direct federal purchases of goodsand services are minuscule in the finance sector Thus we seethis result as evidence that regulatory policy uncertainty drivesfirm-level stock price volatility

ECONOMIC POLICY UNCERTAINTY 1621

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 30: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

These results imply that policy uncertainty accounts for sig-nificant variation in the cross-sectional structure of stock pricevolatilities To see this point consider the estimated changes infirm-level volatilities associated with the change in policy uncer-tainty from 2006 to 2012 Using the results in Table II column(7) we calculate these changes as (0082) (firmrsquos exposure togovernment purchases) (change in overall log EPU) plus (coef-ficient on category-specific log EPU) (change in category-specificlog EPU) Online Appendix Table A1 implements this calculationfor firms in selected industries yielding increases of up to 238 logpoints for financial firms and 139 log points for health care firmsmainly due to the run-up in their respective category-specificEPU indexes and 33 to 46 log points for firms in the ordnanceaircraft and engineering services industries mainly due to theirstrong exposures to government purchases and the rise in overallpolicy uncertainty Comparing JulyndashAugust 2001 to SeptemberndashOctober 2001 (before and after 911) and carrying out the sametype of calculations we find stock price volatility increases of 14ndash15 log points for firms in ordnance aircraft and engineeringServices 112 log points in the finance sector 75 log points inhealth care and tiny responses for firms in most other industriesHence the implied magnitudes are sizable for firms in industrieswith large policy exposures

Table III presents a wide range of additional robustness re-sults for specifications that include firm and year fixed effectsColumns (1) and (2) consider realized volatility and 182-day im-plied volatility to look at longer and shorter uncertainty horizonsyielding very similar results Column (3) adds forecasts from theSurvey of Professional Forecasters of government purchases rel-ative to GDP (interacted with firm-level exposure) as a controland column (4) uses actual future government purchases relativeto GDP (again interacted) as a control Column (5) replaces ourpreferred firm-level exposure measure (calculated by the two-step method described above) with a one-step measure calculateddirectly from the firmrsquos own sales to the federal governmentColumn (6) uses the Belo Gala and Li (2013) measure of indus-try-level exposure to government purchases which exploits theinput-output matrix to capture direct and indirect effects of gov-ernment purchases

Columns (7) and (8) in Table III consider two entirely differ-ent approaches to measuring firm-level exposure to governmentpolicy risks In column (7) we measure exposure by the slope

QUARTERLY JOURNAL OF ECONOMICS1622

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 31: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

TA

BL

EII

I

RO

BU

ST

NE

SS

CH

EC

KS

FO

RO

PT

ION

-IM

PL

IED

ST

OC

KP

RIC

EV

OL

AT

ILIT

YA

ND

PO

LIC

YU

NC

ER

TA

INT

Y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Sp

ecifi

cati

onR

eali

zed

vol

ati

lity

182-d

ay

imp

lied

vol

ati

lity

Ad

dp

urc

hase

fore

cast

Ad

d12

qtr

sfu

ture

pu

rch

ase

sF

irm

-lev

elin

ten

sity

Bel

oet

al

(2013)

inte

nsi

tyB

eta

inte

nsi

ty10-K

risk

mea

sure

$500m

+sa

les

firm

s

Log

(EP

U)

inte

nsi

ty03

46

01

78

01

75

02

58

01

92

04

56

02

83

03

78

02

37

(00

89)

(00

73)

(00

70)

(00

86)

(00

45)

(01

01)

(01

18)

(02

17)

(00

71)

(fed

eral

pu

rch

ase

sG

DP

)

inte

nsi

ty

237

2

274

7

582

8

70

5

142

0

136

061

57

271

6

310

3(1

47

1)

(117

7)

(153

5)

(167

4)

(100

3)

(276

4)

(149

7)

(641

7)

(124

0)

(For

ecast

edfe

der

al

pu

rch

ase

sG

DP

)

inte

nsi

ty326

1

(62

7)

Fir

man

dti

me

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Obse

rvati

ons

1365

78

1365

78

1365

78

737

03

1326

28

1343

81

1333

04

1120

23

427

71

Nu

mber

offi

rms

54

60

54

60

54

60

30

70

52

19

53

74

53

28

37

17

10

56

Not

es

Th

esa

mp

lep

erio

dis

1996ndash2012

Th

ed

epen

den

tvari

able

isth

e30-d

ay

imp

lied

vol

ati

lity

for

the

firm

aver

aged

over

all

days

inth

equ

art

er

exce

pt

that

colu

mn

(1)

use

sth

ere

ali

zed

dail

yvol

ati

lity

over

the

qu

art

er

an

dco

lum

n(2

)u

ses

the

aver

age

182-d

ay

imp

lied

vol

ati

lity

S

eeth

en

otes

toT

able

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1623

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 32: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

coefficient in a regression of the firmrsquos daily stock returns on ourdaily EPU index from 1985 to 1995 which predates the sampleperiod in Table II Using this beta measure of policy risk expo-sure we again find positive and statistically significant effects ofEPU on firm-level volatility In column (8) we use the policy riskexposure measure derived from 10-K filings and plotted over timein Figure VII but now measured at the firm level (averaging overavailable years) We again find sizable effects of EPU on firm-level volatility but the coefficient on the log(EPU) interactionterm is less statistically significant partly due to a smallersample size27 and perhaps partly because this measure reflectsthe firmrsquos perceived exposure to policy risk factors from 2006 on-ward only whereas the regression sample starts in 1996 Column(9) restricts attention to firms with at least $500 million in annualsales These alternative measures and specifications all yieldhighly significant results similar to column (2) in Table II

Finally Online Appendix Table A2 returns to the baselinespecification in Table II column (2) and replaces the keylog(EPU) interaction term by log(EPUX) where X correspondsto the newspaper-based E (Economy) P (Policy) U (Uncertainty)EP EU or PU index These variants yield slope coefficients onthe key log(EPUX) intensity variable that are statistically in-distinguishable from the point estimate in Table II column (2)This highlights how it is the triple combination of the E P and Uterm sets in newspaper articles that drive our results rather thanthe frequency of the individual E P or U term sets or the precisescaling of the EPU index

IVC Investment Rates and Employment Growth

Table IV investigates the contemporaneous relationship be-tween policy uncertainty and firm-level investment rates and em-ployment growth28 We now have data from 1985 to 2012 and as

27 The sample shrinks for several reasons First the Securities and ExchangeCommission did not mandate a risk factors discussion before 2006 so we cannotobtain this measure for firms that delisted before 2006 Second some publicly listedfirms are exempt from the risk factors disclosure requirement and some may notcomply Third our web-scraping and automated text-reading methods may notcapture all relevant 10-K filings perhaps because some firms present their discus-sion of risk factors in an unusual format Fourth it is not always possible to matchdata from 10-K filings to Compustat Our match rates compare favorably to similarefforts by other researchers eg Campbell et al (2014) See Online Appendix E foradditional discussion

QUARTERLY JOURNAL OF ECONOMICS1624

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 33: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

TA

BL

EIV

PO

LIC

YU

NC

ER

TA

INT

YA

ND

FIR

M-L

EV

EL

INV

ES

TM

EN

T

EM

PL

OY

ME

NT

AN

DS

AL

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

IK

IK

IK

IK

E

mp

E

mp

E

mp

E

mp

R

ev

L

og(E

PU

)

inte

nsi

ty

00

32

00

32

00

24

00

29

02

13

02

27

02

20

02

20

01

28

(00

10)

(00

10)

(00

11)

(00

10)

(00

84)

(00

89)

(01

18)

(00

94)

(00

96)

F

eder

al

pu

rch

ase

sG

DP

in

ten

sity

82

0

80

4

121

2

88

5

107

9156

0

31

9109

9203

9

(28

6)

(28

6)

(31

8)

(28

7)

(74

1)

(80

4)

(125

6)

(78

8)

(94

3)

F

orec

ast

edF

eder

al

pu

rch

ase

sG

DP

in

ten

sity

10

1

46

5

(08

28)

(28

9)

L

og(d

efen

seE

PU

)

def

ense

firm

00

02

00

18

(00

04)

(00

17)

L

og(h

ealt

hca

reE

PU

)

hea

lth

firm

00

12

00

05

(00

02)

(00

25)

L

og(fi

n

reg

EP

U)

fin

an

cefi

rm

00

02

00

03

(00

01)

(00

05)

Per

iod

icit

yQ

uart

erly

Qu

art

erly

Qu

art

erly

Qu

art

erly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

Yea

rly

3yrs

Fed

pu

rch

ase

lead

sN

oN

oY

esN

oN

oN

oY

esN

oN

oO

bse

rvati

ons

7083

98

7083

98

4112

05

7083

98

1620

06

1620

06

1072

05

1620

06

1514

73

Nu

mber

offi

rms

216

36

216

36

135

63

216

36

171

51

171

51

115

05

171

51

157

49

Not

es

Th

esa

mp

lep

erio

dru

ns

from

1985

to2012

All

colu

mn

sin

clu

de

afu

llse

tof

firm

an

dti

me

effe

cts

IK

isth

ein

ves

tmen

tra

ted

efin

edas

Cap

Ex

t

Net

Pla

nt

Pro

per

tyan

dE

qu

ipm

ent

ethTHORN t

1

E

mp

isth

eem

plo

ym

ent

gro

wth

rate

mea

sure

das

emp

t

emp

t1

05

emp

tthorn

05

emp

t1

an

d

Rev

isth

eco

rres

pon

din

gre

ven

ue

gro

wth

rate

Fed

eral

pu

rch

ase

sG

DP

in

ten

sity

isth

ech

an

ge

infe

der

al

pu

rch

ase

sG

DP

from

NIP

Ata

ble

sin

the

nex

tqu

art

erin

qu

art

erly

spec

ifica

tion

san

din

the

nex

tyea

rin

an

nu

al

spec

ifica

tion

sm

ult

ipli

edby

the

firm

-lev

elp

olic

yex

pos

ure

inte

nsi

tyvari

able

F

orec

ast

edfe

der

al

pu

rch

ase

sG

DP

in

ten

sity

inst

ead

use

sth

em

ean

fore

cast

edch

an

ge

in(fe

der

al

pu

rch

ase

sG

DP

)fr

omth

eF

eder

al

Res

erve

Ban

kof

Ph

ilad

elp

hia

rsquosS

urv

eyof

Pro

fess

ion

al

For

ecast

ers

dra

win

gon

NIP

Ad

ata

for

the

curr

ent

valu

esan

dfo

reca

std

ata

for

the

futu

revalu

es

See

the

not

esto

Table

IIfo

rad

dit

ion

al

vari

able

defi

nit

ion

sS

tan

dard

erro

rsbase

don

clu

ster

ing

at

the

firm

level

plt

00

1plt

00

5plt

01

ECONOMIC POLICY UNCERTAINTY 1625

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 34: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

before weight by firm sales We use our preferred measure of thefirmrsquos policy exposure intensity and a full set of time and firmeffects in all Table IV specifications Column (1) reports a regres-sion of the firm-level quarterly investment rate on ethlogethEPUTHORNTHORN Intensity and ethfederal purchases

GDP THORN Intensity The former has a sig-nificant negative coefficient of 0032 and the latter has a sig-nificant positive coefficient These results are in line withstandard predictions of investment-under-uncertainty modelsfor example Bernanke (1983) Dixit and Pindyck (1994) andBloom Bond and Van Reenen (2007)

To assess the magnitude of the estimated policy uncertaintyrelationship recall that the EPU index rose 856 log points from2006 to 2012 For a firm that sells 25 of its output to the federalgovernment this EPU change and the coefficient on log(EPU)Intensity in column (1) imply a one-time investment rate drop of068 percentage point (= 0856 0032 025 100) which isabout one sixth of the median firm-level investment rate of42 Although this calculation rests on a large EPU swingthere were several other large EPU moves during the sampleperiodmdashfor example an 82-point fall from 1992 to 1999 a 72-point rise from 1999 to 2001 and a 79-point fall from 2001 to2006 Hence for firms with high exposures to government pur-chases the estimates imply that swings in policy uncertainty in-volve material changes in investment rates

In column (2) we control for ethForecasted Federal PurchasesGDP THORN

Intensity given the forward-looking nature of investment deci-sions and obtain very similar results on the main coefficient ofinterest Adding controls for cash flow and Tobinrsquos q in column (2)yields a coefficient of 030 (010) on ethlogethEPUTHORNTHORN Intensity again

28 We focus on simple linear specifications that do not allow for rich responsedynamics or interactions between uncertainty and the responsiveness of outcomevariables to first-moment driving forces More sophisticated treatments of invest-ment behavior in these respects using other measures of uncertainty include Abeland Eberly (1996) Guiso and Parigi (1999) and Bloom Bond and Van Reenen(2007) There is value in applying these more sophisticated treatments to ourpolicy uncertainty measures but we leave that task to future research For aricher treatment of dynamics in firm-level investment rate responses to our EPUmeasure see Gulen and Ion (2016)

29 Using Compustat data our cash flow measure is operating income beforedepreciation expressed as a ratio to the book value of plant property and equip-ment The numerator of our Tobinrsquos q measure is the market value of equity(common and preferred shares) plus the book value of debt less the value of

QUARTERLY JOURNAL OF ECONOMICS1626

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 35: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

very similar to column (1)29 In column (3) we include the averageethForecasted Federal Purchases

GDP THORN Intensity value in the next 12 quartersas an alternative control for future expectations and again find asignificant negative coefficient In column (4) we add the cate-gory-specific measures and find statistically significant negativeresults for terms involving log changes in the health care EPUindex and the financial regulation EPU index That is the fre-quency of newspaper articles about these types of policy uncer-tainty has additional explanatory power for the investment ratesof firms that operate in sectors most affected by these types ofpolicy

Columns (5) to (8) regress annual firm-level employmentgrowth rates on EPU changes (Compustat lacks quarterly em-ployment data) As with investment rates we find sizable andstatistically significant negative coefficients on policy uncertaintychanges for employment growth rates at firms with high exposureto government policy Consider again an 856 log point increase inthe EPU index and a firm that sells 25 of its output to the fed-eral government Given these values the coefficient of0213 onethlogethEPUTHORNTHORN Intensity in column (5) implies a one-time drop inthe annual employment growth rate of 46 percentage pointswhich is large relative to the mean annual growth rate of 34for firms in the sample The category-specific EPU variables donot have statistically significant effects on employment growthin contrast to the investment results

In column (9) we consider the impact on sales as a placebotest While the real options literature highlights how uncertaintysuppresses demand for input factors with adjustment costs theshort-run impact on output should be smaller according to thisclass of theories Consistent with this prediction the estimatedeffect of ethlogethEPUTHORNTHORN Intensity in column (9) is negative but notstatistically significant while the government purchases variableremains positive and significant Hence our results suggest thatincreases in policy uncertainty are associated with contempora-neous drops in investment rates and employment growth ratesfor firms in policy-exposed sectors but the near-term associationwith their output growth rates is more muted

Finally consider the relationship of policy uncertaintychanges to the cross-sectional structure of investment rates and

inventories and deferred tax credits and the denominator is the book value of plantproperty and equipment

ECONOMIC POLICY UNCERTAINTY 1627

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 36: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

employment growth To do so we return to Online AppendixTable A1 and carry out calculations that parallel the earlierones for stock price volatility Working again with the policy un-certainty changes from 2006 to 2012 the implied quarterly in-vestment rate changes are modest except for a 29 drop for firmsin the health care sector while the annual employment changesare large in several sectors Given the change-on-change natureof the underlying regression specifications these results are one-time changes associated with the total change in the policy un-certainty measures from 2006 to 2012

IVD Policy Uncertainty and Aggregate Economic Activity

We now turn to VAR models that exploit time-series varia-tion at the country level Drawing causal inferences from VARs isextremely challengingmdashin part because policy and policy uncer-tainty can respond to current and anticipated future economicconditions Despite the challenges VARs are useful for charac-terizing dynamic relationships At a minimum they let us gaugewhether policy uncertainty innovations foreshadow weaker mac-roeconomic performance conditional on standard macro andpolicy variables

We start by fitting a VAR to monthly US data from January1985 to December 2014 To recover orthogonal shocks we use aCholesky decomposition with the following ordering the EPUindex the log of the SampP500 index the federal funds rate logemployment and log industrial production Our baseline VARspecification includes three lags of all variables Figure VIII de-picts the model-implied responses of industrial production andemployment to a 90-point upward EPU innovation equal insize to the EPU change from its average value in 2005ndash2006(before the financial crisis and recession) to its average value in2011ndash2012 (a period with major fiscal policy battles and high EPUlevels) Figure VIII shows maximum estimated drops of 11 inindustrial production and 035 in employment These responsesare statistically significant and moderate in size being about onethird as large as a typical business cycle fluctuation Since aggre-gate US investment data are not available at a monthly fre-quency we also estimated an analogous VAR model onquarterly data from 1985 to 2014 using the same type ofCholesky decomposition to identify shocks As shown in Online

QUARTERLY JOURNAL OF ECONOMICS1628

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 37: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

Appendix Figure C7 gross aggregate investment exhibits a peakdecline of about 6 in response to a 90-point EPU innovation

Figure IX shows that the basic character of the impulse re-sponse functions is robust to several modifications of the specifi-cation variable set causal ordering and sample period six lagsinstead of three in the VAR a bivariate VAR (EPU and industrialproduction) a bivariate VAR with reverse ordering including theVIX (after the EPU index) including the EU index (after the EPUindex) dropping the SampP500 index including time trends andusing a sample period that runs from 1920 (when industrial pro-duction data become available) until 1984 These results are inline with the estimated effects of election uncertainty in Julio andYook (2012) and Durnev (2010) despite their distinct empiricalapproaches

A potential concern is whether and to what extent our esti-mated impulse response functions reflect bad news generallyrather than policy uncertainty shocks in particular Includingthe SampP500 stock market index in the VAR somewhat mitigatesthis concern given that stock markets are forward looking and

FIGURE VIII

Industrial Production and Employment Responses to EPU Shock VAR Fit toMonthly US Data

ECONOMIC POLICY UNCERTAINTY 1629

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 38: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

stock prices incorporate many sources of information Our base-line VAR also includes other lsquolsquofirst-momentrsquorsquo variables log em-ployment log industrial production and the fed funds rateStill the EPU index will likely embed first-moment informationnot captured by these variables To investigate this issue we alsoconsidered VARs that include the Michigan Consumer SentimentIndex30 When we place the Michigan index after the EPU indexin the causal ordering the estimated peak effect of a policy un-certainty shock on industrial production falls by about one third

FIGURE IX

US Industrial Production Response to an EPU Shock Alternative SamplesSpecifications and Identification Assumptions

30 The Michigan index reflects phone surveys of consumers and seeks to de-termine how consumers view the short-term economy the long-term economy andtheir own financial situation It takes the difference between the percent answeringpositively and the percent answering negatively for each of five questions thenaverages these differences and normalizes by the base period (December 1968)total The Michigan index has a correlation of0742 with our EPU index Wechose the Michigan index as the more commonly used consumer confidenceindex but other consumer confidence indices are highly correlated with theMichigan indexmdashfor example the Bloomberg confidence index has a correlationof 0943 with the Michigan index and the Conference Board confidence index has acorrelation of 0912 with the Michigan index

QUARTERLY JOURNAL OF ECONOMICS1630

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 39: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

(Online Appendix Figure C8) When we place the Michigan indexfirst in the causal ordering the peak effect shrinks by about halfThese results indicate that conditional on the other variables ourEPU index and the Michigan index contain overlapping informa-tion that has value for predicting future output and employmentmovements

Perhaps this result is unsurprising The Michigan index cap-tures a mix of first-moment and second-moment concerns as ex-pressed by households in survey data The relationship betweenlsquolsquoconfidencersquorsquo and uncertainty is murky and the two concepts aretightly linked at a deep level in some theoretical models for ex-ample Ilut and Schneider (2014) In any event the EPU indexhas several important advantages relative to consumer confi-dence indexes EPU indexes can be extended to many countriespushed back in time by a century or more in some countriescomputed in near real time on a daily basis and parsed inmany ways as illustrated by our category-specific EPU indexes

Figure X shows impulse response functions for a panel VARfit to monthly data from 1985 to 2014 on the 12 countries forwhich we have an EPU index The panel VAR specification par-allels the baseline specification that underlies Figure VI exceptthat we use the unemployment rate in place of log(employment)As before we rely on a Cholesky decomposition to identify shocksand display responses to an upward 90-point EPU innovationwhich is well within the range of EPU movements experiencedby the individual countries The 12-country panel VAR yields re-sults that are similar to the US results in Figure VIII In par-ticular the international panel VAR implies that a 90-point EPUinnovation foreshadows a peak drop in industrial production ofabout 1 and a rise in the unemployment rate of about 25 basispoints Online Appendix Figure C9 shows that the basic characterof the panel VAR results is robust to a variety of alternativespecifications variable sets and weighting methods Other re-searchers who use our EPU indexes in multicountry time-seriesanalyses also find that policy uncertainty shocks foreshadow de-teriorations in macroeconomic outcomesmdashexamples include theInternational Monetary Fund (2012) Colombo (2013) Klossnerand Sekkel (2014) and Nodari (2014)

Broadly speaking we see three ways to interpret this VAR-based evidence Under the first interpretation an upward EPUinnovation corresponds to an unforeseen policy uncertainty shockthat causes the worsening of macroeconomic performance

ECONOMIC POLICY UNCERTAINTY 1631

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 40: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

through real options effects cost-of-capital effects or other mech-anisms Second an upward EPU innovation captures bad newsabout the economic outlook that is not (fully) captured by theother variables in the VAR system and that bad news triggers arise in EPU that has harmful effects on the economy Under thisinterpretation EPU amplifies and propagates a causal impulsethat originates elsewhere Third EPU has no role as either animpulse or a propagation mechanism instead it simply acts as auseful summary statistic for information missing from the othervariables in our systemmdashlog(output) log(employment) or unem-ployment the policy rate log(SampP500) the VIX and consumersentiment31 This third interpretation is hard to fully reconcilewith our firm-level results which suggests that policy uncer-tainty has negative causal effects Itrsquos also worth noting that

FIGURE X

Responses to an EPU Shock in a Twelve-Country Panel VAR

31 Stock and Watson (2012) consider many more variables in much larger andricher time-series models They still find evidence that EPU innovations precededeteriorations in aggregate performance

QUARTERLY JOURNAL OF ECONOMICS1632

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 41: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

our VAR results may understate the importance of policy uncer-tainty shocks as a driving force even under the first interpreta-tion because other variables in the VAR system may respond tonews about future policy uncertainty shocks before they show upin the EPU measure

Clearly there is a need to develop a robust identificationstrategy for assessing the causal role of policy uncertainty in mac-roeconomic performance by for example exploiting close conse-quential democratic elections and exogenous sources of variationin policy uncertainty such as shifts in the outlook for conflict be-tween North and South Korea or events like the UK Brexit voteregarding participation in the European Union In additionlinear VAR systems may be overly restrictive in how theymodel EPU responses to other shocks Perhaps EPU rises in thewake of large negative shocks but responds relatively little tosmall ones Allowing for this type of asymmetry may lead to alarger role for EPU in amplifying and propagating the effects oflarge negative shocks It would also be useful to consider stochas-tic volatility models that allow EPU shocks to directly influencethe future volatility of other shocks including shocks to policyvariables We leave these tasks to future research

At a deeper level the causal role of policy uncertainty is po-tentially quite subtle Sound institutions and policy regimesfoster predictable policy responses even in the face of large neg-ative shocks In this way good institutions and policy regimeslessen the scope for policy to act as a source of uncertainty im-pulses or through uncertain policy responses to amplify andpropagate the effects of other shocks

V CONCLUSION

We develop new measures of economic policy uncertainty forthe United States and 11 other major economies We use thesenew measures to investigate the relationship of policy uncer-tainty to firm-level stock price volatility investment rates andemployment growth and to aggregate investment output andemployment Our findings are broadly consistent with theoriesthat highlight negative economic effects of uncertainty shocksThe results suggest that elevated policy uncertainty in theUnited States and Europe in recent years may have harmed mac-roeconomic performance They also point to sizable effects of

ECONOMIC POLICY UNCERTAINTY 1633

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 42: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

policy uncertainty on the cross-sectional structure of stock pricevolatilities investment rates and employment growth

From a methodological perspective we show how to tapnewspaper archives to develop and evaluate new measures of in-terest to macroeconomists financial economists economic histo-rians and other researchers In this regard itrsquos worth stressingthat newspapers are available for countries around the worldand they have circulated in similar form for decades in mostcountries and for centuries in some countries This ubiquity cou-pled with modern databases and computers offers tremendouspossibilities for drawing on newspaper archives to deepen ourunderstanding of broad economic political and historical devel-opments through systematic empirical inquiries

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qjeoxfordjournalsorg)

Kellogg School of Management

Stanford University Center for Economic and Policy

Research Stanford Institute for Economic Policy

Research and National Bureau for Economic Research

Chicago Booth School of Business and National Bureau

for Economic Research

References

Abel Andrew lsquolsquoOptimal Investment under Uncertaintyrsquorsquo American EconomicReview 73 (1983) 228ndash233

Abel Andrew and Janice Eberly lsquolsquoOptimal Investment with CostlyReversibilityrsquorsquo Review of Economic Studies 63 (1996) 581ndash593

Alexopoulos Michelle and Jon Cohen lsquolsquoThe Power of Print Uncertainty ShocksMarkets and the Economyrsquorsquo International Review of Economics and Finance40 (2015) 8ndash28

Azzimonti Marina lsquolsquoPartisan Conflict and Private Investmentrsquorsquo NBER WorkingPaper 21723 2015

Bachmann Rudiger Steffen Elstener and Eric Sims lsquolsquoUncertainty and EconomicActivity Evidence from Business Survey Datarsquorsquo American EconomicJournal Macroeconomics 5 (2013) 217ndash249

Baker Scott Nicholas Bloom Brandice Canes-Wrone Steven J Davis andJonathan Rodden lsquolsquoWhy Has US Policy Uncertainty Risen since 1960rsquorsquoAmerican Economic Review Papers amp Proceedings 104 (2014) 56ndash60

Baker Scott Bloom Nicholas and Steve J Davis lsquolsquoWhat Triggers Stock MarketJumpsrsquorsquo paper presented at the ASSA Meetings January 2015

Basu Susanto and Brent Bundick lsquolsquoUncertainty Shocks in a Model of EffectiveDemandrsquorsquo NBER Working Paper 18420 2012

QUARTERLY JOURNAL OF ECONOMICS1634

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 43: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

Belo Federico Vito D Gala and Jun Li lsquolsquoGovernment Spending Political Cyclesand the Cross Section of Stock Returnsrsquorsquo Journal of Financial Economics 107(2013) 305ndash324

Bernanke Ben S lsquolsquoIrreversibility Uncertainty and Cyclical InvestmentrsquorsquoQuarterly Journal of Economics 97 (1983) 85ndash106

Bloom Nicholas lsquolsquoThe Impact of Uncertainty Shocksrsquorsquo Econometrica 77 (2009)623ndash685

mdashmdashmdash lsquolsquoFluctuations in Uncertaintyrsquorsquo Journal of Economic Perspectives 28(2014) 153ndash176

Bloom Nicholas Stephen Bond and John van Reenen lsquolsquoUncertainty andInvestment Dynamicsrsquorsquo Review of Economic Studies 74 (2007) 391ndash415

Bloom Nicholas Max Floetotto Nir Jaimovich Itay Saporta and Stephen TerrylsquolsquoReally Uncertain Business Cyclesrsquorsquo working paper Stanford University2014

Born Benjamin and Johannes Pfeifer lsquolsquoPolicy Risk and the Business CyclersquorsquoJournal of Monetary Economics 68 (2014) 68ndash85

Boudoukh Jacob Ronen Feldman Shimon Kogan and Matthew RichardsonlsquolsquoWhich News Moves Stock Prices A Textual Analysisrsquorsquo NBER WorkingPaper 18725 2013

Brogaard Jonathan and Andrew Detzel lsquolsquoThe Asset Pricing Implications ofGovernment Economic Policy Uncertaintyrsquorsquo Management Science 61(2015) 3ndash18

Campbell John L Hsinchun Chen Dan S Dhaliwal Hsin-min Lu and LoganB Steele lsquolsquoThe Information Content of Mandatory Risk Factor Disclosures inCorporate Filingsrsquorsquo Review of Accounting Studies 19 (2014) 396ndash455

Chicago Board Options Exchange lsquolsquoVIX CBOE Volatility Indexrsquorsquo white paper2004

Colombo Valentina lsquolsquoEconomic Policy Uncertainty in the US Does It Matter forthe Euro Arearsquorsquo Economics Letters 121 (2013) 39ndash42

Davis Steven J John Haltiwanger Ron Jarmin and Javier Miranda lsquolsquoVolatilityand Dispersion in Business Growth Rates Publicly Traded versus PrivatelyHeld Firmsrsquorsquo NBER Macroeconomics Annual 21 (2007) 107ndash180

Dixit Avinash K and Robert S Pindyck Investment under Uncertainty(Princeton NJ Princeton University Press 1994)

Durnev Art lsquolsquoThe Real Effects of Political Uncertainty Elections and InvestmentSensitivity to Stock Pricesrsquorsquo working paper McGill University 2010

Federal Open Market Committee Minutes of the December (2009) Meeting avail-able at httpwwwfederalreservegovmonetarypolicyfomcmi-nutes20091216htm

Fernandez-Villaverde Jesus Pablo Guerron-Quintana Keith Kuester andJuan Rubio-Ramirez lsquolsquoFiscal Volatility Shocks and Economic ActivityrsquorsquoAmerican Economic Review 105 (2015) 3352ndash3384

Friedman Milton lsquolsquoThe Role of Monetary Policyrsquorsquo American Economic Review 58(1968) 1ndash17

Gentzkow Matthew and Jesse M Shapiro lsquolsquoWhat Drives Media Slant Evidencefrom US Daily Newspapersrsquorsquo Econometrica 78 (2010) 35ndash71

Giavazzi Francesco and Michael McMahon lsquolsquoPolicy Uncertainty and HouseholdSavingsrsquorsquo Review of Economics amp Statistics 94 (2012) 517ndash531

Gilchrist Simon Jae W Sim and Egon Zakrajsek lsquolsquoUncertainty FinancialFrictions and Investment Dynamicsrsquorsquo NBER Working Paper 20038 2014

Guiso Luigi and Guiso Parigi lsquolsquoInvestment and Demand Uncertaintyrsquorsquo QuarterlyJournal of Economics 114 (1999) 185ndash227

Gulen Huseyin and Mihai Ion lsquolsquoPolicy Uncertainty and Corporate InvestmentrsquorsquoReview of Financial Studies 29 (2016) 523ndash564

Handley Kyle and Nuno Limao lsquolsquoTrade and Investment under PolicyUncertainty Theory and Firm Evidencersquorsquo American Economic JournalPolicy 7 (2015) 189ndash222

Hartman Richard lsquolsquoThe Effects of Price and Cost Uncertainty on InvestmentrsquorsquoJournal of Economic Theory 5 (1972) 258ndash266

Hassett Kevin A and Gilbert E Metcalf lsquolsquoInvestment with Uncertain Tax PolicyDoes Random Tax Policy Discourage Investmentrsquorsquo Economic Journal 109(1999) 372ndash393

ECONOMIC POLICY UNCERTAINTY 1635

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from

Page 44: THE QUARTERLY JOURNAL OF ECONOMICS...THE QUARTERLY JOURNAL OF ECONOMICS Vol. 131 November 2016 Issue 4 MEASURING ECONOMIC POLICY UNCERTAINTY SCOTT R. BAKER NICHOLAS BLOOM STEVEN J

Higgs Robert lsquolsquoRegime Uncertainty Why the Great Depression Lasted So Longand Why Prosperity Resumed after the Warrsquorsquo Independent Review 1 (1997)561ndash590

Hoberg Gerard and Gordon Phillips lsquolsquoProduct Market Synergies andCompetition in Mergers and Acquisitions A Text-Based Analysisrsquorsquo Reviewof Financial Studies 23 (2010) 3773ndash3811

Ilut Cosmin and Martin Schneider lsquolsquoAmbiguous Business Cyclesrsquorsquo AmericanEconomic Review 104 (2014) 2368ndash2399

International Monetary Fund lsquolsquoWorld Economic Outlook Coping with High Debtand Sluggish Growthrsquorsquo IMF Press October 2012

mdashmdashmdash lsquolsquoWorld Economic Outlook Hopes Realities Risksrsquorsquo IMF Press April2013

Julio Brandon and Youngsuk Yook lsquolsquoPolitical Uncertainty and CorporateInvestment Cyclesrsquorsquo Journal of Finance 67 (2012) 45ndash83

mdashmdashmdash lsquolsquoPolicy Uncertainty Irreversibility and Cross-Border Flows of CapitalrsquorsquoJournal of International Economics 103 (2016) 13ndash26

Jurado Kyle Sydney Ludvigson and Serena Ng lsquolsquoMeasuring UncertaintyrsquorsquoAmerican Economic Review 105 (2015) 1177ndash1216

Kelly Bryan Lubos Pastor and Pietro Veronesi lsquolsquoThe Price of PoliticalUncertainty Theory and Evidence from the Option Marketrsquorsquo Journal ofFinance forthcoming (2016)

Klossner Stefan and Rodrigo Sekkel lsquolsquoInternational Spillovers of PolicyUncertaintyrsquorsquo Economics Letters 124 (2014) 508ndash512

Koijen Ralph S J Tomas J Philipson and Harald Uhlig lsquolsquoFinancial HealthEconomicsrsquorsquo Econometrica 84 (2016) 195ndash242

Leduc Sylvain and Zheng Liu lsquolsquoUncertainty Shocks Are Aggregate DemandShocksrsquorsquo Federal Reserve Bank of San Francisco Working Paper May 2015

Nalewaik Jeremy lsquolsquoRegime-Switching Models for Estimating InflationUncertaintyrsquorsquo Board of Governors of the Federal Reserve System WorkingPaper August 2015

Nodari Gabriela lsquolsquoFinancial Regulation Policy Uncertainty and Credit Spreads inthe United Statesrsquorsquo Journal of Macroeconomics 41 (2014) 122ndash132

Oi Walter lsquolsquoThe Desirability of Price Instability under Perfect CompetitionrsquorsquoEconometrica 29 (1961) 58ndash64

Panousi Vasia and Dimitris Papanikolaou lsquolsquoInvestment Idiosyncratic Risk andOwnershiprsquorsquo Journal of Finance 67 (2012) 1113ndash1148

Pastor Lubos and Pietro Veronesi lsquolsquoUncertainty about Government Policy andStock Pricesrsquorsquo Journal of Finance 67 (2012) 1219ndash1264

mdashmdashmdash lsquolsquoPolitical Uncertainty and Risk Premiarsquorsquo Journal of Financial Economics110 (2013) 520ndash545

Rodrik Dani lsquolsquoPolicy Uncertainty and Private Investmentrsquorsquo Journal ofDevelopment Economics 36 (1991) 229ndash242

Scotti Chiara lsquolsquoSurprise and Uncertainty Indexes Real-Time Aggregation ofReal-Activity Macro Surprisesrsquorsquo Journal of Monetary Economics 82 (2016)1ndash19

Shoag Daniel and Stan Veuger lsquolsquoUncertainty and the Geography of the GreatRecessionrsquorsquo AEI Economic Policy Working Paper 2015-07 2015

Stock James and Mark Watson lsquolsquoDisentangling the Channels of the 2007ndash2009Recessionrsquorsquo Brookings Panel on Economic Activity (Spring 2012) 81ndash135

QUARTERLY JOURNAL OF ECONOMICS1636

by guest on Novem

ber 3 2016httpqjeoxfordjournalsorg

Dow

nloaded from