heterogeneouseffectsofregimetypeontrade liberalization...
TRANSCRIPT
Democracy and Trade Policy at the Product Levellowast
Soubhik Bararidagger In Song KimDagger
May 14 2020
Abstract
Despite widespread recognition that trade policies vary significantly across products andtrading partners granular trade policy data is often inaccessible to scholars due to the dif-ficulties in collecting linking and structuring data from multiple sources In this paper weovercome these challenges and construct a dataset of 57 billion observations of applied tar-iff rates between any pair of 136 countries over 20 years Using this data we offer the firstproduct-level analysis of the impact of political regimes on trade liberalization To illustratethe severity of aggregation bias we estimate a series of Bayesian multilevel models at differentlevels of disaggregation at the Harmonized System (HS) 2- 4- and 6-digit levels We findthat democracies tend to enact lower trade barriers than non-democracies However our anal-ysis shows significant variations across products even within the same industry revealing forexample that democracies are often more likely to protect consumer goods Our findings sug-gest that trade politics depends heavily on heterogeneous preferences of individual producersacross specific products and the types of political institutions within which these preferencesare realized
Keywords democracy trade liberalization tariff-line data big data heterogeneous effectsagricultural protection
lowastWe thank the IDB-team at the World Trade Organization for their approval of the re-dissemination of tariff-linedata Financial support from the Amazon Web Services (AWS) Cloud Credits for Research Program is acknowledgedWe thank Weihuang Wong for his valuable insights and contributions to improve the paper Thanks also to TimBuumlthe Devin Caughey Volha Charnysh Raymond Hicks David Lake Lisa Camner McKay Asya MagazinnikEdward Mansfield Helen Milner Rich Nielsen Pablo Pinto Peter Rosendorff Sujeong Shim Alastair Smith RandyStone Jan Stuckatz Rory Truex Ariel White and Bernardo Zacka as well as the seminar participants at PrincetonUniversity Rutgers University Technische Universitaumlt Muumlnchen Washington University in St Louis University ofWisconsin-Madison UCLA (Computational IRWorkshop) University of California Riverside the 2017 InternationalPolitical Economy Society (IPES) and the 2018 European Political Science Association annual meetings An earlierversion of this article was circulated under the title ldquoTrade Liberalization and Regime Type Evidence from a NewTariff-line Datasetrdquo We are grateful for Samir Dutta and Hyungsuk Yoon for their excellent research assistance
daggerPhD Student Department of Government Harvard University Cambridge MA 02138 Emailsbararigharvardedu URL httpwwwsoubhikbarariorg
DaggerAssociate Professor Department of Political Science Massachusetts Institute of Technology Cambridge MA02139 Email insongmitedu URL httpwebmiteduinsongwww
1 Introduction
A rich body of theoretical research on trade liberalization has identified numerous factors that affect
countriesrsquo trade policies Traditionally this literature has focused on the distributional consequences
of trade across industries in which underlying demands for liberalization are driven by factor en-
dowments or sector-specific skills (eg Scheve and Slaughter 2001 Mayda and Rodrik 2005 Luuml
Scheve and Slaughter 2012) More recent firm-level studies of trade politics predict that political
cleavages will arise across firms even within the same industry (eg Kim 2017 Kim and Osgood
2019) In this paper we conduct an empirical evaluation of the related question of whether and how
this demand-side interacts with the characteristics of domestic political institutions which has long
been a source of controversy among social scientists (Rodrik 1995 Morrow Siverson and Tabares
1998 Mansfield Milner and Rosendorff 2000 2002 Milner and Kubota 2005 Kono 2006)
Although canonical political economy models of trade often predict that trade policies will be
highly various and determined endogenously by complex interactions across various political actors
(eg Mayer 1984 Grossman and Helpman 1994 Maggi and Rodriguez-Clare 2007) empirical anal-
ysis of trade policies across different regime types has been constrained by poor data as researchers
have been limited to using high-level aggregate measures of trade policies when evaluating how
underlying trade preferences are translated into policy outcomes As we demonstrate below this
is due in large part to the enormous difficulties in collecting linking and structuring product-level
trade policy data from multiple sources Thus despite the significant variations that indeed exist in
trade policy across products and trading partners many studies often employ Most Favored Nation
(MFN) applied tariff rates or non-tariff barrier ldquocoverage ratiordquo that are averaged across products
(Mansfield and Busch 1995 Gawande and Hansen 1999 Kono 2006) In fact the resulting single
number for a given importer-year observation has often been used to examine how trade policies
differ across regime types (eg Milner and Kubota 2005)
The first contribution of this paper is to address this discrepancy by constructing a dataset of
over 57 billion observations of product-level applied tariff rates that countries apply to their trading
partners We do this by incorporating the universe of preferential rates and the Generalized System
of Preferences (GSP) at the tariff-line level ndash the level at which tariff policy is actually set We
develop a replicable automated procedure to (1) retrieve tariff data from multiple web data sources
1
(2) identify the partner-specific tariff rates for each product and (3) resolve any discrepancies that
arise We then combine our product-level trade policy data for each directed dyad with numerous
country- dyad- and directed dyad-level datasets available in the literature such as measures of
political institutions GATTWTO membership and product-level bilateral trade volume at the
Harmonized System (HS) 6-digit level To the best of our knowledge this is the first database that
combines bilateral trade policies and trade volume at the most granular level comparable across
136 countries over 30 years
Using this data we show that researchers may commit ecological fallacy in studying trade
politics if empirical studies are not done at the appropriate unit of analysis For example within
the United States agricultural sector the 2016 US applied MFN tariff on Onions which is identified
by a HS 6-digit subheading code 071220 is 256 However at the HS 4-digit heading level onions
are classified as Dried vegetables (0712) and given an average tariff rate of 651 More severely
if the analyst chooses the HS 2-digit chapter code classification as the unit of aggregation onions
are simply classified as Vegetables (07) and given the average tariff rate of 435 To date most
researchers only consider the average country-level applied rate in which dried onions and all other
vegetable products would be equally assigned a 334 tariff rate Based on this ldquolowrdquo average
applied tariff rate scholars often believe that the US has mostly eliminated trade barriers and
some may go even so far as to falsely conclude that tariffs are no longer relevant to study trade
politics
Our second contribution is to empirically examine how tariff policy varies by products for coun-
tries with different regime types To exploit the rich structure of this massive dataset we develop a
Bayesian multilevel estimator that explicitly models the correlations across industries while over-
coming computational challenges in estimation and statistical inference with variational approxima-
tions (Jordan Ghahramani Jaakkola et al 1999) To illustrate the severity of potential aggregation
bias when researchers employ aggregated measures of trade policies we estimate the proposed sta-
tistical model at different degrees of disaggregation at the HS2 HS4 and HS6 levels In doing so
we offer the first analysis that yields granular estimates across numerous products and industries
enabling researchers to re-evaluate some of key assumptions in the literature about the relationships
between underlying trade preferences and trade policy-making
We begin our analysis by comparing the MFN trade policies between democracies and non-
2
democratic nations Consistent with Milner and Kubota (2005) we find that democracies are
associated with lower tariffs than non-democracies on average However we find a high level of
heterogeneity across products even within narrowly defined industries Specifically we find that
democracies are more likely to protect consumer goods such as food textile and manufactured
products Meanwhile we find less variability across products within some of industries such as
cotton Furthermore industries such as wood and metal industries as well as intermediate products
tend to get significantly lower tariff rates in democracies compared to non-democracies Our find-
ing provides evidence for highly heterogenous effects of democratic political institutions on trade
liberalization across products consistent with the firm-level theoretical framework In particular
it suggests that trade politics depends heavily on the differences across individual products and
producers within industries We also find that democratic institutions with larger selectorate than
autocracies do not necessarily empower consumers relative to producers thereby calling into ques-
tion whether the underlying assumption of many studies in the Open Economy Politics (OEP)
literature that individual trade preferences (eg given by their factoral or sectoral interests) are
translated into actual trade policies is indeed valid (Lake 2009)
Finally we undertake a dyadic analysis to examine whether the interaction of regime types
between trading partners affects the depth of trade liberalization As Mansfield Milner and
Rosendorff (2000) note measures of bilateral trade barriers across all combinations of country-
pairs are notoriously difficult to collect at the product level thereby constraining researchers to
use bilateral trade volume as a proxy measure for partner-specific trade policy Using our novel
dataset we are able to conduct the most rigorous dyadic analysis of tariff levels to date We con-
sider a total of 90 bilateral Free Trade Agreements (FTAs) that were signed between 1991 and
2012 We compute differences in average applied tariff rates before and after each agreement at
HS2 and HS4 levels We then compare the difference-in-differences between the two institutional
combinations In contrast to existing studies we find little evidence that pairs of democratic nations
tend to undergo deeper trade liberalization than mixed pairs (Mansfield Milner and Rosendorff
2000) However the direction of trade liberalization matters We show that a non-democratic im-
porter engages in shallower trade liberalization when negotiating with a democratic exporter than a
democratic importer does when negotiating with another democracy Democratic importers mean-
while give even deeper reductions in tariffs to non-democratic negotiating partners than they do
3
to other democracies Overall our findings shed new lights on the claim that democratic political
institutions facilitate unilateral and bilateral trade liberalization
The rest of the paper is organized as follows In the next section we provide a detailed descrip-
tion of computational challenges around the usage of tariff-line data and our automated dataset
compilation pipeline to address them Specifically we show that numerous discrepancies exist be-
tween two primary databases that have been widely used in the literature and we explain how we
construct a new dataset that resolves these discrepancies Section 3 presents the empirical findings
from the monadic and dyadic analyses The final section concludes The bilateral trade policy
and volume data at the HS6 level will be fully made available via the Journalrsquos Dataverse reposi-
tory To facilitate future research the source code for constructing the bilateral product-level tariffs
database as well as the estimated product-varying effects of political institutions and their posterior
distributions will be made publicly available at httppoltradegithubio
2 New Database Bilateral Product-level Applied Tariffs
In this section we introduce the new bilateral product-level tariffs database We illustrate the
importance of accounting for the heterogeneity in trade policies across products and trading part-
ners This database will lay an important empirical foundation for our empirical evaluation of the
interaction between democratic political institutions and trade policy-making at the product level
Variation in Tariffs Across Products and Trading Partners While overall tariff rates have
decreased substantially through a slew of bilateral and multilateral trade agreements over the past
decades countries continuously spend enormous resources on negotiating tariff rates In fact tariff
still serves as an important revenue source for many countries even among developed economies
(Hansen 1990 Bastiaens and Rudra 2016) For example US tariff revenue was approximately
$40 billion in 2015 (before the significant increase of applied tariffs under the Trump administra-
tion) which was similar to the revenue from the federal capital gains tax on corporations (Betz
and Pond 2020)1 To be sure there exist various non-tariff barriers (NTB) to trade However
with its simplicity in application and international standardization tariffs serve as highly valuable1In the US about 60 of products are still subject to non-zero tariffs and the mean MFN tariff rate for
these ldquodutiable goodsrdquo is approximately 73 With the tariffs under the Trump administration the applied rates
sometimes go up to 25 or even higher across many products
4
Figure 1 Variations in Ad Valorem Applied Tariff Rates across Trading Partners andIndustries This figure demonstrates how our tariff-line data captures both partner-specific and industry-varyingtrade policies Importers are plotted down each column and exporters are plotted across each row For a given countryand partner our data distinguishes precise tariff rates on more than 100 specific products in various industries (coloredwithin plot) from 1989 to 2015 Increases in applied tariff rates may be attributed to the conversion of specific tariffrates into ad valorem equivalents or actual temporary increases due to ldquobinding overhangrdquo (Pelc 2013 92)
measures for researchers to make systematic comparisons of trade policies across various countries
and products2 In this section we describe the challenges in compiling a large-scale dataset of
bilateral tariff rates We discuss the variation in applied tariff rates our data compilation process
the discrepancies in available data sources and the ways we organize the data for our empirical
analyses
The need for a dataset that captures bilateral tariffs at the product level stems from the substan-
tial heterogeneity in trade polices across industries and partners as Figure 1 shows For example2Getting disaggregated NTB measures is notoriously difficult Also it can be debated whether regulatory policies
that are designed to protect public health and safety should be deemed as barriers of trade further complicating the
use of NTB measures for evaluating trade policies across countries with different norms and cultural backgrounds
5
the first row shows that across industries and over time the MFN tariff rates applied by the US
on imports from China (both members of the WTO) are very different from the preferential rates
applied on imports from Mexico (both members of NAFTA) The columns show that exporters (in
this example China and Mexico) face markedly different tariffs on their products with different
trading partners This heterogeneity exists despite broad membership in the WTO because WTO
members are permitted to enter regional trade agreements under Article XXIV of GATT Enabling
Clause and to lower tariffs for the least developed countries with the GSP That is the rule of ldquonon-
discriminationrdquo does not hold in practice For example in 2013 the US tariffs on cars (Harmonized
Tariff Schedule subheading 87039000) exported by FTA partner South Korea was 15 whereas
it was 25 (the MFN rate) for cars originating from other WTO members Moreover even the
GSP rate for specific products can vary across GSP beneficiaries for strategic reasons As Carnegie
(2015 60) finds Pakistan was partially suspended from the US GSP program in 1996 due to its
violations of workersrsquo rights Indeed we find that the applied rates on gloves (HTS subheading
39262030) given to Pakistan was 3 (the MFN rate) in 1997 instead of the GSP rate of 0 even
though Pakistan remained a GSP beneficiary and still received benefits for many other products
To fully examine the political sources of such heterogeneity researchers must use partner-specific
tariff-line data rather than aggregate tariff measures
Automated Data Collection We develop an automated procedure to create a dataset of bi-
lateral trade policy for each tariff-line product and partner To create our dataset we begin with
two data sources (1) the WTO Integrated Database (IDB) and (2) UNCTAD Trade Analysis
Information System (TRAINS) Both contain applied tariff rates on a variety of products for all
WTO countries from 1996 to 2016 However there are three challenges that limit the use of these
databases by researchers in practice
First to download all product-level tariffs each database requires users to submit numerous
queries to the system for each importer-year pair which in our case amounts to more than 2188
queries (Step 1 in Figure 2) To overcome this difficulty we develop software that automates the
data retrieval process gathering more than 100 gigabytes (GB) of product-level tariff data
The secondmdashand more crucialmdashchallenge is to identify the correct partner-specific rates Specif-
ically both databases specify only the ldquotyperdquo or category of tariff rate that a given importer applies
6
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
Ste
p 1
Web
scr
ape
prod
uct-
leve
l tar
iffs
Ste
p 2
D
isag
greg
ate
topa
rtne
r-sp
ecifi
c du
ties
Ste
p 3
M
erge
two
sour
ces
IDB query form TRAINS query form
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
ener
alrdquo
25
1998
US
A09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoGen
eral
rdquo25
2007
US
AC
UB
JAM
DM
A
0910
4040
ldquoCB
ERA
pr
efer
entia
lrdquo0
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
SP
rdquo3
8
38 m
illio
n pr
oduc
t-le
vel d
utie
s
106
mill
ion
prod
uct-
leve
l dut
ies
47
bill
ion
part
ner-
spec
ific
dutie
s
41
bill
ion
part
ner-
spec
ific
dutie
s
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0ID
B
57
billi
on
mer
ged
dutie
s
gt2 th
ousa
nd
impo
rter
-yea
r que
ries
helliphellip helliphellip helliphellip
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AH
TIJA
MZW
E
0910
4040
ldquoGS
Prdquo
38
helliphellip helliphellip
Figure2
Tariff-lineDataset
CreationThisfig
ureillustrates
theprocessof
creating
ourindu
stry-le
velp
artner-spe
cific
tariffda
taset
Asan
exam
ple
weshow
how
weprod
ucethe1998
USd
utyon
Jamaicanim
portsof
Gingersaffron
turmeric(curcuma)thyme
bayleavesc
urry
andotherspices
(HTSsubh
eading
0910
4040
)Firstwescrape
tariffs
across
allavailableim
portersan
dyearsusingthepu
blic
web
form
sforID
Ban
dTRAIN
SThen
weusethetariffbe
neficiary
description(sho
wnas
desc
ript
ion)
tofin
dallt
ariffswho
sebe
neficiary
grou
pinclud
esJa
maicaAsshow
neach
databa
seismissing
adu
tythat
theothercontainsID
Bcontains
anMFN
dutya
generald
uty
andapreferential
dutyb
utno
ttheGSP
dutyT
RAIN
Scontains
aMFN
duty
andaGSP
dutybu
tno
tthepreferential
dutyFinallyto
select
thedu
tyam
ongstthesecand
idates
mostlik
elyap
pliedin
practicewe
useacustom
merging
algorithm
describe
din
App
endixA2In
this
caseJa
maica
enjoys
azero
tariffdu
eto
apreferential
Caribbe
anBasin
Econo
mic
RecoveryAct
(CBERA)du
tyw
hich
supe
rsedes
both
theMFN
andGSP
rates
7
Issue Year-Importer-Exporter-HS(Product description)
WTO IDB report UNCTAD TRAINS report(asymp AVE)
Solution N obs()
2013-China-India-09041200(Crushed or ground Piper pepper)
10 3 noneMissingReport 1991-Japan-Korea-140490499
(Cod fish)none 10 3
Usenon-missing
235 billion(418)
1997-Australia-Singapore-22082010(Grape wine)
3 3 + $3112L(asymp 127) 3
2005-Canada-Australia-22084010(Rum)
2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3
Use ad valoremequivalent (AVE)computed byUNCTAD
2004-Argentina-Paraguay-87083110(Motor vehicle brakes)
0 3 14
ConflictingReports
1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)
25 0 3
Use lower(preferential)rate
024 billion(425)
Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2
to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate
on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading
69120090) for all partners belonging to the United States Generalized System of Preferences but
does not specify the specific countries included (eg Albania Angola and so on) We use a mix
of hand-coding from WTO and World Bank reference materials and string matching algorithms
applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo
appearing in the original data to its corresponding set of disaggregated country ISO codes Even
when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-
tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an
example tariff-line Appendix A1 describes our data collection and processing in full detail
Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates
two issues that we identify First we find significant differences in data coverage This is problematic
given that researchers tend to rely primarily on either one of the widely used data sources for
empirical research but usually not both Data for 127 importer-years appear only in IDB (but not
TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a
result we find that at least 235 billion observations are missing from one of the databases we
make sure to utilize the available data whenever possible Second IDB returns duties as they are
originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate
8
an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of
alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3
+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise
and informative source to use Third preferential rates may be available from only one source
As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for
United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise
IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade
bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific
preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy
between the two sources we create a dataset of over 57 billion observations of bilateral trade policy
at the product level Appendix A2 details each step in our resolution algorithm In total our
dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from
1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995
(see Figure A1)
3 The Effects of Regime Type on Trade Policy
In this section we demonstrate the significance of employing our granular data Specifically we
examine differences in import tariff policy between democracies and non-democracies We begin
by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN
applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-
racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize
that the goal of this section is to empirically demonstrate high variability in the effects of regime
type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the
unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS
HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability
for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report
is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate
9
31 Monadic Analysis
Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-
Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of
factors with which their country is abundantly endowed and therefore one should expect that trade
liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged
across products they find that democratization in labor-abundant developing countries is associated
with lower trade barriers Other scholars have argued the reverse however suggesting that the need
to win elections makes democratic politicians sensitive to the demands of interest groups who offer
support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile
need appeal to only a very small segment of society to secure their power and therefore might be
less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield
2006)
We begin by examining whether trade policy varies between democracies and non-democracies
across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in
which competing economic interests across sectors determine industry-level trade policy (eg Mayer
1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of
trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting
in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show
political activities of industries such as lobbying interact with economic heterogeneity in import-
penetration and demand elasticity Consequently the canonical model of trade policy predicts
differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)
We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels
to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade
policies Our product-level analysis is also motivated by the possibility that underlying preferences
of various economic actors may interact differently with political institutions at the product-level
a prediction that is consistent with the firm-level theories in international political economy (Kim
2017) The hierarchical structure of the Harmonized System of product classification facilitates
aggregation and disaggregation at different levels of detail For example Brisket cuts a product
in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code
10
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
1 Introduction
A rich body of theoretical research on trade liberalization has identified numerous factors that affect
countriesrsquo trade policies Traditionally this literature has focused on the distributional consequences
of trade across industries in which underlying demands for liberalization are driven by factor en-
dowments or sector-specific skills (eg Scheve and Slaughter 2001 Mayda and Rodrik 2005 Luuml
Scheve and Slaughter 2012) More recent firm-level studies of trade politics predict that political
cleavages will arise across firms even within the same industry (eg Kim 2017 Kim and Osgood
2019) In this paper we conduct an empirical evaluation of the related question of whether and how
this demand-side interacts with the characteristics of domestic political institutions which has long
been a source of controversy among social scientists (Rodrik 1995 Morrow Siverson and Tabares
1998 Mansfield Milner and Rosendorff 2000 2002 Milner and Kubota 2005 Kono 2006)
Although canonical political economy models of trade often predict that trade policies will be
highly various and determined endogenously by complex interactions across various political actors
(eg Mayer 1984 Grossman and Helpman 1994 Maggi and Rodriguez-Clare 2007) empirical anal-
ysis of trade policies across different regime types has been constrained by poor data as researchers
have been limited to using high-level aggregate measures of trade policies when evaluating how
underlying trade preferences are translated into policy outcomes As we demonstrate below this
is due in large part to the enormous difficulties in collecting linking and structuring product-level
trade policy data from multiple sources Thus despite the significant variations that indeed exist in
trade policy across products and trading partners many studies often employ Most Favored Nation
(MFN) applied tariff rates or non-tariff barrier ldquocoverage ratiordquo that are averaged across products
(Mansfield and Busch 1995 Gawande and Hansen 1999 Kono 2006) In fact the resulting single
number for a given importer-year observation has often been used to examine how trade policies
differ across regime types (eg Milner and Kubota 2005)
The first contribution of this paper is to address this discrepancy by constructing a dataset of
over 57 billion observations of product-level applied tariff rates that countries apply to their trading
partners We do this by incorporating the universe of preferential rates and the Generalized System
of Preferences (GSP) at the tariff-line level ndash the level at which tariff policy is actually set We
develop a replicable automated procedure to (1) retrieve tariff data from multiple web data sources
1
(2) identify the partner-specific tariff rates for each product and (3) resolve any discrepancies that
arise We then combine our product-level trade policy data for each directed dyad with numerous
country- dyad- and directed dyad-level datasets available in the literature such as measures of
political institutions GATTWTO membership and product-level bilateral trade volume at the
Harmonized System (HS) 6-digit level To the best of our knowledge this is the first database that
combines bilateral trade policies and trade volume at the most granular level comparable across
136 countries over 30 years
Using this data we show that researchers may commit ecological fallacy in studying trade
politics if empirical studies are not done at the appropriate unit of analysis For example within
the United States agricultural sector the 2016 US applied MFN tariff on Onions which is identified
by a HS 6-digit subheading code 071220 is 256 However at the HS 4-digit heading level onions
are classified as Dried vegetables (0712) and given an average tariff rate of 651 More severely
if the analyst chooses the HS 2-digit chapter code classification as the unit of aggregation onions
are simply classified as Vegetables (07) and given the average tariff rate of 435 To date most
researchers only consider the average country-level applied rate in which dried onions and all other
vegetable products would be equally assigned a 334 tariff rate Based on this ldquolowrdquo average
applied tariff rate scholars often believe that the US has mostly eliminated trade barriers and
some may go even so far as to falsely conclude that tariffs are no longer relevant to study trade
politics
Our second contribution is to empirically examine how tariff policy varies by products for coun-
tries with different regime types To exploit the rich structure of this massive dataset we develop a
Bayesian multilevel estimator that explicitly models the correlations across industries while over-
coming computational challenges in estimation and statistical inference with variational approxima-
tions (Jordan Ghahramani Jaakkola et al 1999) To illustrate the severity of potential aggregation
bias when researchers employ aggregated measures of trade policies we estimate the proposed sta-
tistical model at different degrees of disaggregation at the HS2 HS4 and HS6 levels In doing so
we offer the first analysis that yields granular estimates across numerous products and industries
enabling researchers to re-evaluate some of key assumptions in the literature about the relationships
between underlying trade preferences and trade policy-making
We begin our analysis by comparing the MFN trade policies between democracies and non-
2
democratic nations Consistent with Milner and Kubota (2005) we find that democracies are
associated with lower tariffs than non-democracies on average However we find a high level of
heterogeneity across products even within narrowly defined industries Specifically we find that
democracies are more likely to protect consumer goods such as food textile and manufactured
products Meanwhile we find less variability across products within some of industries such as
cotton Furthermore industries such as wood and metal industries as well as intermediate products
tend to get significantly lower tariff rates in democracies compared to non-democracies Our find-
ing provides evidence for highly heterogenous effects of democratic political institutions on trade
liberalization across products consistent with the firm-level theoretical framework In particular
it suggests that trade politics depends heavily on the differences across individual products and
producers within industries We also find that democratic institutions with larger selectorate than
autocracies do not necessarily empower consumers relative to producers thereby calling into ques-
tion whether the underlying assumption of many studies in the Open Economy Politics (OEP)
literature that individual trade preferences (eg given by their factoral or sectoral interests) are
translated into actual trade policies is indeed valid (Lake 2009)
Finally we undertake a dyadic analysis to examine whether the interaction of regime types
between trading partners affects the depth of trade liberalization As Mansfield Milner and
Rosendorff (2000) note measures of bilateral trade barriers across all combinations of country-
pairs are notoriously difficult to collect at the product level thereby constraining researchers to
use bilateral trade volume as a proxy measure for partner-specific trade policy Using our novel
dataset we are able to conduct the most rigorous dyadic analysis of tariff levels to date We con-
sider a total of 90 bilateral Free Trade Agreements (FTAs) that were signed between 1991 and
2012 We compute differences in average applied tariff rates before and after each agreement at
HS2 and HS4 levels We then compare the difference-in-differences between the two institutional
combinations In contrast to existing studies we find little evidence that pairs of democratic nations
tend to undergo deeper trade liberalization than mixed pairs (Mansfield Milner and Rosendorff
2000) However the direction of trade liberalization matters We show that a non-democratic im-
porter engages in shallower trade liberalization when negotiating with a democratic exporter than a
democratic importer does when negotiating with another democracy Democratic importers mean-
while give even deeper reductions in tariffs to non-democratic negotiating partners than they do
3
to other democracies Overall our findings shed new lights on the claim that democratic political
institutions facilitate unilateral and bilateral trade liberalization
The rest of the paper is organized as follows In the next section we provide a detailed descrip-
tion of computational challenges around the usage of tariff-line data and our automated dataset
compilation pipeline to address them Specifically we show that numerous discrepancies exist be-
tween two primary databases that have been widely used in the literature and we explain how we
construct a new dataset that resolves these discrepancies Section 3 presents the empirical findings
from the monadic and dyadic analyses The final section concludes The bilateral trade policy
and volume data at the HS6 level will be fully made available via the Journalrsquos Dataverse reposi-
tory To facilitate future research the source code for constructing the bilateral product-level tariffs
database as well as the estimated product-varying effects of political institutions and their posterior
distributions will be made publicly available at httppoltradegithubio
2 New Database Bilateral Product-level Applied Tariffs
In this section we introduce the new bilateral product-level tariffs database We illustrate the
importance of accounting for the heterogeneity in trade policies across products and trading part-
ners This database will lay an important empirical foundation for our empirical evaluation of the
interaction between democratic political institutions and trade policy-making at the product level
Variation in Tariffs Across Products and Trading Partners While overall tariff rates have
decreased substantially through a slew of bilateral and multilateral trade agreements over the past
decades countries continuously spend enormous resources on negotiating tariff rates In fact tariff
still serves as an important revenue source for many countries even among developed economies
(Hansen 1990 Bastiaens and Rudra 2016) For example US tariff revenue was approximately
$40 billion in 2015 (before the significant increase of applied tariffs under the Trump administra-
tion) which was similar to the revenue from the federal capital gains tax on corporations (Betz
and Pond 2020)1 To be sure there exist various non-tariff barriers (NTB) to trade However
with its simplicity in application and international standardization tariffs serve as highly valuable1In the US about 60 of products are still subject to non-zero tariffs and the mean MFN tariff rate for
these ldquodutiable goodsrdquo is approximately 73 With the tariffs under the Trump administration the applied rates
sometimes go up to 25 or even higher across many products
4
Figure 1 Variations in Ad Valorem Applied Tariff Rates across Trading Partners andIndustries This figure demonstrates how our tariff-line data captures both partner-specific and industry-varyingtrade policies Importers are plotted down each column and exporters are plotted across each row For a given countryand partner our data distinguishes precise tariff rates on more than 100 specific products in various industries (coloredwithin plot) from 1989 to 2015 Increases in applied tariff rates may be attributed to the conversion of specific tariffrates into ad valorem equivalents or actual temporary increases due to ldquobinding overhangrdquo (Pelc 2013 92)
measures for researchers to make systematic comparisons of trade policies across various countries
and products2 In this section we describe the challenges in compiling a large-scale dataset of
bilateral tariff rates We discuss the variation in applied tariff rates our data compilation process
the discrepancies in available data sources and the ways we organize the data for our empirical
analyses
The need for a dataset that captures bilateral tariffs at the product level stems from the substan-
tial heterogeneity in trade polices across industries and partners as Figure 1 shows For example2Getting disaggregated NTB measures is notoriously difficult Also it can be debated whether regulatory policies
that are designed to protect public health and safety should be deemed as barriers of trade further complicating the
use of NTB measures for evaluating trade policies across countries with different norms and cultural backgrounds
5
the first row shows that across industries and over time the MFN tariff rates applied by the US
on imports from China (both members of the WTO) are very different from the preferential rates
applied on imports from Mexico (both members of NAFTA) The columns show that exporters (in
this example China and Mexico) face markedly different tariffs on their products with different
trading partners This heterogeneity exists despite broad membership in the WTO because WTO
members are permitted to enter regional trade agreements under Article XXIV of GATT Enabling
Clause and to lower tariffs for the least developed countries with the GSP That is the rule of ldquonon-
discriminationrdquo does not hold in practice For example in 2013 the US tariffs on cars (Harmonized
Tariff Schedule subheading 87039000) exported by FTA partner South Korea was 15 whereas
it was 25 (the MFN rate) for cars originating from other WTO members Moreover even the
GSP rate for specific products can vary across GSP beneficiaries for strategic reasons As Carnegie
(2015 60) finds Pakistan was partially suspended from the US GSP program in 1996 due to its
violations of workersrsquo rights Indeed we find that the applied rates on gloves (HTS subheading
39262030) given to Pakistan was 3 (the MFN rate) in 1997 instead of the GSP rate of 0 even
though Pakistan remained a GSP beneficiary and still received benefits for many other products
To fully examine the political sources of such heterogeneity researchers must use partner-specific
tariff-line data rather than aggregate tariff measures
Automated Data Collection We develop an automated procedure to create a dataset of bi-
lateral trade policy for each tariff-line product and partner To create our dataset we begin with
two data sources (1) the WTO Integrated Database (IDB) and (2) UNCTAD Trade Analysis
Information System (TRAINS) Both contain applied tariff rates on a variety of products for all
WTO countries from 1996 to 2016 However there are three challenges that limit the use of these
databases by researchers in practice
First to download all product-level tariffs each database requires users to submit numerous
queries to the system for each importer-year pair which in our case amounts to more than 2188
queries (Step 1 in Figure 2) To overcome this difficulty we develop software that automates the
data retrieval process gathering more than 100 gigabytes (GB) of product-level tariff data
The secondmdashand more crucialmdashchallenge is to identify the correct partner-specific rates Specif-
ically both databases specify only the ldquotyperdquo or category of tariff rate that a given importer applies
6
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
Ste
p 1
Web
scr
ape
prod
uct-
leve
l tar
iffs
Ste
p 2
D
isag
greg
ate
topa
rtne
r-sp
ecifi
c du
ties
Ste
p 3
M
erge
two
sour
ces
IDB query form TRAINS query form
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
ener
alrdquo
25
1998
US
A09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoGen
eral
rdquo25
2007
US
AC
UB
JAM
DM
A
0910
4040
ldquoCB
ERA
pr
efer
entia
lrdquo0
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
SP
rdquo3
8
38 m
illio
n pr
oduc
t-le
vel d
utie
s
106
mill
ion
prod
uct-
leve
l dut
ies
47
bill
ion
part
ner-
spec
ific
dutie
s
41
bill
ion
part
ner-
spec
ific
dutie
s
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0ID
B
57
billi
on
mer
ged
dutie
s
gt2 th
ousa
nd
impo
rter
-yea
r que
ries
helliphellip helliphellip helliphellip
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AH
TIJA
MZW
E
0910
4040
ldquoGS
Prdquo
38
helliphellip helliphellip
Figure2
Tariff-lineDataset
CreationThisfig
ureillustrates
theprocessof
creating
ourindu
stry-le
velp
artner-spe
cific
tariffda
taset
Asan
exam
ple
weshow
how
weprod
ucethe1998
USd
utyon
Jamaicanim
portsof
Gingersaffron
turmeric(curcuma)thyme
bayleavesc
urry
andotherspices
(HTSsubh
eading
0910
4040
)Firstwescrape
tariffs
across
allavailableim
portersan
dyearsusingthepu
blic
web
form
sforID
Ban
dTRAIN
SThen
weusethetariffbe
neficiary
description(sho
wnas
desc
ript
ion)
tofin
dallt
ariffswho
sebe
neficiary
grou
pinclud
esJa
maicaAsshow
neach
databa
seismissing
adu
tythat
theothercontainsID
Bcontains
anMFN
dutya
generald
uty
andapreferential
dutyb
utno
ttheGSP
dutyT
RAIN
Scontains
aMFN
duty
andaGSP
dutybu
tno
tthepreferential
dutyFinallyto
select
thedu
tyam
ongstthesecand
idates
mostlik
elyap
pliedin
practicewe
useacustom
merging
algorithm
describe
din
App
endixA2In
this
caseJa
maica
enjoys
azero
tariffdu
eto
apreferential
Caribbe
anBasin
Econo
mic
RecoveryAct
(CBERA)du
tyw
hich
supe
rsedes
both
theMFN
andGSP
rates
7
Issue Year-Importer-Exporter-HS(Product description)
WTO IDB report UNCTAD TRAINS report(asymp AVE)
Solution N obs()
2013-China-India-09041200(Crushed or ground Piper pepper)
10 3 noneMissingReport 1991-Japan-Korea-140490499
(Cod fish)none 10 3
Usenon-missing
235 billion(418)
1997-Australia-Singapore-22082010(Grape wine)
3 3 + $3112L(asymp 127) 3
2005-Canada-Australia-22084010(Rum)
2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3
Use ad valoremequivalent (AVE)computed byUNCTAD
2004-Argentina-Paraguay-87083110(Motor vehicle brakes)
0 3 14
ConflictingReports
1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)
25 0 3
Use lower(preferential)rate
024 billion(425)
Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2
to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate
on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading
69120090) for all partners belonging to the United States Generalized System of Preferences but
does not specify the specific countries included (eg Albania Angola and so on) We use a mix
of hand-coding from WTO and World Bank reference materials and string matching algorithms
applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo
appearing in the original data to its corresponding set of disaggregated country ISO codes Even
when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-
tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an
example tariff-line Appendix A1 describes our data collection and processing in full detail
Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates
two issues that we identify First we find significant differences in data coverage This is problematic
given that researchers tend to rely primarily on either one of the widely used data sources for
empirical research but usually not both Data for 127 importer-years appear only in IDB (but not
TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a
result we find that at least 235 billion observations are missing from one of the databases we
make sure to utilize the available data whenever possible Second IDB returns duties as they are
originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate
8
an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of
alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3
+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise
and informative source to use Third preferential rates may be available from only one source
As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for
United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise
IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade
bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific
preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy
between the two sources we create a dataset of over 57 billion observations of bilateral trade policy
at the product level Appendix A2 details each step in our resolution algorithm In total our
dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from
1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995
(see Figure A1)
3 The Effects of Regime Type on Trade Policy
In this section we demonstrate the significance of employing our granular data Specifically we
examine differences in import tariff policy between democracies and non-democracies We begin
by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN
applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-
racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize
that the goal of this section is to empirically demonstrate high variability in the effects of regime
type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the
unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS
HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability
for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report
is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate
9
31 Monadic Analysis
Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-
Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of
factors with which their country is abundantly endowed and therefore one should expect that trade
liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged
across products they find that democratization in labor-abundant developing countries is associated
with lower trade barriers Other scholars have argued the reverse however suggesting that the need
to win elections makes democratic politicians sensitive to the demands of interest groups who offer
support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile
need appeal to only a very small segment of society to secure their power and therefore might be
less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield
2006)
We begin by examining whether trade policy varies between democracies and non-democracies
across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in
which competing economic interests across sectors determine industry-level trade policy (eg Mayer
1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of
trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting
in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show
political activities of industries such as lobbying interact with economic heterogeneity in import-
penetration and demand elasticity Consequently the canonical model of trade policy predicts
differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)
We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels
to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade
policies Our product-level analysis is also motivated by the possibility that underlying preferences
of various economic actors may interact differently with political institutions at the product-level
a prediction that is consistent with the firm-level theories in international political economy (Kim
2017) The hierarchical structure of the Harmonized System of product classification facilitates
aggregation and disaggregation at different levels of detail For example Brisket cuts a product
in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code
10
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
(2) identify the partner-specific tariff rates for each product and (3) resolve any discrepancies that
arise We then combine our product-level trade policy data for each directed dyad with numerous
country- dyad- and directed dyad-level datasets available in the literature such as measures of
political institutions GATTWTO membership and product-level bilateral trade volume at the
Harmonized System (HS) 6-digit level To the best of our knowledge this is the first database that
combines bilateral trade policies and trade volume at the most granular level comparable across
136 countries over 30 years
Using this data we show that researchers may commit ecological fallacy in studying trade
politics if empirical studies are not done at the appropriate unit of analysis For example within
the United States agricultural sector the 2016 US applied MFN tariff on Onions which is identified
by a HS 6-digit subheading code 071220 is 256 However at the HS 4-digit heading level onions
are classified as Dried vegetables (0712) and given an average tariff rate of 651 More severely
if the analyst chooses the HS 2-digit chapter code classification as the unit of aggregation onions
are simply classified as Vegetables (07) and given the average tariff rate of 435 To date most
researchers only consider the average country-level applied rate in which dried onions and all other
vegetable products would be equally assigned a 334 tariff rate Based on this ldquolowrdquo average
applied tariff rate scholars often believe that the US has mostly eliminated trade barriers and
some may go even so far as to falsely conclude that tariffs are no longer relevant to study trade
politics
Our second contribution is to empirically examine how tariff policy varies by products for coun-
tries with different regime types To exploit the rich structure of this massive dataset we develop a
Bayesian multilevel estimator that explicitly models the correlations across industries while over-
coming computational challenges in estimation and statistical inference with variational approxima-
tions (Jordan Ghahramani Jaakkola et al 1999) To illustrate the severity of potential aggregation
bias when researchers employ aggregated measures of trade policies we estimate the proposed sta-
tistical model at different degrees of disaggregation at the HS2 HS4 and HS6 levels In doing so
we offer the first analysis that yields granular estimates across numerous products and industries
enabling researchers to re-evaluate some of key assumptions in the literature about the relationships
between underlying trade preferences and trade policy-making
We begin our analysis by comparing the MFN trade policies between democracies and non-
2
democratic nations Consistent with Milner and Kubota (2005) we find that democracies are
associated with lower tariffs than non-democracies on average However we find a high level of
heterogeneity across products even within narrowly defined industries Specifically we find that
democracies are more likely to protect consumer goods such as food textile and manufactured
products Meanwhile we find less variability across products within some of industries such as
cotton Furthermore industries such as wood and metal industries as well as intermediate products
tend to get significantly lower tariff rates in democracies compared to non-democracies Our find-
ing provides evidence for highly heterogenous effects of democratic political institutions on trade
liberalization across products consistent with the firm-level theoretical framework In particular
it suggests that trade politics depends heavily on the differences across individual products and
producers within industries We also find that democratic institutions with larger selectorate than
autocracies do not necessarily empower consumers relative to producers thereby calling into ques-
tion whether the underlying assumption of many studies in the Open Economy Politics (OEP)
literature that individual trade preferences (eg given by their factoral or sectoral interests) are
translated into actual trade policies is indeed valid (Lake 2009)
Finally we undertake a dyadic analysis to examine whether the interaction of regime types
between trading partners affects the depth of trade liberalization As Mansfield Milner and
Rosendorff (2000) note measures of bilateral trade barriers across all combinations of country-
pairs are notoriously difficult to collect at the product level thereby constraining researchers to
use bilateral trade volume as a proxy measure for partner-specific trade policy Using our novel
dataset we are able to conduct the most rigorous dyadic analysis of tariff levels to date We con-
sider a total of 90 bilateral Free Trade Agreements (FTAs) that were signed between 1991 and
2012 We compute differences in average applied tariff rates before and after each agreement at
HS2 and HS4 levels We then compare the difference-in-differences between the two institutional
combinations In contrast to existing studies we find little evidence that pairs of democratic nations
tend to undergo deeper trade liberalization than mixed pairs (Mansfield Milner and Rosendorff
2000) However the direction of trade liberalization matters We show that a non-democratic im-
porter engages in shallower trade liberalization when negotiating with a democratic exporter than a
democratic importer does when negotiating with another democracy Democratic importers mean-
while give even deeper reductions in tariffs to non-democratic negotiating partners than they do
3
to other democracies Overall our findings shed new lights on the claim that democratic political
institutions facilitate unilateral and bilateral trade liberalization
The rest of the paper is organized as follows In the next section we provide a detailed descrip-
tion of computational challenges around the usage of tariff-line data and our automated dataset
compilation pipeline to address them Specifically we show that numerous discrepancies exist be-
tween two primary databases that have been widely used in the literature and we explain how we
construct a new dataset that resolves these discrepancies Section 3 presents the empirical findings
from the monadic and dyadic analyses The final section concludes The bilateral trade policy
and volume data at the HS6 level will be fully made available via the Journalrsquos Dataverse reposi-
tory To facilitate future research the source code for constructing the bilateral product-level tariffs
database as well as the estimated product-varying effects of political institutions and their posterior
distributions will be made publicly available at httppoltradegithubio
2 New Database Bilateral Product-level Applied Tariffs
In this section we introduce the new bilateral product-level tariffs database We illustrate the
importance of accounting for the heterogeneity in trade policies across products and trading part-
ners This database will lay an important empirical foundation for our empirical evaluation of the
interaction between democratic political institutions and trade policy-making at the product level
Variation in Tariffs Across Products and Trading Partners While overall tariff rates have
decreased substantially through a slew of bilateral and multilateral trade agreements over the past
decades countries continuously spend enormous resources on negotiating tariff rates In fact tariff
still serves as an important revenue source for many countries even among developed economies
(Hansen 1990 Bastiaens and Rudra 2016) For example US tariff revenue was approximately
$40 billion in 2015 (before the significant increase of applied tariffs under the Trump administra-
tion) which was similar to the revenue from the federal capital gains tax on corporations (Betz
and Pond 2020)1 To be sure there exist various non-tariff barriers (NTB) to trade However
with its simplicity in application and international standardization tariffs serve as highly valuable1In the US about 60 of products are still subject to non-zero tariffs and the mean MFN tariff rate for
these ldquodutiable goodsrdquo is approximately 73 With the tariffs under the Trump administration the applied rates
sometimes go up to 25 or even higher across many products
4
Figure 1 Variations in Ad Valorem Applied Tariff Rates across Trading Partners andIndustries This figure demonstrates how our tariff-line data captures both partner-specific and industry-varyingtrade policies Importers are plotted down each column and exporters are plotted across each row For a given countryand partner our data distinguishes precise tariff rates on more than 100 specific products in various industries (coloredwithin plot) from 1989 to 2015 Increases in applied tariff rates may be attributed to the conversion of specific tariffrates into ad valorem equivalents or actual temporary increases due to ldquobinding overhangrdquo (Pelc 2013 92)
measures for researchers to make systematic comparisons of trade policies across various countries
and products2 In this section we describe the challenges in compiling a large-scale dataset of
bilateral tariff rates We discuss the variation in applied tariff rates our data compilation process
the discrepancies in available data sources and the ways we organize the data for our empirical
analyses
The need for a dataset that captures bilateral tariffs at the product level stems from the substan-
tial heterogeneity in trade polices across industries and partners as Figure 1 shows For example2Getting disaggregated NTB measures is notoriously difficult Also it can be debated whether regulatory policies
that are designed to protect public health and safety should be deemed as barriers of trade further complicating the
use of NTB measures for evaluating trade policies across countries with different norms and cultural backgrounds
5
the first row shows that across industries and over time the MFN tariff rates applied by the US
on imports from China (both members of the WTO) are very different from the preferential rates
applied on imports from Mexico (both members of NAFTA) The columns show that exporters (in
this example China and Mexico) face markedly different tariffs on their products with different
trading partners This heterogeneity exists despite broad membership in the WTO because WTO
members are permitted to enter regional trade agreements under Article XXIV of GATT Enabling
Clause and to lower tariffs for the least developed countries with the GSP That is the rule of ldquonon-
discriminationrdquo does not hold in practice For example in 2013 the US tariffs on cars (Harmonized
Tariff Schedule subheading 87039000) exported by FTA partner South Korea was 15 whereas
it was 25 (the MFN rate) for cars originating from other WTO members Moreover even the
GSP rate for specific products can vary across GSP beneficiaries for strategic reasons As Carnegie
(2015 60) finds Pakistan was partially suspended from the US GSP program in 1996 due to its
violations of workersrsquo rights Indeed we find that the applied rates on gloves (HTS subheading
39262030) given to Pakistan was 3 (the MFN rate) in 1997 instead of the GSP rate of 0 even
though Pakistan remained a GSP beneficiary and still received benefits for many other products
To fully examine the political sources of such heterogeneity researchers must use partner-specific
tariff-line data rather than aggregate tariff measures
Automated Data Collection We develop an automated procedure to create a dataset of bi-
lateral trade policy for each tariff-line product and partner To create our dataset we begin with
two data sources (1) the WTO Integrated Database (IDB) and (2) UNCTAD Trade Analysis
Information System (TRAINS) Both contain applied tariff rates on a variety of products for all
WTO countries from 1996 to 2016 However there are three challenges that limit the use of these
databases by researchers in practice
First to download all product-level tariffs each database requires users to submit numerous
queries to the system for each importer-year pair which in our case amounts to more than 2188
queries (Step 1 in Figure 2) To overcome this difficulty we develop software that automates the
data retrieval process gathering more than 100 gigabytes (GB) of product-level tariff data
The secondmdashand more crucialmdashchallenge is to identify the correct partner-specific rates Specif-
ically both databases specify only the ldquotyperdquo or category of tariff rate that a given importer applies
6
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
Ste
p 1
Web
scr
ape
prod
uct-
leve
l tar
iffs
Ste
p 2
D
isag
greg
ate
topa
rtne
r-sp
ecifi
c du
ties
Ste
p 3
M
erge
two
sour
ces
IDB query form TRAINS query form
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
ener
alrdquo
25
1998
US
A09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoGen
eral
rdquo25
2007
US
AC
UB
JAM
DM
A
0910
4040
ldquoCB
ERA
pr
efer
entia
lrdquo0
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
SP
rdquo3
8
38 m
illio
n pr
oduc
t-le
vel d
utie
s
106
mill
ion
prod
uct-
leve
l dut
ies
47
bill
ion
part
ner-
spec
ific
dutie
s
41
bill
ion
part
ner-
spec
ific
dutie
s
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0ID
B
57
billi
on
mer
ged
dutie
s
gt2 th
ousa
nd
impo
rter
-yea
r que
ries
helliphellip helliphellip helliphellip
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AH
TIJA
MZW
E
0910
4040
ldquoGS
Prdquo
38
helliphellip helliphellip
Figure2
Tariff-lineDataset
CreationThisfig
ureillustrates
theprocessof
creating
ourindu
stry-le
velp
artner-spe
cific
tariffda
taset
Asan
exam
ple
weshow
how
weprod
ucethe1998
USd
utyon
Jamaicanim
portsof
Gingersaffron
turmeric(curcuma)thyme
bayleavesc
urry
andotherspices
(HTSsubh
eading
0910
4040
)Firstwescrape
tariffs
across
allavailableim
portersan
dyearsusingthepu
blic
web
form
sforID
Ban
dTRAIN
SThen
weusethetariffbe
neficiary
description(sho
wnas
desc
ript
ion)
tofin
dallt
ariffswho
sebe
neficiary
grou
pinclud
esJa
maicaAsshow
neach
databa
seismissing
adu
tythat
theothercontainsID
Bcontains
anMFN
dutya
generald
uty
andapreferential
dutyb
utno
ttheGSP
dutyT
RAIN
Scontains
aMFN
duty
andaGSP
dutybu
tno
tthepreferential
dutyFinallyto
select
thedu
tyam
ongstthesecand
idates
mostlik
elyap
pliedin
practicewe
useacustom
merging
algorithm
describe
din
App
endixA2In
this
caseJa
maica
enjoys
azero
tariffdu
eto
apreferential
Caribbe
anBasin
Econo
mic
RecoveryAct
(CBERA)du
tyw
hich
supe
rsedes
both
theMFN
andGSP
rates
7
Issue Year-Importer-Exporter-HS(Product description)
WTO IDB report UNCTAD TRAINS report(asymp AVE)
Solution N obs()
2013-China-India-09041200(Crushed or ground Piper pepper)
10 3 noneMissingReport 1991-Japan-Korea-140490499
(Cod fish)none 10 3
Usenon-missing
235 billion(418)
1997-Australia-Singapore-22082010(Grape wine)
3 3 + $3112L(asymp 127) 3
2005-Canada-Australia-22084010(Rum)
2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3
Use ad valoremequivalent (AVE)computed byUNCTAD
2004-Argentina-Paraguay-87083110(Motor vehicle brakes)
0 3 14
ConflictingReports
1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)
25 0 3
Use lower(preferential)rate
024 billion(425)
Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2
to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate
on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading
69120090) for all partners belonging to the United States Generalized System of Preferences but
does not specify the specific countries included (eg Albania Angola and so on) We use a mix
of hand-coding from WTO and World Bank reference materials and string matching algorithms
applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo
appearing in the original data to its corresponding set of disaggregated country ISO codes Even
when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-
tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an
example tariff-line Appendix A1 describes our data collection and processing in full detail
Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates
two issues that we identify First we find significant differences in data coverage This is problematic
given that researchers tend to rely primarily on either one of the widely used data sources for
empirical research but usually not both Data for 127 importer-years appear only in IDB (but not
TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a
result we find that at least 235 billion observations are missing from one of the databases we
make sure to utilize the available data whenever possible Second IDB returns duties as they are
originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate
8
an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of
alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3
+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise
and informative source to use Third preferential rates may be available from only one source
As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for
United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise
IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade
bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific
preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy
between the two sources we create a dataset of over 57 billion observations of bilateral trade policy
at the product level Appendix A2 details each step in our resolution algorithm In total our
dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from
1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995
(see Figure A1)
3 The Effects of Regime Type on Trade Policy
In this section we demonstrate the significance of employing our granular data Specifically we
examine differences in import tariff policy between democracies and non-democracies We begin
by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN
applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-
racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize
that the goal of this section is to empirically demonstrate high variability in the effects of regime
type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the
unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS
HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability
for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report
is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate
9
31 Monadic Analysis
Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-
Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of
factors with which their country is abundantly endowed and therefore one should expect that trade
liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged
across products they find that democratization in labor-abundant developing countries is associated
with lower trade barriers Other scholars have argued the reverse however suggesting that the need
to win elections makes democratic politicians sensitive to the demands of interest groups who offer
support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile
need appeal to only a very small segment of society to secure their power and therefore might be
less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield
2006)
We begin by examining whether trade policy varies between democracies and non-democracies
across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in
which competing economic interests across sectors determine industry-level trade policy (eg Mayer
1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of
trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting
in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show
political activities of industries such as lobbying interact with economic heterogeneity in import-
penetration and demand elasticity Consequently the canonical model of trade policy predicts
differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)
We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels
to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade
policies Our product-level analysis is also motivated by the possibility that underlying preferences
of various economic actors may interact differently with political institutions at the product-level
a prediction that is consistent with the firm-level theories in international political economy (Kim
2017) The hierarchical structure of the Harmonized System of product classification facilitates
aggregation and disaggregation at different levels of detail For example Brisket cuts a product
in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code
10
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
democratic nations Consistent with Milner and Kubota (2005) we find that democracies are
associated with lower tariffs than non-democracies on average However we find a high level of
heterogeneity across products even within narrowly defined industries Specifically we find that
democracies are more likely to protect consumer goods such as food textile and manufactured
products Meanwhile we find less variability across products within some of industries such as
cotton Furthermore industries such as wood and metal industries as well as intermediate products
tend to get significantly lower tariff rates in democracies compared to non-democracies Our find-
ing provides evidence for highly heterogenous effects of democratic political institutions on trade
liberalization across products consistent with the firm-level theoretical framework In particular
it suggests that trade politics depends heavily on the differences across individual products and
producers within industries We also find that democratic institutions with larger selectorate than
autocracies do not necessarily empower consumers relative to producers thereby calling into ques-
tion whether the underlying assumption of many studies in the Open Economy Politics (OEP)
literature that individual trade preferences (eg given by their factoral or sectoral interests) are
translated into actual trade policies is indeed valid (Lake 2009)
Finally we undertake a dyadic analysis to examine whether the interaction of regime types
between trading partners affects the depth of trade liberalization As Mansfield Milner and
Rosendorff (2000) note measures of bilateral trade barriers across all combinations of country-
pairs are notoriously difficult to collect at the product level thereby constraining researchers to
use bilateral trade volume as a proxy measure for partner-specific trade policy Using our novel
dataset we are able to conduct the most rigorous dyadic analysis of tariff levels to date We con-
sider a total of 90 bilateral Free Trade Agreements (FTAs) that were signed between 1991 and
2012 We compute differences in average applied tariff rates before and after each agreement at
HS2 and HS4 levels We then compare the difference-in-differences between the two institutional
combinations In contrast to existing studies we find little evidence that pairs of democratic nations
tend to undergo deeper trade liberalization than mixed pairs (Mansfield Milner and Rosendorff
2000) However the direction of trade liberalization matters We show that a non-democratic im-
porter engages in shallower trade liberalization when negotiating with a democratic exporter than a
democratic importer does when negotiating with another democracy Democratic importers mean-
while give even deeper reductions in tariffs to non-democratic negotiating partners than they do
3
to other democracies Overall our findings shed new lights on the claim that democratic political
institutions facilitate unilateral and bilateral trade liberalization
The rest of the paper is organized as follows In the next section we provide a detailed descrip-
tion of computational challenges around the usage of tariff-line data and our automated dataset
compilation pipeline to address them Specifically we show that numerous discrepancies exist be-
tween two primary databases that have been widely used in the literature and we explain how we
construct a new dataset that resolves these discrepancies Section 3 presents the empirical findings
from the monadic and dyadic analyses The final section concludes The bilateral trade policy
and volume data at the HS6 level will be fully made available via the Journalrsquos Dataverse reposi-
tory To facilitate future research the source code for constructing the bilateral product-level tariffs
database as well as the estimated product-varying effects of political institutions and their posterior
distributions will be made publicly available at httppoltradegithubio
2 New Database Bilateral Product-level Applied Tariffs
In this section we introduce the new bilateral product-level tariffs database We illustrate the
importance of accounting for the heterogeneity in trade policies across products and trading part-
ners This database will lay an important empirical foundation for our empirical evaluation of the
interaction between democratic political institutions and trade policy-making at the product level
Variation in Tariffs Across Products and Trading Partners While overall tariff rates have
decreased substantially through a slew of bilateral and multilateral trade agreements over the past
decades countries continuously spend enormous resources on negotiating tariff rates In fact tariff
still serves as an important revenue source for many countries even among developed economies
(Hansen 1990 Bastiaens and Rudra 2016) For example US tariff revenue was approximately
$40 billion in 2015 (before the significant increase of applied tariffs under the Trump administra-
tion) which was similar to the revenue from the federal capital gains tax on corporations (Betz
and Pond 2020)1 To be sure there exist various non-tariff barriers (NTB) to trade However
with its simplicity in application and international standardization tariffs serve as highly valuable1In the US about 60 of products are still subject to non-zero tariffs and the mean MFN tariff rate for
these ldquodutiable goodsrdquo is approximately 73 With the tariffs under the Trump administration the applied rates
sometimes go up to 25 or even higher across many products
4
Figure 1 Variations in Ad Valorem Applied Tariff Rates across Trading Partners andIndustries This figure demonstrates how our tariff-line data captures both partner-specific and industry-varyingtrade policies Importers are plotted down each column and exporters are plotted across each row For a given countryand partner our data distinguishes precise tariff rates on more than 100 specific products in various industries (coloredwithin plot) from 1989 to 2015 Increases in applied tariff rates may be attributed to the conversion of specific tariffrates into ad valorem equivalents or actual temporary increases due to ldquobinding overhangrdquo (Pelc 2013 92)
measures for researchers to make systematic comparisons of trade policies across various countries
and products2 In this section we describe the challenges in compiling a large-scale dataset of
bilateral tariff rates We discuss the variation in applied tariff rates our data compilation process
the discrepancies in available data sources and the ways we organize the data for our empirical
analyses
The need for a dataset that captures bilateral tariffs at the product level stems from the substan-
tial heterogeneity in trade polices across industries and partners as Figure 1 shows For example2Getting disaggregated NTB measures is notoriously difficult Also it can be debated whether regulatory policies
that are designed to protect public health and safety should be deemed as barriers of trade further complicating the
use of NTB measures for evaluating trade policies across countries with different norms and cultural backgrounds
5
the first row shows that across industries and over time the MFN tariff rates applied by the US
on imports from China (both members of the WTO) are very different from the preferential rates
applied on imports from Mexico (both members of NAFTA) The columns show that exporters (in
this example China and Mexico) face markedly different tariffs on their products with different
trading partners This heterogeneity exists despite broad membership in the WTO because WTO
members are permitted to enter regional trade agreements under Article XXIV of GATT Enabling
Clause and to lower tariffs for the least developed countries with the GSP That is the rule of ldquonon-
discriminationrdquo does not hold in practice For example in 2013 the US tariffs on cars (Harmonized
Tariff Schedule subheading 87039000) exported by FTA partner South Korea was 15 whereas
it was 25 (the MFN rate) for cars originating from other WTO members Moreover even the
GSP rate for specific products can vary across GSP beneficiaries for strategic reasons As Carnegie
(2015 60) finds Pakistan was partially suspended from the US GSP program in 1996 due to its
violations of workersrsquo rights Indeed we find that the applied rates on gloves (HTS subheading
39262030) given to Pakistan was 3 (the MFN rate) in 1997 instead of the GSP rate of 0 even
though Pakistan remained a GSP beneficiary and still received benefits for many other products
To fully examine the political sources of such heterogeneity researchers must use partner-specific
tariff-line data rather than aggregate tariff measures
Automated Data Collection We develop an automated procedure to create a dataset of bi-
lateral trade policy for each tariff-line product and partner To create our dataset we begin with
two data sources (1) the WTO Integrated Database (IDB) and (2) UNCTAD Trade Analysis
Information System (TRAINS) Both contain applied tariff rates on a variety of products for all
WTO countries from 1996 to 2016 However there are three challenges that limit the use of these
databases by researchers in practice
First to download all product-level tariffs each database requires users to submit numerous
queries to the system for each importer-year pair which in our case amounts to more than 2188
queries (Step 1 in Figure 2) To overcome this difficulty we develop software that automates the
data retrieval process gathering more than 100 gigabytes (GB) of product-level tariff data
The secondmdashand more crucialmdashchallenge is to identify the correct partner-specific rates Specif-
ically both databases specify only the ldquotyperdquo or category of tariff rate that a given importer applies
6
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
Ste
p 1
Web
scr
ape
prod
uct-
leve
l tar
iffs
Ste
p 2
D
isag
greg
ate
topa
rtne
r-sp
ecifi
c du
ties
Ste
p 3
M
erge
two
sour
ces
IDB query form TRAINS query form
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
ener
alrdquo
25
1998
US
A09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoGen
eral
rdquo25
2007
US
AC
UB
JAM
DM
A
0910
4040
ldquoCB
ERA
pr
efer
entia
lrdquo0
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
SP
rdquo3
8
38 m
illio
n pr
oduc
t-le
vel d
utie
s
106
mill
ion
prod
uct-
leve
l dut
ies
47
bill
ion
part
ner-
spec
ific
dutie
s
41
bill
ion
part
ner-
spec
ific
dutie
s
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0ID
B
57
billi
on
mer
ged
dutie
s
gt2 th
ousa
nd
impo
rter
-yea
r que
ries
helliphellip helliphellip helliphellip
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AH
TIJA
MZW
E
0910
4040
ldquoGS
Prdquo
38
helliphellip helliphellip
Figure2
Tariff-lineDataset
CreationThisfig
ureillustrates
theprocessof
creating
ourindu
stry-le
velp
artner-spe
cific
tariffda
taset
Asan
exam
ple
weshow
how
weprod
ucethe1998
USd
utyon
Jamaicanim
portsof
Gingersaffron
turmeric(curcuma)thyme
bayleavesc
urry
andotherspices
(HTSsubh
eading
0910
4040
)Firstwescrape
tariffs
across
allavailableim
portersan
dyearsusingthepu
blic
web
form
sforID
Ban
dTRAIN
SThen
weusethetariffbe
neficiary
description(sho
wnas
desc
ript
ion)
tofin
dallt
ariffswho
sebe
neficiary
grou
pinclud
esJa
maicaAsshow
neach
databa
seismissing
adu
tythat
theothercontainsID
Bcontains
anMFN
dutya
generald
uty
andapreferential
dutyb
utno
ttheGSP
dutyT
RAIN
Scontains
aMFN
duty
andaGSP
dutybu
tno
tthepreferential
dutyFinallyto
select
thedu
tyam
ongstthesecand
idates
mostlik
elyap
pliedin
practicewe
useacustom
merging
algorithm
describe
din
App
endixA2In
this
caseJa
maica
enjoys
azero
tariffdu
eto
apreferential
Caribbe
anBasin
Econo
mic
RecoveryAct
(CBERA)du
tyw
hich
supe
rsedes
both
theMFN
andGSP
rates
7
Issue Year-Importer-Exporter-HS(Product description)
WTO IDB report UNCTAD TRAINS report(asymp AVE)
Solution N obs()
2013-China-India-09041200(Crushed or ground Piper pepper)
10 3 noneMissingReport 1991-Japan-Korea-140490499
(Cod fish)none 10 3
Usenon-missing
235 billion(418)
1997-Australia-Singapore-22082010(Grape wine)
3 3 + $3112L(asymp 127) 3
2005-Canada-Australia-22084010(Rum)
2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3
Use ad valoremequivalent (AVE)computed byUNCTAD
2004-Argentina-Paraguay-87083110(Motor vehicle brakes)
0 3 14
ConflictingReports
1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)
25 0 3
Use lower(preferential)rate
024 billion(425)
Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2
to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate
on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading
69120090) for all partners belonging to the United States Generalized System of Preferences but
does not specify the specific countries included (eg Albania Angola and so on) We use a mix
of hand-coding from WTO and World Bank reference materials and string matching algorithms
applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo
appearing in the original data to its corresponding set of disaggregated country ISO codes Even
when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-
tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an
example tariff-line Appendix A1 describes our data collection and processing in full detail
Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates
two issues that we identify First we find significant differences in data coverage This is problematic
given that researchers tend to rely primarily on either one of the widely used data sources for
empirical research but usually not both Data for 127 importer-years appear only in IDB (but not
TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a
result we find that at least 235 billion observations are missing from one of the databases we
make sure to utilize the available data whenever possible Second IDB returns duties as they are
originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate
8
an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of
alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3
+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise
and informative source to use Third preferential rates may be available from only one source
As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for
United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise
IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade
bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific
preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy
between the two sources we create a dataset of over 57 billion observations of bilateral trade policy
at the product level Appendix A2 details each step in our resolution algorithm In total our
dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from
1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995
(see Figure A1)
3 The Effects of Regime Type on Trade Policy
In this section we demonstrate the significance of employing our granular data Specifically we
examine differences in import tariff policy between democracies and non-democracies We begin
by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN
applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-
racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize
that the goal of this section is to empirically demonstrate high variability in the effects of regime
type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the
unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS
HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability
for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report
is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate
9
31 Monadic Analysis
Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-
Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of
factors with which their country is abundantly endowed and therefore one should expect that trade
liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged
across products they find that democratization in labor-abundant developing countries is associated
with lower trade barriers Other scholars have argued the reverse however suggesting that the need
to win elections makes democratic politicians sensitive to the demands of interest groups who offer
support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile
need appeal to only a very small segment of society to secure their power and therefore might be
less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield
2006)
We begin by examining whether trade policy varies between democracies and non-democracies
across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in
which competing economic interests across sectors determine industry-level trade policy (eg Mayer
1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of
trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting
in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show
political activities of industries such as lobbying interact with economic heterogeneity in import-
penetration and demand elasticity Consequently the canonical model of trade policy predicts
differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)
We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels
to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade
policies Our product-level analysis is also motivated by the possibility that underlying preferences
of various economic actors may interact differently with political institutions at the product-level
a prediction that is consistent with the firm-level theories in international political economy (Kim
2017) The hierarchical structure of the Harmonized System of product classification facilitates
aggregation and disaggregation at different levels of detail For example Brisket cuts a product
in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code
10
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
to other democracies Overall our findings shed new lights on the claim that democratic political
institutions facilitate unilateral and bilateral trade liberalization
The rest of the paper is organized as follows In the next section we provide a detailed descrip-
tion of computational challenges around the usage of tariff-line data and our automated dataset
compilation pipeline to address them Specifically we show that numerous discrepancies exist be-
tween two primary databases that have been widely used in the literature and we explain how we
construct a new dataset that resolves these discrepancies Section 3 presents the empirical findings
from the monadic and dyadic analyses The final section concludes The bilateral trade policy
and volume data at the HS6 level will be fully made available via the Journalrsquos Dataverse reposi-
tory To facilitate future research the source code for constructing the bilateral product-level tariffs
database as well as the estimated product-varying effects of political institutions and their posterior
distributions will be made publicly available at httppoltradegithubio
2 New Database Bilateral Product-level Applied Tariffs
In this section we introduce the new bilateral product-level tariffs database We illustrate the
importance of accounting for the heterogeneity in trade policies across products and trading part-
ners This database will lay an important empirical foundation for our empirical evaluation of the
interaction between democratic political institutions and trade policy-making at the product level
Variation in Tariffs Across Products and Trading Partners While overall tariff rates have
decreased substantially through a slew of bilateral and multilateral trade agreements over the past
decades countries continuously spend enormous resources on negotiating tariff rates In fact tariff
still serves as an important revenue source for many countries even among developed economies
(Hansen 1990 Bastiaens and Rudra 2016) For example US tariff revenue was approximately
$40 billion in 2015 (before the significant increase of applied tariffs under the Trump administra-
tion) which was similar to the revenue from the federal capital gains tax on corporations (Betz
and Pond 2020)1 To be sure there exist various non-tariff barriers (NTB) to trade However
with its simplicity in application and international standardization tariffs serve as highly valuable1In the US about 60 of products are still subject to non-zero tariffs and the mean MFN tariff rate for
these ldquodutiable goodsrdquo is approximately 73 With the tariffs under the Trump administration the applied rates
sometimes go up to 25 or even higher across many products
4
Figure 1 Variations in Ad Valorem Applied Tariff Rates across Trading Partners andIndustries This figure demonstrates how our tariff-line data captures both partner-specific and industry-varyingtrade policies Importers are plotted down each column and exporters are plotted across each row For a given countryand partner our data distinguishes precise tariff rates on more than 100 specific products in various industries (coloredwithin plot) from 1989 to 2015 Increases in applied tariff rates may be attributed to the conversion of specific tariffrates into ad valorem equivalents or actual temporary increases due to ldquobinding overhangrdquo (Pelc 2013 92)
measures for researchers to make systematic comparisons of trade policies across various countries
and products2 In this section we describe the challenges in compiling a large-scale dataset of
bilateral tariff rates We discuss the variation in applied tariff rates our data compilation process
the discrepancies in available data sources and the ways we organize the data for our empirical
analyses
The need for a dataset that captures bilateral tariffs at the product level stems from the substan-
tial heterogeneity in trade polices across industries and partners as Figure 1 shows For example2Getting disaggregated NTB measures is notoriously difficult Also it can be debated whether regulatory policies
that are designed to protect public health and safety should be deemed as barriers of trade further complicating the
use of NTB measures for evaluating trade policies across countries with different norms and cultural backgrounds
5
the first row shows that across industries and over time the MFN tariff rates applied by the US
on imports from China (both members of the WTO) are very different from the preferential rates
applied on imports from Mexico (both members of NAFTA) The columns show that exporters (in
this example China and Mexico) face markedly different tariffs on their products with different
trading partners This heterogeneity exists despite broad membership in the WTO because WTO
members are permitted to enter regional trade agreements under Article XXIV of GATT Enabling
Clause and to lower tariffs for the least developed countries with the GSP That is the rule of ldquonon-
discriminationrdquo does not hold in practice For example in 2013 the US tariffs on cars (Harmonized
Tariff Schedule subheading 87039000) exported by FTA partner South Korea was 15 whereas
it was 25 (the MFN rate) for cars originating from other WTO members Moreover even the
GSP rate for specific products can vary across GSP beneficiaries for strategic reasons As Carnegie
(2015 60) finds Pakistan was partially suspended from the US GSP program in 1996 due to its
violations of workersrsquo rights Indeed we find that the applied rates on gloves (HTS subheading
39262030) given to Pakistan was 3 (the MFN rate) in 1997 instead of the GSP rate of 0 even
though Pakistan remained a GSP beneficiary and still received benefits for many other products
To fully examine the political sources of such heterogeneity researchers must use partner-specific
tariff-line data rather than aggregate tariff measures
Automated Data Collection We develop an automated procedure to create a dataset of bi-
lateral trade policy for each tariff-line product and partner To create our dataset we begin with
two data sources (1) the WTO Integrated Database (IDB) and (2) UNCTAD Trade Analysis
Information System (TRAINS) Both contain applied tariff rates on a variety of products for all
WTO countries from 1996 to 2016 However there are three challenges that limit the use of these
databases by researchers in practice
First to download all product-level tariffs each database requires users to submit numerous
queries to the system for each importer-year pair which in our case amounts to more than 2188
queries (Step 1 in Figure 2) To overcome this difficulty we develop software that automates the
data retrieval process gathering more than 100 gigabytes (GB) of product-level tariff data
The secondmdashand more crucialmdashchallenge is to identify the correct partner-specific rates Specif-
ically both databases specify only the ldquotyperdquo or category of tariff rate that a given importer applies
6
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
Ste
p 1
Web
scr
ape
prod
uct-
leve
l tar
iffs
Ste
p 2
D
isag
greg
ate
topa
rtne
r-sp
ecifi
c du
ties
Ste
p 3
M
erge
two
sour
ces
IDB query form TRAINS query form
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
ener
alrdquo
25
1998
US
A09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoGen
eral
rdquo25
2007
US
AC
UB
JAM
DM
A
0910
4040
ldquoCB
ERA
pr
efer
entia
lrdquo0
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
SP
rdquo3
8
38 m
illio
n pr
oduc
t-le
vel d
utie
s
106
mill
ion
prod
uct-
leve
l dut
ies
47
bill
ion
part
ner-
spec
ific
dutie
s
41
bill
ion
part
ner-
spec
ific
dutie
s
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0ID
B
57
billi
on
mer
ged
dutie
s
gt2 th
ousa
nd
impo
rter
-yea
r que
ries
helliphellip helliphellip helliphellip
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AH
TIJA
MZW
E
0910
4040
ldquoGS
Prdquo
38
helliphellip helliphellip
Figure2
Tariff-lineDataset
CreationThisfig
ureillustrates
theprocessof
creating
ourindu
stry-le
velp
artner-spe
cific
tariffda
taset
Asan
exam
ple
weshow
how
weprod
ucethe1998
USd
utyon
Jamaicanim
portsof
Gingersaffron
turmeric(curcuma)thyme
bayleavesc
urry
andotherspices
(HTSsubh
eading
0910
4040
)Firstwescrape
tariffs
across
allavailableim
portersan
dyearsusingthepu
blic
web
form
sforID
Ban
dTRAIN
SThen
weusethetariffbe
neficiary
description(sho
wnas
desc
ript
ion)
tofin
dallt
ariffswho
sebe
neficiary
grou
pinclud
esJa
maicaAsshow
neach
databa
seismissing
adu
tythat
theothercontainsID
Bcontains
anMFN
dutya
generald
uty
andapreferential
dutyb
utno
ttheGSP
dutyT
RAIN
Scontains
aMFN
duty
andaGSP
dutybu
tno
tthepreferential
dutyFinallyto
select
thedu
tyam
ongstthesecand
idates
mostlik
elyap
pliedin
practicewe
useacustom
merging
algorithm
describe
din
App
endixA2In
this
caseJa
maica
enjoys
azero
tariffdu
eto
apreferential
Caribbe
anBasin
Econo
mic
RecoveryAct
(CBERA)du
tyw
hich
supe
rsedes
both
theMFN
andGSP
rates
7
Issue Year-Importer-Exporter-HS(Product description)
WTO IDB report UNCTAD TRAINS report(asymp AVE)
Solution N obs()
2013-China-India-09041200(Crushed or ground Piper pepper)
10 3 noneMissingReport 1991-Japan-Korea-140490499
(Cod fish)none 10 3
Usenon-missing
235 billion(418)
1997-Australia-Singapore-22082010(Grape wine)
3 3 + $3112L(asymp 127) 3
2005-Canada-Australia-22084010(Rum)
2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3
Use ad valoremequivalent (AVE)computed byUNCTAD
2004-Argentina-Paraguay-87083110(Motor vehicle brakes)
0 3 14
ConflictingReports
1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)
25 0 3
Use lower(preferential)rate
024 billion(425)
Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2
to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate
on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading
69120090) for all partners belonging to the United States Generalized System of Preferences but
does not specify the specific countries included (eg Albania Angola and so on) We use a mix
of hand-coding from WTO and World Bank reference materials and string matching algorithms
applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo
appearing in the original data to its corresponding set of disaggregated country ISO codes Even
when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-
tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an
example tariff-line Appendix A1 describes our data collection and processing in full detail
Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates
two issues that we identify First we find significant differences in data coverage This is problematic
given that researchers tend to rely primarily on either one of the widely used data sources for
empirical research but usually not both Data for 127 importer-years appear only in IDB (but not
TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a
result we find that at least 235 billion observations are missing from one of the databases we
make sure to utilize the available data whenever possible Second IDB returns duties as they are
originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate
8
an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of
alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3
+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise
and informative source to use Third preferential rates may be available from only one source
As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for
United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise
IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade
bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific
preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy
between the two sources we create a dataset of over 57 billion observations of bilateral trade policy
at the product level Appendix A2 details each step in our resolution algorithm In total our
dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from
1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995
(see Figure A1)
3 The Effects of Regime Type on Trade Policy
In this section we demonstrate the significance of employing our granular data Specifically we
examine differences in import tariff policy between democracies and non-democracies We begin
by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN
applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-
racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize
that the goal of this section is to empirically demonstrate high variability in the effects of regime
type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the
unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS
HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability
for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report
is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate
9
31 Monadic Analysis
Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-
Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of
factors with which their country is abundantly endowed and therefore one should expect that trade
liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged
across products they find that democratization in labor-abundant developing countries is associated
with lower trade barriers Other scholars have argued the reverse however suggesting that the need
to win elections makes democratic politicians sensitive to the demands of interest groups who offer
support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile
need appeal to only a very small segment of society to secure their power and therefore might be
less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield
2006)
We begin by examining whether trade policy varies between democracies and non-democracies
across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in
which competing economic interests across sectors determine industry-level trade policy (eg Mayer
1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of
trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting
in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show
political activities of industries such as lobbying interact with economic heterogeneity in import-
penetration and demand elasticity Consequently the canonical model of trade policy predicts
differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)
We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels
to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade
policies Our product-level analysis is also motivated by the possibility that underlying preferences
of various economic actors may interact differently with political institutions at the product-level
a prediction that is consistent with the firm-level theories in international political economy (Kim
2017) The hierarchical structure of the Harmonized System of product classification facilitates
aggregation and disaggregation at different levels of detail For example Brisket cuts a product
in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code
10
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Figure 1 Variations in Ad Valorem Applied Tariff Rates across Trading Partners andIndustries This figure demonstrates how our tariff-line data captures both partner-specific and industry-varyingtrade policies Importers are plotted down each column and exporters are plotted across each row For a given countryand partner our data distinguishes precise tariff rates on more than 100 specific products in various industries (coloredwithin plot) from 1989 to 2015 Increases in applied tariff rates may be attributed to the conversion of specific tariffrates into ad valorem equivalents or actual temporary increases due to ldquobinding overhangrdquo (Pelc 2013 92)
measures for researchers to make systematic comparisons of trade policies across various countries
and products2 In this section we describe the challenges in compiling a large-scale dataset of
bilateral tariff rates We discuss the variation in applied tariff rates our data compilation process
the discrepancies in available data sources and the ways we organize the data for our empirical
analyses
The need for a dataset that captures bilateral tariffs at the product level stems from the substan-
tial heterogeneity in trade polices across industries and partners as Figure 1 shows For example2Getting disaggregated NTB measures is notoriously difficult Also it can be debated whether regulatory policies
that are designed to protect public health and safety should be deemed as barriers of trade further complicating the
use of NTB measures for evaluating trade policies across countries with different norms and cultural backgrounds
5
the first row shows that across industries and over time the MFN tariff rates applied by the US
on imports from China (both members of the WTO) are very different from the preferential rates
applied on imports from Mexico (both members of NAFTA) The columns show that exporters (in
this example China and Mexico) face markedly different tariffs on their products with different
trading partners This heterogeneity exists despite broad membership in the WTO because WTO
members are permitted to enter regional trade agreements under Article XXIV of GATT Enabling
Clause and to lower tariffs for the least developed countries with the GSP That is the rule of ldquonon-
discriminationrdquo does not hold in practice For example in 2013 the US tariffs on cars (Harmonized
Tariff Schedule subheading 87039000) exported by FTA partner South Korea was 15 whereas
it was 25 (the MFN rate) for cars originating from other WTO members Moreover even the
GSP rate for specific products can vary across GSP beneficiaries for strategic reasons As Carnegie
(2015 60) finds Pakistan was partially suspended from the US GSP program in 1996 due to its
violations of workersrsquo rights Indeed we find that the applied rates on gloves (HTS subheading
39262030) given to Pakistan was 3 (the MFN rate) in 1997 instead of the GSP rate of 0 even
though Pakistan remained a GSP beneficiary and still received benefits for many other products
To fully examine the political sources of such heterogeneity researchers must use partner-specific
tariff-line data rather than aggregate tariff measures
Automated Data Collection We develop an automated procedure to create a dataset of bi-
lateral trade policy for each tariff-line product and partner To create our dataset we begin with
two data sources (1) the WTO Integrated Database (IDB) and (2) UNCTAD Trade Analysis
Information System (TRAINS) Both contain applied tariff rates on a variety of products for all
WTO countries from 1996 to 2016 However there are three challenges that limit the use of these
databases by researchers in practice
First to download all product-level tariffs each database requires users to submit numerous
queries to the system for each importer-year pair which in our case amounts to more than 2188
queries (Step 1 in Figure 2) To overcome this difficulty we develop software that automates the
data retrieval process gathering more than 100 gigabytes (GB) of product-level tariff data
The secondmdashand more crucialmdashchallenge is to identify the correct partner-specific rates Specif-
ically both databases specify only the ldquotyperdquo or category of tariff rate that a given importer applies
6
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
Ste
p 1
Web
scr
ape
prod
uct-
leve
l tar
iffs
Ste
p 2
D
isag
greg
ate
topa
rtne
r-sp
ecifi
c du
ties
Ste
p 3
M
erge
two
sour
ces
IDB query form TRAINS query form
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
ener
alrdquo
25
1998
US
A09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoGen
eral
rdquo25
2007
US
AC
UB
JAM
DM
A
0910
4040
ldquoCB
ERA
pr
efer
entia
lrdquo0
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
SP
rdquo3
8
38 m
illio
n pr
oduc
t-le
vel d
utie
s
106
mill
ion
prod
uct-
leve
l dut
ies
47
bill
ion
part
ner-
spec
ific
dutie
s
41
bill
ion
part
ner-
spec
ific
dutie
s
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0ID
B
57
billi
on
mer
ged
dutie
s
gt2 th
ousa
nd
impo
rter
-yea
r que
ries
helliphellip helliphellip helliphellip
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AH
TIJA
MZW
E
0910
4040
ldquoGS
Prdquo
38
helliphellip helliphellip
Figure2
Tariff-lineDataset
CreationThisfig
ureillustrates
theprocessof
creating
ourindu
stry-le
velp
artner-spe
cific
tariffda
taset
Asan
exam
ple
weshow
how
weprod
ucethe1998
USd
utyon
Jamaicanim
portsof
Gingersaffron
turmeric(curcuma)thyme
bayleavesc
urry
andotherspices
(HTSsubh
eading
0910
4040
)Firstwescrape
tariffs
across
allavailableim
portersan
dyearsusingthepu
blic
web
form
sforID
Ban
dTRAIN
SThen
weusethetariffbe
neficiary
description(sho
wnas
desc
ript
ion)
tofin
dallt
ariffswho
sebe
neficiary
grou
pinclud
esJa
maicaAsshow
neach
databa
seismissing
adu
tythat
theothercontainsID
Bcontains
anMFN
dutya
generald
uty
andapreferential
dutyb
utno
ttheGSP
dutyT
RAIN
Scontains
aMFN
duty
andaGSP
dutybu
tno
tthepreferential
dutyFinallyto
select
thedu
tyam
ongstthesecand
idates
mostlik
elyap
pliedin
practicewe
useacustom
merging
algorithm
describe
din
App
endixA2In
this
caseJa
maica
enjoys
azero
tariffdu
eto
apreferential
Caribbe
anBasin
Econo
mic
RecoveryAct
(CBERA)du
tyw
hich
supe
rsedes
both
theMFN
andGSP
rates
7
Issue Year-Importer-Exporter-HS(Product description)
WTO IDB report UNCTAD TRAINS report(asymp AVE)
Solution N obs()
2013-China-India-09041200(Crushed or ground Piper pepper)
10 3 noneMissingReport 1991-Japan-Korea-140490499
(Cod fish)none 10 3
Usenon-missing
235 billion(418)
1997-Australia-Singapore-22082010(Grape wine)
3 3 + $3112L(asymp 127) 3
2005-Canada-Australia-22084010(Rum)
2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3
Use ad valoremequivalent (AVE)computed byUNCTAD
2004-Argentina-Paraguay-87083110(Motor vehicle brakes)
0 3 14
ConflictingReports
1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)
25 0 3
Use lower(preferential)rate
024 billion(425)
Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2
to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate
on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading
69120090) for all partners belonging to the United States Generalized System of Preferences but
does not specify the specific countries included (eg Albania Angola and so on) We use a mix
of hand-coding from WTO and World Bank reference materials and string matching algorithms
applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo
appearing in the original data to its corresponding set of disaggregated country ISO codes Even
when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-
tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an
example tariff-line Appendix A1 describes our data collection and processing in full detail
Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates
two issues that we identify First we find significant differences in data coverage This is problematic
given that researchers tend to rely primarily on either one of the widely used data sources for
empirical research but usually not both Data for 127 importer-years appear only in IDB (but not
TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a
result we find that at least 235 billion observations are missing from one of the databases we
make sure to utilize the available data whenever possible Second IDB returns duties as they are
originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate
8
an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of
alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3
+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise
and informative source to use Third preferential rates may be available from only one source
As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for
United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise
IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade
bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific
preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy
between the two sources we create a dataset of over 57 billion observations of bilateral trade policy
at the product level Appendix A2 details each step in our resolution algorithm In total our
dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from
1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995
(see Figure A1)
3 The Effects of Regime Type on Trade Policy
In this section we demonstrate the significance of employing our granular data Specifically we
examine differences in import tariff policy between democracies and non-democracies We begin
by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN
applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-
racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize
that the goal of this section is to empirically demonstrate high variability in the effects of regime
type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the
unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS
HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability
for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report
is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate
9
31 Monadic Analysis
Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-
Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of
factors with which their country is abundantly endowed and therefore one should expect that trade
liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged
across products they find that democratization in labor-abundant developing countries is associated
with lower trade barriers Other scholars have argued the reverse however suggesting that the need
to win elections makes democratic politicians sensitive to the demands of interest groups who offer
support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile
need appeal to only a very small segment of society to secure their power and therefore might be
less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield
2006)
We begin by examining whether trade policy varies between democracies and non-democracies
across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in
which competing economic interests across sectors determine industry-level trade policy (eg Mayer
1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of
trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting
in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show
political activities of industries such as lobbying interact with economic heterogeneity in import-
penetration and demand elasticity Consequently the canonical model of trade policy predicts
differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)
We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels
to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade
policies Our product-level analysis is also motivated by the possibility that underlying preferences
of various economic actors may interact differently with political institutions at the product-level
a prediction that is consistent with the firm-level theories in international political economy (Kim
2017) The hierarchical structure of the Harmonized System of product classification facilitates
aggregation and disaggregation at different levels of detail For example Brisket cuts a product
in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code
10
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
the first row shows that across industries and over time the MFN tariff rates applied by the US
on imports from China (both members of the WTO) are very different from the preferential rates
applied on imports from Mexico (both members of NAFTA) The columns show that exporters (in
this example China and Mexico) face markedly different tariffs on their products with different
trading partners This heterogeneity exists despite broad membership in the WTO because WTO
members are permitted to enter regional trade agreements under Article XXIV of GATT Enabling
Clause and to lower tariffs for the least developed countries with the GSP That is the rule of ldquonon-
discriminationrdquo does not hold in practice For example in 2013 the US tariffs on cars (Harmonized
Tariff Schedule subheading 87039000) exported by FTA partner South Korea was 15 whereas
it was 25 (the MFN rate) for cars originating from other WTO members Moreover even the
GSP rate for specific products can vary across GSP beneficiaries for strategic reasons As Carnegie
(2015 60) finds Pakistan was partially suspended from the US GSP program in 1996 due to its
violations of workersrsquo rights Indeed we find that the applied rates on gloves (HTS subheading
39262030) given to Pakistan was 3 (the MFN rate) in 1997 instead of the GSP rate of 0 even
though Pakistan remained a GSP beneficiary and still received benefits for many other products
To fully examine the political sources of such heterogeneity researchers must use partner-specific
tariff-line data rather than aggregate tariff measures
Automated Data Collection We develop an automated procedure to create a dataset of bi-
lateral trade policy for each tariff-line product and partner To create our dataset we begin with
two data sources (1) the WTO Integrated Database (IDB) and (2) UNCTAD Trade Analysis
Information System (TRAINS) Both contain applied tariff rates on a variety of products for all
WTO countries from 1996 to 2016 However there are three challenges that limit the use of these
databases by researchers in practice
First to download all product-level tariffs each database requires users to submit numerous
queries to the system for each importer-year pair which in our case amounts to more than 2188
queries (Step 1 in Figure 2) To overcome this difficulty we develop software that automates the
data retrieval process gathering more than 100 gigabytes (GB) of product-level tariff data
The secondmdashand more crucialmdashchallenge is to identify the correct partner-specific rates Specif-
ically both databases specify only the ldquotyperdquo or category of tariff rate that a given importer applies
6
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
Ste
p 1
Web
scr
ape
prod
uct-
leve
l tar
iffs
Ste
p 2
D
isag
greg
ate
topa
rtne
r-sp
ecifi
c du
ties
Ste
p 3
M
erge
two
sour
ces
IDB query form TRAINS query form
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
ener
alrdquo
25
1998
US
A09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoGen
eral
rdquo25
2007
US
AC
UB
JAM
DM
A
0910
4040
ldquoCB
ERA
pr
efer
entia
lrdquo0
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
SP
rdquo3
8
38 m
illio
n pr
oduc
t-le
vel d
utie
s
106
mill
ion
prod
uct-
leve
l dut
ies
47
bill
ion
part
ner-
spec
ific
dutie
s
41
bill
ion
part
ner-
spec
ific
dutie
s
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0ID
B
57
billi
on
mer
ged
dutie
s
gt2 th
ousa
nd
impo
rter
-yea
r que
ries
helliphellip helliphellip helliphellip
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AH
TIJA
MZW
E
0910
4040
ldquoGS
Prdquo
38
helliphellip helliphellip
Figure2
Tariff-lineDataset
CreationThisfig
ureillustrates
theprocessof
creating
ourindu
stry-le
velp
artner-spe
cific
tariffda
taset
Asan
exam
ple
weshow
how
weprod
ucethe1998
USd
utyon
Jamaicanim
portsof
Gingersaffron
turmeric(curcuma)thyme
bayleavesc
urry
andotherspices
(HTSsubh
eading
0910
4040
)Firstwescrape
tariffs
across
allavailableim
portersan
dyearsusingthepu
blic
web
form
sforID
Ban
dTRAIN
SThen
weusethetariffbe
neficiary
description(sho
wnas
desc
ript
ion)
tofin
dallt
ariffswho
sebe
neficiary
grou
pinclud
esJa
maicaAsshow
neach
databa
seismissing
adu
tythat
theothercontainsID
Bcontains
anMFN
dutya
generald
uty
andapreferential
dutyb
utno
ttheGSP
dutyT
RAIN
Scontains
aMFN
duty
andaGSP
dutybu
tno
tthepreferential
dutyFinallyto
select
thedu
tyam
ongstthesecand
idates
mostlik
elyap
pliedin
practicewe
useacustom
merging
algorithm
describe
din
App
endixA2In
this
caseJa
maica
enjoys
azero
tariffdu
eto
apreferential
Caribbe
anBasin
Econo
mic
RecoveryAct
(CBERA)du
tyw
hich
supe
rsedes
both
theMFN
andGSP
rates
7
Issue Year-Importer-Exporter-HS(Product description)
WTO IDB report UNCTAD TRAINS report(asymp AVE)
Solution N obs()
2013-China-India-09041200(Crushed or ground Piper pepper)
10 3 noneMissingReport 1991-Japan-Korea-140490499
(Cod fish)none 10 3
Usenon-missing
235 billion(418)
1997-Australia-Singapore-22082010(Grape wine)
3 3 + $3112L(asymp 127) 3
2005-Canada-Australia-22084010(Rum)
2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3
Use ad valoremequivalent (AVE)computed byUNCTAD
2004-Argentina-Paraguay-87083110(Motor vehicle brakes)
0 3 14
ConflictingReports
1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)
25 0 3
Use lower(preferential)rate
024 billion(425)
Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2
to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate
on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading
69120090) for all partners belonging to the United States Generalized System of Preferences but
does not specify the specific countries included (eg Albania Angola and so on) We use a mix
of hand-coding from WTO and World Bank reference materials and string matching algorithms
applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo
appearing in the original data to its corresponding set of disaggregated country ISO codes Even
when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-
tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an
example tariff-line Appendix A1 describes our data collection and processing in full detail
Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates
two issues that we identify First we find significant differences in data coverage This is problematic
given that researchers tend to rely primarily on either one of the widely used data sources for
empirical research but usually not both Data for 127 importer-years appear only in IDB (but not
TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a
result we find that at least 235 billion observations are missing from one of the databases we
make sure to utilize the available data whenever possible Second IDB returns duties as they are
originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate
8
an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of
alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3
+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise
and informative source to use Third preferential rates may be available from only one source
As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for
United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise
IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade
bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific
preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy
between the two sources we create a dataset of over 57 billion observations of bilateral trade policy
at the product level Appendix A2 details each step in our resolution algorithm In total our
dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from
1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995
(see Figure A1)
3 The Effects of Regime Type on Trade Policy
In this section we demonstrate the significance of employing our granular data Specifically we
examine differences in import tariff policy between democracies and non-democracies We begin
by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN
applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-
racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize
that the goal of this section is to empirically demonstrate high variability in the effects of regime
type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the
unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS
HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability
for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report
is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate
9
31 Monadic Analysis
Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-
Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of
factors with which their country is abundantly endowed and therefore one should expect that trade
liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged
across products they find that democratization in labor-abundant developing countries is associated
with lower trade barriers Other scholars have argued the reverse however suggesting that the need
to win elections makes democratic politicians sensitive to the demands of interest groups who offer
support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile
need appeal to only a very small segment of society to secure their power and therefore might be
less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield
2006)
We begin by examining whether trade policy varies between democracies and non-democracies
across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in
which competing economic interests across sectors determine industry-level trade policy (eg Mayer
1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of
trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting
in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show
political activities of industries such as lobbying interact with economic heterogeneity in import-
penetration and demand elasticity Consequently the canonical model of trade policy predicts
differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)
We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels
to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade
policies Our product-level analysis is also motivated by the possibility that underlying preferences
of various economic actors may interact differently with political institutions at the product-level
a prediction that is consistent with the firm-level theories in international political economy (Kim
2017) The hierarchical structure of the Harmonized System of product classification facilitates
aggregation and disaggregation at different levels of detail For example Brisket cuts a product
in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code
10
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20
ldquoMos
t Fav
oure
d N
atio
nrdquo6
8
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
ost F
avou
red
Nat
ionrdquo
68
2007
US
A0
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0TR
AIN
S
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
year
imp
exp
code
type
rate
2007
US
AS
GP
0101
9020
ldquoMFN
ap
plie
d ra
terdquo
68
2007
US
AS
GP
0101
9020
ldquoGen
eral
du
tyrdquo
15
2007
US
AS
GP
0101
9020
ldquoFre
e-tr
ade
for
Sin
gapo
rerdquo
0
year
imp
code
type
rate
2007
US
A01
0190
20ldquoM
FN
appl
ied
rate
rdquo6
8
2007
US
A01
0190
20ldquoG
ener
al
duty
rdquo 15
2007
US
A01
0190
20ldquoF
ree-
trad
e fo
r0
Ste
p 1
Web
scr
ape
prod
uct-
leve
l tar
iffs
Ste
p 2
D
isag
greg
ate
topa
rtne
r-sp
ecifi
c du
ties
Ste
p 3
M
erge
two
sour
ces
IDB query form TRAINS query form
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
ener
alrdquo
25
1998
US
A09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoGen
eral
rdquo25
2007
US
AC
UB
JAM
DM
A
0910
4040
ldquoCB
ERA
pr
efer
entia
lrdquo0
year
imp
code
desc
riptio
nra
te
1998
US
A09
1040
40ldquoM
FNrdquo
38
1998
US
A09
1040
40ldquoG
SP
rdquo3
8
38 m
illio
n pr
oduc
t-le
vel d
utie
s
106
mill
ion
prod
uct-
leve
l dut
ies
47
bill
ion
part
ner-
spec
ific
dutie
s
41
bill
ion
part
ner-
spec
ific
dutie
s
year
imp
exp
code
desc
riptio
nra
teso
urce
2007
US
AJA
M09
1040
40ldquoC
BER
A
pref
eren
tialrdquo
0ID
B
57
billi
on
mer
ged
dutie
s
gt2 th
ousa
nd
impo
rter
-yea
r que
ries
helliphellip helliphellip helliphellip
year
imp
exp
code
desc
riptio
nra
te
2007
US
AIT
AJA
MJO
R
0910
4040
ldquoMFN
rdquo3
8
2007
US
AH
TIJA
MZW
E
0910
4040
ldquoGS
Prdquo
38
helliphellip helliphellip
Figure2
Tariff-lineDataset
CreationThisfig
ureillustrates
theprocessof
creating
ourindu
stry-le
velp
artner-spe
cific
tariffda
taset
Asan
exam
ple
weshow
how
weprod
ucethe1998
USd
utyon
Jamaicanim
portsof
Gingersaffron
turmeric(curcuma)thyme
bayleavesc
urry
andotherspices
(HTSsubh
eading
0910
4040
)Firstwescrape
tariffs
across
allavailableim
portersan
dyearsusingthepu
blic
web
form
sforID
Ban
dTRAIN
SThen
weusethetariffbe
neficiary
description(sho
wnas
desc
ript
ion)
tofin
dallt
ariffswho
sebe
neficiary
grou
pinclud
esJa
maicaAsshow
neach
databa
seismissing
adu
tythat
theothercontainsID
Bcontains
anMFN
dutya
generald
uty
andapreferential
dutyb
utno
ttheGSP
dutyT
RAIN
Scontains
aMFN
duty
andaGSP
dutybu
tno
tthepreferential
dutyFinallyto
select
thedu
tyam
ongstthesecand
idates
mostlik
elyap
pliedin
practicewe
useacustom
merging
algorithm
describe
din
App
endixA2In
this
caseJa
maica
enjoys
azero
tariffdu
eto
apreferential
Caribbe
anBasin
Econo
mic
RecoveryAct
(CBERA)du
tyw
hich
supe
rsedes
both
theMFN
andGSP
rates
7
Issue Year-Importer-Exporter-HS(Product description)
WTO IDB report UNCTAD TRAINS report(asymp AVE)
Solution N obs()
2013-China-India-09041200(Crushed or ground Piper pepper)
10 3 noneMissingReport 1991-Japan-Korea-140490499
(Cod fish)none 10 3
Usenon-missing
235 billion(418)
1997-Australia-Singapore-22082010(Grape wine)
3 3 + $3112L(asymp 127) 3
2005-Canada-Australia-22084010(Rum)
2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3
Use ad valoremequivalent (AVE)computed byUNCTAD
2004-Argentina-Paraguay-87083110(Motor vehicle brakes)
0 3 14
ConflictingReports
1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)
25 0 3
Use lower(preferential)rate
024 billion(425)
Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2
to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate
on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading
69120090) for all partners belonging to the United States Generalized System of Preferences but
does not specify the specific countries included (eg Albania Angola and so on) We use a mix
of hand-coding from WTO and World Bank reference materials and string matching algorithms
applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo
appearing in the original data to its corresponding set of disaggregated country ISO codes Even
when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-
tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an
example tariff-line Appendix A1 describes our data collection and processing in full detail
Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates
two issues that we identify First we find significant differences in data coverage This is problematic
given that researchers tend to rely primarily on either one of the widely used data sources for
empirical research but usually not both Data for 127 importer-years appear only in IDB (but not
TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a
result we find that at least 235 billion observations are missing from one of the databases we
make sure to utilize the available data whenever possible Second IDB returns duties as they are
originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate
8
an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of
alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3
+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise
and informative source to use Third preferential rates may be available from only one source
As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for
United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise
IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade
bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific
preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy
between the two sources we create a dataset of over 57 billion observations of bilateral trade policy
at the product level Appendix A2 details each step in our resolution algorithm In total our
dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from
1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995
(see Figure A1)
3 The Effects of Regime Type on Trade Policy
In this section we demonstrate the significance of employing our granular data Specifically we
examine differences in import tariff policy between democracies and non-democracies We begin
by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN
applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-
racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize
that the goal of this section is to empirically demonstrate high variability in the effects of regime
type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the
unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS
HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability
for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report
is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate
9
31 Monadic Analysis
Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-
Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of
factors with which their country is abundantly endowed and therefore one should expect that trade
liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged
across products they find that democratization in labor-abundant developing countries is associated
with lower trade barriers Other scholars have argued the reverse however suggesting that the need
to win elections makes democratic politicians sensitive to the demands of interest groups who offer
support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile
need appeal to only a very small segment of society to secure their power and therefore might be
less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield
2006)
We begin by examining whether trade policy varies between democracies and non-democracies
across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in
which competing economic interests across sectors determine industry-level trade policy (eg Mayer
1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of
trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting
in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show
political activities of industries such as lobbying interact with economic heterogeneity in import-
penetration and demand elasticity Consequently the canonical model of trade policy predicts
differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)
We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels
to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade
policies Our product-level analysis is also motivated by the possibility that underlying preferences
of various economic actors may interact differently with political institutions at the product-level
a prediction that is consistent with the firm-level theories in international political economy (Kim
2017) The hierarchical structure of the Harmonized System of product classification facilitates
aggregation and disaggregation at different levels of detail For example Brisket cuts a product
in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code
10
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Issue Year-Importer-Exporter-HS(Product description)
WTO IDB report UNCTAD TRAINS report(asymp AVE)
Solution N obs()
2013-China-India-09041200(Crushed or ground Piper pepper)
10 3 noneMissingReport 1991-Japan-Korea-140490499
(Cod fish)none 10 3
Usenon-missing
235 billion(418)
1997-Australia-Singapore-22082010(Grape wine)
3 3 + $3112L(asymp 127) 3
2005-Canada-Australia-22084010(Rum)
2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3
Use ad valoremequivalent (AVE)computed byUNCTAD
2004-Argentina-Paraguay-87083110(Motor vehicle brakes)
0 3 14
ConflictingReports
1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)
25 0 3
Use lower(preferential)rate
024 billion(425)
Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2
to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate
on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading
69120090) for all partners belonging to the United States Generalized System of Preferences but
does not specify the specific countries included (eg Albania Angola and so on) We use a mix
of hand-coding from WTO and World Bank reference materials and string matching algorithms
applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo
appearing in the original data to its corresponding set of disaggregated country ISO codes Even
when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-
tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an
example tariff-line Appendix A1 describes our data collection and processing in full detail
Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates
two issues that we identify First we find significant differences in data coverage This is problematic
given that researchers tend to rely primarily on either one of the widely used data sources for
empirical research but usually not both Data for 127 importer-years appear only in IDB (but not
TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a
result we find that at least 235 billion observations are missing from one of the databases we
make sure to utilize the available data whenever possible Second IDB returns duties as they are
originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate
8
an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of
alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3
+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise
and informative source to use Third preferential rates may be available from only one source
As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for
United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise
IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade
bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific
preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy
between the two sources we create a dataset of over 57 billion observations of bilateral trade policy
at the product level Appendix A2 details each step in our resolution algorithm In total our
dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from
1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995
(see Figure A1)
3 The Effects of Regime Type on Trade Policy
In this section we demonstrate the significance of employing our granular data Specifically we
examine differences in import tariff policy between democracies and non-democracies We begin
by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN
applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-
racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize
that the goal of this section is to empirically demonstrate high variability in the effects of regime
type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the
unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS
HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability
for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report
is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate
9
31 Monadic Analysis
Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-
Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of
factors with which their country is abundantly endowed and therefore one should expect that trade
liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged
across products they find that democratization in labor-abundant developing countries is associated
with lower trade barriers Other scholars have argued the reverse however suggesting that the need
to win elections makes democratic politicians sensitive to the demands of interest groups who offer
support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile
need appeal to only a very small segment of society to secure their power and therefore might be
less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield
2006)
We begin by examining whether trade policy varies between democracies and non-democracies
across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in
which competing economic interests across sectors determine industry-level trade policy (eg Mayer
1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of
trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting
in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show
political activities of industries such as lobbying interact with economic heterogeneity in import-
penetration and demand elasticity Consequently the canonical model of trade policy predicts
differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)
We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels
to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade
policies Our product-level analysis is also motivated by the possibility that underlying preferences
of various economic actors may interact differently with political institutions at the product-level
a prediction that is consistent with the firm-level theories in international political economy (Kim
2017) The hierarchical structure of the Harmonized System of product classification facilitates
aggregation and disaggregation at different levels of detail For example Brisket cuts a product
in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code
10
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of
alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3
+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise
and informative source to use Third preferential rates may be available from only one source
As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for
United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise
IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade
bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific
preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy
between the two sources we create a dataset of over 57 billion observations of bilateral trade policy
at the product level Appendix A2 details each step in our resolution algorithm In total our
dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from
1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995
(see Figure A1)
3 The Effects of Regime Type on Trade Policy
In this section we demonstrate the significance of employing our granular data Specifically we
examine differences in import tariff policy between democracies and non-democracies We begin
by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN
applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-
racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize
that the goal of this section is to empirically demonstrate high variability in the effects of regime
type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the
unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS
HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability
for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report
is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate
9
31 Monadic Analysis
Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-
Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of
factors with which their country is abundantly endowed and therefore one should expect that trade
liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged
across products they find that democratization in labor-abundant developing countries is associated
with lower trade barriers Other scholars have argued the reverse however suggesting that the need
to win elections makes democratic politicians sensitive to the demands of interest groups who offer
support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile
need appeal to only a very small segment of society to secure their power and therefore might be
less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield
2006)
We begin by examining whether trade policy varies between democracies and non-democracies
across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in
which competing economic interests across sectors determine industry-level trade policy (eg Mayer
1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of
trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting
in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show
political activities of industries such as lobbying interact with economic heterogeneity in import-
penetration and demand elasticity Consequently the canonical model of trade policy predicts
differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)
We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels
to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade
policies Our product-level analysis is also motivated by the possibility that underlying preferences
of various economic actors may interact differently with political institutions at the product-level
a prediction that is consistent with the firm-level theories in international political economy (Kim
2017) The hierarchical structure of the Harmonized System of product classification facilitates
aggregation and disaggregation at different levels of detail For example Brisket cuts a product
in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code
10
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
31 Monadic Analysis
Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-
Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of
factors with which their country is abundantly endowed and therefore one should expect that trade
liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged
across products they find that democratization in labor-abundant developing countries is associated
with lower trade barriers Other scholars have argued the reverse however suggesting that the need
to win elections makes democratic politicians sensitive to the demands of interest groups who offer
support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile
need appeal to only a very small segment of society to secure their power and therefore might be
less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield
2006)
We begin by examining whether trade policy varies between democracies and non-democracies
across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in
which competing economic interests across sectors determine industry-level trade policy (eg Mayer
1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of
trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting
in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show
political activities of industries such as lobbying interact with economic heterogeneity in import-
penetration and demand elasticity Consequently the canonical model of trade policy predicts
differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)
We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels
to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade
policies Our product-level analysis is also motivated by the possibility that underlying preferences
of various economic actors may interact differently with political institutions at the product-level
a prediction that is consistent with the firm-level theories in international political economy (Kim
2017) The hierarchical structure of the Harmonized System of product classification facilitates
aggregation and disaggregation at different levels of detail For example Brisket cuts a product
in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code
10
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine
animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine
animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while
our full dataset captures trade policy for individual products at the tariff-line level we disaggregate
up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level
explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products
compared to HS2 or HS4 For example in the United States variance between HS2 groupings of
products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4
groupings of products explained 52 and between-HS6 variance explained 68 Second further
disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained
categorization that is internationally comparable as each country may set their own tariff lines using
8-digit or sometimes 10-digit codes independently4
311 Methodology
To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit
model of the observed MFN tariff rate τipt for importer i and product p in year t
τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt
τipt =
τ lowastipt if τ lowastipt ge 0
0 otherwise
(1)
where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero
otherwise We perform separate analyses aggregating products at either the HS2 industry level or
at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate
τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and
use logged values to address the high skewness of tariffs To compare our empirical findings against
existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV
score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner
and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries
although they tend to describe highly similar goods under the common HS6 category which is already highly specific
11
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients
Zit represents a vector of country-level time-varying confounders of regime type and trade policy
log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and
an intercept5 To address missingness in covariate data we create multiple imputed datasets and
conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt
represents log import volume (at the same level of product aggregation as the tariff rate) All
covariates are lagged by 1 year
We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed
comparative advantages which captures technological differences across countries industries h (that
product p belongs to) and time t7 For example this allows us to account for the possibility
that developing and developed countries may use different production technologies even when they
produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and
εipt is idiosyncratic error assumed to be drawn from a Normal distribution
ηiiidsim N (0Ση) θt
iidsim N (0Σθ) εiptiidsim N (0 σ2
ε ) (2)
To be sure countries may have different domestic institutions that aggregate trade preferences
across various sectors meaning that trade policies of certain sectors tend to be highly correlated
For example the US Congress established the Agricultural Policy Advisory Committee (APAC)
and other advisory committees within the Department of Agriculture to provide advice on the
administration and implementation of US trade policy To account for heterogeneous political
processes across products we model the product-varying effects hierarchically Specifically we
allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg
Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use
ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log
trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade
volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally
intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total
exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2
level
12
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
correlations within a broader sector k (eg food sector) that operates differently from other sectors
(eg textile sector)
γp sim N (φk[p]Σγ) (3)
φk sim N (0Σφ) (4)
where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-
tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal
distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-
cients vary based on the sector k to which the product belongs which increases the plausibility of
the exchangeability assumption for the product-specific effects
Our analysis requires substantial computational resources For example in the HS2 analysis we
examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26
years (1990 to 2015) We overcome computational challenges by estimating the parameters of the
HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan
program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run
four separate Markov chains Our posterior sample combines the chains from the imputed datasets
While we focus specifically on the posterior means and credible intervals of our quantity of interest
when we present our findings below we also make the entire posterior samples publicly available
Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of
HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the
Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that
examines the relationships between regime type and MFN trade policy covering both developing
and developed nations
312 Empirical Results
Our quantity of interest is the product-specific effect of democracy on trade policy The model
given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter
space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)
method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter
values with accuracy within a reasonable length of time
13
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
1920
21
22
23
24
25
26
27
2829
30
31
32
3334
3536
373839
4041
42
43
44
45
46
47
48
49
50
5152
53
5455
5657
585960
6162
6364
65
66
67
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
8788
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus2
minus1
0
1
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate
product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9
Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of
the posterior distribution of the main effect of democracy β (marked by the dotted horizontal
line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN
tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota
(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing
nations is associated with trade liberalization Notably this finding holds even though we include
a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy
variable Dit by γDEMp
14
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
existing studies do Thus the conclusion that on average democracies impose lower tariffs than
non-democracies appears to be robust
It is important to note however that the HS2 results reveal significant heterogeneity in the
effects of democracy across industries Visual inspection of Figure 3 showcases several industries
for which the effect of democracy differs notably from the posterior mean of the main effect β
(the dotted line) Animal Vegetable and Food products have positive estimates for the
majority of industries in these sectors the effects turn out to be statistically significant (marked
by black vertical lines) That is democracies are more protective of agricultural consumer goods
than non-democracies are This finding is consistent with agricultural protection in democracies
that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998
Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that
democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and
Metals industries have estimates that are significantly lower than the main effect suggesting that
democracies engage in deeper liberalization of these industries For comparability with Milner and
Kubota (2005) we conduct further analysis with only less-developed countries and we find similar
results for the average effect of democracy across all products see Figure C1 in Appendix C
Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-
anced product categories for example there are only two HS2 chapters for MachElec (ma-
chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that
when considering HS4 product groups as opposed to HS2 groups the estimated main effect of
democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially
increases We find stronger effects of democratic political institutions on agricultural protection at
the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or
chilled (0707) - stronger in fact than any other products We also find much stronger evidence
that democracies are more protective of textile products On the other hand intermediate goods
tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates
for the products that belong to Minerals Metals and MachElec This is consistent with
Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average
tariff rate of 1003
15
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
0102
03
04
05
06
07
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
25 2627
28
29
30
31
32
33
34
353637
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
5455
56
57
5859
60
61 62
63
64
65
6667
6869
70
71
72
73
74
75
7678
79
80
81
8283
84
85
86
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box
intra-industry trade make it relatively easier for countries to liberalize intermediate goods than
finished products Our findings add nuance to this claim by showing that such effects are even
more pronounced among democracies holding economic size comparative advantages institutional
memberships and trade volumes constant
Finally we present our estimates from the HS 6-digit level across three distinct set of granular
products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-
tial variations in estimated effects even across similar products within the narrowly defined meat
industry We find that consumer products such as beef pork and lamb tend to be more protected in
democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that
there are industries such as cotton that exhibit relatively lower variation across products although
we see some differences between generic vs specialized cotton products
We conduct a more systematic analysis of the difference between consumption goods and in-
16
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
0201
1002
0120
0201
3002
0210
0202
2002
0230
0203
11 0203
1202
0319
0203
2102
0322
0203
2902
0410
0204
2102
0422
0204
2302
0430
0204
41 0204
4202
0443
0204
5002
0500
0206
1002
0621 02
0622
0206
2902
0630
0206
4102
0649
0206
80 0206
9002
0711
0207
1202
0713
0207
1402
0721
0207
2202
0723
0207
2402
0725
0207
2602
0727
0207
3202
0733
0207
3402
0735
0207
3602
0739
0207
4102
0742
0207
4302
0750
0208
1002
0820
0208
3002
0840
0208
5002
0890
0209
00 0210
1102
1012
0210
1902
1020
0210
9002
1091
0210
9302
1099
beef
frozen beef
pork
sheep
horse
edible animal organs
poultry
fatty livers
fatty livers
other meat
seasoned meat
minus4
minus2
0
2
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(a) HS2 chapter 02 Meat products
5201
0052
0210
5202
9152
0299
5203
0052
0411
5204
1952
0420
5205
1152
0512
5205
1352
0514
5205
1552
0521
5205
2252
0523
5205
2452
0525
5205
2652
0527
5205
2852
0531
5205
3252
0533
5205
3452
0535
5205
4152
0542
5205
4352
0544
5205
4552
0546
5205
4752
0548
5206
1152
0612
5206
1352
0614
5206
1552
0621 52
0622
5206
2352
0624
5206
2552
0631
5206
32 5206
3352
0634
5206
3552
0641
5206
42 5206
4352
0644
5206
4552
0710
5207
9052
0811
5208
1252
0813
5208
1952
0821
5208
2252
0823
5208
2952
0831
5208
3252
0833
5208
3952
0841
5208
4252
0843
5208
4952
0851
5208
5252
0853 52
0859
5209
1152
0912
5209
1952
0921
5209
2252
0929
5209
3152
0932
5209
3952
0941
5209
4252
0943
5209
4952
0951
5209
5252
0959
5210
1152
1012
5210
1952
1021
5210
2252
1029
5210
3152
1032
5210
3952
1041
5210
42 5210
4952
1051
5210
52 5210
5952
1111
5211
1252
1119
5211
2152
1122
5211
2952
1131
5211
3252
1139
5211
4152
1142
5211
4352
1149
5211
5152
1152
5211
5952
1211
5212
1252
1213
5212
1452
1215
5212
2152
1222
5212
2352
1224
5212
25
generic cotton
specialized cotton
minus2
minus1
0
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS2 chapter 52 Cotton products
Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products
7301
1073
0120
7302
1073
0220
7302
3073
0240
7302
9073
0300
7304
1073
0420
7304
2173
0429
7304
31 7304
3973
0441 73
0449 73
0451
7304
5973
0490
7305
1173
0512
7305
1973
0520
7305
3173
0539 73
0590
7306
1073
0620
7306
3073
0640
7306
5073
0660
7306
9073
0711
7307
1973
0721
7307
2273
0723
7307
2973
0791
7307
9273
0793
7307
9973
0810
7308
20 7308
30 7308
4073
0890
7309
0073
1010
7310
2173
1029
7311
0073
1210
7312
9073
1300
7314
1173
1412
7314
1373
1414
7314
1973
1420
7314
3073
1431
7314
3973
1441
7314
4273
1449
7314
5073
1511 73
1512
7315
1973
1520
7315
8173
1582
7315
8973
1590
7316
0073
1700
7318
1173
1812
7318
1373
1814
7318
1573
1816
7318
1973
1821
7318
2273
1823
7318
2473
1829
7319
1073
1920
7319
3073
1990
7320
1073
2020
7320
9073
2111
7321
1273
2113
7321
8173
2182
7321
83 7321
9073
2211
7322
1973
2290
7323
1073
2391
7323
92 7323
9373
2394
7323
9973
2410 73
2421
7324
2973
2490
7325
1073
2591
7325
9973
2611
7326
1973
2620
7326
90
sheetpiling
railwayrails
switch blades
tubespipestubepipefittings
buildingstructures
tankscasks
containers
wire
steelminuscontainingcloths
chainsanchors
nailstacks
screwsbolts
washersrivets
sewingneedles
springs
steel appliances
other
Trump 20182020 tariffs
Intermediate goods
Consumption goods
minus2
minus1
0
1
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)
17
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
termediate goods based on the differences across steel products To do this we create a mapping
from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based
on main end use (eg consumption vs intermediate) We find that democracies are more likely to
liberalize intermediate goods within the steel industry (shaded in blue box) However we find that
consumption goods such as steel appliances (shaded in red box) tend to receive higher protection
Interestingly we find that the set of steel products on which the Trump administration imposed
high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to
liberalize
Our findings are consistent with Betz and Pond (2019) who show that democracies are more
likely to protect consumer goods than non-democracies While it is well understood that consumers
incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-
ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer
interests do not get translated into trade policy-making in democracies especially when it comes to
consumer goods More generally our findings call into question the validity of the key assumptions
made in the literature that individual preferences (given by their factoral or sectoral interests) are
key determinants of actual trade policy-making Taken together the significant variability across
products suggest that the current empirical understanding of how political institutions interact with
underlying set of preferences of actors in economy is incomplete at best
32 Dyadic Analysis
Does the interaction of regime type between trading partners affect the depth of trade liberalization
We make three contributions to the analysis of this question First we directly analyze trade policies
between country-pairs rather than using a proxy measure such as trade volume The standard
gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade
which includes barriers to market access between the trading partners By using applied tariffs
as the dependent variable therefore our analysis returns more direct estimates of the relationship
between regime type and the choice of trade policy Second we distinguish the direction of trade
policy between importing and exporting countries A direct test of the hypothesis that pairs of
democracies are more likely to engage in liberalization requires researchers to examine the interactive
effect in two directions (1) whether a democratic importer is more likely to liberalize when its export
18
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can
achieve freer market access when its negotiating import partner is a democracy instead of a non-
democracy Finally we investigate heterogeneity across industries The findings from the monadic
analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of
political institutions as well as by political pressures that vary across interest groups Consequently
we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific
political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine
the complexity of bilateral trade policy outcomes across industries
321 Methodology
We employ a difference-in-differences identification strategy Specifically we examine the industry-
specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral
Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after
FTAs between dyads with different regime types
Our proposed hierarchical linear model for the change in trade policy before and after an FTA
between importer i and exporter j is given by
∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt
+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)
where p indexes product at the chosen level of disaggregation The proposed model in equation (5)
distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity
IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt
and DNONDEMNONDEMijt are defined similarly
For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff
reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-
tively This accounts for the possibility of anticipation effects as well as phase-in periods that are
prevalent in trade agreements To minimize excessive extrapolation into the future we focus on
the comparison of tariff rates immediately before and after each trade agreement by setting L = 1
19
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t
ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year
tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log
population and log GDP in year t Zijt represents dyad-level covariates including logged total trade
volume between the two countries log of the partner-specific mean tariff imposed by the importer
across all industries whether at least one of the pair is a major power whether both parties were
GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-
thermore to account for the fact that democracies might have lower underlying tariff rates to begin
with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each
product p Finally ξp is an product-specific intercept As in the monadic analysis we model the
prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp
]to be Normally distributed
γp sim N (φk[p]Σγ)
φk sim N (0Σφ)(6)
The quantities of interest are the differences in the degree of trade liberalization between demo-
cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is
a democracy and the other is not)
E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βDEMNONDEM + γDEMNONDEMp (7)
E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM
ijt ] = βNONDEMDEM + γNONDEMDEMp (8)
where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is
a democracy and the partner is not and equation (8) compares a dyad with two democracies to a
mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-
tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently
fit the HS4-level model12
11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check
the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is
significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots
20
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
322 Empirical Results
We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements
Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two
parties to the agreement and in which both parties are sovereign states to have a conceptually
cleaner analysis of interactions between regime type We therefore include agreements such as
the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA
for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44
are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties
are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies
and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in
Appendix B13
Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-
stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade
negotiations are not a random sample from the population of all possible dyads and therefore
we emphasize that our estimand is not the difference in tariff reduction between the population
of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff
reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This
interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the
formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes
in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade
agreements
In order to make a direct comparison between our analysis and prior research we estimate an
undirected version of equation (5) of the form
∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt
+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt
+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)
13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes
revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We
periodically check the underlying databases for changes and will update our analysis as the data are refreshed
21
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
HS4HS2
minus10
minus05
00
05
larrD
eepe
r re
duct
ions
S
hallo
wer
red
uctio
nsrarr
Undirected
HS4
HS2
HS4
HS2
Democratic ImporterNonminusDemocratic Exporter
NonminusDemocratic ImporterDemocratic Exporter
Directed
Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would
which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM
ijt ] = βMIXED +
γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic
pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents
our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find
any difference in the applied tariff rates of mixed dyads and democratic dyads14
To shed light on this finding we decompose the direction of trade liberalization among FTA
partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of
the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4
levels data respectively First we examine whether democratic importers are able to engage in
deeper trade liberalization when their counterpart is a democracy rather than a non-democracy
This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former
engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given
the small number of FTAs involving non-democracy pairs in our data
22
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff
reductions when their export partner is a non-democracy rather than a democracy the magnitude is
substantially larger when disaggregating to the HS4 level Second we consider whether democratic
exporters can achieve better market access when their import partner is a democracy or a non-
democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)
in the panel we find that tariffs reductions are smaller when non-democratic importers are paired
with democratic exporters than when democratic importers are paired with democratic exporters
Again the magnitude is larger with more granular tariff data These results suggest that the finding
in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions
than mixed dyads might be due to the fact that non-democratic importers give shallower concessions
to democratic exporters than democratic importers give to democratic exporters
Finally we explore the complex bilateral strategic incentives among FTA partners at granular
levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in
panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)
of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions
than pairs of democracies The estimated effects are significant for agriculture and metals industries
with very small overall differences across industries This pattern is even more prominent with HS4-
level results whereby most HS4 estimates are negative and statistically significant Similarly panel
(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff
reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4
industries Importantly unlike our monadic analysis we do not find significant heterogeneities when
we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively
constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This
suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level
interests rather than highly heterogeneous interest of individual producers15
It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South
Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation
Korean government formed several task forces organized around broad industry groupings within relevant agencies
such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare
23
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243 44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 6364656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 8889 90 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 04 05
06 07 08 09 10 11 121314 15
16171819202122 2324
25 26
27
28 29 3031 32 3334353637 38
39 40
41 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 7172 73 74 75 76 787980 81 82 83
84 85
86 878889
90 91 92939495 96 97
minus25
minus20
minus15
minus10
minus05
00
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(a) Democratic Importer Non-Democratic Exporter
0102
03 0405
06 07 08 09 10 11 121314 15
161718192021222324
25 2627
28 29 3031 32 3334353637 38
39 4041 4243
44 454647 48 49
50 51 52 53 54 55 565758 59 60 61 62 63
64656667
68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85
86 87 888990 91 92939495 96 97
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
TransportMiscellan
minus05
00
05
Harmonized System 2minusdigit
larrD
eepe
r
Sha
llow
errarr
0102 03 0405
06 07 08
09
10 11
12
1314
15
161718
19
2021
22
23
24
2526
27
28
2930
31
32
33
3435
36
37 38
3940
41 42
43
44
4546
4748
49
50
5152
53 5455
56
57
58
59
60 6162
63
6465
6667
68
69
7071
72
7374
75
76
78
79
80
81
82
8384 85
86
87
888990 91 929394
95
96
97
minus05
00
05
10
Harmonized System 4minusdigit
larrD
eepe
r ta
riff r
educ
tions
Sha
llow
er ta
riff r
educ
tions
rarr
(b) Non-Democratic Importer Democratic Exporter
Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)
24
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
causal effects of democratization on bilateral trade liberalization across industries We leave for
future research this challenging task of investigating how exactly changes in political institutions
translate into trade policy outcomes
4 Concluding Remarks
In this paper we present a novel dataset with nearly 6 billion observations of product-level applied
tariff rates that countries apply to their trading partners incorporating the universe of preferential
rates and the Generalized System of Preferences To do so we combine and augment existing
datasets available from the WTO and UNCTAD and we resolve conflicting information between
the two Our dataset lays an important empirical foundation for investigating trade politics at a
much more granular level than has previously been done
To illustrate the importance of product-specific tariff measurement we examine an enduring
question in international political economy whether there are systematic differences in trade policy
between countries with different regime types Consistent with prior work we find that democra-
cies have lower tariff rates than non-democracies on average However focusing on the country-
or sector-level average elides substantial heterogeneity We document that democracies are more
protective than non-democracies for many industries such as agriculture and textile industries
Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-
ization We find significant variations across products even within the same industry Specifically
we show that consumer products are more likely to be protected in democracies whereas intermedi-
ate goods receive much deeper liberalization compared to the average level across other products
Our findings demonstrate various instances of aggregation bias with the use of increasingly granular
units of analysis
Our data also allows researchers in the fields of international and comparative political economy
to track fine-grained temporal changes in partner-specific trade policy In particular we examine
whether interactions between regime types at the dyad level results in differences in the degree of
bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences
design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a
mixed pair with a democracy and a non-democracy However by disentangling who is the recipient
of concessions we show that the difference between democratic pairs and mixed pairs is due in
25
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
large part to shallower concessions granted by non-democratic importers to democratic partners
but not vice-versa Put another way democratic importers grant trade concessions to democratic
exporters and non-democratic exporters but democratic exporters win more trade concessions
from democratic importers than from non-democratic importers Future studies would benefit from
investigating these empirical findings with a theoretical focus on the direction of trade liberalization
Our dataset can be combined with industry-level covariates such as import and export concen-
tration as well as country-specific industry structures to further explore linkages between political
institutions and industry-level trade policies It can also be used to study the increasing complex-
ity of product-level trade policy that affects the deepening of global supply chain and production
networks In addition as other scholars have pointed out there exists significant variation in insti-
tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and
Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to
assemble are likely to result in different configurations of demands for trade protection Research
into the relationships between political institutions and trade policies continues to be relevant as
policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-
ing trade agreements in response to pressures from their constituents The question is not so much
whether there will be more or less liberalization but rather which products and industries will be
most exposed to a review of trade policies This article presents findings that usefully contribute
to this research agenda
26
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press
Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank
Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340
Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289
Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323
Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt
Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600
mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http
peopletamuedu~apondresearchhtml
Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press
Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37
Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309
Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress
27
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159
Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331
Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850
Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551
Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212
Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47
Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233
Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20
Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming
Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384
Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244
Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654
Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406
Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749
28
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321
mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513
Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430
Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985
Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143
Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661
Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329
Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102
Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press
Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297
Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier
Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222
Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292
Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press
29
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Appendix A Tariff-line Dataset
A1 Bilateral Tariff Data Collection and Processing
A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System
There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present
To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2
Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000
A11 WTO IDB Duty Collection and Processing
We perform the following steps to collect and process IDB duties
Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg
We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013
16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg
1
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Year Imp Code Full description Type Reported rate
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585
Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg
Year Imp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates
Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
We find that both IDB duty types stipulate South Africa as a beneficiary
A12 UNCTAD TRAINS Duty Collection and Processing
Likewise we perform the following corresponding steps for TRAINS tariffs
Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg
17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg
2
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)
Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary
Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem
A2 Tariff Merging Algorithm
A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad
3
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers
The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps
Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg
Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
02 497 centskg + 197 (asymp 197)
2013 USA ZAF 62011330ldquoGeneralduty raterdquo
80 529 centskg + 585 (asymp 585)
In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values
Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg
Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)
2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo
002 497 centskg + 197 (asymp 2122)
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
051 00
Since there is only a single preferential duty we select it as the best TRAINS candidate
4
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg
Year Imp Exp Code Original description Final applied rate Source
2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo
497 centskg + 197 (asymp 197) IDB
2013 USA ZAF 62011330
ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo
00 TRAINS
Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line
The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year
18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size
5
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Year
JPNUSABRAPHLARGCHECANKORCOLNZL
NORAUSMEXZAFIDN
CHLBOLEUNURYIND
EGYPAKNIC
ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI
MDGISL
PERLKA
GTMTURBGDDOM
ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV
PNGJORBFAMACHKGSENBENZMBNERATG
MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ
MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI
KNAZWEGABGUYVCTUKR
FJIRUSAREMDV
HTICPVAGOKHMGHAVUTSUR
DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE
WSMYEMSYCAFGLBRLIE
PRT
Impo
rter
WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)
Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail
6
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Appendix B List of Bilateral FTAs
Table B1 List of Bilateral Free Trade Agreements
Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009
Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008
Table B1 Continued on next page
7
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012
Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009
Table B1 Continued on next page
8
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012
9
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Animal
dem demdem nondemnondem dem
Vegetable FoodProd Minerals
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Fuels Chemicals PlastiRub HidesSkin
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Wood TextCloth Footwear StoneGlas
10 0 10years from agreement
00
05
10
15
20
25
30
aver
age
duty
acro
ss p
rodu
cts
(log
)
Metals
10 0 10years from agreement
MachElec
10 0 10years from agreement
Transport
10 0 10years from agreement
Miscellan
Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure
10
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
Appendix C Alternative Specifications and Diagnostics
01
02
03
04
05
0607
0809
1011
12
13
14
1516
17
18
1920
2122
23
24
25
26272829
30
31
32
33
3435
36
37
3839
40
41
42
43
44
45
46
47
48
4950
5152
53
5455
56
57
58
59
60
6162
63
64
65
6667
68
6970
71
72
7374
75
76
78
79
80
81
82
8384
85
86
87
88
89
Animal
Vegetable
FoodProd
Minerals
Fuels
Chemicals
PlastiRub
HidesSkin
Wood
TextCloth
Footwear
StoneGlas
Metals
MachElec
Transport
minus10
minus05
00
05
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
LDC only
Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate
11
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2829
30
31
3233
34
3536
37
38
39
4041
42
43
44
45
46
47
48
4950
5152
53
545556
57
5859
60
6162
63
64
65
6667
68
6970
71
72
73
74
75
76
78
79
80
81
8283
84
85
86
87
88
89
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
minus3
minus2
minus1
0
1
2
Harmonized System 2minusdigit Industry
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
All WTO
(a) HS 2-digit level
AnimalVegetable
FoodProdMinerals
FuelsChemicals
PlastiRubHidesSkin
WoodTextCloth
FootwearStoneGlas
MetalsMachElec
Transport
01
02
03
04
05
0607
08
09
10
11
12
13
14
15
1617
18
19
20
21
22
23
24
2526
27
28
29
30
31
32
33
34
3536
37
38
39
40
41
42
4344
45
46
47
48
49
50
51
52
53
54
55
56
57
5859
6061 62
63
64
65
66
67
68
69
70
71
72
73
74
75
7678
79
80
81
8283
84 8586
87
88
89
minus4
minus2
0
2
4
Harmonized System 4minusdigit Product
larrD
emoc
raci
es h
ave
low
er ta
riffs
Dem
ocra
cies
hav
e hi
gher
tarif
fsrarr
(b) HS 4-digit level
Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does
12
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-
02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus03
00
03
minus05
00
05
10
00
05
10
15
iteration
(a) HS 2-digit (HMC)
0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)
500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
00
04
08
12
00
05
10
minus10
minus05
00
iteration
(b) HS 4-digit (HMC)
020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)
200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000
minus01
00
01
02
minus06
minus03
00
minus15
minus10
minus05
00
iteration
(c) HS 6-digit (VB)
Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification
13
- Title Page
- 1 Introduction
- 2 New Database Bilateral Product-level Applied Tariffs
- 3 The Effects of Regime Type on Trade Policy
-
- 31 Monadic Analysis
-
- 311 Methodology
- 312 Empirical Results
-
- 32 Dyadic Analysis
-
- 321 Methodology
- 322 Empirical Results
-
- 4 Concluding Remarks
- References
- A Tariff-line Dataset
-
- A1 Bilateral Tariff Data Collection and Processing
-
- A11 WTO IDB Duty Collection and Processing
- A12 UNCTAD TRAINS Duty Collection and Processing
-
- A2 Tariff Merging Algorithm
-
- B List of Bilateral FTAs
- C Alternative Specifications and Diagnostics
-