capital structure and product markets interactions evidence from business cycles

8/9/2019 Capital Structure and Product Markets Interactions Evidence From Business Cycles

1/26

Journal of Financial Economics 68 (2003) 353378

Capital structure and product markets

interactions: evidence from business cycles$

Murillo Campello

Department of Finance, University of Illinois at Urbana-Champaign, 340 Wohlers Hall,1206 South Sixth Street, IL 61820, USA

Received 18 December 2000; received in revised form 28 February 2002

Abstract

This paper provides firm- and industry-level evidence of the effects of capital structure on

product market outcomes for a large cross-section of industries over a number of years. The

analysis uses shocks to aggregate demand as surrogates for exogenous changes in the product

market environment. I find that debt financing has a negative impact on firm (relative-to-industry) sales growth in industries in which rivals are relatively unlevered during recessions,

but not during booms. In contrast, no such effects are observed for firms competing in high-

debt industries. At the industry level, markups are more countercyclical when industry debt is

high. The cyclical dynamics I find for firm sales growth and for industry markups are

consistent with Chevalier and Scharfsteins (American Economic Review (1996)) prediction

that firms which rely heavily on external financing are more likely to cut their investment

in market share building in response to negative shocks to demand and that the competitive

outcomes resulting from such actions depend on the financial structures of their industry

rivals.

r 2003 Elsevier Science B.V. All rights reserved.

JEL classification: G32; D43; E32

$This paper is a revised version of Chapter I of my doctoral dissertation at the University of Illinois. I

thank the members of my dissertation committee: Charlie Calomiris, Charlie Kahn, George Pennacchi,

and Mike Weisbach. I am indebted to Charlie Hadlock, Bill Schwert (the editor), and two anonymous

referees for their valuable suggestions. Comments from Naveen Khanna, Peter MacKay, Vojislav

Maksimovic (WFA discussant), Gordon Phillips, and Allen Poteshman were also very helpful. I also

benefited from comments of seminar participants at Arizona State University, Case Western Reserve

University, College of William & Mary, Federal Reserve Bank of Cleveland, Michigan State University,

Penn State University, Purdue University, SUNY at Stony Brook, Western Finance Association 2000meetings, University of Illinois at Urbana-Champaign, and University of Utah. All remaining errors are

mine.

E-mail address: [email protected] (M. Campello).

0304-405X/03/$ - see front matter r 2003 Elsevier Science B.V. All rights reserved.

doi:10.1016/S0304-405X(03)00070-9


2/26

Keywords: Product markets; Capital structure; Business cycles; Industry rivalry; Markups

1. Introduction

Recent theoretical work on capital structure has emphasized not only the

contracting problems among agents within the firm (e.g., managers and investors),

but also the implications of financing decisions for agents outside the firm (e.g.,

competitors and consumers). This literature stresses that a firms mode of financing

influences its conduct in the product market as well as the conduct of other market

participants, thereby influencing competitive outcomes.1 Although these theories

have received much attention, only a few of their implications have been empiricallyexamined. Directly testing those ideas is a challenging task since it is difficult to

establish whether or not the competitive outcomes associated with a firms financing

decisions were already anticipated by the firms managers. Another concern with the

interpretation of the empirical relation between capital structure and product market

behavior is the possibility that both a firms financial structure and its competitive

performance may be jointly influenced by unobserved (or unmodeled) factors arising

from the market environment. A study of competitive performance following capital

structure changes, for instance, may assign a spurious causality to capital structure if

both performance and recapitalizations are influenced by underlying trends in

industry concentration, excess capacity, and growth (see Kovenock and Phillips,1997; and Zingales, 1998).

One way to mitigate concerns about the endogenous nature of the relation

between capital structure and competitive performance is to look at performance

following events affecting product market participants incentives (or ability) to

compete under existing financing arrangements. This paper uses such an approach to

establish a link between firm financing and product market outcomes. Assuming that

the exact timing, magnitude, and consequences of macroeconomic shocks are not

fully anticipated by product market participants and their financiers, I examine the

sensitivity of markups and sales growth to financial leverage following shocks to

aggregate demand.2 My main focus, however, is on the differences in these

1A partial list of seminal theoretical papers includes Benoit (1984), Titman (1984), Maksimovic (1986,

1988), Brander and Lewis (1986), Bolton and Scharfstein (1990), and Maksimovic and Titman (1991).

More recent papers are Williams (1995), Chevalier and Scharfstein (1996), Dasgupta and Titman (1998),

and Maurer (1999).2 In some of my tests, I allow for the possibility that firms could endogenize macroeconomic effects in

their financing decisions. Notice, however, that we do not know whether (and how) firms would account

for those effects in their capital structure. The scant literature on this issue supports that even if

macroeconomic expectations could be incorporated in financing decisions one would still observe the side

effects of debt. That literature suggests, for example, that optimal financial structures are unfeasible

owing to practical issues such as tax-induced (and other institutional) biases that prevent firms from using

their capital structure as an aggregate risk-sharing mechanism (see, e.g., Gertler and Hubbard, 1993) and

lack of coordination among agents (Lamont, 1995).

M. Campello / Journal of Financial Economics 68 (2003) 353378354


3/26

sensitivities across industries in which rivals are more debt-financed and industries

in which rivals are less debt-financed. Such intertemporal and cross-sectional

contrasts help sidestep the problems that usually affect the interpretation of the

empirical relation between capital structure and competitive performance. Inparticular, after conditioning on the phase of the business cycle and on competitors

leverage, my tests show that firm debt leads to sales underperformance in some

industries in certain states of demand realization, whereas under other sets of

circumstances these debt-led losses are either nonexistent or reversed. While such

findings are consistent with extant theories on the effects of capital structure on

competitive performance, the contrasts I use make it difficult to identify alternative

stories (or biases) capable of explaining how debt financing could lead to poor

performance in some specific industries and periods precisely along the lines of the

results I report.

I present two sets of results characterizing the dynamics of the interactions

between capital structure and product market outcomes. First, I use industry-level

data to conduct a direct test of Chevalier and Scharfsteins (1996) theory of markup

cyclicality. Consistent with their model, I find that markups are more countercyclical

in industries in which firms use more external financing. My estimates suggest that

the markup of a hypothetical all-debt industry increases by approximately 42%

more than that of a zero-debt industry in response to a 1% decline in gross

domestic product (GDP). My conclusions about debt-led markup countercyclicality

are robust to the inclusion of controls for the level of industry capacity utilization,

market concentration, and product demand cyclicality, among other estimationconsiderations.

I then use a panel data set containing quarterly information from firms in 71

industries covering over two decades to study the impact of debt financing on sales

performance at the firm level. My empirical strategy focuses on the differences in

responses of firm salesleverage sensitivity to macroeconomic shocks across lowdebt

and highdebt industries. The results I obtain show that reliance on debt financing

can significantly depress a firms (relative-to-industry) sales growth in industries in

which rivals are less leveraged as economic conditions worsen. Comparing the

performance of two firms in a low-debt industry, one firm with a debt-to-asset ratio

10% above the industry average and the other with a debt-to-asset ratio 10% belowthat average, I find that the industry-adjusted quarterly sales growth of the more

indebted firm is 1.3% lower than that of its unlevered rival following a 1% decline in

GDP. In contrast, no such effects are observed in industries in which rivals are

relatively more indebted just prior to a similar aggregate shock. These results are

robust to changes in the estimation methodology, to changes in the proxies for

macroeconomic activity, to the inclusion of further macroeconomic conditioning

variables, and also hold across different subperiods of my sample. My conclusions

are also robust to concerns that the results I observe could be driven by the potential

endogeneity between the financing decisions of firms in a particular industry and that

industrys product demand cyclicality.I interpret my findings on cross-industry differences in cyclical movements in

markups and in sales-debt sensitivities as consistent with the theory ofChevalier and

M. Campello / Journal of Financial Economics 68 (2003) 353378 355


4/26

Scharfstein (1996). Specifically, my findings agree with the prediction that firms

which rely more heavily on external financing are more likely to reduce their

investment in market share building during downturns and that the competitive

outcomes resulting from such actions are jointly determined by the firms and by itsrivals capital structures. My firm-level results, however, cannot dismiss a state-

contingent form of Telsers (1966) long purse argument as they suggest that,

conditional on a low demand realization, indebted firms lose more (less) market

share when rivals are relatively unlevered (leveraged).

This study adds to the evidence on interactions between financial structure

and product markets presented in the pioneering work of Chevalier (1995a, b),

Phillips (1995), and Kovenock and Phillips (1997). The papers empirical strategy

is most similar to that of Chevalier and Scharfstein (1996), Zingales (1998),

and Khanna and Tice (2000), who look at these interactions during periods when

the competitive environment is affected by exogenous events (e.g., oil shocks,

market deregulation, and entry). While those studies report time- and industry-

specific results, this paper shows that the argument that a firms financial

structure influences its competitive performance is considerably more pervasive

and economically significant than previously thought. To my knowledge, this

is the first study to present evidence of a link between capital structure, product

markets, and business cycles, allowing for both firm-level and macroeconomic

inferences.3

The rest of the paper is organized as follows. Section 2 provides an overview of

some of the main arguments relating capital structure and product markets, and itdiscusses testable hypotheses. In Section 3, I examine whether industry-wide use of

debt financing influences the behavior of markups over the business cycle. In Section

4, I use firm-level data to examine the relation between firm capital structure and

product market performance via a testing strategy that emphasizes cross-sectional

differences in changes in the sensitivity of sales performance to debt following

macroeconomic shocks. Section 5 concludes the paper.

2. Financial structure and product markets: some theories and hypotheses

The goal of this paper is to empirically examine the argument that capital

structure influences a firms (as well as its rivals) incentives to compete in the

product market, thereby influencing competitive outcomes. Unfortunately, few of

the theoretical studies proposing this argument make clear predictions about how

their results would manifest in the data. Moreover, only some of those predictions

can be simultaneously tested under a single empirical framework, making it difficult

to dismiss one argument in favor of another. In the subsequent analysis, I study

whether the ideas presented in this brief review section are economically significant

3Subsequent work by Campello and Fluck (2003) provides detailed evidence of these linkages using

firm-level data from manufacturers during the 199091 recession.



5/26

and pervasive using fairly general empirical models relating debt financing to

product market outcomes.4

One of the prominent theories in the literature about the interactions

between capital structure and product markets is the long purse or predationargument. The basic story, put forth by Telser (1966) and formalized by Bolton

and Scharfstein (1990), stresses that dependence on outside financing can hinder a

firms ability to fight competition, which in turn prompts financially un-

constrained rivals to pursue predatory market strategies. In empirically

examining this argument, researchers often take some measure of debt financing

as a proxy for financial fragility and study whether that measure is associated with

competitive outcomes such as diminished profits and market share losses. The testing

strategy used in this paper goes a step further in that it only takes indebtedness to be

a measure of financial fragility under a set of specific conditions: (1) the firm is

highly-leveraged in relation to its industry rivals, (2) the firm operates in a market

where the use of debt financing is generally low, and (3) there is a negative shock to

demand.

Financial fragility may also hinder competitiveness in models in which firms

use pricing as a means of investing in long-term market share building instead of

as a means of maximizing single-period profits (e.g., Gottfries, 1991; Chevalier

and Scharfstein, 1996). Chevalier and Scharfstein argue that liquidity-

constrained firms are less inclined to invest in market share building when

their liquidation probability is high. In recessions, externally financed firms

will be more inclined to boost price-cost margins at the expense of futuresales.5 Since prices are strategic complements, their rivals will also inflate

prices, but by an extent given by their own finances. As those rivals also raise

their prices, the market share losses of the more constrained firms are reduced. It

follows that the degree of firm markup cyclicality should depend both on the firms

own financial constraints as well as on the financial status of rivals in the industry. At

the industry level, markups should be less cyclical if firms use more external

financing.

An alternative line of argument suggests that debt could contribute to firm

competitive performance. According to the strategic commitment theory

of Brander and Lewis (1986), a firms decision to use debt works as a commitmentto more aggressive behavior in the product market. This threat is credible because

of the option-like payoffs associated with debt under limited liability and induces

the firms unlevered rivals to accommodate by reducing their output (also

see Maksimovic, 1986). If ex post this dynamic is found to work in the data,

4Related theoretical papers not reviewed here include Titman (1984), Rotemberg and Scharfstein

(1990), and Maksimovic and Titman (1991). See Maksimovic (1995) for a detailed review of the theoretical

literature.5Firms can do so if switching costs allow them to extract more of their customers reservation utility

without necessarily losing their business in the short run. See Dasgupta and Titman (1998) and Campello

and Fluck (2003) for models with a similar flavor.



6/26

then one should observe market share gains for the more leveraged firms.

One difficulty in testing this idea, however, is that the availability of such gains

should ex ante lead to a prisoners dilemma-like behavior in which rival firms would

use more debt.Finally, debt financing could be related to performance in a noncausal sense. In

the industry equilibrium models of Maksimovic and Zechner (1991) and Williams

(1995), the optimal set of financial contracts for each firm in a industry is jointly

determined with various industry characteristics, such as the number of firms,

riskiness of projects, and technologies. Williams proposes a stable industry

equilibrium supporting the coexistence of profitable firms that are large, make more

fixed-capital investments, and use more debt, with smaller firms that are less

profitable and use low debt. While my experiments are not designed to formally test

this theory, I briefly discuss whether my results could be ascribed to such an

equilibrium-type story.6

3. Industry-level evidence on markup movements over the business cycle

Chevalier and Scharfstein (1996) predict that if most firms in an industry are

externally (internally) financed, then the industry markup will be more counter-

cyclical (procyclical). In this section, I use industry-level data to test this hypothesis.

The results presented here are relevant in their own rightextending the analysis in

Chevalier and Scharfstein (1995, 1996)but they are also useful for the firm-leveltests conducted in Section 4. Specifically, I explore the cyclical cross-industry

differences in the relationship between competitive outcomes and financial structure

that are identified in this section to address the potential for estimation biases in my

firm-level tests.

3.1. Data and methodology

Detailed industry-level data are typically compiled by governmental agencies and

are published at the two-digit standard industry classification (SIC) level. To

perform the tests of this section, I gather data from 20 two-digit SIC classified

manufacturing industries. These industries are listed in Table 1 below. The

most important data series I need for my experiments are industry markup and

leverage.

There is no consensus in the literature about the best measure of industry markup.

Because the proxy proposed by Bils (1987) is less likely to overstate the degree of

markup countercyclicality (see Rotemberg and Woodford, 1991), and thus less likely

to allow for unwarranted conclusions about the impact of capital structure on

markups over the cycle, I choose that proxy for my tests. In contrast to other

6See MacKay and Phillips (2002) for a detailed empirical examination of industry-equilibrium theories.



7/26

Table1

Industry

summarystatistics.

Industriesaredefinedatthetwo-digitSICcodelevelandthesampleperiodis1984:IV1996:IV.

Leverageiscomputed

quarterly

astheas

set-weightedaveragedebt-to-assetratiousingfirm-leveldatafrom

COMPUSTAT.

Capacityutilizat

ionisavailablefromtheFederal

Reserves

Statistica

lReleaseG.1

7.

Four-firmconcentrationratiosaretakenfromTable2inRotembergandSaloner(1986).Thedurableversusnondurabled

ichotomy

isfromS

harpe(1994).

SIC

Industrydescription

Le

verageratio

Capacity

utilization(%)

Four-firm(%)

concentrationratio

Sharpes(1994)d

ichotomy

Mean

(Std.

dev.)

Mean

(Std.

dev.)

20

Foods

0.26

0.03

83.3

8

0.90

3

4.50

Nondurabl

e

21

Tobacco

0.27

0.03

N/A

N/A

7

3.60

Nondurabl

e

22

Textilemills

0.37

0.04

87.2

1

3.16

3

4.10

Nondurabl

e

23

Apparel

0.22

0.05

81.4

3

2.38

1

9.70

Nondurabl

e

24

Lumber

0.31

0.03

85.5

9

4.28

1

7.60

Durable

25

Furniture/fixtures

0.23

0.05

80.8

8

3.14

2

1.60

Durable

26

Paper

0.28

0.03

89.4

6

2.02

3

1.20

Nondurabl

e

27

Printing/publishing

0.24

0.02

84.2

1

3.76

1

8.90

Nondurabl

e

28

Chemicals

0.18

0.01

80.4

3

2.31

4

9.90

Nondurabl

e

29

Petroleum

0.15

0.01

88.2

9

3.49

3

2.90

Nondurabl

e

30

Rubberandplastics

0.25

0.05

86.1

0

2.94

6

9.10

Durable

31

Leather

0.20

0.07

78.7

9

4.92

2

4.50

Nondurabl

e

32

Stone,clay,andgas

0.35

0.06

78.3

1

3.15

3

7.40

Nondurabl

e

33

Primarymetals

0.22

0.02

83.6

7

6.49

4

2.90

Durable

34

Fabricatedmetals

0.27

0.02

78.8

1

3.35

2

9.10

Durable

35

Industrialmachinery

0.16

0.02

78.9

3

5.38

3

6.30

Durable

36

Electricalmachinery

0.15

0.02

81.3

2

3.63

4

5.00

Durable

37

Transport.

Equipment

0.26

0.09

76.3

4

3.72

5

0.10

Durable

38

Instruments

0.19

0.03

N/A

N/A

4

7.80

Nondurabl

e

39

Misc.manufactures

0.20

0.06

75.5

9

3.81

N/A

Durable

N/A:No

tavailable.



8/26

markup measures, Bilss proxy allows for the marginal cost of one hour of labor to

change with the level of hours worked in the industry.7 This is relevant for tests using

data from manufacturers since hours worked vary over the business cycle and firms

in this sector are required by law to pay a premium for overtime hours. Similarly toother markup proxies, however, Bilss measure might do a relatively poor job in

capturing productivity shifts over the cycle.8

The measure of industry markup I operationalizea slightly simplified version of

Bils (1987) is computed in terms of log deviations from trend:

Markupt ln Pt=MCt

ln Pt ln wt

ln 1 rV0 Ht

ln Ht jtNt=Yt

trend constant; 1

where P is price, w is hourly average wages, r is the legal overtime premium (set to

50% by the Fair Labor Standards Act of 1938), H is weekly average hours, V(H) is

overtime hours, j is the number of weeks in the period, N is the number of

production workers, and Y is gross output. The change in overtime hours with

respect to a unit change in average hours, V(H), equals one when the weekly average

hours of work in the industry exceeds 40 hours and equals zero otherwise.

To construct the markup series, I gather industry price data, P, from the Bureau of

Labor Statistics (BLS) Producer Price Indexes. N, H, and w are taken from the

BLSs National Employment, Hours and Earnings. These publicly available series are

compiled on a monthly basis, and I use their quarterly averages. Y is collected from

the Bureau of Economic Analysis and is available on an annual basis only. To obtainY at a quarterly frequency, I interpolate each one of the industry series with

quarterly gross national product (GNP). I can only match all of these series

beginning in the fourth quarter of 1984.9

To measure industry leverage, on a quarterly basis, I compute the asset-weighted

average long-term debt-to-asset ratio of firms with identical two-digit SIC codes in

COMPUSTATs Primary, Supplementary, Tertiary, Full Coverage, and Research

tapes (all at book values). The sample period for the tests in this section is from

1984:IV through 1996:IV. Summary statistics for various industry characteristics are

reported in Table 1.

7The production technology is of the form Y=Haf(factors other than H), where Y is output and H is

hours worked. If the cost of labor is a function of hours worked, W(H), the cost of marginally increasing

hours, holding all other inputs, including employment, N, at their cost-minimizing values (indicated by *),

is then:

MC dCosts=dH

dH=dY Y; H; etc: 1=a

H=Y

dCosts=dH

W H

W0 H

H

1=a

HN=Y

:

Marginal costs include straight-time wages, movements along the marginal wage schedule, and a

productivity term.8As noted by a referee, this is a potentially important limitation. Some of the results below should thus

be interpreted with caution.9The exceptions are SICs 36 and 38, for which pre-1987 data are discarded because of the well-known

difficulties in reconciling information from the 1972 and the 1987 SIC code structure.



9/26

To test whether markups are more countercyclical in industries in which rivals are

more externally financed, I regress industry markup movements on a measure of

aggregate activity, industry leverage, and a term interacting activity and leverage.

Macroeconomic movements are proxied by changes in the negative of log real GDP.Increases in this measure represent economic slowdowns, the consequences of which

I want to emphasize in this paper. To minimize simultaneity between markup and

leverage, the latter variable enters the specification in lag form. The empirical model

can be written as

Markupi;t Z a DLog GDP t

bLeveragei;t1

l Leveragei;t1 DLog GDP t

ei;t: 2

In all estimations of this model, I use HuberWhite heteroskedasticity-consistent

errors, allowing for within-period error clustering (see Rogers, 1993).

3.2. Results

The first estimation reported in Table 2 examines whether markups are

countercyclical, as shown in previous studies in the macroeconomics literature. I

simply regress markup movements on the proxy for aggregate activity and a constant

via ordinary least squares (OLS). The results point to markup countercyclicality in

my data. The estimates suggest that a 1% decline in GDP brings about a 10%

increase in (relative-to-trend) industry price-cost margins. This result is statistically

significant at the 1% level.In the next set of regressions, I include industry-level variables. Since I cannot

reject the presence of industry-fixed effects via standard tests, I estimate Eq. (2) with

data in first differences (see Hsiao, 1986). The OLS-fixed effects (OLS-FE) estimates

displayed in the second row of Table 2 are the most important results of this section.

The significantly positive coefficient for the interaction term indicates that higher

industry-wide indebtedness leads markups to increase faster during downturns.

Consistent with Chevalier and Scharfstein (1996), these estimates suggest that

negative shocks to demand prompt firms to raise price-cost margins more (or cut

them less) in industries with more externally financed competitors.

To gauge the macroeconomic implications of these results, note that estimates inthe second row of the table imply that the markup of a hypothetical all-debt

industry would increase by nearly 42% more than that of a zero-debt industry in

response to a 1% decline in GDP. This finding points to feedback effects going from

the corporate sector to the overall economy, with potential policy implications. For

instance, this evidence suggests that past firm financing decisions may become a

source of inflationary pressures during recessions.

3.3. Interpretation and robustness

A potential problem with the interpretation of the results reported in the secondrow of Table 2 is that the business cycle measure I use may not distinguish between

changes in markups that are caused by demandshocks from those that are caused by



10/26

Table2

Industry

debtandmarkupmovementsoverthebusinesscycle.ThedependentvariableisbasedonBilss(1987)

measureofindustrymarkupmovement[see

Eq.

(1)].Industriesaredefinedatthetwo-d

igitSICcodelevelandthesampleperiodis1984:IV1996:IV.

Estim

ationsincludeaconstantterm(c

oefficients

omitted).

Themeasureofcyclicalityisthen

egativeofthelogrealGDP.

Leverageiscomputedquarterlyforeach

industryastheasset-weightedave

ragedebt-

to-assetratiousingfirm-levelbookdata.C

apacityutilizationistheindustry

quarterlyproductionasapercentageofnormalproduction.

Costshocksare

proxiedb

ythechangeinfuelandenergyp

riceindextoindustrycommodity

PPI.Thedurables-nondurablesdichotomyisfromSharpe(1994).T

hehigh

four-firm

concentrationdummyisbasedon

datafromRotembergandSaloner

(1986).

Fixedeffects(FE)estimationseliminateindustryeffectsbydifferencing

thedata.

Two-stageleastsquares(2SLS)areusedtocontrolforbiasescause

dbydifferencingifleverageisonlyweaklyexogenous.Thesetofin

struments

includestwolagsofleverage(inlevels).All

estimationsuseHuber-Whitehete

roskedasticity-consistentt-statisticsallowingforwithin-perioderrorclustering.

Estimato

r

Dlog(GDP)t

Leveragei

,t1

Levi

,t1

[Dlog(GDP)

t

]

DEnergy

price

t

Capacity

utili,

t1

Durables

dum

my

Concentration

dummy

Adj-R

2

Obs.

1.OLS

10.0

764

0.046

924

(5.8

36)

2.OLS-F

E

6.1456

1.0453

146.8777

0.035

924

(3.5

42)

(2.856)

(2.7

12)

3.OLS-F

Ea

6.9705

1.0727

137.7774

0.9345

0.051

832

(4.0

47)

(2.999)

(2.5

87)

(3.259)

4.OLS-F

Eb

7.0551

1.0447

128.4153

0.0047

0.055

842

(4.1

03)

(4.638)

(3.2

48)

(0.709)

5.OLS-F

E

6.1507

1.0479

146.6471

0

.0016

0.031

924

(3.9

05)

(3.266)

(2.7

54)

(1

.120)

6.OLS-F

E

6.0988

1.1362

151.0229

0.0015

0.033

924

(3.7

17)

(3.441)

(2.7

64)

(1.070)

7.2SLS-FE

9.0880

0.9587

237.9982

0.036

859

(4.8

29)

(1.433)

(2.4

81)

aExclu

desSICs35and37.

bExclu

desSICs21and38.



11/26

cost shocks. For instance, if following a supply side shock such as an increase in

energy prices, GDP falls and unemployment increases, then markups could appear

to be countercyclical even when changes in aggregate demand are of second order.

Following Rotemberg and Woodford (1996) and Ghosal (2000), I control for costshifts in the estimation of Eq. (2) by adding a proxy for changes in energy prices.

This variable is measured as the change in the ratio of the fuel price index to the

producer price index (PPI) of industrial commodities. As noted by Ghosal, changes

in energy prices will affect not only costs but also demand in some industries. Using

that authors approach, I therefore exclude industries that are likely to fall in this

category (SICs 35 and 37) from the estimation. The results for the new specification,

shown in the third row of Table 2, support my previous conclusions.

Although I design my tests to mitigate concerns with various types of estimation

biases, admittedly, I am unable to obtain bias-free results. In turn, I examine some of

the most compelling cases for biases potentially affecting my conclusions about debt-

led markup countercyclicality.

One could argue, for instance, that the level of utilized capacity may influence

markup movements and also help explain countercyclicality through leverage if debt

is used to finance expansions. In response, in an alternative specification to Eq. (2) I

add lagged industry capacity utilization as a control. The data on industry capacity

utilization I use are quarterly averages of monthly series taken from the Federal

Reserves Statistical Release G.17 (data for SICs 21 and 38 are not available).

Including this variable is particularly useful in that it may also capture other

underlying trends in the industries sampled. Results displayed in the fourth row ofTable 2 show that my conclusions are unaffected by the inclusion of that extra

control.

Alternatively, suppose that sales are more sensitive to the level of aggregate

activity in some industries than in others and that firms in cycle-sensitive industries

choose (ex ante) to carry low debt. In this case, too, markups could fall more in less

leveraged industries during recessions. But if this story drives my results, I would

expect the evidence of debt-led markup countercyclicality to disappearor, at least,

the magnitude of the coefficients to be affectedonce I explicitly account for cycle-

sensitivity in my empirical specification. Following the approach used by Chevalier

and Scharfstein (1995) in addressing a similar concern, I add an indicator variablefor cycle-sensitive industries to Eq. (2). This dummy is based on the dichotomy

proposed by Sharpe (1994), which groups industries according to the covariance

between their sales and the GNP. The set of high covariance industries includes all of

the durable goods industries (except SICs 32 and 38) plus SIC 30. I refer to these

industries as durables. In the fifth row of Table 2, I report the results obtained

after including the durables dummy. The size and significance of the debt-cycle

coefficients suggest that my results cannot be explained away by that alternative

story.

Another explanation for why price-cost margins may rise in recessions is that firms

in noncompetitive markets may be more likely to collude when gains from deviationstrategies are not promising (see Rotemberg and Saloner, 1986; and Rotemberg and

Woodford, 1991). Chevalier and Scharfsteins (1996) theory of financial structure



12/26

determining markup movements over the cycle questions this view. But if

concentration and debt are positively correlated, then my results could be reflecting

that collusion story, and I would be too aggressive in rejecting the null of capital

structure irrelevance. To address this concern, I modify Eq. (2) by including adummy variable based on the same four-firm concentration ratios used by

Rotemberg and Saloner. I assign to the high concentration dummy those

industries with ratios above the median ratio in Table 2 of their paper. The estimates

in the sixth row of Table 2 below suggest that capital structure explains industry

markup cyclicality despite differences in market concentration.

Finally, because data differencing can bias the OLS estimator if regressors are only

weakly exogenous, I perform a two-stage stage least squares (2SLS) estimation of

Eq. (2) in which leverage is instrumented by two lags of itself, with the interaction

term being modified accordingly.10 The 2SLS estimates in the seventh row ofTable 2

strongly support my previous conclusions about debt-induced markup counter-

cyclicality. In unreported regressions, I use changes in the unemployment rate as the

proxy for cycles, obtaining qualitatively similar results. The same applies when I use

log changes in prices (rather than markups) as the dependent variable in Eq. (2).

These robustness checks make it difficult to dismiss the evidence of a link between

corporate financing and markup cyclicality stemming from debt-induced behavior in

product markets.

4. Firm-level evidence on market share building over the business cycle

The endogeneity of financing decisions at the firm level makes it difficult to

identify a causal relationship in which the direction goes from capital structure to

competitive performance. Any testing of the theory on product market and capital

structure interactions has to explicitly accommodate the need to address concerns

with endogenous and other estimation biases. Unfortunately, this is not an easy task.

One way to minimize such concerns is to look at competitive performance following

exogenous events altering both the product market environment and rival firms

incentives (or ability) to compete under an existing set of financing arrangements (see

Zingales, 1998). The firm-level tests of this section use such an approach.In identifying events suitable for a my testing strategy, I consider a number of

issues. First, there must be an exogenous, real-side shock to the competitive

environment. Second, the shock must allow for unanticipated effects of financial

structure. In other words, the nature of the shock should make it difficult for firms

and their financiers to fully endogenize all of its consequences in their financing

arrangements (contracts). Third, the event should not be industry-specific, but rather

affect a large cross-section of industries at the same time. This last requirement

generates useful cross-sectional contrasts for the relation between a firms

performance and its rivals finances.

10Following Arellano (1989), I instrument the differenced regressors with their lagged levels.

Estimations with longer and more complex lag structures yield qualitatively similar results.



13/26

I find that periods surrounding shocks to aggregate demand provide the

conditions needed for my proposed tests. In particular, consider the post-war

recession episodes in the United States. These have been typically accompanied by

contractionary measures by the monetary authority as well as tougher credit termsfor borrowers.11 Since debt is likely to bind firm behavior in periods when income

from operations is unusually low, and because it is difficult to renegotiate debt

contracts when credit is tight, those concurrent macroeconomic movements should

provide for crisp results on the relation between capital structure and product

markets. Specifically, concerns with simultaneity problems affecting the interpreta-

tion of the empirical relation between firm performance and capital structure will be

lessened around these exogenous shocks to demand and credit: The shocks will affect

all firms in a market, but according to the theory, those firms should respond

differently to the shocks depending on their own financing and on the financing of

their rivals. I examine these cross-sectional differences and their intertemporal

evolution in the tests that follow.

4.1. Data

I set out to put together firm-, industry-, and macro-level data for a period

long enough to capture a number of shocks to aggregate activity. Issues concerning

data availability and consistency lead me to restrict the analysis to the 1976:

I1996:IV period. The firm-level data (all inflation-deflated) are obtained from

the COMPUSTAT tapes. Using the quarterly tapes, I retrieve book data ontotal assets; long-term debt; sales; and plant, property, and equipment (PPE). I use

a firms primary three-digit SIC code to identify its product market, restricting

the analysis to firms in the 200-399 SIC range (manufacturing). Following

Clarke (1989), I discard firms assigned to SICs that fail to identify economically

meaningful markets.12 Firms with negative equity or with sales growth

exceeding 200% in one quarter are also discarded. Finally, because my

estimations use industry-adjusted data, only industry-quarters with a minimum

of ten firms are retained in the sample. After these screening criteria are imposed, I

obtain a panel of 128,133 firm-quarters distributed across 71 three-digit SIC

industries. The industry and macro-level data come from various sources and aredescribed below.

4.2. Measuring firm performance

In examining the interactions between product markets and capital structure,

empirical research has often linked price-setting behavior with some aspect of debt

11Evidence on such co-movements in financial and real macroeconomic variables is provided in

Bernanke and Blinder (1992), Christiano, et al. (1996), and Lown and Morgan (2001), among others.12Some firms may be too diversified or lack enough counterparts to make up a three-digit industry.

These firms are assigned to three-digit codes ending with zero. Other codes combine miscellaneous and

not elsewhere classified businesses and seem unfit for inclusion in my tests. See Clarke (1989) and Kahle

and Walkling (1997) for more details.



14/26

financing (see, e.g., Chevalier, 1995b; Phillips, 1995; Chevalier and Scharfstein,

1996).13 In some settings, pricing decisions should reasonably reflect how a firms

financial status affects its competitive behavior. More generally, though, firms can

implement a number of alternative policies that significantly affect product marketoutcomes but that may not be reflected in how they price their products. Examples of

such policies are decisions about fixed investments, research and development

expenditures, and use of promotions. One way to build a practical measure of

performance that incorporates information from the combined effects of pricing and

other market strategies is to look at changes in the firms share of its industry sales.

In this section, I use a firms relative-to-industry sales growth (which I improperly

refer to as market share growth) to gauge its performance in the product market.

This proxy can be seen as a suitable compromise for the proposed analysis: It can be

consistently measured across many industries and periods, but it is too general to pin

down the particular mechanisms contributing to competitive behavior in each of the

markets sampled.

4.3. Estimation methodology

4.3.1. A two-step procedure

The estimation methodology I use is very similar to that of Kashyap and Stein

(2000) and Campello (2002). The idea is to estimate the effect of macro-level shocks

on micro-level (firm) sensitivities via a two-step procedure that combines cross-

sectional and times series regressions. The approach sacrifices estimation efficiencybut reduces the likelihood of Type I inference errors; that is, it reduces the odds of

concluding that capital structure matters when it does not.14

The initial step of the two-stage estimation consists of a purely cross-sectional

regression that is meant to produce a time-specific estimate of the sensitivity of

product market performance to financial structure. At each quarter, t, I regress the

log change of firm is sales on four lags of itself, four lags of the log change in PPE,

the lagged log of total assets, and the lagged leverage ratio:

DLog Sales i;t ZX

4

k1akDLog Sales i;tk

X4

k1

bkDLog PPE i;tklLog Assets i;t1

dLeveragei;t1 ei;t: 3

13Exceptions are Kovenock and Phillips (1997) and Zingales (1998), who study product market

outcomes such as plant closings and market exit.14The alternative one-step regressionwith Eq. (4) nested in Eq. (3)would impose a more constrained

(efficient) parametrization. However, results from tests of coefficient stability show that the data strongly

reject parameter constancy across different time periods and industry groups, weakening the case for

efficiency. An added advantage of the approach used is that it relaxes the restriction of homogeneous error

structure across time.



15/26

Lags of sales growth are included to control for firm-specific characteristics that

contribute to performance over time. Changes in PPE and total assets are also

included as controls on grounds that the leverage coefficient may be biased if the

model fails to control for investment spending (which might have been financed withdebt) or for the portion of leverage that is explained by a higher borrowing capacity.

My focus is on the estimate of the sensitivity of sales growth to leverage (d). To help

reduce the potential for reverse causality between sales performance and debt, I

compute leverage as the ratio of the book value of long-term debt to total assets.

This measure best suits my tests for at least two reasons. First, in contrast to market

values, book values of debt are less sensitive to capital markets assessments about

future cash flows. Second, while leverage changes are likely to reflect changes in

expectations about future product market outcomes, long-term leverage levels are

more likely to reflect the cumulative effect of past decisions.

Before estimating Eq. (3), I adjust all observations of the dependent variable for

their industry-specific sales-weighted means, so that this variable measures the firms

sales growth relative to that of its competitors.15 The right-hand side variables are

also industry-adjusted, and thus the average rivals leverage is the metric used to

measure a firms indebtedness. These adjustments facilitate the interpretation of the

regression results and minimize the potential for biases induced by unobserved

industry effects. After each run of Eq. (3), the estimated leverage coefficient is

collected and stacked in a time series vector, dt.

In the second stage, I regress the resulting time series of first-stage coefficients

that is, the dt vectoron four lags of a measure of economic activity that proxies forthe worsening of macroeconomic conditions (DActivity), plus a constant and a time

trend:

dt Z X4

k1

fkDActivitytk gTrendt ut 4

I use two alternative proxies for negative shocks to activity in all estimations

performed: (1) the negative of the change in log real GDP, and (2) the change in the

unemployment rate ( 100).16 Because my tests should help distinguish between real

and financial explanations for the observed relation between sales growth and debt, Ialso check whether the proxy for economic activity retains significant predictive

power after conditioning on other macroeconomic factors. To do so, I estimate

15The COMPUSTAT tapes do not contain data on the entire universe of firms. The tapes usually lack

data on very small industry players, and thus the market share figures for the firms in my sample are likely

to be slightly inflated. These data coverage limitations should not systematically bias my conclusions in

any particular direction.16To see how this procedure accounts for the error contained in the first step, assume that the true dt

*

equals what is estimated from the first-step run (dt) plus some residual (et): dt*=dt+et. One would like to

estimate Eq. (4) as dt*=Z+Xb+ot, where the error term would only reflect the errors associated with the

specification of the model. However, the empirical version of Eq. (4) uses dt (rather than dt*) on the right-

hand side. Consequently, so long as E[X0e]=0, Z will absorb the mean ofet, while ut will be a mixture ofetand o t. That is, the measurement errors of the first step will increase the total error variance in the second

step, but will not bias the coefficient estimates in b.



16/26

alternative versions of Eq. (4) that include additional macro variables: (1) the

Federal Reserve funds rate (Fed funds), (2) the spread between the rates paid on six-

month commercial paper and 180-day T-bills (paper-bill spread), and (3) the

consumer price index (CPI).17 These series are obtained from the Federal ReservesStatistical Release H.15 and from the BLS. In a bivariate version of Eq. (4), I add

the change in the Fed funds rate to the right-hand side as a proxy for the stance of

monetary policy. In a multivariate version of Eq. (4), I also include the changes in

the paper-bill spread and changes in the CPI.

To gauge the economic relevance and the statistical significance of shocks to

demand on the sensitivity of sales growth to debt, I compute both the sum of the

coefficients for the four lags of the measure of economic activitythat is, the Sfks

in Eq. (4)as well as the p-value of this sum. If feedback effects are not too large,

that sum yields a good first-order approximation of the impact of the right-hand side

variables of interest. I also report p-values for the rejection of the hypothesis that the

four lags of the measure of economic activity do not help forecast the sensitivity of

sales growth to debt. To assess the impact of debt on sales growth over a longer

horizon (four-quarters ahead), I also estimate a version of Eq. (3) in which the

dependent variable is computed as [Log(Sales)t+3Log(Sales)t1]/4. To ensure the

robustness of my results, at each pass, I estimate each of the two versions of Eq. (3)

both via OLS and 2SLS. In all estimations, I use Newey and Wests (1987)

heteroskedasticity- and autocorrelation-consistent standard errors.

4.3.2. Cross-industry comparisons

Although the use of exogenous shocks to the competitive environment such as

recessions help lessen concerns about endogeneity in the performance-capital

structure relation, the estimations are not free from simultaneity biases arising from

unobservable firm characteristics that could influence both a firms debt before the

recession begins and its sales performance afterwards. Fortunately, the theories

considered provide for a way of addressing this problem. Those theories propose

that the extent to which a firms financing influences its competitive performance

should be a function of the financial status of its competitors, and thus a firms

performance-capital structure relation should differ across industries depending onrivals finances. Accordingly, in the tests of this section I compare estimates of firm

sales-debt sensitivities from industries in which rivals are relatively more leveraged

with those from industries in which rivals are less leveraged.18 Focusing on

differences in responses of sales-debt sensitivities to macroeconomic shocks across

17The paper-bill spread serves as a proxy for the economy-wide perceived risk of default (see, e.g.,

Bernanke, 1990). Korajczyk and Levy (2003) find that this spread has predictive power over firms

target leverage.18Recall, the results in Section 3 already indicate that rivals finances influence product market

dynamics. Those results implied less competition in the price-setting dimension during recessions when the

average industry competitor is more indebted. The interpretation was that an indebted firm facing a

recession is able to increase prices more without loosing customers when its rivals are also leveraged and

are therefore expected to behave in the same way.



17/26

sets of firms in different industry debt categories will render irrelevant the concerns

that my two-step estimated coefficients could be biased in the levels.19

Implementing a test that emphasizes those contrasts is a straightforward task with

my two-step procedure. At each quarter, t, I rank the sample industries according totheir leverage ratios and assign firms in the bottom and top quintiles of that industry

ranking to low-debt and high-debt industries, respectively. Next, also on a

quarterly basis, I estimate Eq. (3) separately for firms competing in low-debt

industries and for firms in high-debt industries as of the previous period (t1),

saving the estimated leverage coefficients from each of these cross-sectional

regressions in individual time series vectors. These series are used as the dependent

variables in separate estimations of Eq. (4), and the results from these estimations

are then compared.

To facilitate comparisons of results across the industry groups, I report not only

the individual estimated responses of sales-debt sensitivities to macro shocks for

firms in the low-debt and in the high-debt groups, but also the differences in these

responses. Standard errors for the difference coefficients are estimated via a

seemingly unrelated regression (SUR) system that combines the two industry groups

(p-values reported).

4.4. Results

Fig. 1 provides descriptive evidence of remarkably distinct dynamics for the

relation between a firms sales performance and its capital structure conditional onrivals finances and on the state of aggregate demand. The figure shows the cyclical

behavior of the first-step estimated sales-debt sensitivities (dt) of low- and high-debt

industries for the 1977:II1996:IV period; where I isolate the cyclical from secular

movements in dt using the Hodrick and Prescott (1980) decomposition. Notice that

the sales-debt sensitivity of firms in low-debt industries trails below its trend twice:

During the downturns of the early 1980s, and in a more prolonged period that starts

with the 199091 recession. The cyclical behavior of high-debt industries dt is

markedly different.

Table 3 presents estimates of the impact of negative shocks to macroeconomic

activity on firm sales-debt sensitivities for both low- and high-debt industries and for

different combinations of first- and second-step specifications. Panel A displays

results from models in which the dependent variable is the series of first-step dts

estimated via OLS, while results in Panel B are for dts estimated via 2SLS. Across

rows, the top half of each panel features the negative of the change in log real GDP

as the proxy for shocks to activity, while in the bottom half the change in the

unemployment rate is the relevant measure of change in aggregate conditions. Each

panel displays six pairs of estimated responses of the one-quarter ahead dt to shocks

to economic activity [i.e., the Sfks in Eq. (4)] along with the associated SUR results

19This is true unless there exists some reason why those biases should be systematically more

pronounced in certain industries than in others precisely along the lines of the sample partition I use. I

consider one such possibility in detail below.



18/26

for differences in responses across industry groups (i.e., the low-debt Sfks minus the

high-debt Sfks). To conserve space, I only report the difference coefficients for the

four-quarters ahead dts.

Focusing first on the results for the one-quarter dts of Panel A, note that for each

of the regression pairs the impact of the macro variable of interest is negative for

firms operating in low-debt industries. These coefficients are statistically significant

at the 2% level in all but one of the estimations. In contrast, for firms in high-debt

industries, none of the coefficients is significantly different from zero, with most of

them returning a positive sign. The differential impact of economic downturns onsales-debt sensitivities across high- and low-debt industries reported in Table 3 (see

the Low High columns) is negative for all regression pairs. The differences are

statistically significant at the 2.9% level or better in 9 of the 12 cases. Somewhat

stronger results are reported in Panel B, which uses 2SLS-estimated dts. These

findings suggest that, following negative shocks to activity, highly leveraged firms

lose market share in industries in which rivals are relatively unleveraged, but that

those losses are reversed in booming periods. In stark contrast, such competitive

dynamics are not observed in high-debt industries.

The exclusion test p-values indicate that changes in the state of the economy are

particularly relevant in predicting the sensitivity of firm sales growth to leverage inlow-debt industries. For multivariate estimations, the business cycle measures have

marginal predictive power at the 4% level or better in all cases. This shows that

-0.12

-0.08

-0.04

0.00

0.04

0.08

0.12

Jun

-77

Jun

-7

8

Jun

-7

9

Jun

-80

Jun

-81

Jun

-82

Jun

-83

Jun

-84

Jun

-85

Jun

-86

Jun

-87

Jun

-88

Jun

-89

Jun

-90

Jun

-91

Jun

-92

Jun

-93

Jun

-94

Jun

-95

Jun

-96

Time

t

Low-debt industr ies High-debt industr ies

Fig. 1. This figure shows the cyclical behavior of the first-step estimated sales-debt sensitivity of firms in

low-debt and those in high-debt industries over the 1977:II1996:IV period. Quarterly estimates of the

sales-debt sensitivities are obtained from the estimation of Eq. (3) in the text via 2SLS. The estimations are

performed separately over the two industry subsamples with the coefficients returned for leverage (d) saved

into two separate times series. Firm-quarters are assigned to either low- and high-debt industries

conditional on whether their three-digit SIC defined industries were ranked in the bottom or in the top

quintile of the industry-wise leverage debt-to-asset ranking, respectively, as of the quarter preceding the

estimation of Eq. (3). Cyclical and secular movements in dt are isolated using the Hodrick and Prescott

(1980) decomposition (bandwidth of 1600).



19/26

Table 3

Sales-debt sensitivities and the business cycle: two-stage estimator. The dependent variable is the sensitivity

of sales growth to leverage [estimated either via ordinary least squares (OLS) or two-stage least squares

(2SLS)] for firms operating in industries categorized as low- versus highly-leveraged. In each estimation,

the dependent variable is regressed on four lags of a measure of aggregate economic activity, a constant,

and a time trend. In the bivariate regressions, the change in the Fed funds rate is added. In the multivariate

regressions the changes in Fed funds rate, the change in the CPI, and the change in the spread between the

rates on commercial paper and the six-month T-bill are included. The sample period is 1976:I1996:IV.

The sum of the coefficients for the four lags of the economic activity measure is shown along with the p-

value for the sum. Exclusion test rows report the p-values for the rejection of the hypothesis that the four

lags of the economic activity measure do not forecast the sensitivity of sales growth to debt.

Heteroskedasticity- and autocorrelation-consistent errors are computed with a Newey and West (1987)

lag window of size four in the one-quarter ahead regressions. The standard errors for the difference of the

sum of the four lags of the economic activity measure are computed with a seaming unrelated regression

(SUR) system that estimates industry group regressions jointly.

One-quarter d Four-quarters d

Low-debt High-debt Low High Low High

Panel A: Dependent variable is the OLS-estimated sensitivity of sales growth to leverage

1. Economic activity measure is the (negative) change of log real GDP

Univariate Sum of coefficients 4.753 1.637 6.390 15.016

Summation test (p-value) 0.009 0.466 0.015 0.004

Exclusion test (p-value) 0.099 0.588

Bivariate Sum of coefficients 6.345 1.610 7.955 19.084



Multivariate Sum of coefficients 6.300 1.835 8.135 19.546Summation test (p-value) 0.001 0.416 0.002 0.001


2. Economic activity measure is the change in the rate of unemployment ( 100)







Multivariate Sum of coefficients 11.420 0.643 10.777 44.666

Summation test (p-value) 0.003 0.917 0.137 0.006Exclusion test (p-value) 0.000 0.818

Panel B: Dependent variable is the 2SLS-estimated sensitivity of sales growth to leverage

1. Economic activity measure is the (negative) change of log real GDP










2. Economic activity measure is the change in the rate of unemployment ( 100)



20/26

weakening of economic activity alone predicts the increase in the sales-debt

sensitivity beyond what other macroeconomic factors would predict.

4.5. Economic interpretation

Thus far I have mostly emphasized the statistical significance of my results. Of

course, the most interesting question is whether the estimates I report implyeconomically meaningful magnitudes. I describe alternative strategies to measure the

product market consequences of financial structure implied by my firm-level

estimates in turn.

In order to gauge the economic significance of the results in Table 3, first consider

the relative performance of two hypothetical firms competing in a low-debt industry.

Assume that these firms are equal in all dimensions except capital structure: One firm

has a debt-to-asset ratio 10% above the industry average ratio, while its rival has a

debt-to-asset ratio 10% below that average. Now consider the estimate of the impact

of four lags of the change of the log real GDP on the sales-debt sensitivity of firms in

low-debt industries in the multivariate specification of Panel A (=6.3). Thisestimate implies that the industry-adjusted sales growth of the more indebted firm is

expected to be nearly 1.3% (=0.063 0.2) lower than that of its unlevered rival

following a 1% decline in GDP. This difference is nontrivial considering that in a

regression of quarterly industry sales growth on four lags of changes in GDP for

low-debt industries over the 197696 period I find that total industry sales fall by

1.2% following a 1% decline in GDP.

One can also use the results from the same specification to draw inferences about

the sales growth of similarly highly-leveraged firms competing in industries with

significantly different average debt levels. In this case, too, the imputed effect of the

shock to activity is noteworthy. Four quarters following a 1% decline in GDP, thequarterly sales growth of the firm in the low-debt industry should underperform that

of its high-debt industry counterpart by 0.8%.










Table 3 Continued

One-quarter d Four-quarters d

Low-debt High-debt Low High Low High



21/26

In sum, my firm-level tests show that highly-leveraged firms lose market share in

recessions (but not in booms) only in industries in which rivals are relatively

unleveraged. Recall from Section 3 that markups also fall more in low-debt

industries during recessions. These firm- and industry-level cyclical dynamics seemconsistent with Chevalier and Scharfsteins (1996) argument that financially

constrained firms invest less in market share in low states and that the extent to

which market share is lost depends on the financial status of their industry rivals.

Also note that the firm-level results of this section cannot dismiss a state-contingent

form of the long purse story, as they suggest that unconstrained firms capture market

share from their indebted rivals in low-demand states. Finally, the nature of the time

and cross-sectional variations in performance-debt relation makes it difficult to

ascribe my findings to the industry equilibrium theories discussed in Section 2.

4.6. Robustness

To check that my conclusions do not hinge on any particular choice of lag

structure, I also estimate the above two-step procedure using two, six, and eight lags

of the right-hand side macroeconomic variables, as well as with the inclusion of

contemporaneous (lag 0) observations. All of these specifications yield qualitatively

similar results (unreported). The same applies when I use other alternative measures

of monetary policy. My conclusions also hold for various subperiods examined in

the 19761996 span.

My testing strategy focuses on the responses of sales-debt sensitivities tomacroeconomic shocks. I choose to emphasize the observed cross-sectional

differences in those responses (rather than their levels) because of concerns with

underlying characteristics that could influence both a firms debt before the recession

and its performance afterwards. On theoretical grounds, I partitioned the sample

industries according to their debt usage. However, one could argue that if product

demand is more sensitive to the level of aggregate activity in some markets than in

others, then the average debt in industries with higher (lower) demand cyclicality

should be lower (higher).20 Now suppose costs of financial distress are higher in

more cycle-sensitive industries andthat the costs associated with financial distress for

the more leveraged firms in a industry include market share losses (see Opler andTitman, 1994). Then one could observe more negative sales-debt sensitivities for high

leverage firms in low-debt industries relative to that of similarly leveraged firms in

high-debt industries following a decline in activity just because low-debt industries

are more cycle-sensitive. This alternative story suggests that my industry leverage-

based partitions could be inadvertently picking up differences in cycle sensitivities

across industries.

To verify the extent to which that story could affect my conclusions, I need to

check whether my findings on differences in sales-debt sensitivities across low- and

high-debt industries remain after controlling for cross-industry differences in the

20This argument supposes that the cyclicality of product demand might be an industry-wide

characteristic.



22/26

cycle-sensitivity of product demand. The proxy for industry demand cycle-sensitivity

I employ follows from Sharpe (1994), who classifies industries into two groups

(durables and nondurables) based on the covariance between industry sales and

the GNP (see Section 3).21 I use Sharpes dichotomy as a conditioning variable in mytests. The idea is as follows. Suppose the alternative story is true. Then,

unconditional comparisons between the coefficients for Sfks across low- and

high-debt industries would just reflect the magnitude of the differences in the

responses of sales-debt sensitivities to aggregate shocks across cycle-sensitive and

cycle-insensitive industries. Consider in turn partitioning the sample according to

whether the observations belong to cycle-sensitive or cycle-insensitive industries and

subsequently performing the two-step procedure separately for each of the

subsamples. Then, under that same competing story, the conditional comparisons

across low- and high-debt industries should yield much smaller (perhaps insignif-

icant) magnitudes for the estimated differences in Sfks.

Results from the multivariate version of Eq. (4) where the dependent variable is

the one-quarter dt are shown in the first (durables) and the second (nondurables)

rows of each of the two panels of Table 4. Consistent with my earlier findings, in

both subsamples, sales growth responds negatively to leverage following a negative

aggregate shock in industries in which rivals carry low debt, but not in high-debt

industries. In the durables regressions, the coefficients for the low-debt equation are

negative and statistically significant at the 3% level or better in all cases. These same

coefficients are also negative but less statistically significant in the nondurables

regressions. As before, all coefficients in the high-debt equations are insignificantlydifferent from zero. Most important, all of the difference coefficients in Table 4

retain the right sign, and their magnitudes are very similar to those displayed in

Table 3, demonstrating that the case for cycle-sensitivity of demand influencing

industry debt does not alter my conclusions.22

5. Concluding remarks

Recent theoretical research has proposed that capital structure can affect firm

performance in the product markets because financing arrangements can alter afirms incentives (or ability) to compete. In this paper, I investigate this claim by

examining the impact of debt financing on industry markups and on firm sales

growth using data from a large cross-section of industries over a number of years.

My tests show that debt has a negative impact on relative-to-industry firm sales

growth in industries in which rivals are relatively unlevered during recessions, but

21The durables (high GNP-sales growth covariance) subsample contains 75,423 firm-quarters in 36

different three-digit SICs, while the nondurables (low covariance) subsample contains 80,360 firm-quarters

in 35 industries.22A referee suggests yet another scenario: Both the industry average leverage and a firms deviation from

that average (i.e., its choice of a niche) may be correlated with the cycle-sensitivity of demand.

Admittedly, tackling this argument would be more difficult absent a firm-level measure of demand cycle-

sensitivity.



23/26

Table4

Sales-deb

tsensitivitiesandthebusinesscyc

le:subsamples.Thedependentvariableistheone-quartersensitivity

ofsalesgrowthtoleverageratio[estimated

eithervia

ordinaryleastsquares(denoteit

bydOLS)ortwo-stageleastsquare(d2SLS)],

forfirmsoperatinginindustriescategorizedaslow-vers

ushighly-

leveraged

.Ineachestimation,thedependen

tvariableisregressedonfourlags

ofameasureofaggregateeconom

icactivity,

fourlagsofthechangeintheFed

fundsrat

e,fourlagsofthechangeintheCPI,fourlagsofthechangeinth

espreadbetweentheratesonco

mmercialpaperandthesix-monthT-bill,a

constant,

andatimetrend.

Thesampleperiodis1976:I1996:IV.

Thesumofthecoefficientsforthefourlagsoftheeconomicactivitymeasureisshownalong

withthep

-valueforthesum.

Exclusiontest

rowsreportthep-valuesforthere

jectionofthehypothesisthatthefourlagsoftheeconomicactivitym

easuredo

notforecastthesensitivityofsalesgrowthtodebt.Heteroskedasticity-andautocorrelation-consistenterrorsarecomputedwithaNeweyandWest

(1987)lag

windowofsizefour.Thestandarderrorsforthedifferenceofthesumofth

efourlagsoftheeconomicactiv

itymeasurearecomputedwitha

seemingly

unrelated

regression(SUR)systemthatestimatesindustrygroupregressionsjointly.DurablesandnondurablesindustrydefinitionsarebasedonSharpe

(1994).

dOLS

d2SLS

Low-debt

High-debt

Low

High

L

ow-debt

High-debt

Lo

w

High

PanelA:

Economicactivitymeasureisthe(

negative)changeoflogrealGDP

1.Durables

Sumofcoefficients

7.606

0.713

6.894

5.824

1.022

6.846

Summationtest(p-value)

0.002

0.804

0.054

0.012

0.728

0.088

Exclusiontest(p-value)

0.016

0.155

0.156

0.145

2.Nondu

rables

Sumofcoefficients

2.085

3.846

5.931

2.745

5.474

8.219


0.227

0.214

0.107

0.093

0.182

0.056


0.108

0.098

0.099

0.111

PanelB:

Economicactivitymeasureisthechangeintherateofunemployment

(

100)

1.Durables

Sumofcoefficients

15.8

78

6.400

9.478

8.641

3.095

5.546


0.000

0.438

0.342

0.030

0.740

0.628


0.000

0.421

0.076

0.311

2.Nondu

rables

Sumofcoefficients

3.936

5.571

9.508

6.857

13.3

93

20.2

50


0.524

0.529

0.346

0.295

0.227

0.093


0.000

0.480

0.014

0.345



24/26

not during booms. At the industry level, I find that markups are more

countercyclical (i.e., they increase more in recessions) when industry debt is high.

These results seem consistent with Chevalier and Scharfsteins (1996) prediction that

financially constrained firms have greater incentives to boost short-term profits at theexpense of future sales in response to negative shocks to demand.

While results from inter-industry studies are inherently difficult to interpret, they

can provide valuable insights.23 I take my findings as evidence that capital structure

influences competitive performance in a systematic way. Overall, my results suggest

that firms financing choices have implications for cyclical macro dynamics, in

general, and for the dynamics of strategic interactions in product markets, in

particular. This paper contributes to the literature in suggesting that those dynamics

should be accounted for in future work by both macroeconomists and financial

economists.

References

Arellano, M., 1989. A note on the Anderson-Hsiao estimator for panel data. Economics Letters 31,

337341.

Benoit, J., 1984. Financially constrained entry in a game with incomplete information. Rand Journal of

Economics 15, 490499.

Bernanke, B., 1990. On the predictive power of interest rates and interest rate spreads. New England

Economic Review November/December, 5168.

Bernanke, B., Blinder, A., 1992. The federal funds rate and the channels of monetary transmission.American Economic Review 82, 901921.

Bils, M., 1987. The cyclical behavior of marginal cost and price. American Economic Review 77, 838855.

Bolton, P., Scharfstein, D., 1990. A theory of predation based on agency problems in financial contracting.

American Economic Review 80, 93106.

Brander, J., Lewis, T., 1986. Oligopoly and financial structure. American Economic Review 76, 956970.

Campello, 2002. Internal capital markets in financial conglomerates: evidence from small bank responses

to monetary policy. Journal of Finance 57, 27732805.

Campello, M., Fluck, Z., 2003. Market share, leverage, and competition: theory and empirical evidence.

Working paper, University of Illinois.

Chevalier, J., 1995a. Capital structure and product-market competition: empirical evidence from the

supermarket industry. American Economic Review 85, 415435.

Chevalier, J., 1995b. Do LBO supermarkets charge more? an empirical analysis of the effects of LBOS onsupermarket pricing. Journal of Finance 50, 10951112.

Chevalier, J., Scharfstein, D., 1995. Liquidity constraints and the cyclical behavior of markups. American

Economic Review (Papers and Proceedings) 85, 390396.

Chevalier, J., Scharfstein, D., 1996. Capital-market imperfections and countercyclical markups: theory

and evidence. American Economic Review 85, 703725.

Christiano, L., Eichenbaum, M., Evans, C., 1996. The effects of monetary policy shocks: evidence from

the flow of funds. Review of Economics and Statistics 78, 1634.

Clarke, R., 1989. SICS as delineators of economic markets. Journal of Business 62, 1731.

Dasgupta, S., Titman, S., 1998. Pricing strategy and financial policy. Review of Financial Studies 11,

705737.

23As noted by Schmalensee (1989, p.952), ycross-section studies rarely if ever yield consistent

estimates of structural parameters, but they can produce useful stylized facts to guide theory

constructiony.



25/26

Gertler, M., Hubbard, R.G., 1993. Corporate financial policy, taxation, and macroeconomic risk. Rand

Journal of Economics 17, 286303.

Ghosal, V., 2000. Product market competition and the industry price-cost markup fluctuation: role of

energy price and monetary changes. International Journal of Industrial Organization 18, 415444.Gottfries, N., 1991. Costumer markets, capital markets imperfections, and real price rigidity. Economica

58, 313323.

Hsiao, C., 1986. Analysis of Panel Data. Cambridge University Press, Cambridge, England.

Hodrick, R., Prescott, E., 1980. Postwar U.S. business cycles: an empirical investigation. Discussion paper

451. Carnegie-Mellon University.

Kahle, K., Walkling, R., 1997. The impact of industry classification on financial research. Journal of

Financial and Quantitative Analysis 31, 309335.

Kashyap, A., Stein, J., 2000. What do a million observations on banks say about the transmission of

monetary policy? American Economic Review 90, 407428.

Khanna, N., Tice, S., 2000. Strategic responses of incumbents to new entry: the effect of ownership

structure, capital structure and focus. Review of Financial Studies 3, 749779.

Korajczyk, R., Levy, A., 2003. Capital structure choice: macroeconomic conditions and capitalconstraints. Journal of Financial Economics, forthcoming.

Kovenock, D., Phillips, G., 1997. Capital structure and product market behavior. Review of Financial

Studies 10, 767803.

Lamont, O., 1995. Corporate-debt overhang and macroeconomic expectations. American Economic

Review 85, 11061117.

Lown, C., Morgan, D., 2001. The credit cycle and the business cycle: new findings using the survey of

senior lender officers. Discussion paper. Federal Reserve Bank of New York.

MacKay, P., Phillips, G., 2002. Is there an optimal industry financial structure? Working paper. University

of Maryland.

Maksimovic, V., 1986. Optimal capital structure in oligopolies. Unpublished doctoral dissertation.

Harvard University.

Maksimovic, V., 1988. Capital structure in repeated oligopolies. Rand Journal of Economics 19, 389407.Maksimovic, V., 1995. Financial structure and product market competition. In: Jarrow, R., Maksimovic,

V., Ziemba, W. (Eds.), Handbook of Finance. North-Holland, Amsterdam, pp. 887920.

Maksimovic, V., Titman, S., 1991. Financial reputation and reputation for product quality. Review of

Financial Studies 2, 175200.

Maksimovic, V., Zechner, J., 1991. Agency, debt, and product market equilibrium. Journal of Finance 46,

16191643.

Maurer, B., 1999. Innovation and investment under financial constraints and product market competition.

International Journal of Industrial Organization 17, 455.

Newey, W., West, K., 1987. A simple positive semi-definite, heteroskedasticity, and autocorrelation

consistent covariance matrix. Econometrica 55, 703708.

Opler, T., Titman, S., 1994. Financial distress and corporate performance. Journal of Finance 49,

10151040.

Phillips, G., 1995. Increased debt and industry product markets: an empirical analysis. Journal of

Financial Economics 37, 189238.

Rogers, W., 1993. Regression standard errors in clustered samples. Stata Technical Bulletin 13, 1923.

Rotemberg, J., Saloner, G., 1986. A supergame theoretic model of business cycles and price wars during

booms. American Economic Review 76, 390407.

Rotemberg, J., Scharfstein, D., 1990. Shareholder value maximization and product market competition.

Review of Financial Studies 3, 367391.

Rotemberg, J., Woodford, M., 1991. Markups and the business cycles. Macroeconomics Annual 6,

63129.

Rotemberg, J., Woodford, M., 1996. Imperfect competition and the effect of energy price increases on

economic activity. Journal of Money Credit and Banking 28, 549577.Schmalensee, R., 1989. Inter-industry studies of structure and performance. In: Schmalensee, R., Willig,

R. (Eds.), Handbook of Industrial Organization, Vol. 2. North-Holland, Amsterdam, pp. 9521005.



26/26

Sharpe, S., 1994. Financial market imperfections, firm leverage, and the cyclicality of employment.

American Economic Review 84, 10601074.

Telser, L., 1966. Cutthroat competition and the long purse. Journal of Law and Economics 9, 259277.

Williams, J., 1995. Financial and industrial structure with agency. Review of Financial Studies 8, 431474.Titman, S., 1984. The effect of capital structure on a firms liquidation decision. Journal of Financial

Economics 13, 137152.

Zingales, L., 1998. Survival of the fittest of the fattest? exit and financing in the trucking industry. Journal

of Finance 53, 905938.


capital structure and product markets interactions evidence from business cycles

Documents