Industry Characteristics and Debt Contracting
Dan Amiram, Alon Kalay, and Gil Sadkay
July 22, 2014
Abstract
This paper provides empirical evidence for theory that suggests industry characteristics,
in addition to …rm-level characteristics, a¤ect debt contracts. We address the empirical
challenges that arise when testing these theories by using a proprietary dataset of
time-varying and forward-looking measures of industry characteristics. The characteristics
we examine include growth, sensitivity to aggregate shocks, and industry structure, all
measured at the six-digit NAICS level. Our results, obtained using industry …xed-e¤ect
regressions and controlling for …rm-level measures of credit risk, suggest that lenders
demand compensation for industry-level growth and sensitivity attributes. However,
lenders can bene…t from industry structure, which re‡ects certain aspects of competition.
Our results also suggest that industry characteristics a¤ect debt contracts by informing
lenders about expected loss and the systematic risk premium.
JEL classi…cation: G31, G32, G33, M21
Keywords: Debt, Probability of Default, Loss Given Default, Industry Characteristics
We would like to thank IBISWorld Inc. for providing their data. We thank Hassan Bakiriddin and
Kathleen Dryer for their invaluable help obtaining and utilizing the IBIS data. We also want to thank Mary
Barth, Phil Berger, Zahn Bozanic, Peter Demerjian, Trevor Harris, Colleen Honigsberg, Ed Owens, Bugra
Ozel, Ethan Rouen, Ronnie Sadka, Ayung Tseng, Regina Wittenberg-Moerman, and workshop participants
at the Columbia University Burton Conference, SUNY Binghamton, and Tel-Aviv University for their
valuable comments and suggestions. We appreciate the …nancial support of Columbia Business School.
All errors are our own.
y
Dan Amiram and Alon Kalay are from Columbia Business School. Gil Sadka is from the University of
Texas at Dallas. e-mail addresses: [email protected], [email protected], [email protected].
1
Introduction
Despite theoretical and anecdotal evidence highlighting the importance of industry-level
analyses to lenders, the empirical literature on debt contract design has focused almost
exclusively on …rm-level forces. Our study aims to …ll this gap in the literature by using
proprietary data on three time-varying, forward-looking industry characteristics. Speci…cally,
we aim to understand which industry forces shape the design of debt contracts and whether
industry characteristics a¤ect debt contracts through expected loss, risk premiums, or both.
Industry characteristics can a¤ect debt contracts through their relation with expected
loss and risk premiums, the two primary determinants of the compensation lenders demand
(Elton et al., 2001). The probability of default and the loss given default determine the
expected loss that the lender must be compensated for. Lenders demand a risk premium for
the systematic risk that arises from the correlation of expected payo¤s across loans.
Industry characteristics can inform banks about expected losses. Industry characteristics
a¤ect …rms’ fundamentals (Maksimovic and Zechner, 1991; MacKay and Phillips, 2005),
which in turn a¤ect both the probability of default (Altman, 1968) and loss given default
(Amiram, 2013). Therefore industry characteristics may partially determine expected loss
through their relation with …rm fundamentals. One reason why industry characteristics
may correlate with debt contract terms when controlling for …rm-level measures is because
…rm-level measures are measured with error.
Banks use industry-level information to
estimate the probability of default, loss given default, or both (Moody’s 2005), potentially
to address this concern. As a result, industry characteristics can relate to the debt contract
terms, after controlling for …rm-level measures of expected loss. Elton et al. (2001) show
that expected loss accounts for up to 25% of spreads. Given the quality of existing …rm-level
measures of probability of default and loss given default, measurement error in expected loss
is not likely to be the primary reason why industry characteristics are associated with debt
contracts.
More interestingly, theory predicts that industry characteristics relate to the systematic
risk premium lenders demand. The systematic risk premium a¤ects debt contracts even
1
when expected loss is perfectly controlled for. When economic conditions in an industry
are poor there is a higher likelihood of bankruptcy contagion among …rms, and bankruptcy
contagion is mitigated in more competitive industries (Jarrow and Yu, 2001; Das et al., 2007).
Since defaults are correlated across …rms within an industry and the correlation increases
during periods of poor economic performance, lenders will require a higher premium when
lending to …rms in industries that are expected to perform poorly. However, contagion
concerns decline in more competitive industries (Jarrow and Yu, 2001). Therefore lenders
may require lower premiums in more competitive industries. Additionally, Shleifer and
Vishny (1992) hypothesize that the value of assets in liquidation depends on the health of
the industry. That is, assets in liquidation are sold at a slower pace and for lower prices
when an industry is in distress, as there are fewer potential buyers with …nancial resources
to purchase assets from distressed …rms. This, in turn, increases the correlation of loss
given default across …rms in the same industry, and the correlation depends on industry
performance (see Acharya, Bharath, and Srinivasan (2007) for empirical evidence). Therefore
lenders may demand a higher risk premium to hold loans in industries that are expected to
perform poorly. In sum, theory predicts that industry characteristics a¤ect the correlation of
expected payo¤s across loans, which in turn can a¤ect systematic risk premiums. Whether
industry characteristics ultimately a¤ect risk premiums depends on the ability of banks to
diversify their loan portfolios across industries. Thus, it is an empirical question.1
To better understand the risk premium argument, consider the following example. A bank
considers two loans with identical expected loss. The bank knows that each …rm defaults in
two out of 100 equally likely states of the world. Thus each loan has the same probability of
default, 2%. Further assume that the loss given default is identical for both …rms. Thus the
two loans have the same expected loss. However, the …rst loan defaults at times other loans in
the economy do not default, while the second loan defaults at times where many loans default.
Therefore, the loss on the second loan will occur when the overall health of the bank’s loan
portfolio is deteriorating. As a result, the bank will require a higher spread to compensate
1
In Section 2, we discuss why lenders care about systematic risk and why lenders in the private debt
market may not fully diversify their portfolio across industries.
2
for the systematic risk associated with the second loan. Since defaults are correlated across
…rms within an industry, and the correlation varies with industry characteristics, industry
characteristics relate to the systematic risk premium lenders demand. This relation exists
even when …rm-level expected loss can be measured without error and perfectly controlled
for. In Section 2, we extend this example to explain state-contingent loss given default.
In this paper, we employ a novel dataset to examine how variation in industry attributes
relate to debt contract terms at loan origination. This dataset enables us to examine the
relation controlling for a set of …rm-level attributes (fundamentals) and …rm-level measures
of loss given default and probability of default. To the extent that we adequately control for
expected loss, our …ndings shed light on whether industry characteristics relate to the risk
premium discussed above. Providing evidence on the role of industry characteristics in debt
contracts is important for several reasons. First, debt is a crucial source of external …nancing
for …rms, and it is important to understand the factors that shape debt contracts. To date,
there is scant empirical evidence to support the theoretical predictions that industry-level
forces a¤ect the systematic risk for which lenders are compensated and to explain the
extent to which debt contracts address industry-level frictions. Second, debt contracts
have implications for …rms’ future operations and value (Chava and Roberts, 2008; Nini,
Su…, and Smith, 2012). Therefore understanding the impact of industry forces on debt
contracting enhances our knowledge of how industry forces a¤ect real decisions and value.
Third, information about which industry-level characteristics matter to capital providers is
limited. Overall, the importance of this research question stems from the need to better
understand the forces that shape debt contracts (Su… and Roberts, 2009).
Our data is produced by IBISWorld Inc. (henceforth IBIS), a leading data provider. Its
data is used by prominent …nancial companies such as Bank of America, Deutsche Bank, and
American Express. IBIS produces industry-level reports, which cover public, private, and
international …rms in an industry. The data includes approximately 700 industries, where
an industry is de…ned at the six-digit NAICS classi…cation level. As part of its reports,
IBIS produces industry ratings, which measure various performance aspects of an industry.
The ratings are proprietary forward-looking scores that provide quantitative estimates of
3
the future performance of an industry, across three dimensions. The …rst dimension is
structure. Industry structure measures elements such as the amount of competition, product
di¤erentiation, barriers to entry, regulatory protection, and openness to international trade.
We note, this measure does not explicitly measure the level of concentration in the industry,
which is an alternative proxy for industry structure and competition (e.g., Hoberg and
Phillips 2010; Hoberg and Phillips, 2012; Valta, 2012). The second dimension is the expected
future performance of the industry or industry growth. The third dimension is industry
sensitivity, or the sensitivity of an industry to external economic forces such as changes in
raw material prices or GDP per capita. The three dimensions are rated (scored) separately
and then weighted and combined to derive an overall industry rating. IBIS analyzes how
these characteristics a¤ect the di¢ culty of the business operating environment.2 Higher
scores imply a more di¢ cult operating environment. In our validation tests, we empirically
show that the various ratings relate to the theoretical constructs they are meant to capture
(see Section 3.2).
The IBIS data allows us to overcome the challenges researchers face in measuring industry
characteristics such as industry structure, growth, and sensitivity. These challenges likely
account for the scant empirical evidence to date.3 One challenge is that there are very
few ways a researcher can measure these industry characteristics. Second, detailed industry
speci…c measures (e.g., for a …rm’s six-digit NAICS membership) are unavailable and cannot
be computed using CRSP and Compustat due to the limited amount of …rms at the six-digit
NAICS level. Third, potential empirical measures based on Compustat and CRSP rely on
U.S. public …rms, despite the fact that signi…cant industry activity occurs internationally
and in private …rms. Fourth, Compustat-based measures are backward looking, making
2
IBIS uses the term “risk” to de…ne the various industry characteristics. However, these are not risks
in the classical economic sense as they do not measure the variance or covariance of each attribute. The
risk scores describe the relative performance of an industry across several dimensions. IBIS measures an
industry’s relative structure via a measure called structural risk and its expected growth using a metric
called growth risk. Finally, IBIS measures industry sensitivity to outside economic forces via a metric called
sensitivity risk. We refer to these metrics as characteristics throughout the paper.
3
There is intra-industry evidence related to the e¤ect of collateral risk and liquidation values on debt
contract terms (Benmelech, Garmaise, and Moskowitz 2005; Benmelech 2009; Benmelech and Bergman,
2008, 2009, 2011). We discuss these papers in more detail in Section 2.
4
them less useful when attempting to answer the questions raised in this paper. Fifth,
the lack of time-varying data forces researchers to employ either industry-…xed e¤ects or
other time-invariant measures to estimate the relation between industry a¢ liation and
the dependent variable of choice. This limits the interpretation of potential results to a
relation between industry membership and loan contracts, because the relation may be due
to the characteristics of …rms in certain industries, as opposed to the industry-level forces
discussed above. For example, …rms in some industries may simply have more leverage and
higher spreads as a result. The IBIS dataset allows us to address these empirical challenges
and improves the identi…cation strategy employed. Speci…cally, using industry …xed-e¤ect
regressions allows us to attribute a relation between the industry characteristics examined
and the debt contract terms to the industry level-forces discussed above.
Based on the theoretical arguments discussed, we expect loans issued in industries with
poorer industry ratings (higher ratings), lower expected growth (higher industry growth
ratings), and higher sensitivity ratings to have higher spreads, smaller loan amounts, shorter
maturities, and more covenants. We also expect them to be more likely to use explicit
collateral. However, the e¤ect of industry structure on the debt contract terms is more
nuanced and is ultimately an empirical question. On the one hand, competition is expected
to reduce the risk of default contagion in the industry (Jarrow and Yu, 2001; Das et al.
2007). On the other hand, it can increase the probability of default.
Consistent with our predictions, we …nd that increases in overall industry ratings,
industry growth ratings, and industry sensitivity ratings result in higher spreads, shorter
maturities and smaller loans. An interquartile increase in the overall industry rating results
in an increase in loan spreads of approximately 15 basis points, which is economically
signi…cant. Our main …ndings hold after controlling for …rm-level probability of default,
…rm-level loss given default, and other …rm-level fundamentals. We further …nd a negative
relation between loan spreads and industry structure. At a minimum, our results suggest that
industry characteristics inform lenders about expected loss beyond the information available
in common …rm-level variables. To the extent that our …rm-level variables adequately
measure expected loss, our results suggest that part of the e¤ect of industry characteristics on
5
debt contracts is via the systematic risk premium. Moreover, the negative relation between
industry structure and spreads implies that competition reduces the risk of default contagion
in the industry (Jarrow and Yu, 2001). Therefore, this result also suggests that part of the
e¤ect of industry characteristics on debt contracts is via risk premiums.
One of the advantages of the IBIS measures is that they are available at the six-digit
NAICS classi…cation level. To further illustrate the advantage of the IBIS data, we generate
alternative measures for these industry characteristics using Compustat three-digit SIC
codes.4 To measure industry growth, we employ the growth in industry-level operating
earnings.
To measure industry structure, we employ industry-level pro…t margins as
industrial organization theory suggests more competitive industries have lower margins.
To measure industry sensitivity, we employ the correlation between industry-level earnings
growth and aggregate GDP growth. Finally, we employ the Hoberg and Philips (2011)
industry concentration measure. Our …ndings suggest that the relation between the IBIS
industry measures and loan spreads is incremental to these alternative measures. Adding the
alternative measures to our …xed-e¤ects regressions does not materially change our results.
Moreover, the Compustat based measures do not have a signi…cant association with loan
spreads.
It is important to note that there are various industry characteristics that de…ne the
economics of an industry. We examine three speci…c dimensions (industry structure, industry
growth, and industry sensitivity). Valta (2012) examines the link between spreads and a
fourth characteristic, industry concentration. While these four characteristics may be related
to each other, each one describes a di¤erent aspect of the industry.5 For example, an industry
can be competitive and have high stable expected future growth, due to an expected increase
in the demand for the product. Moreover, industry growth may, or may not, be sensitive to
aggregate shocks to demand. Therefore the relation between these di¤erent characteristics
is not clear ex ante. Furthermore, each of these characteristics has di¤erent implications
4
Using three-digit SIC codes ensures that there are enough …rms in each industry while maintaining a
relatively precise de…nition of an industry. This level of granularity is still signi…cantly less than the six-digit
NAICS codes employed by IBIS.
5
We distinguish competition as captured by concentration, and competition as measured by industry
structure. We discuss this distinction in further detail throughout the paper.
6
for the risk lenders bear. Consequently, the relation between industry characteristics and
debt contracts may vary across the industry characteristics. Therefore, when examining the
relation between industry characteristics and debt contracting, it is important to examine
the e¤ect of each industry characteristic on debt contracting, as opposed to focusing on one
characteristic. Thus, while Valta (2012) provides an important …rst step, our paper aims to
further our understanding of the role industry characteristics play in debt contracting.
This paper contributes to the literature in several ways. First, it is the …rst study
to examine how expected industry performance (growth) and sensitivity to external factors
a¤ect the design of debt contracts. While prior literature focuses on …rm-level characteristics,
and industry concentration, this paper uses a robust industry …xed-e¤ect research design to
document that industry characteristics a¤ect debt contract design. We …nd that industry
growth and industry sensitivity matter more for debt contract design, relative to industry
structure. Second, this paper introduces a novel dataset that provides forward-looking
industry-level performance assessments. This dataset may help answer questions that have
empirically challenged related prior research. Third, the paper provides new evidence on
how di¤erent types of industry characteristics deferentially a¤ect various contract terms.
Fourth, the results provide initial evidence that the e¤ect of industry on debt prices does
not necessarily result from the e¤ect of industry characteristics on expected loss but rather
via the e¤ect on the systematic risk premium. Fifth, since industry characteristics are
priced when we include industry …xed-e¤ects in our regressions, our …ndings imply that
temporal variations in industry characteristics are important for debt pricing and that
industry dummies do not fully control for the e¤ects of industry characteristics on debt
contracts. Finally, our results explain why lenders employ industry-level analyses in addition
to …rm-level analyses in the credit decision process.
Our paper proceeds as follows.
Section 2 discusses our motivation, hypothesis
development, and empirical predictions in more detail. Section 3 presents our data and
empirical results. Section 4 concludes.
7
2
Motivation and Hypothesis Development
2.1
Motivation
Theory shows that industry-level forces a¤ect …rms’ interactions with capital providers
through distinct channels, which di¤er from the familiar …rm-level forces (e.g., Maksimovic
and Zechner, 1991; Shleifer and Vishny, 1992; Williams, 1995). Anecdotal evidence also
suggests that lenders commit signi…cant resources to analyzing a borrower’s industry. During
the credit decision process, a lender will assess various industry characteristics and trends
in the borrower’s industry in addition to the borrower’s …rm-level (business) fundamentals
(Bobrow et al., 2007). For example, lenders analyze the industry’s strengths and weaknesses,
growth opportunities, technological shifts, labor status, regulatory environment, and
competitiveness (Bobrow et al., 2007). This analysis complements analysis of the borrower’s
fundamentals and projections of the borrower’s position in the industry. Furthermore,
the industry analysis is an integral piece of the loan o¤ering materials discussed in the
credit meeting (Page et al., 2007). Despite the importance that lenders place on industry
characteristics and the theoretical importance of industry characteristics to debt contracts,
there is limited empirical evidence on how industry characteristics shape debt contracts.
The lack of empirical evidence presumably arises from the empirical challenges involved in
examining this relation.
As we note above, industry characteristics can a¤ect debt contract terms through their
relation with the two primary determinants of the compensation lenders demand: expected
loss and risk premiums (Elton et al., 2001). The probability of default and the loss given
default determine the expected loss that the lender must be compensated for. The risk
premium lenders demand is for the systematic risk due to the correlation of expected payo¤s
across loans.
Industry characteristics can inform banks about expected losses. Prior studies such
as Maksimovic and Zechner (1991) and MacKay and Phillips (2005) suggest industry
characteristics a¤ect …rms’fundamentals, which in turn a¤ect both the probability of default
8
(Altman, 1968) and loss given default (Amiram, 2013). Therefore, industry characteristics
can inform banks about expected losses. One reason why industry characteristics may
correlate with debt contract terms, after controlling for …rm-level measures, is the
measurement error in the …rm-level measures. Banks use industry-level information to
estimate the probability of default, loss given default, or both (Moody’s 2005), potentially
to address this concern. As a result, industry characteristics can relate to the debt contract
terms, after controlling for …rm-level measures of expected loss. As we note above, Elton
et al. (2001) show that expected loss accounts for up to 25% of spreads. Given the quality
of …rm-level measures of probability of default and loss given default, measurement error in
expected loss is not likely to be the primary reason why industry characteristics are associated
with debt contracts.
In addition to expected loss, theory predicts that industry characteristics relate to the
systematic risk premium lenders demand. First, industry characteristics are predicted to
a¤ect the systematic risk related to the probability of default. Jarrow and Yu (2001)
hypothesize that the probability of default depends on the …rm’s industry structure.
Speci…cally, they suggest that when economic conditions in the industry deteriorate, the
likelihood of bankruptcy contagion among …rms rises. Bankruptcy contagion is also more
likely in less competitive industries. Empirical validation of this idea is o¤ered by Das et
al. (2007), who …nd defaults tend to cluster in certain industries, especially less competitive
ones. Page et al. (2007) similarly report that loan defaults cluster during recessions and that
private loan defaults tend to cluster in industries as well. Because the likelihood of default
within an industry is correlated across …rms (in the same industry) and the correlation
increases during periods of poor economic performance, lenders will require a higher premium
when lending to …rms in industries that are expected to perform poorly. Thus industry
characteristics may relate to the systematic risk associated with the probability of default
and hence shape debt contracts.
Second, industry characteristics are predicted to a¤ect the systematic risk associated
with the loss given default. Shleifer and Vishny (1992) hypothesize that the value of assets
in liquidation depends on the health of the industry. That is, assets are sold faster and at
9
a higher price in healthier industries. When an industry is in …nancial distress, there are
fewer buyers with less …nancial resources to purchase assets from defaulted …rms. This in
turn increases the correlation of loss given default across …rms in the same industry, and the
correlation also increases during periods of poor industry performance. Acharya, Bharath,
and Srinivasan (2007) provide empirical support for this theory. They show loss given default
is higher in distressed industries due to the lower values of the …rm’s implicit collateral or
the …rm’s asset values during distressed times. Therefore, lenders may demand a higher risk
premium to hold loans in industries that are expected to perform poorly. Thus, industry
characteristics relate to the systematic risk associated with the loss given default, and hence
shape debt contracts.
To understand the risk premium arguments, consider the following example. A bank
considers two loans with identical expected loss. The bank knows that each …rm defaults in
two out of a possible 100 equally likely states of the world. Thus, each loan has the same
unconditional probability of default (2%). Further assume that the loss given default is 50%
for both …rms. Thus, the two loans have the same expected loss of 1% (50% * 2%). However,
the …rst loan defaults when other loans in the economy do not default, while the second loan
defaults in states of the world where many loans default. Therefore, the loss on the second
loan will materialize when the overall health of the bank’s loan portfolio is deteriorating.
As a result, the bank will require a higher spread to compensate for the systematic risk
associated with the second loan. Defaults are correlated across …rms within an industry, and
the correlation varies with industry characteristics (Jarrow and Yu, 2001; Das et al., 2007).
Therefore, if lenders cannot diversify across industries, industry characteristics will relate to
the systematic risk premium lenders demand.
The example above can be extended to describe state-contingent loss given default.
Again, assume that the …rms default in two out of 100 equally likely states of the world.
Assume, too, that the two loans default at the same time. Therefore, not only do the loans
have the same expected loss, but they also default at the same time. The loans di¤er as
follows. In the …rst default state, the …rst loan has a 100% loss given default, and the second
loan has a 0% loss given default. In the second default state, the …rst loan has a 0% loss given
10
default, and the second loan has a 100% loss given default. Thus the loans di¤er in their
loss given default, contingent on the particular default state. Hence, while the expected loss
given default is the same for both loans (0.5*0% + 0.5*100% = 50%), the actual loss given
default (the realization) varies across the loans. Further assume, other loan losses in the
economy are 100% in the …rst default state and 0% in the second default state. Therefore,
the loss on the …rst loan (which has a 100% loss given default in the …rst default state) will
occur when the overall health of the bank’s loan portfolio is deteriorating. In contrast, the
loss on the second loan (which has a 100% loss given default in the second default state)
occurs when the other loans in the bank’s portfolio are performing well. As a result, the
bank will require a higher spread to compensate for the systematic risk associated with the
…rst loan. Losses are correlated across …rms within an industry, and the correlation varies
with industry characteristics (Shleifer and Vishny, 1992; Acharya, Bharath, and Srinivasan,
2007). Therefore, if lenders cannot diversify across industries, industry characteristics will
relate to the systematic risk premium lenders demand.
Whether industry characteristics a¤ect risk premiums ultimately depends on the ability
of banks to diversify their loan portfolios across industries. Banks can diversify the risk of
their portfolios using various tools, including alternative lending strategies, credit default
swaps, collateralized loan obligations, and loan syndication. However, the structure of
the loan market may prevent banks from fully diversifying their industry exposure. For
example, lenders have private information about their borrowers. As a result, information
asymmetries across lenders and investors exist. This in turn creates moral hazard and adverse
selection problems when banks attempt to sell their loans (Su… 2007). This makes portfolio
diversi…cation costly. Additionally, industry and regional expertise limits the pool of …rms
a bank can lend to. Thus it is di¢ cult for banks to fully diversify their portfolios across
industries.6
Collateral and liquidation values are also central to debt contracting theory (Aghion and
6
We further note, banks are sensitive to systematic risk due to regulatory requirements (i.e., banks are
more likely to fail when loan failures are correlated, and the bank is required to retain more capital when
holding a less diversi…ed portfolio). Banks may also be sensitive to systematic risk if the banks’managers
are risk averse.
11
Bolton, 1992; Hart and Moore, 1994), since the optimal debt contract depends on how costly
it is for creditors to liquidate assets at the time of default. However, as Benmelech et al.
(2005) note, empirical evidence on this issue is scarce. The growing evidence in this literature
utilizes special settings to examine the link between liquidation values, collateral, and debt
characteristics. Benmelech et al. (2005) focus on the ability to redeploy property assets
(redeployability) as determined by commercial zoning regulation and …nd that properties
that are more deployable receive larger loans, longer maturities, and lower interest rates.
Benmelech (2009) …nds that asset salability in the 19th century railroad industry led to
longer debt maturities. Benmelech and Bergman (2009) utilize a sample of loans in the
airline industry to show that collateral and redeployability are negatively correlated with
yield spread. Benmelech and Bergman (2011) use a novel dataset of secured debt tranches
issued by U.S. airlines which includes a detailed description of the underlying assets that
serve as collateral. Their results suggest the occurrence of bankruptcy has a sizeable impact
on the cost of debt …nancing for other (remaining) industry participants. The authors
further demonstrate how the collateral channel can create contagion e¤ects which amplify
the business cycle during industry downturns and thus lead to increases in the cost of external
debt …nancing for the entire industry. In contrast to these important contributions, which
focus on intra-industry variation, our study focuses on changes across di¤erent industries over
time. This allows us to draw broader conclusions about the role of industry characteristics
as a whole.
2.2
IBISWorld Data
The dataset we use is produced by IBIS. IBIS is a leading industry data provider whose
data is used by leading institutions such as Bank of America, Deutsche Bank and American
Express. IBIS produces industry-level reports which cover public, private, and international
…rms. The analysis is conducted for each six-digit NAICS industry, across approximately 700
industries. The industry reports include information about industry-level attributes such as
growth, sensitivity, value-add, and wages.
12
The IBIS Industry Risk Rating report evaluates the inherent risks associated with
a particular industry.
Industry risk is de…ned as “the di¢ culty, or otherwise, of
the business operating environment.” The reports include proprietary forward-looking
industry characteristic ratings (scores), which quantify the relative level of various industry
characteristics that a¤ect a …rm’s operating environment (e.g., industry growth and industry
sensitivity). The industry characteristic rating is forward looking, and the methodology used
to create it is designed to identify and quantify attributes inherent in speci…c industries
both now and 12 to 18 months into the future. IBIS states: "Industry-based information
would, for example, enable the examination of a loan book (portfolio) with regards to risk,
which would enable a more sophisticated assessment of risk spread and pricing to risk.
Alternatively, individual exposures can be better evaluated using an assessment of structure
and key drivers of change in the industry of the exposure." IBIS uses the term “risk”to de…ne
the various industry characteristics. However, these are not risks in the classical economic
sense as they do not measure the variance or covariance of each attribute. Rather, the
industry ratings describe the relative performance of an industry (in levels) across various
dimensions.
IBIS relies on both public and proprietary sources to produce its reports. It categorizes
its data sources into four groups.
(1) Publicly available sources include reports from
agencies such as the U.S. Census Bureau and U.S. International Trade Commission. (2)
Industry-speci…c publicly available resources include reports issued by trade associations,
industry federations (e.g., The National Retail Federation), and major industry players. (3)
Direct industry contacts include information obtained from industry speci…c conferences and
clients. (4) In-house databases provide statistics and analyses on 700 U.S. industries and
more than 2,000 business environment reports covering the U.S. and world macroeconomic
variables and demographic and consumer trends. Forecasts made by IBIS, which are an
integral part of the industry ratings, rely on the above sources and are further supported by
in-house data and economic modeling, analyst knowledge of industry operating conditions
(e.g., competition, barriers to entry, life cycle), and expected future movements of key
external drivers (IBISWorld, 2012).
13
To calculate the overall industry rating, IBIS assesses the operating environment of the
industry across three dimensions. First, it considers the industry’s structural forces, such as
the amount of competition and product di¤erentiation, potential barriers to entry, and the
industry’s openness to international trade. This characteristic is called industry structure.
This measure does not explicitly capture industry concentration, which is a separate element
of competition. Second, it incorporates the expected future performance of the industry or
industry growth. As IBIS states, “a high revenue growth rate in an industry is associated with
a lower overall di¢ culty of operation in the industry. That is, success comes relatively easily
in a quickly growing market while a contracting market tests management’s skill to a much
greater level, as each dollar earned requires a greater level of e¤ort for the average company."
The growth rating is a weighted function of both historical growth rates (25%) and forecasted
growth rates (75%), estimated using IBIS’s in-house models. The third dimension is how
external economic forces such as changes in GDP per capita and input costs a¤ect …rms
in the industry. This characteristic is called industry sensitivity. The industry sensitivity
rating includes two broad groups of sensitivities: macroeconomic sensitivities (e.g., GDP per
capita) and not-independently quanti…able sensitivities (e.g., changes in consumer tastes).
The rating is a function of both the potential change in the sensitivities identi…ed for the
industry and the signi…cance of the sensitivities to the industry. The three dimensions are
rated separately and then weighted and combined to derive an overall industry rating. In
its analysis, IBIS aims to measure the di¢ culty of the operating environment and refers
to the ratings as risks. Higher scores imply a more di¢ cult environment. Higher industry
structure ratings measure a more competitive operating environment. Higher growth ratings
indicate lower expected future growth. Higher sensitivity ratings imply that an industry’s
performance is more sensitive to external shocks. Finally, a higher overall industry rating
implies the …rm operates in a more di¢ cult operating environment due to the (three)
characteristics of its industry. While these ratings may not capture risks in the classical
economic sense, as they do not measure the variance or covariance of the characteristics of
the …rms in the industry, they describe the relative level of the industry characteristics we
aim to examine. We therefore refer to these risk as characteristics throughout the paper.
14
IBIS has internally tested the e¢ cacy of its models. IBIS used its overall industry ratings
to gauge the di¢ culty of the operating environment for each of the S&P 500 companies
between 2006 and 2009 and to project these companies’ ability to generate an operating
pro…t. The results show that the operating environment, as captured by IBIS’s ratings, plays
a statistically signi…cant role in a company’s ability to turn a pro…t (IBISWorld, 2012). In
sum, these ratings can be used to provide empirical evidence on how industry characteristics
a¤ect debt contracting at loan initiation.7
The IBIS dataset allows us to address the empirical challenges researchers face when
attempting to analyze the relation between debt contracts and industry characteristics. First,
IBIS data is available at the six-digit NAICS classi…cation level, providing detailed industry
speci…c measures. Second, the estimates capture di¤erent industry characteristics, which
allows for di¤erential predictions. Third, the estimates are forward looking, which increases
the statistical power and construct validity of our tests. Fourth, using the IBIS measures
avoids the endogeneity involved with aggregating backward-looking …rm-level characteristics
to measure industry-level characteristics. Fifth, the estimates account for international and
private …rms, avoiding data selection biases present in CRSP and Compustat. Finally, the
measures vary over time, which allows us to employ industry …xed-e¤ects in our analysis.
Thus, the identi…cation strategy used in this paper utilizes time-series variation in industry
characteristics to identify the relation between industry characteristics and debt contracting.
This signi…cantly reduces concerns related to correlated omitted variables.
2.3
Empirical Predictions
Designing private debt contracts is complicated. A lender and a borrower can negotiate terms
related to the price of the loan over LIBOR (spread), the maturity of the loan, the size of the
loan, and the covenants that restrict the borrower’s behavior and help protect the lender’s
interests. These loan characteristics are considered substitutes rather than complements
(Bradley and Roberts, 2004). Speci…cally, a lender and a borrower can negotiate a lower
7
We further validate the various industry ratings in Section 3.2
15
interest rate if the lender is willing to accept more covenants. Empirically, however, since the
contract terms are simultaneously determined, it is not a straightforward exercise to identify
the magnitude of the e¤ect of industry characteristics on each one of the contract terms
separately. Therefore we study the e¤ects of industry characteristics on all …ve of the main
debt contract terms: spread, loan size, maturity, collateral, and covenants. The rationale
behind this approach is to identify the change across most of the decision variables available
to the lender at the time of the contract.
Our main predictions relate to the e¤ect of industry characteristics on loan pricing,
since the professional literature claims that the pricing of the loan is the most important
characteristic of the loan contract (Page et al., 2007). The granularity of the IBIS industry
data allows us to examine di¤erent predictions based on the industry characteristic being
examined.
As discussed above, theory predicts that industry characteristics a¤ect the
correlation of expected payo¤s across loans in poorly performance industries. Therefore,
lenders are likely to require higher risk premiums when lending to …rms in poorly performing
industries. This leads us to our …rst prediction; the industry rating is positively related to the
spread lenders require for providing a loan. This line of reasoning holds for industry ratings
related to expected future performance, or growth, and for ratings relating to economic forces
external to the industry, or sensitivity.
The prediction with respect to the industry structure rating is more nuanced. On the
one hand, higher industry structure scores, which are mainly driven by higher levels of
competition, increase the probability of default and thus can increase the spread lenders
demand on the loan. This argument is consistent with the negative relation between spreads
and industry concentration documented by Valta (2012). On the other hand, prior literature
shows that defaults are less clustered in competitive industries (Das et al., 2007). This implies
that increased competition (higher industry structure ratings) may reduce the systematic
risk premium lenders require for originating loans. If the risk premium e¤ect dominates,
the industry structure rating will be negatively correlated with spread. In sum, the relation
between industry structure and debt contracts is not obvious, and is ultimately an empirical
question.
16
Since we view spread and other loan characteristics such as term, size, collateral and
covenants as substitutes rather than complements we predict that the overall industry rating,
industry growth, and industry sensitivity are negatively correlated with term and size, and
positively correlated with the use of collateral and covenants. The relation between industry
structure and the other loan characteristics may have the opposite sign.8
3
Data and Empirical Evidence
3.1
Sample Construction
To create our sample, we begin by downloading debt contract terms from Dealscan. We
download data for the following: loan spread (AllInDrawn), maturity, loan size, number of
covenants, and the use of collateral for all …rm-level facilities issued after 2003, which we
can link to Michael Roberts’s linking …le.9 We begin our sample in 2003 because the IBIS
industry data is available beginning in 2003. Data are available for 9,114 loans issued to 3,099
distinct …rms during the sample period. We then proceed to collect six-digit NAICS codes
from Compustat and link the contract terms to the IBIS industry ratings using the six-digit
NAICS codes available on Compustat. We can link 6,036 loans (2,051 …rms) to the IBIS
industry data. Finally, we require CRSP and Compustat data for a set of control variables
commonly used in the literature, which include …rm-level estimates of the probability of
default and loss given default. These additional data requirements further reduce our sample
to 3,654 loans (1,332 …rms).
Table 1 presents summary statistics for the IBIS industry ratings, the debt contract
terms, and the control variables used in our empirical analysis. The IBIS industry ratings
are assigned values using a scale of 1 to 9, where higher scores indicate riskier industries (more
8
The relation between industry characteristics and the use of explicit collateral is slightly more nuanced.
We discuss this relation in more detail in Section 3.3.3
9
We require an observation to have data available for all …elds except the number of covenants. If data
is missing for the number of covenants, we assume that the number of covenants in the contract is zero. In
untabulated tests we …nd similar results when we require data for the number of covenants to be available
as well.
17
di¢ cult operating environments).10 The mean (median) overall industry rating in our sample
is 4.63 (4.57), with an interquartile range of 0.96. Industry sensitivity has the lowest mean
(median) rating of 4.39 (4.21) and an interquartile range of 1.66. The industry structure
rating has the highest mean (median) score of 5.29 (5.29) and an interquartile range of 1.46.
Industry growth has a mean (median) rating of 4.45 (4.78) and an interquartile range of 1.05.
In Figure 1, we plot the cross-sectional average ratings over time. For each year in our sample,
we compute the average rating for each industry characteristic across all six-digit NAICS
industries (hereafter industries) in our sample. We then plot the average rating over time
for each characteristic. Figure 1 highlights that there is variation in the various industry
ratings over time. The …gure also reveals that a majority of the variation in the overall
industry rating comes from variation in industry growth and sensitivity. As the competitive
forces in an industry tend to be relatively more time-invariant, the industry structure rating
is much more stable over time. To further examine the variation in industry ratings, we also
plot the cross-sectional dispersion in each industry rating over time. Each year we compute
the standard deviation of each rating across all the industries covered by IBIS. We then
plot the standard deviation over time. The result is displayed in Figure 2. Once again,
the …gure highlights that industry structure is the most stable industry characteristic, while
industry sensitivity varies considerably across industries. It is also interesting to note that
the dispersion in the industry growth rating across industries has grown signi…cantly over
time.
Table 2 reports the results from an AR(1) model for each industry rating, estimated for
all available industry-year pairs in the IBIS data set, and all industry-year pairs in our …nal
sample (the debt sample). The results in Table 2 reveal a similar pattern. The industry
structure rating is very stable over time with coe¢ cients above 0.9 and R-squared values
ranging between 0.85 and 0.88. Industry growth and sensitivity vary more over time, with
coe¢ cients ranging between 0.47 and 0.71 and R-squared values ranging from 0.21 to 0.57.
The overall industry rating has a coe¢ cient of approximately 0.70 and a R-squared value of
10
For example, a higher industry growth rating implies that the industry is more likely to experience
declines in growth and pro…tability going forward. Higher industry sensitivity ratings imply that the industry
is more likely to be shocked by changes in factors external to the industry.
18
approximately 0.45. Since industry characteristics do not change dramatically from period
to period, it is not surprising that the coe¢ cients and R-squares from the AR(1) models are
all fairly high. However, as Figures 1 and 2 suggest, there is su¢ cient variation over time to
identify the relation between debt contract terms and industry characteristics using industry
…xed-e¤ects.
Table 2, Panel C, reports correlations for the industry ratings and the probability of
default. As discussed, the correlations between the di¤erent industry characteristics are
not obvious ex ante. The overall industry rating is positively correlated with the various
ratings by construction. Furthermore, the industry structure rating and sensitivity rating
are positively related. In contrast, industry growth is negatively related to industry structure,
and to the industry sensitivity rating. However, we note that all the ratings are positively
related to the probability of default. Thus, while the di¤erent IBIS industry ratings capture
di¤erent industry characteristics, they all capture an element of risk to lenders.
3.2
Validating the IBIS Ratings
As mentioned above, IBIS validates its overall industry rating internally. As an additional
validation test, we test how the various ratings relate to the theoretical constructs they
are meant to capture. First, we test whether industry growth is associated with future
growth in industry EVA. Second, we test whether industry structure is associated with future
changes in pro…t margins, which are expected to be lower in more competitive industries.
Speci…cally, we measure industry margins as industry-wide EVA divided by industry-wide
revenues. Finally, we test whether industry sensitivity is related to the correlation between
industry-wide EVA growth and GDP growth. While the actual ratings are estimated ex
ante, we test whether these measures capture ex post outcomes. The results are reported in
Table 3.
The results in Table 3 show that the industry ratings are indeed associated with the
outcomes in a predictable fashion. The industry growth rating is negatively related to future
three-year growth in EVA. In a similar vein, industries considered more competitive have
19
lower future pro…t margins. Finally, industries considered more sensitive to aggregate shocks
(i.e., those with higher sensitivity ratings) have a higher correlation between future industry
EVA growth and future changes in GDP. These tests support the notion that the various
ratings indeed capture the industry characteristics described in section Section 2.2.
3.3
3.3.1
Industry Characteristics and Debt Contracting
Research Design
As we discuss in section 2, our empirical analysis focuses on the relation between industry
characteristics and loan spread, in addition to the other contact terms that serve as
substitutes for loan spreads. We begin the analysis by documenting the relation between loan
spread and the various industry characteristics. Our model includes a variety of …rm-level
and facility-level controls. Speci…cally, we estimate the following model:
Loan_Attributel;t =
+
IBIS_Ratingj;t +
+
Controls +
P Di;t +
+
Industry_dummies + "
Other_Loan_Attributesl;t
(1)
LGDi;t
The subscript l represents loan. The subscript j represents industry. And the subscript
i represents …rm.
IBIS Rating equals the rating for the industry characteristic being
examined (e.g., growth, sensitivity). Our loan-level controls (Other_Loan_Attributes)
include maturity, loan size, the number of covenants, and an indicator variable that receives
the value of one if the loan has speci…c collateral. When examining the other loan attributes
(e.g. loan size), we use the same model except that spread becomes an additional control
variable included in (Other_Loan_Attributes).11 Our …rm-level controls include …rm size,
return on assets (ROA), operating pro…t margins, tangibility, and leverage. We further
11
When examining the use of collateral, we use a logit model, instead of an OLS model, and do not include
industry …xed-e¤ects.
20
include …rm-level controls for the probability of default (P D) as de…ned in Hillegeist et
al.
(2004), and the expected loss given default (LGD) as de…ned in Amiram (2013).
Our market-level control equals the aggregate stock market risk premium (extracted from
Kenneth French’s web site). We include this variable to control for time varying systematic
changes in risk premiums over our sample period. The variables are de…ned in detail in the
appendix and in Table 1.
In addition to these controls, we include industry …xed-e¤ects (dummies). These dummies
capture the average cross-sectional e¤ect of industry membership on spread. Therefore, the
coe¢ cients on our industry ratings capture the e¤ects of changes in the various characteristics
on loan spreads, rather than the average cross-sectional e¤ect.
Our ability to include
industry …xed-e¤ects when analyzing this relation signi…cantly improves our ability to
identify the e¤ect of an industry characteristic on spreads. It is much more likely that changes
in industry characteristics a¤ect debt contract terms, as opposed to a reverse causality
explanation. Furthermore, the industry dummies remove the e¤ect of time-invariant industry
characteristics on loan spreads. Therefore, any documented relation is not likely to result
from the characteristics of …rms in certain industries, as opposed to the industry-level forces
discussed above. For example, because …rms in some industries simply have more leverage
and higher spreads as a result (a correlated omitted variable concern). Finally, if the
industry ratings explain di¤erences in loan spreads, then our …ndings suggest that industry
dummies do not fully capture the e¤ects of industry characteristics on debt contracts. It is
also important to note that the granularity of the IBIS data allows us to include industry
…xed-e¤ects based on the six-digit NAICS code of the borrower. Thus, our typical regression
includes close to 350 industry dummies.
3.3.2
Main Findings
We predict that higher ratings are associated with higher loan spreads to compensate lenders
for taking increased risk. The results in Table 4 are consistent with this prediction. Loans
in industries with higher overall industry ratings have higher spreads. The coe¢ cient of
21
15.37 (t-statistic of 2.54) suggests that an interquartile increase in the overall industry
rating results in an approximately 15 basis points increase in loan spread. Our …ndings
hold after controlling for …rm-level measures of probability of default and loss given default,
which are both positively and signi…cantly associated with loan spread. At a minimum, our
results suggest that industry characteristics inform lenders about expected loss beyond the
information available in common …rm-level variables. To the extent that our …rm-level
variables adequately measure expected loss, our results suggest that part of the e¤ect
of industry characteristics on debt contracts is via the systematic risk premium.
Put
di¤erently, the results suggest that industry characteristics a¤ect the correlation of expected
payo¤s across loans. It is also worth noting that a signi…cant relation between spread and
industry …xed-e¤ect coe¢ cients, in a similar regression, does not necessarily lead to the same
conclusions due to the correlated omitted variable concern.
The positive relation between the overall industry rating and loan spread is driven mostly
by industry growth and sensitivity. In contrast, industry structure is negatively related to
loan spread. The relation is weakly signi…cant at the 10% level. This …nding is consistent
with the risk premium hypothesis, that is, that bankruptcy contagion is less pronounced
in more competitive industries. Our …ndings imply that, while higher industry structure
ratings may imply more risk for a single …rm’s operations, it is bene…cial for a debt holder
to lend money to a …rm in a more competitive industry where bankruptcy contagion is less
likely.
In Table 4, we employ the individual industry ratings as well as the overall industry
rating which is a weighted-average of the three industry characteristic ratings. In Table 6,
we include all three industry ratings simultaneously in the regression model. Using all three
ratings simultaneously allows the individual characteristics to receive di¤erent weights, which
also likely di¤er from the weights employed by IBIS to generate the overall industry rating.
The results in Table 6 are consistent with our …ndings in Table 4. The industry growth
and industry sensitivity ratings are both positively associated with spread. These results
suggest that lenders demand higher spreads when lending to …rms in industries with weaker
growth prospects, and in industries that are more sensitive to external shocks. In contrast,
22
the relation between spread and industry structure is once again negative. However, the
t-stat declines to 1.60 in this speci…cation.
3.3.3
Other Loan Terms
Table 5, Panel A, documents the empirical relation between the industry ratings and loan
size. As in Table 4, we control for the other debt characteristics: spread, maturity, the
number of covenants, and the use of collateral. Consistent with our hypothesis, riskier
loans are generally smaller. The coe¢ cient on the overall industry rating is negative and
statistically signi…cant. When we break out the overall industry rating into its di¤erent
components we …nd that the relation between the industry rating and loan size is driven
largely by industry sensitivity (sensitivity to aggregate shocks). The coe¢ cient on industry
structure is insigni…cant, which is consistent with our prediction that industry structure
di¤ers from industry sensitivity and industry growth.
However, we do note that the
magnitude of the coe¢ cients on industry structure appear to be similar to those for industry
sensitivity.
Table 5, Panel B, documents the empirical relation between the industry ratings and
debt maturity. As in Table 4 and Panel A, we control for the other debt characteristics:
spread, loan size, the number of covenants and the use of collateral. Consistent with our
…ndings in Table 4, we …nd that the overall industry rating is negatively related with loan
maturities. When analyzing the di¤erent rating components, we …nd that the results are
driven mainly by industry growth. These …ndings suggest that lenders provide loans with
shorter maturities to …rms in industries with higher industry growth ratings (lower expected
growth). The coe¢ cient for industry structure is positive but statistically insigni…cant. This
result is consistent with the …ndings in prior tables that industry structure includes positive
aspects that can bene…t lenders.
In sum, the relation between the industry ratings and debt contracting is consistent across
the three characteristics examined so far: spreads, loan size, and maturity. On average, loans
issued to …rms with higher industry ratings have higher spreads, smaller loan amounts and
23
shorter maturities. For completeness, we also examine the relation between the various
industry characteristics and the number of covenants in the loan contract. We …nd no
evidence of a relation between the industry ratings and the number of covenants used in the
debt contract. For brevity, these …ndings are not tabulated.
We also test the relation between industry characteristics and the use of explicit collateral.
The use of collateral di¤ers from the other contract terms examined because it tends to be
relatively time-invariant with respect to the …rm’s industry. Consistently, we do not …nd a
robust relation between industry characteristics and the use of explicit collateral, when we
include industry …xed e¤ects. In untabulated results, we employ a pooled logit speci…cation
to further examine the relation between industry characteristics and the use of explicit
collateral. We …nd that industries with higher overall industry ratings tend to use more
explicit collateral. When we decompose the IBIS overall industry rating, we …nd that this
relation is driven mostly by industry structure. While these results are consistent with our
predictions and highlight the di¤erences across the industry characteristics, we caution that
these results are more prone to correlated omitted concerns because we cannot use industry
…xed-e¤ects in this analysis. However, taken together with our results related to spreads,
loan size and maturity, they support out main predictions and …ndings.
The results in Table 6 are similar to those presented in Table 5. Industry growth is
negatively correlated with the term of the loan. Industry sensitivity is negatively related
to loan size. Finally, the results using industry structure are weaker, and have the opposite
sign for the case of maturity. Taken together, our …ndings in Table 5 and 6 suggest that the
various industry characteristics are indeed di¤erent and have varying implications for debt
contracts. These results also suggest that at a minimum the various industry characteristics
inform lenders about expected loss beyond the information available in common …rm-level
variables. To the extent that our …rm-level variables adequately measure expected loss, these
results also suggest that part of the e¤ect of industry characteristics on debt contracts is via
risk premiums.
24
3.4
Productivity and Industry Characteristics
The IBIS industry ratings examine di¢ culties in the operating environment that arise from
the various industry characteristics. We expect the industry characteristics to matter more
in less productive industries. In these industries, declines in productivity are more likely to
result in default and thus the correlation of expected payo¤s across loans will increase. To
test our hypothesis, we sort …rms based on the productivity of employees in their industry.
Speci…cally, we employ two alternative measures of employee productivity: value added
per employee and average wage per employee. To sort observations (industries), we use the
average productivity during our entire sample period. Thus, an industry is classi…ed as either
a low or high productivity industry and the classi…cation does not vary across periods. An
industry is de…ned as a high productivity industry when the productivity measure is above
the sample median. We then estimate the regression in Table 4 for the two groups separately.
We expect debt contracting to be less sensitive to risk in more productive industries, where
employees earn higher wages. The results from this analysis are reported in Table 7.
Consistent with our hypothesis, we …nd that the positive association between the overall
industry rating and loan spread is driven largely by industries with lower productivity per
employee. For example, for industries with high value added per employee, the coe¢ cient
is 11.50 and statistically insigni…cant (t-statistic of 1.30). Alternatively, for industries with
lower employee productivity, the relation between loan spread and the overall industry rating
is positive (coe¢ cient of 24.87) and statistically signi…cant (t-statistic of 2.94). We …nd
similar results using wages to proxy for productivity. Using nonparametric simulations, we
also …nd that the di¤erence in the coe¢ cients across the wage groups is statistically positive
with a p-value of 0.016 and the di¤erence in the coe¢ cients across the value added groups
is statistically positive with a p-value of 0.073.
3.5
Alternative Industry Measures
As discussed, one of the advantages of the IBIS industry ratings is that they are available
at the six-digit NAICS classi…cation level. Due to the limited number of …rms in the
25
six-digit NAICS classi…cation level, we cannot create alternative measures for these industry
characteristics at the same level of speci…city using available Compustat data. Therefore,
in order to illustrate the advantage of the IBIS data we generate alternative measures using
three-digit SIC codes. Using the three-digit SIC classi…cation ensures a su¢ cient number of
…rms in each industry, while maintaining a relatively speci…c de…nition of an industry. To
measure industry growth, we employ the growth in industry-level operating earnings (using
the most recent period before the loan date). To estimate industry structure, we employ
industry-level pro…t margins (using operating income after depreciation scaled by revenues).
We follow industrial organization theory, which suggests that more competitive industries
have lower margins. To measure industry sensitivity, we employ the correlation between
industry-level earnings growth and aggregate GDP growth. We follow a similar approach
to the one used in our validation tests, and employ two years of forward earnings growth
and three years of prior growth to estimate …ve year rolling correlations. For brevity, and
because our main prediction relates to the e¤ect of industry characteristics on loan pricing,
we only report results using spreads. The …ndings are reported in Table 8.
Our …ndings imply that the relation between the IBIS industry characteristics and
spread is incremental to the industry level characteristics that can be measured using
available Compustat data. The results are largely unchanged when the alternative measures
are included in the regressions. The only exception is the result for industry structure,
which becomes marginally insigni…cant. Moreover, the alterative measures are statistically
insigni…cant in the regression models. These …ndings illustrate the advantage of employing
the IBIS data in our analyses.
3.6
Robustness Tests
We perform a variety of robustness tests (untabulated). First, we exclude all …nancial …rms
from our sample (SIC codes between 6000 and 6999). This reduces our …nal sample size by
approximately 110 observations. We …nd similar results using this smaller sample. Therefore,
our results are not attributable to …nancial …rms. Second, we cluster our standard errors
26
by industry as opposed to by …rm. Our results remain unchanged using this approach.
Third, we use GDP, industrial production, and aggregate market-to-book ratios as separate
alternative proxies for market risk premiums (as opposed to the aggregate stock market risk
premium). We …nd similar results using these proxies. Fourth, we include loan-class …xed
e¤ects (Drucker and Puri, 2009). Our results again remain largely unchanged. Finally,
we include the Hoberg and Phillips (2011) HHI industry concentration measure in our
loan spread regressions.12 Following Valta (2012), we include an indicator variable equal
to one for observations in the lowest quartile of the HHI distribution in a given year.
Low concentration captures one element of competition that is not speci…cally captured
by the IBIS industry structure measure. Consistent with Valta (2012), we …nd that the
concentration measure is positively related to loan spreads. Furthermore, the IBIS measure
of industry structure becomes more negative when the concentration measure is added to
the regression (reinforcing our main …ndings in Section 3.3.2). This result also suggests
that competition as measured by concentration di¤ers from the IBIS measure of industry
structure.
4
Conclusions
In this paper, we utilize a novel dataset provided by IBISWorld Inc. to examine how
industry characteristics shape debt contracts. Using detailed industry characteristic ratings
based on the …rm’s six-digit NAICS classi…cation, we empirically measure the relation
between industry characteristics and the following debt contract terms: loan spread, loan
size, maturity, the number of covenants, and the use of explicit collateral. Employing
industry …xed-e¤ect regressions, we show that loans issued in industries (a) with lower
growth prospects and (b) with higher levels of exposure to macroeconomic shocks have
higher spreads, shorter maturities, and smaller loan amounts. We further show that the
relation between the overall industry rating and spreads is more pronounced in less productive
12
We do not include the Hoberg and Phillips (2011) measure in our main regressions because its inclusion
signi…cantly reduces our sample size.
27
industries.
While prior literature focuses on …rm-level characteristics and industry concentration,
this paper uses a robust industry …xed-e¤ect research design to document that industry
characteristics have an e¤ect on debt-contract design that is incremental to that of common
…rm-level variables. At a minimum, our results suggest that industry characteristics inform
lenders about expected loss beyond the information available in common …rm-level variables.
To the extent that our …rm-level variables adequately measure expected loss, our results
suggest that part of the e¤ect of industry characteristics on debt contracts is via risk
premiums. Moreover, we …nd that industry growth and industry sensitivity matter more
for debt contract design, relative to industry structure.
28
References
Acharya, Viral V., Sreedhar T. Bharath, and Anand Srinivasan, 2007, Does industry-wide
distress a¤ect default …rms?
Evidence from creditor recoveries, Journal of Financial
Economics, 85, 787-821
Altman, Edward I., 1968, Financial ratios, discriminant analysis and the prediction of
corporate bankruptcy, Journal of Finance, 23, 589-609
Amiram, Dan, 2013, Debt Contracts, Loss Given Default and Accounting Information,
Working paper - Columbia University
Benmelech, Efraim, Mark J. Garmaise, and Tobias J. Moskowitz, 2005, Do Liquidation
Values A¤ect Financial Contracts? Evidence from Commercial Loan Contracts and Zoning
Regulation, The Quarterly Journal of Economics, 120(3), 1121–1154.
Benmelech, Efraim and Nittai K. Bergman, 2008, Liquidation values and the credibility
of …nancial contract renegotiation: evidence from US airlines, The Quarterly Journal of
Economics, 123(3), 1635-1677.
Benmelech, Efraim, 2009, Asset Salability and Debt Maturity:
Evidence from
Nineteenth-Century American Railroads. Review of Financial Studies, 22(4), 1545–1584.
Benmelech, Efraim and Nittai K. Bergman, 2009, Collateral pricing, Journal of Financial
Economics, 91(3), 339–360.
Benmelech, Efraim and Nittai K. Bergman, 2011, Bankruptcy and the Collateral
Channel, Journal of Finance, 66(2), 337-378.
Bobrow, B., Tech, M., Reeding, L., Spiro, A., and Ganz, E., 2007, The Primary Market,
in The Handbook of Loan Syndications and Trading, Taylor, A., and Sansone, A., eds.
(McGrow-Hill, 2007)
Bradley, Michael, and Michael R. Roberts, 2004, The structure and pricing of bond
covenants, Working Paper - University of Pennsylvania.
Chava, Sudheer, and Michael R. Roberts, 2008, How does …nancing impact investment?
The role of debt covenants, The Journal of Finance, 63(5), 2085-2121.
29
Das, Sanjiv R., Darrell Du¢ e, Nikunj Kapadia, and Leandro Saita, 2007, Common
failings: How corporate defaults are correlated, Journal of Finance, 62, 93-117
Drucker, Steven and Manju Puri, 2009, On Loan Sales, Loan Contracting, and Lending
Relationships, Review of Financial Studeis, 22 (7), 2835-2872
Elton, Edwin J., Martin J. Gruber, Deepak Agrawal, and Christopher Mann, 2001,
Explaining the rate spread on corporate bonds, Journal of Finance, 56, 247-278
Fama, Eugene F., and Kenneth R. French, 1997, Industry costs of equity, Journal of
Financial Economics, 43, 153-193
Hou, Kewei, and David T. Robinson, 2006, Industry concentration and average stock
returns, Journal of Finance, 61, 1927-1956
Jarrow, Robert A., and Fan Yu, 2001, Counterparty risk and the pricing of defaultable
securities, Journal of Finance, 56, 1765-1799
Hillegeist, Stephen A., Elizabeth K. Keating, Donald P. Cram, and Kyle G. Lundstedt,
2004, Assessing the probability of bankruptcy, Review of Accounting Studies, 9, 5-34
Hoberg, Gerard and Gordon Phillips, 2010, Real and …nancial industry boom and bust,
Journal of Finance, 65, 45-86.
Hoberg, Gerard and Gordon Phillips, 2011, Text-Based Network Industries and
Endogenous Product Di¤erentiation, Working Paper.
Hoberg, Gerard and Gordon Phillips, 2012, The stock market, product uniqueness, and
comovement of peer …rms, Working Paper.
IBISWorld, 2012, Risk Rating White Paper
Li, Feng, Russell Lundholm, and Michael Minnis, 2013, A measure of competition based
on 10-K …lings, Journal of Accounting Research, 51, 399-436
MacKay, Peter, and Gordon M. Philips, 2005, How does industry a¤ect …rm …nancial
structure, Review of Financial Studies, 18, 1433-1466
Maksimovic, Vojislav, and Josef Zechner, 1991, Debt, agency costs, and industry
equilibrium, Journal of Finance, 46, 1619-1645
30
Moody’s, 2005, Dynamic prediction of LGD modeling methodology
Nini, Greg, David C. Smith, and Amir Su…, 2009, Creditor control rights and …rm
investment policy, Journal of Financial Economics 92(3),400-420.
Page, S., Swa¢ eld, P., Taylor, A., Yang, R., Vaky, P., Miller, S., Krase, S., and Ganz,
E., 2007, An Overview of the Loan Market, in The Handbook of Loan Syndications and
Trading, Taylor, A., and Sansone, A., eds. (McGrow-Hill, 2007)
Roberts, Michael R., and Amir Su…, 2009, Financial contracting: A survey of empirical
research and future directions, Annual Review of Finance and Economics, 1.1, 207-226.
Shleifer, Andrei and Robert W. Vishny, 1992, Liquidation values and debt capacity: A
market equilibrium approach, Journal of Finance, 47, 1343-1366
Su…, Amir, 2007, Information asymmetry and …nancing arrangements: Evidence from
syndicated loans, The Journal of Finance, 62(2), 629-668
Valta, Philip, 2012, Competition and the cost of debt, Journal of Financial Economics,
105, 661-682
Williams, Joseph T.,1995, Financial and industrial structure with agency, Review of
Financial Studies 8, no. 2: 431-474.
31
Appendix: Variable De…nitions
Variable
Structure
Growth
Sensitivity
Overall Rating
Spread
Collateral
Maturity
Loan Size
# of Covenants
De…nition
The Industry Structure Rating, labeled "Structural Risk" by IBIS, measures
the industry’s structural forces, such as the amount of competition
and product di¤erentiation, potential barriers to entry, and openness to
international trade. Industries are assigned ratings using a scale of 1 to 9,
where higher scores indicate more competitive industries (industries with
more di¢ cult operating environments). Data source: IBIS.
The Industry Growth Rating, labeled "Growth Risk" by IBIS, measures
the expected future growth of the industry. Expected growth is a weighted
function of both historical growth rates (25%), and forecasted growth rates
(75%), estimated using IBIS’s in-house models. Industries are assigned
ratings using a scale of 1 to 9, where higher scores indicate lower expected
future growth (industries with more di¢ cult operating environments). Data
source: IBIS.
The Industry Sensitivity Rating, labeled "Sensitivity Risk" by IBIS,
measures the sensitivity of …rms in an industry to external economic shocks
such as changes in GDP per capita and input costs. The rating includes
two broad groups of sensitivities: macroeconomic sensitivities (e.g., GDP
per capita) and not-independently quanti…able sensitivities (e.g., changes
in consumer tastes). Industries are assigned ratings using a scale of 1 to 9,
where higher scores indicate more sensitive industries (industries with more
di¢ cult operating environments). Data source: IBIS.
The Overall Industry Rating, labeled "Overall Risk" by IBIS, is the
weighted average rating of Industry Structure, Industry Growth, and
Industry Sensitivity. It equals 50% of the sensitivity score, 25% of the
growth score and 25% of the structure score. By construction, the overall
rating also has a scale of 1 to 9, where higher scores indicate more di¢ cult
operating environments. Data source: IBIS.
The spread between the interest rate on the loan and the relevant LIBOR
rate, per dollar of loan, measured in basis points. Data source: Dealscan.
An indicator variable that receives the value of one if the debt is
collateralized and zero otherwise. Data source: Dealscan.
The term of the loan in months. Data source: Dealscan.
The natural log of the amount of the loan in dollars. Data source: Dealscan.
The number of …nancial and net worth covenants reported on Dealscan.
32
Variable
LGD
PD
ROA
Pro…t Margin
Tangible
Leverage
V W _RET
Firm Size
De…nition
We de…ne LGD following Amiram (2013): LGD = 0:292 + 0:063
LT A + 0:018 ST T OLDEBT + 0:003 IN T AN GIBLE_RAT IO 0:005
N ET _W ORT H 0:907 ROA. The annual variables used to estimate LGD
are de…ned as follows: ROA is income before extraordinary items divided
by total assets. Net worth is total assets minus total liabilities scaled by
common shares outstanding. Intangible ratio is de…ned as intangibles scaled
by property plant and equipment. ST T OLDEBT is the ratio of short-term
to long-term debt. LT A is the log of total assets. Computed values above 1
and below 0 are assigned missing values. Data source: Compustat Annual
…le.
The monthly estimated probability of default based on the
Black–Scholes–Merton model. We follow the approach used in Hillegeist et
al. (2004).
Annual operating income after depreciation divided by beginning of period
total assets. Data source: Compustat Annual …le.
Annual operating income after depreciation divided by beginning of period
total assets. Data source: Compustat Annual …le.
Total assets, minus intangible assets, divided by total assets. Data source:
Compustat Annual …le.
Total debt (short-term debt plus long-term debt) divided by total assets.
Data source: Compustat Annual …le.
Annual value weighted market excess returns extracted from Kenneth
French’s webpage.
Firm size equals the log of the …rm’s market value of equity. Data source:
Compustat Annual …le.
33
34
Table 1: Summary Statistics
This table contains summary statistics for the variables employed in our analysis. The overall industry rating (Overall)
is the weighted average score of the IBIS six-digit industry characteristic ratings: industry growth (Growth), industry
sensitivity (Sensitivity) and industry structure (Structure). We obtain data for the number of covenants, loan size,
maturity, use of explicit collateral and spread from Dealscan. Spread equals the spread between the interest rate on
the loan and the relevant LIBOR rate, per dollar of loan, measured in basis points. Collateral is de…ned as an indicator
variable that receives the value of one if the debt is collateralized and zero otherwise. Maturity is the term of the loan
in months. Loan size equals the natural log of the amount of the loan in dollars. The number of covenants equals the
number of …nancial and net worth covenants reported on Dealscan. If no data is available we assume the number of
covenants in the contracts is zero. The variables are measured per facility. We de…ne loss given default (LGD) following
Amiram (2013): LGD = 0:292 + 0:063 LT A + 0:018 ST T OLDEBT + 0:003 IN T AN GIBLE_RAT IO 0:005
N ET _W ORT H 0:907 ROA. The variables used to estimate LGD are de…ned as follows: ROA is de…ned as income
before extraordinary items divided by total assets. Net worth is de…ned as total assets minus total liabilities scaled
by common shares outstanding. Intangible ratio is de…ned as intangibles scaled by property plant and equipment.
ST T OLDEBT is de…ned as the ratio of short-term to long-term debt. LT A is de…ned as the log of total assets.
Computed LGD values above 1 and below 0 are assigned missing values. Probability of default (P D) is the estimated
probability of bankruptcy based on the Black–Scholes–Merton model. The estimation follows Hillegeist et al. (2004).
Firm size equals the log of the …rm’s market value of equity. Return on assets (ROA) is de…ned as operating income
after depreciation divided by beginning of period total assets. Pro…t margins are de…ned as operating income after
depreciation divided by beginning of period total assets. Tangible equals the …rm’s total assets, minus its intangible
assets, divided by total assets. Leverage is de…ned as total debt (short-term debt plus long-term debt) divided by total
assets. All the continuous contract terms and control variables, other than LGD, are winzorized at the 1% level.
Overall Growth Sensitivity Structure # of Covenants
Loan Size
Maturity Spread
Mean
4.63
4.45
4.39
5.29
2.89
18.87
50.89
189.24
Median
4.57
4.78
4.21
5.29
3.00
19.11
60.00
175.00
std
0.73
1.22
1.32
1.03
1.40
1.55
21.25
136.77
25%
4.09
4.10
3.50
4.51
2.00
17.91
36.00
87.50
75%
5.05
5.15
5.16
5.97
4.00
19.92
60.00
250.00
LGD
PD
Collateral Firm Size
ROA
Pro…t Margin Tangible Leverage
Mean
0.67
0.02
0.70
7.13
0.10
0.10
0.81
0.30
Median
0.66
0.00
1.00
7.03
0.09
0.09
0.88
0.29
std
0.12
0.08
0.46
1.70
0.10
0.18
0.20
0.21
25%
0.59
0.00
0.00
6.00
0.05
0.04
0.70
0.15
75%
0.74
0.00
1.00
8.26
0.15
0.16
0.98
0.42
Table 2: Industry Rating Correlations
Panels A and B of this table report the slope coe¢ cients
and related R2 s from AR(1) models of the IBIS industry
ratings. Overall is the weighted average score of the IBIS
six-digit industry characteristic ratings: industry growth
(Growth), industry sensitivity (Sensitivity) and industry
structure (Structure). Panel A reports results for our sample.
Panel B reports results for all industries covered by IBIS.
Panel C reports correlations for the industry ratings and the
probability of default (de…ned in Table 1) in our sample.
The above (below) diagonal reports the Spearman (Pearson)
correlations. All correlations signi…cant at the 5% level are
highlighted in bold.
Panel A: Debt Sample
Overall Growth Sensitivity Structure
Coe¢ cient
0.65
0.56
0.71
0.96
2
R
0.41
0.23
0.57
0.88
Panel B: Full IBIS Sample
Overall Growth Sensitivity Structure
Coe¢ cient
0.69
0.47
0.70
0.92
R2
0.48
0.21
0.49
0.85
Panel C: Correlation Table Debt Sample
Overall Growth Sensitivity Structure P D
Overall
0.22
0.87
0.30
0.05
Growth
0.15
-0.09
-0.22
0.06
Sensitivity
0.89
-0.18
0.07
0.03
Structure
0.39
-0.29
0.18
0.09
PD
0.05
0.01
0.04
0.04
35
36
Obs.
R2
Industry Structure
Industry Sensitivity
Industry Growth
3,966
6.3%
EV Ai;t!t+3
-0.021***
[-7.74]
Corri;t
4!t+1
3,966
2.2%
0.066***
[7.36]
( EV Ai;t!t+1 ; GDPi;t!t+1 )
Dependent Variable
-0.0006*
[-1.70]
3,966
0.1%
Industry Marginsi;t!t+3
Table 3: The Relation between the IBIS Industry Ratings and Future Economic Outcomes
This table reports the results for the OLS regressions of future growth in industry Economic Value-Add (EVA),
future changes in industry margins, and the correlation of future industry EVA growth and GDP, on the IBIS
industry ratings. Future EVA growth is computed as the average percentage growth in EVA over the next three
years for each industry observation (i) and year (t). Industry margins = Industry EVA / Industry revenues.
Changes in future industry margins are computed as the average change in industry margins over the next
three years, for each industry observation (i) and year (t). The correlation between future industry EVA growth
and future GDP growth equals the correlation between next year’s EVA growth and the change in next year’s
GDP, for each industry (i) and year (t), computed over the time period t 4 and t + 1 (with a minimum of
three years; t 2, t + 1). Industry wide EVA and revenues are provided by IBIS. t-statistics based on robust
standard errors clustered by industry are reported below the coe¢ cients.
Table 4: Loan Spreads and Industry Characteristics
This table reports results for the OLS regressions of loan spreads on the IBIS
industry ratings, and various control variables. V W _RET is the value weighted
market excess returns extracted from Kenneth French’s webpage. All the
remaining variables are de…ned in the appendix. Excluding PD, the …rm-level
control variables are estimated using Compustat data for the …scal year end
before the date of the loan. PD is estimated monthly, and we employ the
estimate for the month of the loan. V W _RET is estimated during the year
of the loan. All the regression models include industry …xed-e¤ects based on
the …rm’s six-digit NAICS industry classi…cation. t-statistics based on robust
standard errors clustered by …rm are reported below the coe¢ cients.
37
Dependent Variable: Loan Spread
Overall Rating
15.37**
[2.54]
Growth
6.723**
[2.08]
Sensitivity
8.523**
[2.23]
Structure
Maturity
Loan Size
# of Covenants
Collateral
Firm Size
ROA
Pro…t Margin
Tangible
Leverage
PD
LGD
V W _RET
Obs.
R2
0.0899
[0.64]
-19.49***
[-5.95]
0.786
[0.31]
86.46***
[14.27]
-15.58***
[-3.74]
-127.1***
[-3.12]
-40.98*
[-1.75]
-12.42
[-0.66]
71.34***
[4.13]
184.1***
[3.75]
142.4***
[2.93]
0.188
[0.01]
3,654
0.506
0.0909
[0.64]
-19.79***
[-6.06]
0.711
[0.28]
86.29***
[14.27]
-15.67***
[-3.75]
-126.7***
[-3.11]
-37.92
[-1.63]
-11.17
[-0.60]
70.46***
[4.06]
188.7***
[3.82]
143.5***
[2.92]
1.717
[0.12]
3,654
0.506
38
0.0800
[0.57]
-19.50***
[-5.94]
0.785
[0.31]
86.54***
[14.25]
-15.48***
[-3.72]
-129.1***
[-3.17]
-40.86*
[-1.75]
-13.05
[-0.69]
71.53***
[4.14]
185.3***
[3.78]
141.9***
[2.94]
0.892
[0.06]
3,654
0.506
-19.89*
[-1.66]
0.0751
[0.53]
-19.94***
[-6.06]
0.730
[0.29]
86.17***
[14.21]
-15.46***
[-3.69]
-130.2***
[-3.21]
-36.95
[-1.59]
-11.76
[-0.62]
70.55***
[4.09]
191.6***
[3.88]
142.2***
[2.91]
1.598
[0.11]
3,654
0.505
Table 5: Loan Size, Maturity, and Industry Characteristics
This table reports results for the OLS regressions of loan size (Panel A) and
maturity (Panel B) on the IBIS industry ratings, and various control variables.
V W _RET is the value weighted market excess returns extracted from Kenneth
French’s webpage. All the remaining variables are de…ned in the appendix.
Excluding PD, the …rm-level control variables are estimated using Compustat
data for the …scal year end before the date of the loan. PD is estimated
monthly, and we employ the estimate for the month of the loan. V W _RET
is estimated during the year of the loan. All the regression models include
industry …xed-e¤ects based on the …rm’s six-digit NAICS industry classi…cation.
t-statistics based on robust standard errors clustered by …rm are reported below
the coe¢ cients.
39
Panel A: Dependent Variable: Loan Size
Overall Rating
-0.096**
[-2.31]
Growth
-0.005
[-0.22]
Sensitivity
-0.062**
[-2.37]
-0.107
[-1.45]
Maturity
0.008*** 0.008*** 0.008*** 0.008***
[6.43]
[6.51]
[6.45]
[6.52]
# of Covenants 0.082*** 0.083*** 0.082*** 0.083***
[3.89]
[3.93]
[3.87]
[3.95]
Spread
-0.002*** -0.002*** -0.002*** -0.002***
[-6.61]
[-6.75]
[-6.61]
[-6.80]
Collateral
-0.188*** -0.187*** -0.189*** -0.189***
[-2.68]
[-2.66]
[-2.68]
[-2.68]
Firm Size
0.586*** 0.586*** 0.585*** 0.586***
[17.43]
[17.38]
[17.39]
[17.42]
ROA
0.528
0.545
0.539
0.548
[1.51]
[1.55]
[1.53]
[1.56]
Pro…t Margin
0.021
-0.003
0.024
-0.003
[0.10]
[-0.02]
[0.12]
[-0.02]
Tangible
-0.234
-0.236
-0.229
-0.231
[-1.24]
[-1.25]
[-1.22]
[-1.23]
Leverage
0.225
0.232
0.223
0.232
[1.25]
[1.29]
[1.24]
[1.29]
PD
-0.204
-0.247
-0.204
-0.249
[-0.55]
[-0.67]
[-0.55]
[-0.68]
LGD
-0.057
-0.059
-0.052
-0.062
[-0.17]
[-0.18]
[-0.16]
[-0.19]
V W _RET
-0.086
-0.112
-0.087
-0.130
[-0.69]
[-0.90]
[-0.69]
[-1.05]
Obs.
3,654
3,654
3,654
3,654
R2
0.668
0.667
0.668
0.668
Structure
40
Panel B: Dependent Variable: Loan Term
Overall Rating
-2.610***
[-2.82]
Growth
-2.023***
[-4.44]
Sensitivity
-0.864
[-1.48]
Structure
1.783
[1.03]
Loan Size
3.266*** 3.299*** 3.294*** 3.335***
[5.53]
[5.63]
[5.54]
[5.61]
# of Covenants
0.283
0.290
0.292
0.299
[0.74]
[0.77]
[0.76]
[0.78]
Spread
0.003
0.003
0.003
0.003
[0.64]
[0.65]
[0.57]
[0.54]
Collateral
10.57*** 10.60*** 10.60*** 10.66***
[7.59]
[7.60]
[7.60]
[7.62]
Firm Size
0.824
0.858
0.805
0.801
[1.11]
[1.15]
[1.08]
[1.07]
ROA
2.374
1.793
2.783
2.880
[0.35]
[0.26]
[0.41]
[0.42]
Pro…t Margin
2.822
2.399
2.539
2.141
[0.87]
[0.74]
[0.78]
[0.65]
Tangible
-8.174** -8.518** -8.138** -8.268**
[-2.25]
[-2.36]
[-2.23]
[-2.26]
Leverage
6.023**
6.174**
6.093**
6.207**
[2.28]
[2.33]
[2.30]
[2.34]
PD
-9.541*
-9.859*
-10.16*
-10.79**
[-1.80]
[-1.84]
[-1.93]
[-2.04]
LGD
-17.17*** -17.37*** -17.17*** -17.20***
[-2.67]
[-2.72]
[-2.67]
[-2.68]
V W _RET
-7.526*** -7.376*** -7.922*** -8.033***
[-3.06]
[-3.03]
[-3.22]
[-3.24]
Obs.
3,654
3,654
3,654
3,654
R2
0.296
0.298
0.294
0.294
41
Table 6: Debt Contract Terms and Multiple Industry Ratings
This table reports results for the OLS regressions of loan spread,
loan size, and maturity on the various IBIS industry ratings. All
the regressions employ the three industry ratings simultaneously,
and include various control variables. V W _RET is the value
weighted market excess returns extracted from Kenneth French’s
webpage. All the remaining variables are de…ned in the appendix.
Excluding PD, the …rm-level control variables are estimated using
Compustat data for the …scal year end before the date of the loan.
PD is estimated monthly, and we employ the estimate for the
month of the loan. V W _RET is estimated during the year of
the loan. All the regression models include industry …xed-e¤ects
based on the …rm’s six-digit NAICS industry classi…cation.
t-statistics based on robust standard errors clustered by …rm are
reported below the coe¢ cients.
42
Dependent Variable:
Spread
Loan Size
Term
Growth
5.393*
[1.67]
7.651**
[1.97]
-19.35
[-1.60]
0.003
[0.12]
-0.062**
[-2.34]
-0.106
[-1.44]
-0.002***
[-6.65]
-1.927***
[-4.15]
-0.547
[-0.92]
1.571
[0.94]
0.003
[0.71]
3.288***
[5.59]
Sensitivity
Structure
Spread
Loan Size
Maturity
# of Covenants
Collateral
Firm Size
ROA
Pro…t Margin
Tangible
Leverage
PD
LGD
V W _RET
Obs.
R2
-19.57***
[-5.98]
0.100
[0.71]
0.863
[0.34]
85.89***
[14.24]
-15.56***
[-3.73]
-125.8***
[-3.09]
-40.92*
[-1.75]
-11.44
[-0.61]
71.24***
[4.11]
182.8***
[3.70]
141.3***
[2.92]
-4.061
[-0.28]
3,654
0.508
43
0.008***
[6.48]
0.082***
[3.89]
-0.190***
[-2.69]
0.585***
[17.33]
0.541
[1.54]
0.025
[0.12]
-0.225
[-1.20]
0.223
[1.24]
-0.205
[-0.55]
-0.056
[-0.17]
-0.104
[-0.83]
3,654
0.668
0.278
[0.73]
10.60***
[7.61]
0.854
[1.15]
1.766
[0.26]
2.622
[0.81]
-8.494**
[-2.36]
6.091**
[2.31]
-9.491*
[-1.77]
-17.24***
[-2.70]
-6.930***
[-2.75]
3,654
0.298
44
Table 7: Loan Spread, Productivity and Industry Characteristics
This table reports results for the OLS regressions of loan spreads on the IBIS overall industry rating, and various control
variables. Observations are sorted based on the average wage per employee and value-add per employee in the industry.
These metrics are computed using industry total value-add, wages, and number of employees available in the IBIS data
set. High productivity industries have value-add or wages per employee above the sample median. V W _RET is the
value weighted market excess returns extracted from Kenneth French’s webpage. All the remaining variables are de…ned
in the appendix. Excluding PD, the …rm-level control variables are estimated using Compustat data for the …scal year end
before the date of the loan. PD is estimated monthly, and we employ the estimate for the month of the loan. V W _RET
is estimated during the year of the loan. All the regression models include industry …xed-e¤ects based on the …rm’s
six-digit NAICS industry classi…cation. t-statistics based on robust standard errors clustered by …rm are reported below
the coe¢ cients.
45
Dependent Variable: Spread
High Value Added Low Value Added
High Wage
Low Wage
Per Employee
Per Employee
Per Employee Per Employee
Overall Rating
11.50
24.87***
9.075
30.23***
[1.30]
[2.94]
[1.06]
[3.65]
Maturity
0.250
-0.197
0.352*
-0.233
[1.35]
[-0.97]
[1.82]
[-1.22]
Loan Size
-25.94***
-12.91***
-27.20***
-12.46***
[-7.31]
[-2.63]
[-6.89]
[-2.74]
# of Covenants
-0.566
3.251
2.038
-0.395
[-0.15]
[0.98]
[0.54]
[-0.12]
Collateral
84.24***
89.48***
86.88***
85.20***
[9.81]
[10.49]
[10.45]
[9.70]
Firm Size
-9.371*
-21.13***
-13.63***
-17.35***
[-1.79]
[-3.60]
[-2.60]
[-3.05]
ROA
-167.8***
-112.8**
-92.71*
-156.6***
[-3.17]
[-2.01]
[-1.58]
[-2.75]
Pro…t Margin
-31.24
-49.03**
-41.96
-49.93***
[-0.94]
[-2.29]
[-1.19]
[-2.42]
Tangible
14.46
-18.13
-14.80
-5.448
[0.55]
[-0.73]
[-0.54]
[-0.21]
Leverage
62.27***
90.74***
33.58
109.6***
[2.63]
[4.01]
[1.44]
[4.87]
PD
270.2***
154.3***
255.3***
137.4***
[2.69]
[2.80]
[2.97]
[2.43]
LGD
109.9
165.4***
201.0***
76.50
[1.36]
[2.95]
[2.56]
[1.40]
V W _RET
-9.173
5.259
-6.448
11.50
[-0.42]
[0.24]
[-0.30]
[0.54]
Obs.
1,763
1,855
1,754
1,864
2
R
0.495
0.544
0.479
0.558
46
Table 8: Alternative Compustat-Based Industry Measures
This table reports results for the OLS regressions of loan spread on the IBIS industry ratings, including the alternative
industry measures computed using data available in Compustat. The alternative industry growth measure (Alt. Growth)
equals the growth in industry-level operating earnings during the year before the loan origination date. The alternative
industry structure measure (Alt. Structure) equals industry-level pro…t margins at the time of loan initiation. The
alternative industry sensitivity measure (Alt. Sensitivity) equals the correlation between industry-level earnings growth
and aggregate GDP growth. Five year rolling correlations are calculated using two years of forward earnings growth
and three years of prior growth. All the additional control variables employed in Tables 4 through 7 are included in the
regressions. These variables are de…ned in the appendix. Excluding PD, the …rm-level control variables are estimated
using Compustat data for the …scal year end before the date of the loan. PD is estimated monthly, and we employ the
estimate for the month of the loan. V W _RET is estimated during the year of the loan. All the regression models include
industry …xed-e¤ects based on the …rm’s six-digit NAICS industry classi…cation. t-statistics based on robust standard
errors clustered by …rm are reported below the coe¢ cients.
47
R2
Obs.
Additional Controls
LGD
PD
Collateral
# of Covenants
Loan Size
Maturity
Alt. Structure
Alt. Growth
Alt. Sensitivity
Structure
Sensitivity
Growth
Overall Rating
0.090
[0.64]
-19.49***
[-5.95]
0.786
[0.31]
86.46***
[14.27]
184.1***
[3.75]
142.4***
[2.93]
YES
3,654
0.506
15.37**
[2.54]
-0.524
[-0.08]
-3.117
[-0.45]
-185.1
[-1.47]
0.096
[0.68]
-19.33***
[-5.81]
0.480
[0.19]
87.10***
[14.15]
175.0***
[3.53]
144.9***
[2.95]
YES
3,522
0.505
15.21**
[2.52]
0.091
[0.64]
-19.79***
[-6.06]
0.711
[0.28]
86.29***
[14.27]
188.7***
[3.82]
143.5***
[2.92]
YES
3,654
0.506
6.723**
[2.08]
-0.578
[-0.08]
-3.131
[-0.45]
-175.4
[-1.40]
0.096
[0.68]
-19.62***
[-5.91]
0.466
[0.18]
86.93***
[14.17]
180.2***
[3.60]
146.2***
[2.94]
YES
3,522
0.504
6.704**
[2.18]
0.080
[0.57]
-19.50***
[-5.94]
0.785
[0.31]
86.54***
[14.25]
185.3***
[3.78]
141.9***
[2.94]
YES
3,654
0.506
8.523**
[2.23]
Dependent Variable: Loan Spread
-0.078
[-0.01]
-2.564
[-0.39]
-202.3
[-1.59]
0.088
[0.62]
-19.35***
[-5.80]
0.458
[0.18]
87.24***
[14.14]
175.8***
[3.55]
144.7***
[2.95]
YES
3,522
0.504
8.192**
[2.13]
0.075
[0.53]
-19.94***
[-6.06]
0.730
[0.29]
86.17***
[14.21]
191.6***
[3.88]
142.2***
[2.91]
YES
3,654
0.505
-19.89*
[-1.66]
-16.24
[-1.44]
0.535
[0.08]
-1.476
[-0.24]
-198.4
[-1.58]
0.082
[0.58]
-19.75***
[-5.92]
0.475
[0.19]
86.95***
[14.12]
182.9***
[3.66]
145.2***
[2.93]
YES
3,522
0.503
Figure 1: This …gure plots the time trend of the cross-sectional average IBIS ratings.
Overall is the weighted average score of the three di¤erent industry rating categories:
Industry growth, Industry sensitivity, and Industry structure. Each year, we calculate the
cross-sectional average of each industry rating score and then plot the averages over time.
48
Figure 2: This …gure plots the cross-sectional dispersion in the IBIS ratings across industries,
over time. Overall is the weighted average score for the three di¤erent industry rating
categories: industry growth, Industry sensitivity, and industry structure. Each year, we
calculate the cross-sectional standard deviation of each industry rating and plot the results
by year.
49