customer concentration and loan contract terms* · 2018-08-06 · customer concentration and loan...
TRANSCRIPT
Customer Concentration and Loan Contract Terms*
Murillo Campello Janet GaoCornell University & NBER Cornell University
[email protected] [email protected]
This Draft: April 1, 2015
Abstract
Recent research argues that firms enjoy operating efficiencies when dealing with fewer, larger
customers. It ignores, however, how firms’ creditworthiness is affected by the exposure to those
customers. We study pricing and non-pricing features of bank loan contracts to gauge how the
credit market evaluates firms’ customer-base profile. We show that higher customer concen-
tration leads to increases in interest rate spreads and in the number of restrictive covenants
featured in newly-initiated as well as renegotiated bank loans. Concentration also reduces the
maturity of those loans. The duration and depth of the relationship between firms and their
banks are also negatively affected by customer concentration.
Key words: Customer Concentration, Bank Loans, Contract Terms, Financial Distress, Instru-
mental Variables.
JEL classification: G21, G30, G32.
*We are thankful to Ted Fee, Erasmo Giambona, and Sudheer Chava for sharing their data. We also
thank Kenneth Ahern, Kevin Aretz, Jean-Noel Barrot, Martijn Cremers, Sudipto Dasgupta, Jerry
Hoberg, Andrew Karolyi, Tomislav Ladika, Rafael Matta, Pamela Moulton, Justin Murfin, Maureen
O’Hara, and Felipe Silva for their valuable input. Comments from seminar participants at Cornell Uni-
versity, McGill–HEC Montreal, University of Notre Dame, and USC Finance, Organizations and Markets
conference are also appreciated.
Customer Concentration and Loan Contract Terms
Abstract
Recent research argues that firms enjoy operating efficiencies when dealing with fewer, larger
customers. It ignores, however, how firms’ creditworthiness is affected by the exposure to those
customers. We study pricing and non-pricing features of bank loan contracts to gauge how the
credit market evaluates firms’ customer-base profile. We show that higher customer concen-
tration leads to increases in interest rate spreads and in the number of restrictive covenants
featured in newly-initiated as well as renegotiated bank loans. Concentration also reduces the
maturity of those loans. The duration and depth of the relationship between firms and their
banks are also negatively affected by customer concentration.
Key words: Customer Concentration, Bank Loans, Contract Terms, Financial Distress, Instru-
mental Variables.
JEL classification: G21, G30, G32.
1 Introduction
U.S. manufacturers attribute over one-third of their sales to “large customers.” A con-
centrated customer base is often cited as a positive factor in analyst reports, management
forecasts, and even IPO prospectuses, as it is believed to enhance firms’ profitability by
reducing overhead costs. Notably, these arguments find support in recent academic re-
search (e.g., Patatoukas (2012) and Irvine et al. (2013)). Relying on major customers has
shortcomings, nonetheless. Major customers demand lower prices, purchase irregularly,
and may delay payments (Fee and Thomas (2004), Kelly et al. (2013), and Murfin and
Njoroge (2013)).1 Shocks to major customers are also known to reverberate through the
supply chain, affecting suppliers (Cohen and Frazzini (2008) and Kolay et al. (2012)).
While these problems are important, the literature has not examined whether a close
association with fewer, larger customers exposes firms to costs and risks that may ul-
timately affect their access to credit. This problem becomes pressing as the level of
customer concentration has increased in recent years.
This paper examines how the credit market evaluates a firm’s customer-base profile,
showing how customer concentration and customer financial status affect a firm’s cred-
itworthiness. It does so considering both pricing and non-pricing features of bank loan
contracts, as well as firm–bank relationships. Our contract-level analysis allows us to
assess how informed lenders modify the terms of their credit offerings in response to the
evolving nature of firms’ customer-base profile. We first model the interplay between
customer concentration, firm investment choices, and loan contract terms in a simple
theoretical framework. The model is useful in providing predictions that can be taken to
the data. We then empirically examine the impact of customer concentration on several
features of bank loan contracts, including interest rate spreads, maturity, and the number
of restrictive covenants. We also examine the impact of customer concentration on the
length and depth of the relationships between firms and their banks. Our results are new
in revealing significant costs associated with firms’ reliance on large customers. We show
that these costs are manifested along various credit dimensions, pointing to limitations
to deeper integration among firms along their supply chain.
Our model characterizes firms’ incentives to invest in projects that enhance relation-
ships with their customers, showing how this affects the firm’s credit access. Relationship-
specific investments have long been described in the existing literature (see Bolton and
1These behaviors have attracted the attention of the financial press, with reports that large, powerfulfirms such as Walmart and P&G “abuse” their suppliers when paying for products. See Wall StreetJournal article: “Small Firms’ Big Customers Are Slow to Pay” (June 6, 2012).
1
Scharfstein (1998) for a theoretical framework and Allen and Phillips (2000) for early
empirical work). These projects involve investments in R&D, unique fixed assets, and
modifications to standard production processes. Critically, relationship-specific projects
may be less desirable from lenders’ perspective because their uniqueness engenders higher
risks and lower resale values in liquidation. Our analysis shows that the higher the im-
portance of major customers, the greater the operational gains from relationship-specific
investments, but the lower the expected credit quality of firms undertaking those invest-
ments. The model implies that higher customer concentration prompts banks to impose
costlier, stricter loan contract terms.
To test our model’s predictions, we gather information on bank loan terms from LPC–
Dealscan and merge that information with data on corporate customers from Compustat’s
Segment Database. Our data collection produces a comprehensive sample of 3,375 loans
granted to 1,110 individual firms in the manufacturing sector over 25 years. We add to
these data information on corporate failures, product differentiation, and CDS spreads.
Our baseline results can be summarized as follows. A more concentrated customer
base increases both the interest rate spreads and the number of restrictive covenants fea-
tured in new (or renewed) bank loans. Customer concentration also reduces the maturity
of those loans. These effects are statistically and economically significant. Controlling for
bank identity, industry effects, macroeconomic conditions, and firm characteristics, a one-
standard-deviation increase in customer concentration leads to 10 basis points higher in-
terest spreads on bank loans; or a 6% higher loan markups compared to an average spread
of 179 basis points. The same shift leads to 0.2 additional restrictive loan covenants; com-
pared to the sample mean of 1.8 covenants. It also leads to a reduction in loan maturity by
2 months; compared to average maturity of 46 months. These magnitudes are significant
given the high level of competition in the market for manufacturing lending.
We also examine whether customer concentration affects the length and depth of a
firm’s banking relationships. Banks lend less to firms with concentrated customer bases
over time. They also abbreviate the duration of their relationships with those firms.
Estimates of the relation between customer concentration and borrowing terms may be
subject to empirical biases. In particular, one could argue that unobserved characteristics
might lead a firm’s customer concentration to increase and its credit terms to deterio-
rate. This is a tall order in light of the documented positive relation between customer
concentration and firm profitability — a relation that we verify in our data. To alleviate
those concerns, however, we experiment with testing approach that exploits M&A waves
in customers’ industries (downstream mergers) as an instrument for customer concen-
2
tration. Downstream M&A activity is a plausible instrument for two reasons. First, it
is related to customers’ own growth prospects (see Fee and Thomas (2004) and Erel et
al. (2014)) and following mergers in customer industries, suppliers face higher customer
concentration (inclusion restriction). Second, downstream M&A activity is not a policy
variable for suppliers and is unlikely to affect their borrowing terms through channels
other than customer–supplier linkages (exclusion restriction). Bearing in mind concerns
that suppliers’ industry-level, time-varying dynamics could influence customers’ M&A
activity and credit terms, we further account for industry-year-fixed effects in our tests.
We go further and incorporate in our test strategy downstream M&A activity that is
triggered by acts by the Federal Government that change entry, prices, and other ele-
ments of the competitive environment of the industries affected.2 Our IV estimations
confirm the prior that following high levels of M&A activity in downstream industries,
firms observe higher customer-base concentration, which then leads to costlier, stricter
borrowing terms, as well as shorter banking relationships.
Our theoretical analysis predicts that deeper integration along the supply chain can
fundamentally change firms’ operations, especially the types of investments they make.
When firms choose to conduct relationship-specific investments, loan terms will be more
adversely affected by customer concentration, since those investments embed higher risk
exposure for creditors. We test this prediction empirically using a proxy for the specificity
of the investment firms make that is based on the uniqueness of the inputs they use in their
production processes (cf. Giannetti et al. (2011)). Our results point to a strong effect of
customer concentration on loan contract terms in industries where relationship-specific
investments are more prevalent. In industries with low levels of relationship-specific in-
vestments, in contrast, the relation between customer concentration and loan terms is
quite weak. Aside from changing the types of investments, a deeper supply-chain inte-
gration may also change firms’ financing. In particular, as firms trade with concentrated
customers and deliver those customers’ preferred products, they may receive payments
at a faster rate so as to finance their customized production cycle. Indeed, we find a
negative relation between customer concentration and the ratio of account receivables to
sales, indicating that firms facing more concentrated customer bases gain by operating
with shorter receivables cycles.
We dig deeper into the meaning of our results by examining whether and how the
2Examples of regulatory interventions that we use include the Natural Gas Wellhead Decontrol Actof 1989, the Energy Policy Act of 1992, the Trucking Industry and Regulatory Reform Act of 1992, andthe Telecommunications Act of 1996.
3
financial conditions of a firm’s large customers affect the firm’s credit terms. Customers
in worse financial shape may, for example, face difficulties in maintaining purchase agree-
ments and paying on time, eventually burdening their suppliers. Confirming the logic of
this argument, we find that loan spreads increase even more and the number of restrictive
covenants is even higher when a firm’s large customers are likely to be distressed. Large
customers’ financial distress further reduces a firm’s loan maturity, and the length and
depth of its banking relationships.
Our empirical investigation further characterizes the channels through which customer
concentration affects the credit terms offered by banks. As highly-regulated intermedi-
aries, banks are particularly concerned about loan underperformance. If higher customer
concentration is associated with higher loan failure rates, banks should naturally impose
stricter loan terms. To examine this link, we identify loan failures by matching our sam-
ple with a corporate bankruptcy database used in Chava and Jarrow (2004) and Chava
et al. (2011). We find a positive, significant relation between customer concentration
and loan failure rates. To wit, a one-standard-deviation increase in a firm’s customer
concentration is associated with a 2-percentage point increase in the likelihood that the
firm files for bankruptcy before paying its loan. This impact is sizable when compared
to the sample average loan failure rate of 6.1%. These results confirm that customer
concentration is an important risk concern for banks’ lending decisions.
Our paper is related to several strands of literature. First, it speaks to a growing
literature on the relation between customer concentration and profitability. Patatoukas
(2012) argues that having large customers helps firms achieve economies of scale by lower-
ing overhead costs. Irvine et al. (2013) further show that the beneficial effects of customer
concentration vary according to suppliers’ size and age. Fee and Thomas (2004) report
that customers gain additional bargaining power over suppliers after horizontal mergers.
Greene et al. (2013) show that when customers become more powerful they demand
better trade terms. Like previous papers, we show that customer concentration is indeed
associated with higher profitability. Using creditors’ perspective, however, we show that
customer concentration ultimately bears negative implications for firm creditworthiness,
leading banks to impose costlier, stricter loan terms.
Our study is also related to existing work on how supply-chain relationships affect
firms’ financial policies. Titman and Wessel (1988) and Allen and Phillips (2000) show
that firms tend to procure unique assets when they rely on major customers. These
firms have lower leverage because customer liquidation imposes high redeployment costs
for their relationship-specific assets (see also Kale and Shahrur (2007) and Banerjee et
4
al. (2008)). We push this research forward by showing how several different features of
debt contracting — e.g., interest rates, maturity, and covenants — relate to the firm’s
customer base.
Finally, our study is related to the recent literature on the determinants of bank loan
terms (examples are Graham et al. (2008), Roberts and Sufi (2009), Hertzel and Officer
(2012), Valta (2012), and Cen et al. (2014)). Closer to our study, Valta finds that firms
in competitive industries face higher loan spreads because competition increases cash flow
risk. Hertzel and Officer report that firms face higher spreads following industry-rivals’
bankruptcies, especially in competitive industries. Cen et al. find that supply-chain rela-
tions may reduce informational asymmetries between firms and creditors in the long run.
None of these papers consider the effect of customer concentration or customer distress
on loan terms.
The paper proceeds as follows. Section 2 contains our theoretical motivation. Sec-
tion 3 describes our data and methodology. Section 4 reports univariate analyses. Sec-
tion 5 contains our baseline regression results. Section 6 reports our instrumental variable
analysis. Section 7 shows how relationship-specific investments alter the relation between
customer concentration and loan terms. Section 8 explores cross-sectional differences in
the effect of customers’ financial health on suppliers’ borrowing terms. Section 9 re-
ports tests of the effect of customer concentration on bank loan failure rates. Section 10
concludes.
2 Theoretical Framework
We analyze the relation between customer concentration and bank credit using a
simple theoretical framework. In it, we model the interplay between the customer, the
supplier firm, and the supplier’s bank, keeping the focus on the relations we want to study
empirically. We do not model industry dynamics. In turn, we take customer concentra-
tion as given, capturing the crux of our supply-chain story of a deep association between
a firm and its major customer. The model delivers several testable implications.
2.1 Setting
The base model contains a firm, a major customer, and a bank (we allow for multiple
firms and firm heterogeneity below). The setting has two time periods. At t = 0, the
firm faces two mutually exclusive projects. Both projects require initial investment I and
5
have a payoff at t = 1. The firm has no funds, so it needs to borrow from the bank.
Project A produces output αI with probability p, and 0 with probability 1− p. Project
B is safe and produces βI with probability 1.
Project B is designed as the optimal investment choice from an atomistic perspective
in which a firm’s relationships along its supply chain are ignored; that project gives the
highest expected payoff (β > pα). Our analysis allows project choice to have different
impacts on the firm’s relationship with its major customer and bank. Project B can be
thought of as a standard technology with ex-ante known ability to generate stable cash
flows and with high resale value; this is the project that is preferred by the bank. Project
A, in contrast, engenders relationship-specific investments the firm makes to fulfill the
needs of its major customer. These relationship-specific investments can involve expendi-
tures with R&D, unique assets, customization, and modifications to standard production
processes. In conducting such investments, firms often develop innovative production pro-
cesses and face higher uncertainty of success. The output from these investments may be
also difficult to transfer to alternative users. Accordingly, project A is riskier for the firm,
with lower success rate and lower resale value. At the same time, we allow for it to create
synergistic gains for the major customer that are ultimately shared by the firm. Assume
that the major customer derives per-unit value VA from project A and VB from project B.
We allow for the major customer to prefer the relationship-specific project: pαVA > βVB.
At t = 1, the firm sells a proportion µ of its output to the major customer; the re-
mainder, 1− µ, is sold to a set of small customers. We assume the non-major customers
pay a price 1 per unit of output from project B, but only η ≤ 1 per unit of output
from project A.3 In order to incentivize the firm to take project A, the major customer
will offer different prices for different project outputs. The major customer offers δA per
unit for project A and δB per unit for project B; where δA > δB. This price schedule
is the outcome of bargaining between the firm and the major customer. It reflects the
terms in the sales contract agreed upon by the parties and it is binding. It can include
future transaction prices, speed of payment, and can also reflect variations in overhead
costs during the production process. In other words, in our analysis, prices are a simple
representation of the monetary and non-monetary exchanges executed between the firm
and its customers. As such, the analysis does not specify the price-setting process.
3Non-major customers apply a discount to the output derived from project A because it is tailoredfor the need of the major customer (less standard output).
6
2.2 Base Analysis
To ease the exposition, we momentarily assume the bank can observe firm’s choice of
project and its customer concentration µ. The bank also knows that the firm will default
on its loan payment at t = 1 with probability 1− p if it chooses project A. Accordingly,
the bank imposes the rate R = rp
if the firm chooses the risky project, and the risk-free
rate r (1 ≤ r ≤ β) if the firm chooses the safe project. The firm chooses between the two
projects to maximize its total value at t = 1 given bank rates R and r as follows:
max{pα(µδA + (1− µ)η)I − pRI, β(µδB + 1− µ)I − rI}. (1)
The firm will choose to invest in project A if pα(µδA+(1−µ)η)−pR > β(µδB+1−µ)−r.Simplifying this condition, the firm chooses project A if the customer’s offer satisfies the
following:
pαδA − βδB >(1− µ)(β − pαη) + pR− r
µ. (2)
The major customer will also benefit from project A if the following holds:
pα(VA − δA) > β(VB − δB). (3)
Therefore, the firm and its major customer will be in agreement and choose project
A if:(1− µ)(β − pαη) + pR− r
µ< pαδA − βδB < pαVA − βVB. (4)
Conversely, project B will be chosen if:
(1− µ)(β − pαη) + pR− rµ
> pαδA − βδB > pαVA − βVB. (5)
The conditions above imply that for high levels of customer concentration µ (that is,
µ > β−pαη+pR−rβ−pαη−βVB+pαVA
), the firm and the customer will agree on the relationship-specific
investment, project A. For low levels of µ, the standard project B will be selected.
2.3 Firm Heterogeneity
To make the model realistic and deliver testable predictions, we allow for many firms
in the economy. Moreover, we allow firms to be heterogeneous in their ability to suc-
cessfully invest in the relationship-specific project. Finally, we relax the assumption that
7
banks have perfect information about firms. Instead, we only assume that banks can
observe (ex-post) firms’ project choices, and that they know the general distribution of
“firm quality” (ability to succeed in the relationship-specific project). We cut clutter
in the model’s notation by associating the distribution of firm quality with the success
probability parameter p, denoting F (p) as a function of p ⊂ [p, p]; where p (p) is the
lowest (highest) probability of success in the distribution F (p).
In this economy, banks set loan prices based on the expected probability of default.
This captures the fact that commercial banks are subject to regulatory capital require-
ments and cannot diversify away the default risk of their loans. We assume a competitive
banking market. Without loss of generality, we model the lending decision of only one
bank.
We take that major customers have private information about their needs for cus-
tomization and know their suppliers’ ability to successfully deliver the relationship-specific
project A. This reflects the common assumption that major customers have information
advantage over banks regarding projects their suppliers undertake, as customers can bet-
ter understand input transactions and trade credit works as a monitoring tool (see, e.g.,
Biais and Gollier (1997) and Burkart and Ellingsen (2004)). Finally, we assume that
major customers always prefer project A, which is equivalent to setting pαVA > βVB.
To derive the equilibrium of this setting, we start by examining the incentive-compatible
agreement between a firm and its major customer. A firm will agree to take project A if
its customer offers a sufficiently high price (as in Eq. (2)). As firms differ in their qual-
ity, major customers of better firms can offer a higher price, leading firms to undertake
project A. Major customers of worse firms, on the other hand, cannot provide such a
price given the limited expected payoff from production. It follows that for every level
of customer concentration µ, a separating equilibrium exists where there is a threshold
value p∗ such that the “better firms” (those whose p > p∗) will choose the risky project
and the “worse firms” (whose p < p∗) will choose the safe project.
In this environment, the bank charges a break-even rate R = 1E[p|p≥p∗] if the firm takes
the risky project, and the risk-free rate r if the firm takes the safe project. The only
condition needed for the equilibrium is that the better firms may prefer to take the risky
project and the worse firms prefer to take the safe project:
pα(µδA + (1− µ)η)− pR > β(µδB + 1− µ)− r, ∀p > p∗ (6)
pα(µδA + (1− µ)η)− pR < β(µδB + 1− µ)− r, ∀p < p∗. (7)
8
Note that R decreases with p∗ ( ∂R∂p∗
< 0), suggesting that when the bank knows only
better firms undertake the risky project, it is less worried about default and will thus
charge a lower rate.
The equilibrium threshold p∗ will satisfy the following break-even condition:
(1− µ)(β − p∗αη) + p∗R− rµ
= p∗αVA − βVB. (8)
This expression describes the trade-off between projects for the marginal firm. The left-
hand side presents the cost of undertaking project A. The first term captures the loss of
sales to the non-major customers, while the other terms capture the additional cost (mark
up) of the bank loan. The right-hand side presents the benefit of undertaking project A,
which is also the maximum level of “inducement” the major customer could offer. The
threshold value p∗ is determined by equating the costs to the benefits of project A for
the firm.
Eq. (8) can be written as F = (p∗αVA−βVB)µ+(p∗αη−β)(1−µ)− (p∗R−r). Given∂R∂p∗
< 0, it follows that ∂F∂p∗
> 0 and ∂F∂µ
> 0. Using the Implicit Function Theorem, we
have ∂p∗
∂µ< 0, which implies that the quality of firms taking the risky project declines with
customer concentration. The end result is that a higher level of customer concentration,
µ, prompts more marginal firms to invest in the relationship-specific project A (lower p∗),
prompting the bank to charge a higher mark up interest rate (R− r).
2.4 Empirical Predictions
The model delivers several empirical implications and it is worth collecting them in a
subsection. To wit, as µ increases, the firm’s payoff depends more on its large customer.
That large customer can then more easily “induce” the firm to undertake the relationship-
specific, risky project. For larger µ, lower quality firms will choose the risky project. It
follows that the quality threshold p∗ declines with customer concentration. A lower p∗
indicates higher overall failure rates for firms that choose to conduct relationship-specific
projects. Anticipating the higher default rates that are associated with those projects, the
bank will require costlier, stricter terms for its loans. We write these central predictions
as follows:
Hypothesis 1 Banks will impose costlier, stricter loan contract terms on firms with
higher customer-base concentration.
9
Hypothesis 2 Firms with higher customer-base concentration will experience higher loan
failure rates.
The logic of our model implies that once a firm conducts relationship-specific invest-
ments for the major customer, the firm’s ability to pay back its bank loan will be more
closely tied to the financial conditions of the customer. In case its large customer enters
into financial distress, for example, the firm may face a cash flow shortage and default
on its loans. On seeing worsening financial conditions of major customers, banks should
further impose costlier and stricter loan contract terms. This natural extension of our
model can be stated as follows:4
Hypothesis 3 Banks will impose costlier, stricter loan contract terms on firms that face
customers in worse financial conditions.
The model predicts that large customers make firms conduct relationship-specific
investments, which leads banks to impose costlier loan terms. The empirical data will
present heterogeneity in firm investment and the intensity of relationship-specific projects
will depend on the industry in which firms operate. In taking our theory to the data,
one can think of industries with more intensive relationship-specific investments as hav-
ing lower levels of p (a downward shift in the distribution of p), as firms face riskier
investments and lower resale values for those investments. In these industries, banks will
impose even higher interest rates and stricter contract terms as customer concentration
increases. In industries with lower requirements for relationship-specific investments, in
contrast, loan contract terms will be relatively less sensitive to customer concentration.
This hypothesis can be stated as follows:
Hypothesis 4 The relation between customer concentration and loan contract terms will
be stronger in industries that involve more relationship-specific investments.
We note that the model focuses on a general notion of firms’ “borrowing costs” for
simplicity. The intuition easily extends to various features of standard loan terms, in-
cluding interest rate spread, maturity, and the presence of restrictive covenants. Making
these contract features costlier and stricter for the firm is meant to deter risk-taking. Our
empirical tests will revolve around each of these observable outcomes: loan markups, loan
maturity, loan covenants, and loan failures. We will also examine how customer concen-
tration affects derivative measures of the economic links between supplier firms and their
banks: depth and duration of lending relationships.
4In an omitted appendix, we formalize an extension of our model containing large customers’ finances.
10
3 Sample Construction and Empirical Methodology
We identify firms’ major customers using Compustat’s Segment Customer database.
Statement of Financial Accounting Standard (SFAS) No.14 requires firms to report all
customers that represent 10% of a firm’s total sales. The Segment database collects
information including the names of the customers and their assigned sales figures. In
identifying customer relations, we focus on recurring customers and exclude those that
appear for fewer than three times for a firm in the sample period. We focus on manufac-
turers (SIC 2000–3999) to ease the data collection process, to facilitate comparisons across
firms (reduce uncontrolled heterogeneity), and because firms operating in this sector re-
late more naturally to our supply-chain story. Information from the U.S. input/output
matrix suggests that supplier–customer links in the manufacturing sector feature firms
on both ends of the relationship.5
We extract bank loan contract information from LPC–Dealscan and link loan-level
data to Compustat firm data following Chava and Roberts (2008). We treat each loan
facility as an independent contract. We examine revolvers and term loans since these
types of loans contain more detailed information on the pricing and the restrictiveness of
bank credit.
We construct our final sample by combining the customer and bank loan databases
from 1985 through 2010. For a firm to be included in the sample, we require it to have
available customer information, basic loan features, and information on standard vari-
ables such as size, leverage, and market-to-book. We glean into how banks update loan
pricing and other contracting features by focusing on newly initiated (or renegotiated)
loans during the year when the firms report customer information. Following the existing
literature (e.g., Lin et al. (2011), Valta (2012), and Hertzel and Officer (2012)), we do
not repeatedly account for the same loans for the years after initiation. As a result, our
dataset has a panel structure in which individual firms appear sparsely (more on this
shortly). Our sample consists of 3,375 loans granted to 1,110 individual manufacturers.
3.1 Customer Concentration
The literature does not provide a consistent way to measure customer concentration.
Previous studies use a variety of measures, most of which are computations based on
the percentage of firm sales to major customers (examples are Banerjee et al. (2008)
5Over two-thirds of output in those industries is sold as intermediary goods to other manufacturers,the remainder largely goes to bulk retailers.
11
and Patatoukas (2012)). These approaches fit conveniently our theoretical analysis of
customer concentration, which is based on the parameter µ. We experiment with several
measures of concentration meant to capture the importance of a firm’s large customers.
Our first measure of concentration is based on the percentage of sales that a firm as-
signs to its major customers (similar to Banerjee et al. (2008)). In particular, we define
CustomerSales as the sum of the percentage sales coming from the set of customers the
firm reports as “major customers” (i.e., those with at least 10% of total sales). Custom-
erSales is computed as:
CustomerSalesi =
ni∑j=1
%Salesij,
where ni is the number of firm i ’s major customers, and %Salesij = Sales of i to jTotal Sales of i
, is the
percentage sales from firm i to customer j over i ’s total sales. A high level of Customer-
Sales means that a large proportion of a firm’s sales goes to its major customers.
Our second measure is the sales-weighted size of a firm’s major customers. This mea-
sure is more nuanced than the first in that it gives more importance to major customers
that also happen to be larger firms; likely more significant supply-chain partners. We
define CustomerSize as the firm’s percentage sales to major customers weighted by the
size of those customers. CustomerSize is computed as follows:
CustomerSizei =
ni∑j=1
%Salesij × Sizej,
where Sizej is the size (defined by log of total assets) of customer j. A high level of
CustomerSize means that a firm relies more heavily on fewer, large-sized customers.
Our third measure follows Patatoukas (2012), who defines concentration based on the
notion of a Herfindahl index of sales to large customers:
CustomerHHIi =
ni∑j=1
%Sales2ij.
Expanding on this approach, we also compute the Gini coefficient of a firm’s sales to
its large customers, a measure we call CustomerGini. Conversely, a simpler approach is
based on the percentage sales a firm assigns to its single largest customer, CustomerMax
= maxj=1,...,ni
%Salesij. Alternatively, the simple count of the number of large customers
12
provide an indication of the extent to which a firm deals with few, major customers,
CustomerCount = ni.
These different measures provide unique insights about the extent to which a firm is
engaged in trades with major customers in its supply-chain. While we conduct the bulk
of our analysis based on the first two measures, we often report results for other measures
as well.6
3.2 Borrowing Terms
Chava and Roberts (2008), Roberts and Sufi (2009), and Campello et al. (2011) de-
scribe the elements of the LPC–Dealscan dataset that are relevant for our analysis. We
follow the methodology in Campello et al. and measure three contract features of bank
loan terms. The first is loan spread (LoanSpread). LoanSpread is the “All-in-Drawn”
spread (in basis-points) over LIBOR. “All-in-Drawn” spread is computed as the sum of
coupon and annual fees on the loan in excess of six-month LIBOR. The second feature
is loan maturity (LoanMaturity). LoanMaturity is the number of months until maturity.
Finally, we count the total number of restrictive covenants present in the loan facility
(LoanCovenants).
3.3 Banking Relationships
In addition to adjusting the terms of new loan facilities, banks can also react to a
firm’s customer concentration by abbreviating their relationships with the firm. If cus-
tomer concentration is related to excessive credit risk or undesirable investment choices,
banks can reduce the amount of funds lent to firms or even terminate their relations. We
design empirical measures of firm–bank relationships to capture these dynamics.7
Each time a firm discloses its customer concentration, we look forward in the sam-
ple window searching for subsequent loan arrangements (renewed relations in the future)
with its current banks. We measure these future relations using two methods. First,
we measure the additional amount of lending extended by the bank to the firm given
the information of customer concentration (FutureLoans). FutureLoans is defined as the
total dollar amount of loans issued by the same bank following the currently observed
6In Appendix B, for example, we report the central tests of the paper (on loan markups) for all sixmeasures.
7A related paper is Bharath et al. (2011), who focus on the relative importance of a particular bankcompared to other banks. We get qualitatively similar results using the measure employed by thoseauthors.
13
agreement, scaled by the dollar amount of current loans. Higher values of FutureLoans
suggest that the bank maintains a stronger relationship with the firm after the disclosure
of information about customer concentration.
Our second measure of banking relationship is the length of the future banking re-
lationship, defined as the number of months in which the bank continues to lend to the
firm in the future (FutureDuration). For each year a firm receives a bank loan, Future-
Duration counts the number of months until the last occurrence of the firm receiving a
loan from the current bank. Similar to FutureLoans, FutureDuration also reflects a bank’s
commitment to the lending relationship. However, it emphasizes the length rather than
the intensity of the relationship.
Naturally, both measures of banking relationships suffer from attrition bias, as we
observe shorter future duration and less future lending as we approach the end of the
sample. We thus restrict our banking relationship tests to fiscal years prior to 2007.8
3.4 Loan Failures
To corroborate our argument that a more concentrated customer base is associated
with worse creditworthiness, we also examine the relation between loan failure rates and
customer concentration. If customer concentration is associated with a higher likelihood
of loan failure, banks will naturally impose stricter loan terms ex-ante.
We examine this conjecture by using bankruptcy data from Sudheer Chava’s collec-
tion of corporate bankruptcies.9 This database provides a comprehensive coverage of
bankruptcy filings over the 1962–2010 window. We match these bankruptcy data with
bank loan information, and identify a loan failure event if the borrower files for bankruptcy
prior to an existing loan maturity date. For this case, we assign an indicator variable
LoanFailure to 1. If there is no bankruptcy before the maturity of the loan, then Loan-
Failure is set to 0. We match the loan failure variable with firms’ customer information
and identify 241 loan failures in our sample.
3.5 Empirical Methodology
We estimate panel regression models for our baseline tests that build on a large litera-
ture on bank loans (see, among others, Graham et al. (2008), Campello et al. (2011), and
Lin et al. (2011)). Our models regress loan term variables on customer concentration
8Nonetheless, our results are unaffected if we do not impose this time window constraint.9We thank Sudheer Chava for kindly providing these updated data.
14
measures together with standard firm-level, loan-level, and macro-level controls. The
specifications also feature bank effects, capturing firm–bank pairings. The model can be
written as follows:
LoanTermi,k,t = β0 + β1CustomerConcentrationi,t + β2FirmCharacteristici,t
+ β3MacroV ariablet + β4LoanCharacteristick,t +∑g
Industryg +∑h
Bankh + εi,k,t,
(9)
where i indicates the supplier, k indicates newly initiated loans, t indicates the year of
the loan initiation; LoanTerm ∈ {LoanSpread, LoanMaturity, LoanCovenants}, and Cus-
tomerConcentration ∈ {CustomerSales, CustomerSize}. Borrowing terms and customer
concentration may vary significantly across industries. We thus include industry-fixed
effects (Industryg). Differences of borrowing terms can also arise from banks’ screening
technology. Some banks may be able to better detect firms’ credit quality or to more
closely monitor the firms. These banks may select firms with lower customer concentra-
tion and impose looser borrowing terms. We thus include bank-fixed effects (Bankh).
We report heteroskedasticity-robust errors clustered by industry.
Firm characteristics include proxies for profitability, size, age, tangibility, market-to-
book, leverage, and credit ratings. Macroeconomic conditions are measured by the credit
spread, the term spread, and the GDP growth rate. Loan characteristics include loan
maturity, loan size, and loan spread. We also include a dummy variable for loan type
(term loans or revolvers). A detailed definition of the variables is provided in Appendix A.
Our model predicts that customer concentration has negative implications for firm
borrowing terms. Therefore, we expect the coefficient on customer concentration, β1, to
be positive in the regressions for loan spreads and for the number of covenants. In the
regression for loan maturity, we expect that coefficient to be negative.
We estimate analogous models for the link between customer concentration and future
firm–bank relationships as follows:
BankingRelationshipi,h,t = β0+β1CustomerConcentrationi,t+β2FirmCharacteristici,t
+ β3MacroV ariablet +∑g
Industryg +∑h
Bankh + ui,h,t, (10)
where h indicates the lending bank and BankingRelationship ∈ {FutureLoans, FutureDu-
ration}. We expect customer concentration to hamper firms’ future relationships with
15
Figure 1. The frequency of firms’ distinct appearances in the sample. This figure shows thenumber of firms who appear in the sample for a certain number of distinct observations. The horizontalaxis shows the number of distinct observations. The vertical axis shows the number (frequency) of firms.
their banks. Therefore, we expect the coefficient β1 to be negative in both banking
relationship regressions.
3.6 Data Structure and Summary Statistics
Similar to prior studies on contracts features (Graham et al. (2008), Campello et
al. (2011), and Lin et al. (2011)), the unit of observation in our baseline tests is a loan
contract. As such, we only observe variations in a firm’s customer concentration if the
firm signs new contracts in different years. This results in few recurrences for each firm.
Figure 1 plots the histogram of firms’ distinct observations (entries) in the sample. The
distribution is highly skewed, with few firms appearing in the sample more than two or
three times. Indeed, nearly 50% of the firms appear in the sample only once. Given this
data structure, similar to prior studies in the loan literature, we are unable to include
firm-fixed effects in our regressions.
Table 1 reports the summary statistics of the suppliers’ characteristics, customer
concentration, loan terms, and banking relationship measures in our sample. The firms
sampled attribute about 30% of their sales to major customers. These firms, on average,
have total assets of 590 million, asset tangibility of 26%, and leverage of 29%. These
figures are similar to those in Campello et al. (2011), among others, who report total asset
of 680 million, tangibility of 33%, and leverage of 29%. The average loan contracts in our
16
0.135
0.140
0.145
0.150
0.155
0.160
0.165
1 2 3 4 5 6 7 8 9 10
Rank of Customer Concentration
Profitability
0.085
0.095
0.105
0.115
0.125
0.135
0.145
0.155
0.165
1 2 3 4 5 6 7 8 9 10
Rank of Customer Concentration
Sales Growth
Figure 2. The relation between customer concentration and firm operating performance.The left panel shows the relation between customer concentration and profitability; the right panel showsthe relation between concentration and sales growth. Customer concentration is measured by the totalpercentage sales to all major customers, CustomerSales. The decile ranking of CustomerSales is shownon the horizontal axes.
sample have spreads of 179 bps over LIBOR, maturity of 46 months, and 1.8 covenants.
Table 1 About Here
4 Univariate Analysis
We start our investigation by characterizing the very phenomenon of customer con-
centration, which is still understudied. Prior research points to significant gains in con-
centrating sales to a small group of buyers. These benefits arise from the argument that
firms can achieve economies of scale and superior operating efficiency (cf. Patatoukas
(2012) and Irvine et al. (2013)). It is important that we verify these benefits in our data.
Otherwise, one could attribute the worsening of borrowing terms that we document to
potentially negative effects of customer concentration on operating performance. Along
similar lines, we conjecture that customer concentration may be associated with other
firm characteristics that influence their credit terms. Although our multivariate analyses
are designed to address concerns about confounding heterogeneity effects, it is impor-
tant that we have a basic understanding of these relations. As we show next, customer
concentration is related to firm fundamentals such as size, age, and technology.
4.1 Customer Concentration and Firm Operating Performance
We depict the relation between customer concentration and operating performance in
Figure 2. Following Patatoukas (2012), we rank firms into deciles according to their cus-
17
0.25
0.27
0.29
0.31
0.33
0.35
0.37
0.39
0.41
0.43
0.45
1 2 3 4 5 6 7 8 9 10
Rank of Firm Size
Customer Concentration
0.25
0.27
0.29
0.31
0.33
0.35
0.37
0.39
0.41
0.43
0.45
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
Firm Age (in Years)
Customer Concentration
Figure 3. The relation between customer concentration with firm size and age. The leftpanel shows the relation of size with customer concentration, where the horizontal axis shows the decileranking of firm size. The right panel shows the relation of firm age with customer concentration, wherethe horizontal axis shows firm age. Customer concentration is measured by the total percentage sales toall major customers, CustomerSales.
tomer concentration measure CustomerSales and plot the average operating performance
of firms in each decile. The left (right) panel shows the average sales growth (profitabil-
ity) of firms in each customer concentration level. Sales performance and profitability
both increase with customer concentration. Firms in the lowest deciles of customer con-
centration, for example, observe average annual sales growth of about 9%, while those in
the highest deciles observe 13% or higher growth (nearly 50% higher growth rates). The
patterns we document in Figure 2 are consistent Patatoukas’s argument that firms with
concentrated customer bases enjoy improved performance (see also Irvine et al. (2013)).
Important for our purposes, those patterns show that firms with high customer concen-
tration are not “worse firms” who observe low profits and should naturally face costlier,
stricter loan terms.
4.2 Customer Concentration and Firm Characteristics
Customer concentration can be correlated with important firm characteristics such as
size, age, leverage, and market-to-book. We explore the relation of customer concentra-
tion with these firm characteristics, since they may also affect credit terms.
Figure 3 shows that customer concentration is negatively related to firm size and
age, indicating that smaller, younger firms tend to deal more frequently with major
customers.10 This association can lead to spurious correlation between customer concen-
10To depict the process of customer-base concentration across firms’ “life cycle,” we only include inthe figure firms whose life spans exceed 10 years. Yet, including firms who exist in the sample for fewer
18
0.20
0.21
0.22
0.23
0.24
0.25
0.26
0.27
0.28
1 2 3 4 5 6 7 8 9 10
Rank of Customer Concentration
Leverage
1.5
1.6
1.7
1.8
1.9
2.0
2.1
2.2
2.3
2.4
2.5
1 2 3 4 5 6 7 8 9 10
Rank of Customer Concentration
Market-to-Book
Figure 4. The relation between customer concentration with firm leverage and market-to-book. The left panel shows the relation between customer concentration and firm leverage. Theright panel shows the relation between concentration and market-to-book. Customer concentration ismeasured by the total percentage sales to all major customers, CustomerSales. The decile ranking ofCustomerSales is shown on the horizontal axes.
tration and loan terms, since smaller, younger firms also tend to face more informational
problems, hence higher borrowing costs. It is thus important to control for firm size and
age effects in our tests.
Figure 4 provides further insights into firms that operate with higher levels of cus-
tomer concentration. Concentration is associated with lower leverage ratios. It is also
associated with higher market-to-book ratios. Notably, research shows that firms with
lower leverage and higher market-to-book are able to command lower interest rate spreads
in their bank loans (e.g., Graham et al. (2008), Campello et al. (2011), and Lin et al.
(2011)). These findings corroborate the argument that firms with major customers are
not underperforming businesses that are naturally prone to receive costlier, stricter loan
terms from their banks.
The concentration of firms’ customer bases may well vary across industries. Figure 5
shows the distribution of customer concentration in each manufacturing industry. It
shows that firms in most manufacturing industries have customer concentration around
20% to 30%. However, in industries such as furniture and fixtures, customer bases reach
nearly 50%, while in the food industry the average is only about 17%. These patterns
suggest that it is necessary for account for industry effects in our analysis.
than 10 years does not change our inferences.
19
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Distribution of Customer Concentration by Industry
(p25) Customer Concentration (Median) Customer Concentration (p75) Customer Concentration
Figure 5. The dispersion of customer concentration across industries. This figure shows the 25percentile, median, and 75 percentile levels of customer concentration for each manufacturing industry.Industry is classified using 2-digit SIC code, and customer concentration is measured by CustomerSales,total percentage sales to major customers.
4.3 Customer Concentration and Relationship-Specific Invest-
ment
Our model implies that higher customer concentration may prompt a higher level of
relationship-specific investment. To wit, Eq. (8) shows that a higher µ induces a lower
p∗, meaning that higher customer concentration will induce more firms to undertake the
relationship-specific project. This result seems sensible: an important customer can more
easily contract with the firm to invest in customized projects that are suitable to its par-
ticular needs. Firms who do not have to cater to a major customer, on the other hand,
only need to manufacture standardized products. We study this implication by exam-
ining the relation between customer concentration and relationship-specific investment.
While it is difficult to measure relationship-specific investment, we gauge the uniqueness
of firms’ investment and production following the existing literature.
Our first measure focuses on the level of differentiated inputs firms use. In particular,
we examine how specific are the inputs they employ in their production process. Firms
that use more differentiated inputs have been shown to offer more differentiated products
to their customers, and we use this metric to gauge of the depth of their relationship.
20
Figure 6. The relation between customer concentration and relationship-specific invest-ments. The left panel shows the relation of customer concentration with firms’ uniqueness of inputs,measured by the percentage of firms’ inputs from industries producing differentiated goods; the middlepanel shows the relation of concentration with the originality of patents; and the right panel shows therelation of concentration with R&D intensity. Customer concentration is measured by the total percent-age sales to all major customers, CustomerSales. The decile ranking of CustomerSales is shown on thehorizontal axes.
Giannetti et al. (2011) provide detailed information on industries’ use of differentiated
inputs. We follow Giannetti et al.’s approach and assign a firm to a given level of differ-
entiated inputs usage according to the industry in which it operates. Our second measure
is based on the uniqueness of firms’ patents. We gather information on firms’ patents
from the NBER database and focus on “patent uniqueness,” measured by the width of
different patents cited in the creation of firms’ granted patents (see Hall et al. (2001) for
details). Firms with more unique patents are thought to develop and sell more unique,
customized products to their customers. Finally, following prior literature (e.g., Allen
and Phillips (2000) and Kale and Shahrur (2007)), we also measure relationship-specific
investments using firms’ R&D intensity, defined as the ratio of firms’ R&D expenditure
scaled by total assets.
Figure 6 points to a positive relation between firms’ relationship-specific investments
and customer concentration. The plots in the figure suggest that firms that have higher
customer concentration use more specific inputs for production, produce more original
patents, and invest more in R&D. These patterns are consistent with the economics of
our model. They are also consistent with theories of optimal stakeholder investment and
the limits of the firm, which predict that a firm’s relations with its important stakeholders
will involve relationship-specific investments (e.g., Titman (1984) and Hart (1995)).
21
5 Regression Analysis
We estimate panel regressions of loan spreads, the number of restrictive covenants,
loan maturity, and future banking relationship variables (length and depth) on each of
our measures of customer concentration. In each estimation run, we first control for firm-
level characteristics, industry-fixed effects, and bank-fixed effects. We then augment the
model with macro variables as controls. In the last round, we further include loan-level
characteristics in the set of controls. Tables 2 through 5 present results for 22 such models.
Table 2 shows the results for regressions of loan spreads on alternative measures of
customer concentration, CustomerSales and CustomerSize. Both measures attract signif-
icant and positive coefficients across all estimations, suggesting that firms with a higher
customer concentration are asked to pay higher spreads in their next bank loan facilities.
The most conservative set of estimates in the table (column (6), featuring the full set
of controls) suggests that a one-standard-deviation increase in customer concentration
is associated with a 10 basis points increase in the loan spread. This amounts to a 6%
increase relative to the average loan spread of 179 basis points. Our results suggest that
banks consider customer concentration as a negative factor affecting firms’ credit quality;
a factor that is priced into loan mark-ups.
Table 2 About Here
Table 3 shows results for the number of restrictive covenants. Both measures of cus-
tomer concentration attract positive and statistically significant coefficients across all
regressions, suggesting that firms with high customer concentration have more restrictive
covenants written in their new loan contracts. The estimates from column (6) indicate
that a one-standard-deviation increase in customer concentration is associated with a 0.2
increase in the number of restrictive covenants, which accounts for a 12% increase relative
to the average number of covenants (1.8 covenants) for the loans in the sample.
Table 3 About Here
Table 4 shows results for loan maturity. In this set of regressions, only term loans
are used, since revolvers do not set fixed loan maturities. Both measures of customer
concentration attract negative coefficients, suggesting that firms with higher customer
concentration receive loans with shorter maturity. The statistical significance of these
estimates is less strong. In economic terms, however, a one-standard-deviation increase
22
in customer concentration is associated a 2-month reduction in loan maturity; compared
to the average maturity of 46 months.
Table 4 About Here
We also study the impact of customer concentration on length and depth of the
relationship of the supplier firm and its bank. The results are reported in Table 5. In
economic terms, the most conservative estimates in the table suggest that a one-standard-
deviation increase in customer concentration is associated with a decline in the amount
of future lending extended by the firm’s bank equivalent to 10% of the sample mean.
A one-standard-deviation increase in concentration is associated with 8 fewer months of
future relations with the bank; a 15% drop from the mean.
Table 5 About Here
The results in Tables 2 through 5 are internally coherent and consistent with the pre-
dictions of our simple theoretical model. They show that a more concentrated customer
base is associated with costlier, stricter loan terms for the firm’s loans, including higher
interest rate spreads, more restrictive covenants, and shorter maturities. Customer con-
centration is also associated with the deterioration of firm–bank associations, represented
by shorter relations and less lending by the bank in the future.
6 An Instrumental Variables Approach
Although common in the loan contracting literature, regressions such as those per-
formed in the previous section are subject to several concerns about estimation biases.
Concerns about endogeneity biases, in particular, may arise from the fact the model
lacks an explicit source of exogenous variation in concentration. To allay those concerns,
this section conducts instrumental variables (IV) tests that exploit sharp shifts in the
concentration of a firm’s customer base.
We use aggregate merger and acquisition activity in customers’ industries (down-
stream M&A) as an instrument in assessing the impact of customer concentration on
suppliers’ loan terms and banking relationships. Our instrumental approach implies that
suppliers will face a more concentrated customer base following M&A waves in their cus-
tomers’ industries (inclusion restriction). Existing research supports such prior (e.g., Fee
23
and Thomas (2004), and Bhattacharya and Nain (2011)) and we verify that this is indeed
the case in the tests below. The approach also assumes that downstream M&A affects
suppliers’ borrowing terms through customer–supplier links (exclusion restriction). This
is a plausible assumption since downstream M&A activity (among customers) is not a
policy variable for the supplier. Yet, that activity may be influenced by suppliers’ indus-
try dynamics affecting both suppliers’ customer base and their financial conditions (e.g.,
industry profit declines can be correlated with both a firm’s customer concentration and
credit condition). To control for these effects, as we discuss below, we further introduce
suppliers’ industry-level, time-varying effects in our tests.
6.1 Measuring M&A Activity in the Customer Industry
We gather information on M&A deals from SDC database and apply the following
data filters following Ahern and Harford (2014): 1) only include completed deals; 2) both
the acquirer and target are U.S. firms; 3) the acquirer can be matched with a Compustat
identifier; 4) the acquirer purchases at least 20% of the target during the transaction, and
owns more than 51% after the transaction; and 5) the acquirer does not buy its suppliers.
Finally, we exclude suppliers who are in the same 2-digit SIC industry as their customers.
We use the transaction values of M&As scaled by the acquirers’ total sales as a proxy
for acquisition activity. An industry-level 5-year mean acquisition activity is measured
as the average acquisition of firms in the industry over the past five years. Each of the
firms in our sample supplies products to a portfolio of customers, and those customers
may be in different industries. For each sample firm, we gauge the potential impact of
downstream M&A activity on customer concentration by taking the average of the 5-year
acquisition activity across the industries to which the firm’s customers belong. We refer
to this variable as CustumerM&A. The variable is defined as follows:
CustomerM&Ai =1
ni
ni∑j=1
Industry Average(Acquisitionj
Salesj).
6.2 IV Specification and Results
We use two-stage least square regressions to reassess the impact of customer concen-
tration on loan terms. In the first stage, we regress customer concentration measures
(CustomerSales and CustomerSize) on CustumerM&A, together with a full set of con-
trols. In the second stage, we regress borrowing terms and banking relationship variables
24
on the projected customer concentration measures, together with controls. The two-stage
system for loan terms can be written as follows:
CustomerConcentrationi,k,t = β0 + β1CustomerM&Ai,t + Controls+ εi,k,t [First stage] (15a)
LoanTermsi,k,t = β2 + β3 ̂CustomerConcentrationi,k,t + Controls+ νi,k,t, [Second stage] (15b)
where i indicates the supplier, k indicates loans, t indicates the fiscal year. LoanTerms in-
cludes the loan term variables LoanSpread, LoanMaturity, and LoanCovenants. Customer-
Concentration indicates customer concentration measures. ̂CustomerConcentration de-
notes the predicted value of customer concentration from the first-stage regression (Eq.
(15a)), reflecting the variation of customer concentration induced by the variation M&A
activity in customers’ industries. Controls contains the set of control variables used in
our baseline regressions, including firm-level controls, macro variables, loan-level controls,
and bank-fixed effects.
Our IV tests further account for unobserved industry dynamics that can drive both
downstream merger waves and firms’ credit terms. Economic and technological shocks to
an industry, for example, can lead to merger waves (Harford (2005)). These industry-wide
shocks may further propagate along the supply chain and affect all firms in the system. To
control for industry-level, time-varying dynamics that may confound our results, we fur-
ther incorporate supplier industry-year-fixed effects in our specifications. The fixed effects
approach removes any unobservable shock that is common to an industry in a given time
period, thus preventing it from contaminating the exclusion restriction of our instrument.
The two-stage system for banking relationships can be written as:
CustomerConcentrationi,h,t = β0 + β1CustomerM&Ai,t + Controls+ εi,h,t [First stage] (16a)
BankingRelationshipi,h,t = β2 + β3 ̂CustomerConcentrationi,h,t + Controls+ ηi,h,t, [Second stage] (16b)
where h indicates the lending bank. BankingRelationship is the measures for bank rela-
tionships: FutureLoans and FutureDuration.
In the first stage (Eqs. (15a) and (16a)), we expect β1 to be positive, indicating that
suppliers experience increases in customer concentration following high levels of M&A
activity in the customers’ industries. In the second stage, we expect β3 to be positive for
the loan spreads and covenants regressions, and negative for the loan maturity and bank
relationship regressions.
Table 6 shows the first-stage regression results for customer concentration on cus-
tomer industry acquisition activity (CustumerM&A). The instrument loads significantly
25
positively in all models. For brevity, we only present results for our first customer con-
centration measure, CustomerSales (the results are similar for CustomerSize). Results
in Table 6 are consistent with the prior of our identification strategy implying that firms
face more concentrated customer bases following high levels of M&A activity in their cus-
tomers’ industries. In economic terms, a one-standard-deviation increase in downstream
M&A contributes to a 3.6-percentage points increase in customer concentration, which is
a 11% increase relative to the average customer concentration. Notably, the F -statistics
from the first-stage regressions pass the weak identification tests at the 1% level. The
Kleinberg-Paap statistics pass the under-identification tests.
Table 6 About Here
Table 7 shows the second-stage regression results of loan terms and banking relation-
ships on the instrumented customer concentration. Consistent with our OLS results, the
instrumented customer concentration is positive and statistically significant in the loan
spread and loan covenants regressions. Also consistent with the OLS results, the coeffi-
cients associated with customer concentration are negative and statistically significant in
the loan maturity and future banking relationship regressions.
Table 7 About Here
Our IV results confirm our baseline findings that increases in customer concentration
lead to higher loan spreads and more loan covenants for supplier firms. Higher customer
concentration also leads to lower loan maturity, shorter banking relationships, and less
lending extended by the same bank in the future. The economic and statistical signif-
icance of the impact of customer concentration on all the outcome variables we study
are magnified under the IV approach. In all, the evidence we gather suggests that while
customer concentration may make supply chain relations more efficient and increase firm
profits, a deeper exposure to a small set of customers also has negative consequences for
a firm’s relations with its creditors.
6.3 Strategy Verification: Focusing on Regulatory-Induced M&A
Activity
The merger activity we exploit in our IV tests are spontaneous and one could be
concerned that unobserved developments or trends affecting the supplier firms we study
26
could: 1) lead to changes in the contracting terms those firms receive from their banks,
and simultaneously 2) trigger M&A activity among their customers in downstream indus-
tries. It is ultimately hard to completely rule in or rule out such a claim, but we take one
extra step at ensuring that this is not a concern for the results derived from our IV tests.
We do so designing a test strategy that relies solely on downstream M&A activity that
is associated with acts by the Federal Government that change entry, prices, and other
elements of the competitive environment of the downstream industry affected. Ample
literature shows that these regulatory actions have led to spikes in the M&A activity
in the industries affected (e.g., Schoenberg and Reeves (1999), Andrade et al. (2001),
Bruner (2004), and Ovtchinnikov (2010)). We use these induced spikes in M&A in a
difference-in-differences framework.
There were eight major nation-wide regulatory actions applicable to our analysis in
the sample period we examine: the Natural Gas Wellhead Decontrol Act (1989), the
Cable Television Consumer Protection and Competition Act (1992), the Energy Pol-
icy Act (1992), the Federal Energy Regulatory Commission (FERC) Order 636 (1992),
the Negotiated Rates Act (1992), the Trucking Industry and Regulatory Reform Act
(1992), the Telecommunications Act (1996), and the FERC Order 888 (1996).11 The af-
fected (“deregulated”) industries include natural gas, cable TV, trucking, telephone, and
wholesale, defined at the 4-digit SIC level. We verify claims by previous researchers that
these Federally-mandated Acts and Orders affecting the competitive environment had
the end result of triggering M&A activity. We do so by comparing the M&A activity in
the industries affected by regulation with the M&A activity in other industries that share
the same 3-digit SIC codes. As shown in Figure 7, the average level of M&A activity
(acquisition deal value scaled by total sales of acquirers) in the deregulated industries
increased 3- fold following a regulatory innovation. By comparison, the M&A activity in
other industries under the same 3-digit SIC code did not exhibit any change.
The next step of our analysis is to establish supplier–customer links for the industries
affected by the eight regulatory events listed above. Our specific goal is to identify the
suppliers of these deregulated industries. We do so using the input–output (I–O) ma-
trix provided by the Bureau of Economic Analysis.12 To closely capture input–output
relations amongst industries in our sample period, we use the I–O matrix of 1987 for
the regulatory event taking place in 1989, and the I–O matrix of 1992 for the events of
1992 and 1996. We first identify the suppliers of the deregulated industries, and examine
11See Asker and Ljungqvist (2010) for a detailed summary of these deregulation events.12The matching of 4-digit SIC industries with I-O industries comes from Joseph Fan’s website.
27
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
-3 -2 -1 0 1 2 3
Dea
l V
alu
e/S
ales
Years around Deregulation
Merger Acitivity around Deregulation
Control Industries Deregulated Industries
Figure 7. M&A activity around deregulation. This figure shows the industry average level ofM&A activity around deregulation events. The horizontal axis shows the number of years past thederegulation event, and the vertical axis shows the industry average level of M&A activity, measuredby acquisition value scaled by sales. Deregulated industries are classified at the 4-digit SIC level, whilecontrol industries include other firms in the same 3-digit SIC industries as the deregulated firms, butwith different 4-digit SIC codes.
the changes in firm-level customer concentration and loan contract terms for firms in the
supplier industries.13 In other words, we consider the suppliers of deregulated industries
as our treated group. We define our control group as the other firms in the same 3-digit
SIC industries as these suppliers, but with different 4-digit SIC codes (non-deregulated
industries). In this way, we are able to keep similarity between treated and control groups
both economically and in terms of sample size. We further restrict the event period to
be from 3 years before to 3 years after each regulatory event.
Using a differences-in-differences panel setup, we examine how regulatory-induced
downstream mergers affect a firm’s customer concentration and the loan contract terms
it receives. Formally, we estimate the following regressions:
Yi,k,t = β0 + β1Treatedi + β2Postt + β3Treatedi × Postt + Controls+ εi,k,t, (17)
where i indicates the borrower, k indicates loans, t indicates the fiscal year. Outcome
variable Y includes CustomerSales and LoanSpread. Treated is a dummy variable indicat-
13We restrict suppliers to be in upstream industries that attribute over 3% sales to the deregulatedindustries.
28
ing whether a firm’s customer industry experienced a Federally-mandated Act or Order
affecting its competitive environment. Post is a dummy variable covering the period fol-
lowing the regulatory event. Finally, Controls contains the set of control variables used
in our baseline regressions, including firm-level controls, macro-level variables, loan-level
controls, industry-fixed effects, and bank-fixed effects.
Table 8 reports the results from Eq. (17). Columns (1) and (2) show the results for
customer concentration. The interaction term Treated×Post loads positively, indicating
that the mergers driven by deregulation events increase the customer-base concentration
for suppliers. Columns (3) and (4) show the results for loan markups, where the in-
teraction term Treated × Post also loads positively. This result suggests that creditors
become concerned about suppliers whose customers concentrate following regulatory acts,
imposing higher interest rates on their loans.
Table 8 About Here
The evidence we gather from tests around regulatory events are consistent with our
prior results on the relation between customer concentration and credit terms. In particu-
lar, the study of downstream M&A activity triggered by Federally-mandated Acts and Or-
ders supports the results of our larger IV-based tests, which suggest that firms’ customer
concentration can lead to costlier, stricter loan contract terms imposed by their banks.
7 Deepening Relations Along the Supply Chain
7.1 Relationship-Specific Investments
According to our model, the relationship-specific nature of the investments firms make
may be a function of the importance of its major customers. In Section 4, we reported a
robust association between customer concentration and relationship-specific investments
(RSI). In this section, we examine whether the relation between customer concentration
and bank loan terms is modulated by the nature (or specificity) of the investments made
by firms.
Recent research by Giannetti et al. (2011) suggests that firms that use more differen-
tiated inputs also supply more differentiated products to their customers. We follow their
analysis in using input differentiation as a proxy for the prevalence of RSI across different
industries. We start by classifying firms into high and low RSI categories. To do so,
29
we rank the industries in our sample and partition our data so that we have a balanced
number of firms across high, middle, and low levels of RSI (tirciles). We then conduct
subsample analysis using our baseline regression model (Eq. (9)) for firms in high-RSI
industries and low-RSI industries. We expect customer concentration to be a particularly
important determinant of loan contract terms in high-RSI industries, and relatively less
so in low-RSI industries.
The results are reported in Table 9. Column (1) shows that customer concentration
has a strong positive impact on loan spreads across firms in high-RSI industries (consid-
erably larger than the baseline estimates of column (3) of Table 2). For the subsample
of low-RSI industries, however, the coefficient on customer concentration is economi-
cally negligible and statistically insignificant (see column (4)). This is consistent with
the conjecture that customer concentration affects firms’ credit profile as a function of
relationship-specific investments in supply-chain relations. Additionally, we also find that
customer concentration leads to shorter loan maturities in high-RSI industries. Our tests,
however, do not show a measurable impact of RSI on the association between customer
concentration and covenants.
Table 9 About Here
7.2 Account Receivables Cycle
As firms operate in a more integrated fashion along their supply chain, they may not
only conduct different types of investments, but may also present different financing pat-
terns. In our setting, one would argue that firms dealing with a concentrated customer
base (delivering their preferred products) may receive payments in a more expeditious
basis as a way to finance their production cycle. This follows from our model’s hypothesis
that suppliers and large customer share mutual benefits from deepening their relations.
One way to empirically examine this conjecture is to study the relation between re-
ceivables and customer concentration. We do so using our baseline model and report
the results in Table 10. The dependent variables are account receivables, which we scale,
alternatively, by the cost of goods sold (COGS ) and total sales. The proxies we examined
provide a gauge for the firm’s “account receivables cycle.”
We find results indicating that customer concentration is negatively related to ac-
count receivables cycle. The estimates in Table 10 suggest that a one-standard-deviation
increase in customer concentration is associated with a 1.3% drop in receivables relative
30
to cost of goods sold (compared to the mean of 27%), and a 0.7% drop in receivables
relative to total sales (compared to the mean level of 16%). These results are interesting
in suggesting that firms with concentrated customer bases partially offset their costlier,
stricter bank facilities with more favorable receivables cycle from their large customers.
Table 10 About Here
8 Customer Financial Condition
We study the financial conditions of major customers to better understand how cus-
tomer concentration may trigger worse credit terms for supplier firms. Prior evidence
suggests that when facing financially-distressed customers, firms tend to grant deep con-
cessions to maintain product market relationships (Wilner (2000)). One could expect
a firm whose large customers are in worse financial shape to receive worse loan terms,
including higher spreads, more covenants, and shorter maturity.
8.1 Measuring Large Customers’ Financial Conditions
We design two measures of large customers’ financial conditions. One measure is sim-
ply based on customers’ leverage, the other is based on customers’ probability of default.
When a customer has a higher level of indebtedness, it may have difficulty maintaining
existing purchase schedules or paying suppliers on time. These problems will be more
relevant for suppliers when that customer is large. Accordingly, we construct a measure
of customer financial condition that aggregates customers’ leverage for each supplier.
CustomerLeverage is defined as:
CustomerLeveragei =
ni∑j=1
%Salesij × Leveragej,
where Leveragej is the leverage of customer j. Higher values of CustomerLeverage indi-
cate that the firm’s large customers are more indebted.
Beyond assessing leverage, we directly gauge customers’ financial distress using their
probability of default based on Merton’s (1974) model. We follow Bharath and Shumway
(2008) and employ a reduced form model to calculate customers’ distance to default (DD).
Higher distance to default indicates that a firm is less likely to default. For each supplier,
we measure the average default likelihood of its major customers using its percentage
31
sales to these customers and refer to this variable as CustomerDefault. The variable is
defined as follows:
CustomerDefaulti =
ni∑j=1
%Salesij × (1−DDj),
where DDj is the predicted distance to default for customer j, scaled by 100 so it is within
the range of 0 to 1. Similar to CustomerLeverage, higher values of CustomerDefault in-
dicate that the firm faces a more financially-distressed customer base.
8.2 Results
We report the impact of customers’ finances on firms’ borrowing terms in Table 11,
where we estimate models that resemble those of our baseline tests.14 Columns (1) and
(2) of Table 11 show the relation between loan spreads and customer financial condition
measures CustomerLeverage and CustomerDefault. Both measures attract significant,
positive coefficients, suggesting that firms facing large, financially-distressed customers
experience higher interest spreads on their new loans. Columns (3) and (4) show the
relation between the number of restrictive covenants and customers’ financial condition.
The results suggest that banks impose more restrictive covenants for firms with large,
more financially-troubled customers. Finally, despite weaker statistical significance, re-
sults in columns (5) and (6) suggest that firms with more distressed customers are offered
shorter maturities on their loans. Notably, the economic magnitudes of the coefficients
in Table 11 imply that the negative effects of customer concentration on loan terms are
even larger than those of our baseline tests in Tables 2 through 4.
Table 11 About Here
The results of this section are important in confirming the logic behind our base find-
ings. A deeper relation with a small set of large customers has negative consequences for
a firm’s relations with its creditors. And more so the more financially unhealthy those
customers are. In the next two sections we discuss how fears associated with exposure
and contagion may explain banks’ decision to offer worse credit terms when firms sell
their output to larger customers.
14Results on the duration and depth of bank relationships are omitted for brevity, but are readilyavailable.
32
9 Customer Concentration and Bank Exposure
As a last step of our study, we examine the relation between firm credit risk and
customer concentration. We do this to substantiate the argument that firms with higher
customer concentration face more adverse credit terms because of their deteriorated cred-
itworthiness. Banks are highly regulated intermediaries and the exposure to loan risk
bears tremendous costs for them. If customer concentration increases the likelihood of
loan failure, it should be associated with costlier, stricter bank loan terms. Yet, these
results could be ameliorated if insurance contracts (such as CDS) are available.
In this section, we match our customer and loan datasets with the Sudheer Chava’s
bankruptcy database (updated from Chava and Jarrow (2004)) to test the idea that cus-
tomer concentration is related to loan failures. We also match our sample with data
from CDS market to assess if the costs banks impose on firms with higher customer
concentration are affected by the degree to which creditors can resort to CDS insur-
ance. We also examine the syndication structure of our sample. Finally, we provide a
back-of-the-envelope calculation of the aggregate credit losses associated with customer
concentration, showing how they evolve over time.
9.1 The Likelihood of Loan Failures
The variable LoanFailure indicates whether the company files for bankruptcy before a
loan matures in our dataset. To test the idea that a more concentrated customer base is
associated with higher likelihood of supplier loan failure, we estimate the following logit
regression model:
LoanFailurei,k,t = β0 + β1CustomerConcentrationi,t + β2FirmCharacteristici,t
+ β3MacroV ariablet + β4LoanCharacteristick,t +∑g
Industryg +∑h
Bankh + εi,k,t,
(18)
where CustomerConcentration includes CustomerSales, CustomerSize, CustomerLever-
age, and CustomerDefault. The model features the same sets of controls used in our
baseline regressions.
Table 12 reports results for CustomerSales and CustomerSize. Both customer con-
centration variables have positive and significant coefficients, suggesting that firms with
more concentrated customer bases are more likely to fail during the existence of a loan
33
60
80
100
120
140
160
180
200
220
1 2 3 4 5 6 7 8 9 10
Rank for Customer Concentration
Loan Spreads with Customer Concentration
Spreads for firms without CDS Spreads for firms with CDS
Figure 8. Loan spreads for firms with and without CDS trading. This figure shows therelation between loan spreads and customer concentration for firms that have and have no CDS tradingduring the sample period. The horizontal axis shows the ranks for customer concentration, measured byCustomerSales.
contract, exposing their banks to higher loan default risk. To help interpret our estimates,
note that the coefficient for CustomerSize in column (6) suggests that a one-standard-
deviation increase in customer concentration is associated with a 2-percentage points
increase in loan failure rates. This is a sizable effect, especially in comparison to the av-
erage loan failure rate of 6.1%. In unreported tables, we further show that firms are more
likely to fail on their loans as their customers’ financial condition deteriorates (proxied
by CustomerLeverage and CustomerDefault).
Table 12 About Here
9.2 The Impact of CDS Insurance on Loan Markups
The results from Table 12 show that firms with more concentrated customer bases
are more likely to default on their loans. This begs the question: With the emergence
of CDS markets, can banks hedge away firm default risk using CDSs? To answer this
question, we collect CDS information from Datastream and classify firms into two groups
depending on whether they have CDS traded on their debt. We then examine whether
banks are less concerned about default risk caused by customer concentration when they
can hedge their credit exposures using CDSs. Figure 8 provides evidence supporting this
34
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
1 2 3 4 5 6 7 8 9 10
Rank of Customer Concentration
Share of Lead Arranger (%)
Figure 9. The relation of customer concentration with lead arranger shares. This figureshows the relation between customer concentration and the percentage shares of the syndicated loanstaken by lead arrangers. The horizontal axis shows the ranks for customer concentration, measured byCustomerSales.
conjecture. For firms that have traded CDS contracts, loan spreads do not vary much
with customer concentration. In contrast, for firms that do not have CDSs, loan spreads
increase as firms face higher levels of customer concentration.
9.3 Loan Syndication Structure and Customer Concentration
A potential concern for focusing on the syndicated loan market is that firms facing
concentrated customer bases may self-select into this market in order to avoid intense bank
monitoring. This could potentially cause biases in our estimation due to the following
scenario: firms with higher customer concentration could prefer a syndicated structure
where the lead bank takes a smaller percentage of the facility. In such a scenario, the lead
bank may impose higher spreads simply as a way to compensate for lower monitoring
effort. However, we do not observe any correlation between customer concentration and
lead bank shares; the correlation coefficient is less than 1%. Figure 9 corroborates this
claim. The lack of a meaningful relation between customer concentration and lead bank
shares casts doubt on a self-selection story.15
15We further control for lead bank shares in our analyses. Our results are robust to this additionalcontrol.
35
0
5
10
15
20
25
30
35
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Estimated Loss from Default
Driven by Customer Concentration (Billion $)
(Mean) Loss from Default (p5) Loss from Default (p95) Loss from Default
Figure 10. Loss from default driven by customer concentration over time. This figure showsthe estimated loss from loan failures over time in billion dollars. The solid line shows the estimated lossdriven by the average concentration of borrowers’ customer bases. The dotted lines show the 5% and95% confidence interval of such losses. Customer concentration is measured by CustomerSales.
9.4 The Impact of Customer Concentration on Aggregate Loan
Losses
Finally, we investigate if the additional default hazard driven by customer concentra-
tion generates significant monetary losses. To provide a rough estimate of such losses,
we take the product of four pieces of information: 1) the average loan face value, 2) the
average loss given default (see Altman and Suggitt (2000)), 3) the additional default rate
driven by customer concentration (from in Table 12), and 4) the average customer concen-
tration for our sample firms. Our back-of-the-envelope calculation suggests that customer
concentration leads to around $2 million loss per loan.16 Figure 10 depicts the evolution
of such losses over time. As customer concentration increases over the past two decades, it
leads to increasing losses from loan failures, ranging from around $10 billion per year in the
early-1990s to $20 billion per year in the mid-2000s. Our loan failure analysis helps fur-
ther explain why lenders impose costlier, stricter terms on loans offered to suppliers with
concentrated customer bases, even though those suppliers themselves are more profitable.
16To be precise, we arrive at this number by multiplying average customer concentration 30%, themarginal effect from the loan failure estimate 0.09 (generated by estimates in column (3) in Table 12),the loss given default 20% (from Altman and Suggitt (2000)), and the average loan face value $312 million.
36
10 Concluding Remarks
Recent literature argues that customer concentration can increase firms’ operational
efficiency, leading to significant increases in profitability. While recognizing those gains,
we examine the potential costs associated with deeper relationships between suppliers
and customers.
Our study looks at how credit markets respond to customer-base concentration. We
do so examining the credit terms offered by banks. Specifically, using detailed informa-
tion on bank loans, we study how customer-base concentration affects a firm’s borrowing
terms, including loan interest spread, maturity, and the number of restrictive covenants.
We further gauge the impact of customer profile on the length and intensity of suppliers’
relations with their banks. Our tests show that higher levels of customer concentra-
tion increase the interest rates and the number of restrictive covenants on a firm’s bank
loans. Customer concentration also reduces loan maturities and the duration and depth
of banking relationships.
Exploring the cross-sectional variation in customer characteristics, we further show
that when firms have larger, more financially-distressed customers, they face even worse
loan terms. We also show direct evidence that a more concentrated customer base af-
fects a firm’s credit risk. In particular, suppliers with higher customer concentration and
more financially-troubled customers are more likely to default on their loans. This result
corroborates the argument that while customer concentration may make supply chain re-
lations more efficient and profitable, a deeper exposure to a small set of large customers
has negative consequences for the firm’s relations with its creditors.
Understanding the trade-offs associated with customer concentration is important
as the modern business-to-business economy experiences increasing levels of customer
concentration. Integration along the supply chain is fraught with contracting prob-
lems between customers and suppliers. These problems allow powerful customers to
set trade terms to their own advantage, while leaving suppliers “held-up” due to costly,
relationship-specific investments they may be forced to make. Our paper sheds light on
one additional cost induced by this contracting problem. This phenomenon is reflected
in the increased borrowing costs and tightened debt contract terms required by banks.
Future research would benefit from understanding how bank contracting can complement
or substitute for supply chain contracting. Overall, our analyses suggest that the cost re-
lated to customer concentration deserves better understanding and may give new insights
into important research topics, such as the issue of optimal concentration and integration
along the productive supply chain and the limits of the modern corporation.
37
References
Ahern, K., Harford, J., 2014. “The importance of industry links in merger waves.”Journal of Finance forthcoming.
Allen, J., Phillips, G., 2000. “Corporate equity ownership, strategic alliances, andproduct market relationships.” Journal of Finance 55, 2791-2815.
Altman, J., Suggitt, H., 2000. “Default rates in the syndicated bank loan market: Amortality analysis.” Journal of Banking & Finance 24, 229-253.
Andrade, G., Mitchell, M., Stafford, E., 2001. “New evidence and perspectives onmergers.” Journal of Economic Perspectives 15, 103-120.
Asker, E., Ljungqvist, A., 2000. “Competition and the structure of vertical relationshipsin capital markets.” Journal of Political Economy 118, 599-647.
Banerjee, S., Dasgupta, S., Kim, Y., 2008. “Buyer-supplier relationships and the stake-holder theory of capital structure.” Journal of Finance 63, 2507-2552.
Barrot, J. N., 2014. “Trade credit and industry dynamics: Evidence from truckingfirms.” Working Paper.
Bharath, S., Dahiya, S., Saunders, A., Srinivasan, A., 2011. “Lending relationships andloan contract terms.” Review of Financial Studies 24, 1141-1203.
Bharath, S., Shumway, T., 2008. “Forecasting default with the Merton distance todefault model.” Review of Financial Studies 21, 1339-1369.
Bhattacharya, S., Nain, A., 2011. “Horizontal acquisition and buying power: A productmarket analysis. ” Journal of Financial Economics 99, 97-115.
Biais, B., Gollier, C., 1997. “Trade credit and credit rationing.” Review of FinancialStudies 10, 903-937.
Bolton, P., Scharfstein, D., 1998. “Corporate finance, the theory of the firm, and orga-nizations.” Journal of Economic Perspectives 12, 95-114.
Bruner, R. F., 2004. “Applied mergers and acquisition.” John Wiley & Sons, Inc.
Burkart, M., Ellingsen, T., 2004. “In-kind finance: A theory of trade credit.” Journalof Economic Perspectives 94, 569-590.
Campello, M., Lin, C., Ma, Y., Zou, H., 2011. “The real and financial implications ofcorporate hedging.” Journal of Finance 66, 1615-1647.
Cen, L., Dasgupta, S., Elkamhi, R., Pungaliya, R., 2014. “Reputation and loan contractterms: The role of principal customers.” Working Paper.
Chava, S., Stefanescu, C., Turnbull, S., 2011. “Modeling the loss distribution.” Man-agement Science 57: 1267-1287.
Chava, S. and Jarrow, R., 2004. “Bankruptcy Prediction with Industry Effects.” Reviewof Finance 8, 537-569.
Chava, S., Roberts, M., 2008. “How does financing impact investment? The role of debtcovenants.” Journal of Finance 63, 2085-2121.
38
Cohen, L., Frazzini, A., 2008. “Economic links and predictable returns.” Journal ofFinance 63, 1977-2011.
Erel, I., Jang, Y., Weisbach, M. S., 2014. “Do acquisitions relieve target firms’ financialconstraints?” Journal of Finance forthcoming.
Fee, E., Thomas, S., 2004. “Sources of gains in horizontal mergers: Evidence fromcustomer, supplier, and rival firms.” Journal of Financial Economics 74, 423-460.
Giannetti, M., Burkart, M., Ellingsen, T., 2011. “What you sell is what you lend?Explaining trade credit contracts.” Review of Financial Studies 24, 1261-1298.
Graham, J. R., Li, S., Qiu, J., 2008. “Corporate misreporting and bank loan contract-ing.” Journal of Financial Economics 89, 44-61.
Greene, D., Kini, O., Shenoy, J., 2013. “Buyer power in conglomerate acquisitions.”Working Paper.
Hall, B., Jaffe, A., Trajtenberg, M., 2001. “The NBER patent citations data file:Lessons, insights and methodological tools.” Working Paper.
Harford, J., 2005. “What drives merger waves?” Journal of Financial Economics 77,529-560.
Hart, O., 1995. “Firms, contracts, and financial structure.” Oxford University Press.
Hertzel, M., Officer, M., 2012. “Industry contagion in loan spreads.” Journal of Finan-cial Economics 103, 493-506.
Irvine, P., Park, S., Yildizhan, C., 2013. “Customer-base concentration, profitabilityand distress across the corporate life cycle.” Working Paper.
Kale, J., Shahrur, H., 2007. “Corporate capital structure and the characteristics ofsuppliers and customers.” Journal of Financial Economics 83, 321-365.
Kelly, B., Lustig, H., Van Nieuwerburgh, S., 2013. “Firm volatility in granular net-works.” Chicago Booth Research Paper.
Kolay, M., Lemmon, M., Tashjian, E., 2012. “Spillover effects in the supply chain:Evidence from Chapter 11 filings.” Working Paper.
Lin, C., Ma, Y., Malatesta, P., Xuan, Y., 2011. “Ownership structure and the cost ofcorporate borrowing.” Journal of Financial Economics 100, 1-23.
Merton, R., 1974. “On the pricing of corporate debt: The risk structure of interestrates.” Journal of Finance 29, 449-470.
Murfin, J., Njoroge, K., 2013. “The implicit costs of trade credit borrowing by largefirms.” Working Paper.
Ovtchinnikov, A., 2013. “Merger waves following industry deregulation.” Journal ofCorporate Finance 21, 51-76.
Patatoukas, P. N., 2012. “Customer-base concentration: Implications for firm perfor-mance and capital markets.” The Accounting Review 87, 363-392.
39
Roberts, M. R., Sufi, A., 2009. “Renegotiation of financial contracts: Evidence fromprivate credit agreements.” Journal of Financial Economics 93, 159-184.
Schoenberg, R., Reeves, R., 1999. “What determines acquisition activity within anindustry?” European Management Journal 17, 93-98.
Titman, S., Wessels, R., 1988. “The determinants of capital structure choice.” Journalof Finance 43, 1-19.
Valta, P., 2012. “Competition and the cost of debt.” Journal of Financial Economics105, 661-682.
Wilner, B., 2000. “The exploitation of relationships in financial distress: The case oftrade credit.” Journal of Finance 55, 153-178.
40
Appendix A Variable Definitions
LoanSpread : All-in-Drawn loan spread over LIBOR
LoanCovenants : Total number of covenants on the loan package
LoanMaturity : Loan maturity in months
FutureLoans : The total dollar amount of loans issued by the same bank in the future,scaled by the current loan amount
FutureDuration: The number of months until the last loan extended by the same bank
CustomerSales : Total percentage sales to all major customers
CustomerSize: Total size of all major customers, weighted by the firm’s percentage salesto customer
CustomerLeverage: Leverage of major customers, weighted by the firm’s percentagesales to customer
CustomerDefault : The inverse distance to default (1-distance to default) of major cus-tomers predicted by Bharath and Shumway (2008)’s distance-to-default model,weighted by the firm’s percentage sales to customer
Size: Log of total assets (AT)
Age: Years after a firm’s first appearance in Compustat database
Profitability : Operating income (OIBDP)/total assets
Tangibility : Property, plant, and equipment (PPENT)/total assets
M/B : (Stock price (PRCC)×shares outstanding (CSHO) + total assets – book equity(CEQ))/total assets
Leverage: (Long-term debt (DLTT) + current debt (DLC))/total assets
Ratings : A dummy variable that equals 1 if the firm has a bond rating, 0 otherwise
CreditSpread : Yield spread between average AAA-rated corporate bonds and averageBBB-rated corporate bonds
TermSpread : Yield spread between 10-year Treasury bond and 3-month Treasury bills
GDPGrowth: Quarterly average of GDP growth rate of the year
LoanSize: Log of total loan amount (in dollars)
LoanType: A dummy variable that equals 1 if the loan is a term loan, 0 if loan is arevolver
41
Appendix B
Table A1Loan Spreads and Alternative Measures of Customer ConcentrationThis table shows the relation between loan spreads and alternative measures of customer concentration.The dependent variable is All-in-Drawn loan spread (LoanSpread). All regressions use industry-fixedeffects and bank-fixed effects. Industry is classified as 2-digit SIC industry and the lending banks areclassified by their ultimate parents. Robust-clustered t-statistics are shown in parentheses. Customer-Sales: the sum of the percentage sales to the set of customers a firm reports as “major customers”.Columns (4) through (6) show the regression results for CustomerSize: the total size of all major cus-tomers, weighted by the firm’s percentage sales to these customers. CustomerHHI is measured as thesum of squared percentage sales to major customers; CustomerMax is the highest percentage sales tomajor customers; CustomerCount is the total number of a firm’s major customers; and CustomerGiniis the Gini coefficient of a firm’s sales to its customers.
CustomerConcentration is:
Dep. Var.: LoanSpread (1) (2) (3) (4) (5) (6)Measure of Concentration: CustomerSales CustomerSize CustomerHHI CustomerGini CustomerMax CustomerCount
CustomerConcentration 40.89*** 5.28*** 72.99*** 7.38** 53.10* 6.55*(3.32) (3.72) (5.12) (2.14) (1.83) (1.96)
Size -17.69*** -17.97*** -17.60*** -17.42*** -18.18*** -17.61***(-5.40) (-5.43) (-6.68) (-6.01) (-6.93) (-6.39)
Age -0.60*** -0.61*** -0.78*** -0.82*** -0.85*** -0.82***(-2.88) (-3.07) (-3.77) (-4.04) (-4.33) (-4.05)
Profitability -246.94*** -273.44*** -257.76*** -260.61*** -263.00*** -261.58***(-3.89) (-4.23) (-3.96) (-3.99) (-3.71) (-3.87)
Tangibility -27.66 -22.23 -8.96 -3.84 -0.02 -3.34(-0.77) (-0.62) (-0.42) (-0.17) (-0.00) (-0.14)
M/B -3.75 -1.83 -4.99 -4.96 -4.93 -4.96(-0.90) (-0.40) (-1.55) (-1.56) (-1.44) (-1.53)
Leverage 128.86*** 131.47*** 122.48*** 120.42*** 117.93*** 120.01***(8.49) (8.21) (10.84) (10.68) (9.90) (10.27)
Ratings 35.74*** 35.24*** 41.52*** 41.93*** 42.92*** 42.36***(4.68) (4.62) (6.96) (6.95) (7.77) (7.17)
CreditSpread 81.53*** 76.32*** 88.35*** 89.42*** 88.03*** 89.36***(5.49) (4.97) (6.43) (6.49) (6.35) (6.45)
TermSpread 3.40 4.42* 2.49 2.66 2.65 2.64(1.31) (1.69) (1.27) (1.36) (1.33) (1.33)
GDPGrowth -4.79* -5.18** -3.96 -3.67 -4.14 -3.74(-1.96) (-2.08) (-1.65) (-1.55) (-1.68) (-1.57)
LoanMaturity 5.91 6.48 8.18* 7.99* 8.01* 8.02*(1.50) (1.61) (1.88) (1.83) (1.78) (1.83)
LoanSize -15.77*** -16.02*** -16.04*** -15.67*** -15.74*** -15.71***(-5.66) (-5.69) (-6.65) (-6.45) (-6.41) (-6.45)
LoanType 45.78*** 44.87*** 48.80*** 49.04*** 48.93*** 49.05***(6.86) (6.76) (7.07) (7.06) (7.05) (7.04)
Observations 2,983 2,983 2,983 2,983 2,983 2,983R-squared 0.61 0.62 0.57 0.57 0.57 0.57
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
42
Table 1Summary StatisticsThis table shows the summary statistics of the firm characteristics variables, customer variables, and loanterm variables. The sample spans the 1985–2010 window, featuring manufacturing firms (SIC 2000–3999)in Compustat that have available firm characteristics and loan term variables. All continuous variablesexcept leverage and the number of covenants (count) are winsorized within 5 and 95 percentile. Leverageis restricted to the 0–1 range.
Variables Mean Std. Dev. p25 p50 p75 #Obs
LoanSpread 179.16 109.33 75 175 275 3,375
LoanMaturity 46 22 29 48 60 3,384
LoanCovenants 1.80 1.65 0 2 3 3,454
FutureLoans 4.43 5.41 1.08 2.39 5.67 1,410
FutureDuration 57.22 46.29 24 44 74 1,410
CustomerSales 0.30 0.20 0.15 0.23 0.39 3,454
CustomerSize 2.72 1.86 1.29 2.05 3.48 3,358
CustomerLeverage 0.09 0.08 0.03 0.06 0.11 3,358
CustomerDefault 0.30 0.26 0.12 0.20 0.38 3,358
Size 6.38 1.62 5.13 6.41 7.61 3,454
Age 18.41 16.45 5 12 31 3,454
Profitability 0.12 0.11 0.08 0.13 0.18 3,436
Tangibility 0.26 0.15 0.15 0.24 0.37 3,451
M/B 1.71 1.06 1.12 1.38 1.91 3,454
Leverage 0.33 0.21 0.18 0.31 0.45 3,454
Ratings 0.48 0.50 0 0 1 3,454
LoanType 0.28 0.45 0 0 1 3,454
43
Table 2Loan Spreads and Customer ConcentrationThis table shows the relation between loan spreads and customer concentration. The dependent variableis All-in-Drawn loan spread (LoanSpread). All regressions use industry-fixed effects and bank-fixedeffects. Industry is classified as 2-digit SIC industry and the lending banks are classified by their ultimateparents. Robust-clustered t-statistics are shown in parentheses. Columns (1) through (3) show theregression results for CustomerSales: the sum of the percentage sales to the set of customers a firmreports as “major customers”. Columns (4) through (6) show the regression results for CustomerSize:the total size of all major customers, weighted by the firm’s percentage sales to these customers.
Dep. Var.: LoanSpread CustomerConcentration is CustomerSales CustomerConcentration is CustomerSize
(1) (2) (3) (4) (5) (6)
CustomerConcentration 52.56*** 47.14** 40.89*** 7.05*** 5.90** 5.28***(3.15) (2.67) (3.32) (3.78) (2.84) (3.72)
Size -27.58*** -28.28*** -17.69*** -28.11*** -28.64*** -17.97***(-10.44) (-9.60) (-5.40) (-10.36) (-9.99) (-5.43)
Age -0.78*** -0.96*** -0.60*** -0.80*** -0.98*** -0.61***(-3.07) (-4.30) (-2.88) (-3.29) (-4.56) (-3.07)
Profitability -272.89*** -260.23*** -246.94*** -292.40*** -284.38*** -273.44***(-3.20) (-3.62) (-3.89) (-3.53) (-4.12) (-4.23)
Tangibility -20.65 -4.89 -27.66 -16.13 -1.90 -22.23(-0.79) (-0.24) (-0.77) (-0.63) (-0.09) (-0.62)
M/B -11.07*** -6.58* -3.75 -9.23** -4.60 -1.83(-3.09) (-1.88) (-0.90) (-2.20) (-1.07) (-0.40)
Leverage 135.13*** 146.97*** 128.86*** 138.60*** 149.30*** 131.47***(7.92) (10.58) (8.49) (7.53) (10.01) (8.21)
Ratings 38.08*** 37.98*** 35.74*** 37.94*** 38.06*** 35.24***(5.60) (5.80) (4.68) (5.55) (5.83) (4.62)
CreditSpread 87.21*** 81.53*** 82.11*** 76.32***(5.37) (5.49) (5.17) (4.97)
TermSpread 1.74 3.40 2.98 4.42*(0.75) (1.31) (1.33) (1.69)
GDPGrowth -3.96 -4.79* -4.18 -5.18**(-1.47) (-1.96) (-1.60) (-2.08)
LoanMaturity 5.91 6.48(1.50) (1.61)
LoanSize -15.77*** -16.02***(-5.66) (-5.69)
LoanType 45.78*** 44.87***(6.86) (6.76)
Observations 3,045 3,045 2,983 2,965 2,965 2,903R-squared 0.45 0.52 0.61 0.45 0.52 0.62
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
44
Table 3Number of Restrictive Covenants and Customer ConcentrationThis table shows the relation between the number of restrictive covenants written on bank loans andcustomer concentration. The dependent variable is the number of loan covenants (LoanCovenants). Allregressions use industry-fixed effects and bank-fixed effects. Industry is classified as 2-digit SIC industryand the lending banks are classified by their ultimate parents. Robust-clustered t-statistics are shownin parentheses. Columns (1) through (3) show the regression results for CustomerSales. Columns (4)through (6) show the regression results for CustomerSize.
Dep. Var.: LoanCovenants CustomerConcentration is CustomerSales CustomerConcentration is CustomerSize
(1) (2) (3) (4) (5) (6)
CustomerConcentration 1.04*** 1.01*** 0.95*** 0.12*** 0.11*** 0.09***(4.85) (4.81) (2.75) (4.98) (4.89) (3.31)
Size -0.24*** -0.23*** -0.22*** -0.24*** -0.24*** -0.18***(-4.20) (-4.38) (-4.08) (-4.17) (-4.35) (-6.31)
Age -0.00 -0.00 0.00 -0.00 -0.00 0.01**(-1.02) (-1.20) (1.13) (-1.14) (-1.31) (2.58)
Profitability 0.79 0.83 1.59*** 0.66 0.73 1.57**(1.15) (1.28) (3.24) (1.07) (1.18) (2.54)
Tangibility -0.53 -0.39 -0.45 -0.58 -0.46 -0.18(-1.11) (-0.82) (-0.71) (-1.15) (-0.94) (-0.37)
M/B -0.15*** -0.13*** -0.06* -0.13** -0.12** -0.04(-3.04) (-3.06) (-1.66) (-2.41) (-2.59) (-0.72)
Leverage 0.73** 0.72** -0.20 0.74** 0.73** -0.17(2.47) (2.70) (-0.79) (2.66) (2.84) (-0.88)
Ratings 0.20** 0.19** 0.16 0.20** 0.19** 0.08(2.32) (2.18) (1.37) (2.23) (2.14) (1.09)
CreditSpread 1.46*** 1.01*** 1.38*** 1.05***(4.03) (4.09) (3.89) (3.55)
TermSpread -0.11* -0.11* -0.10* -0.11*(-1.90) (-1.86) (-1.75) (-2.06)
GDPGrowth 0.21*** 0.23*** 0.21*** 0.23***(4.56) (5.25) (4.76) (5.38)
LoanSpread 0.68*** 0.71***(11.81) (25.99)
LoanMaturity 0.21*** 0.27***(3.82) (5.49)
LoanSize 0.13*** 0.12**(3.54) (2.17)
LoanType 0.06 0.06(1.52) (1.28)
Observations 3,055 3,055 2,983 2,975 2,975 2,903R-squared 0.17 0.19 0.33 0.17 0.19 0.28
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
45
Table 4Loan Maturity and Customer ConcentrationThis table shows the relation between the bank loan maturity (in months) and customer concentration.The dependent variable is loan maturity (LoanMaturity). Only term loans are considered for this test.All regressions use industry-fixed effects and bank-fixed effects. Industry is classified as 2-digit SICindustry and the lending banks are classified by their ultimate parents. Robust-clustered t-statistics areshown in parentheses. Columns (1) through (3) show the regression results for CustomerSales. Columns(4) through (6) show the regression results for CustomerSize.
Dep. Var.: LoanMaturity CustomerConcentration is CustomerSales CustomerConcentration is CustomerSize
(1) (2) (3) (4) (5) (6)
CustomerConcentration -5.83* -5.70* -8.33** -0.77 -0.68 -1.04**
(-1.83) (-2.03) (-2.27) (-1.71) (-1.71) (-2.12)
Size -0.33 -0.28 -2.30 -0.38 -0.34 -2.35
(-0.26) (-0.22) (-1.73) (-0.29) (-0.25) (-1.74)
Age -0.24*** -0.20** -0.16*** -0.23** -0.19** -0.14***
(-3.16) (-2.86) (-3.10) (-2.91) (-2.61) (-2.93)
Profitability 44.20*** 45.07*** 49.95*** 42.79*** 44.72*** 49.80***
(4.63) (4.27) (4.40) (4.05) (3.90) (4.06)
Tangibility -18.46** -18.27** -19.26** -18.61** -18.24** -19.82**
(-2.44) (-2.48) (-2.61) (-2.37) (-2.37) (-2.57)
M/B -2.29 -2.91* -3.19** -2.49* -3.17** -3.43**
(-1.61) (-2.01) (-2.34) (-1.76) (-2.23) (-2.53)
Leverage 12.07 9.95 4.04 12.40 10.46 4.07
(1.68) (1.38) (0.55) (1.60) (1.34) (0.52)
Ratings 5.09 4.50 2.78 4.79 4.14 2.20
(1.57) (1.43) (0.90) (1.49) (1.30) (0.75)
CreditSpread -2.63 -5.57 -1.77 -4.75
(-0.49) (-0.94) (-0.31) (-0.76)
TermSpread -1.40 -0.91 -1.59* -1.10
(-1.71) (-1.17) (-1.85) (-1.37)
GDPGrowth 2.06** 2.26** 2.14** 2.38**
(2.21) (2.46) (2.22) (2.55)
LoanSpread 4.01 4.70**
(1.74) (2.28)
LoanSize 4.25*** 4.49***
(3.21) (3.43)
Observations 903 903 894 879 879 870
R-squared 0.23 0.27 0.29 0.23 0.27 0.30
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
46
Table 5Future Bank Relationships and Customer ConcentrationThis table shows the relation between future bank relationships to a firm and the firm’s customer concen-tration. The dependent variable is future bank relationships, as measured by the total amount of lendingextended by the same bank scaled by the current amount of lending (FutureLoans) and the durationof banking relationships in the future (FutureDuration). All regressions use industry-fixed effects andbank-fixed effects. Industry is classified as 2-digit SIC industry and lending banks are classified by theirultimate parents. Robust-clustered t-statistics are shown in parentheses. Columns (1) and (2) show theregression results for CustomerSales. Columns (3) and (4) show the regression results for CustomerSize.
CustomerConcentration is CustomerSales CustomerConcentration is CustomerSize
(1) (2) (3) (4)Dep. Var.: FutureLoans FutureDuration FutureLoans FutureDuration
CustomerConcentration -1.81** -38.88*** -0.23** -4.90***(-2.44) (-4.96) (-2.63) (-5.63)
Size 0.10 -0.76 0.05 -0.88(0.34) (-0.33) (0.18) (-0.37)
Age -0.02 -0.27** -0.02 -0.26**(-1.13) (-2.67) (-0.90) (-2.54)
Profitability 6.43*** 66.46*** 6.02*** 63.63***(3.20) (3.79) (2.94) (3.62)
Tangibility 2.72 26.68*** 2.85 28.42***(1.57) (3.32) (1.59) (3.36)
M/B 0.87*** 2.29 0.84*** 2.18(3.61) (1.33) (3.58) (1.41)
Leverage -1.25 -1.72 -1.49 -2.81(-1.25) (-0.19) (-1.47) (-0.31)
Ratings 0.32 11.67** 0.27 11.79**(0.60) (2.81) (0.49) (2.79)
CreditSpread -0.52 -24.60* -0.19 -21.48(-0.33) (-1.90) (-0.12) (-1.57)
TermSpread 0.39** 3.22** 0.35* 3.18**(2.15) (2.34) (2.03) (2.31)
GDPGrowth -0.17 -1.06 -0.17 -0.98(-0.94) (-1.24) (-0.99) (-1.12)
Observations 1,271 1,271 1,239 1,239R-squared 0.16 0.13 0.15 0.13
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
47
Table 6First-Stage Results for IV EstimationsThis table shows the first-stage regression results of a firm’s customer concentration (CustomerSales) onthe average customers’ industry acquisition activity (CustomerM&A). All regressions use industry-year-fixed effects and bank-fixed effects. Robust-clustered t-statistics are shown in parentheses.
(1) (2) (3) (4) (5)Dep. Var. in the Second Stage: LoanSpread LoanCovenants LoanMaturity FutureLoans FutureDuration
CustomerM&A 5.82*** 5.63*** 6.61*** 4.30*** 4.30***(3.62) (3.61) (3.04) (4.62) (4.62)
Size -0.01 -0.01 -0.01 -0.02 -0.02(-0.97) (-0.68) (-0.37) (-1.67) (-1.67)
Age -0.00* -0.00 -0.00 -0.00** -0.00**(-1.77) (-1.60) (-1.19) (-2.45) (-2.45)
Profitability 0.06 0.09 0.04 -0.13 -0.13(1.08) (1.53) (0.21) (-0.86) (-0.86)
Tangibility -0.05 -0.04 -0.08 0.00 0.00(-0.54) (-0.45) (-0.73) (0.04) (0.04)
M/B -0.01 -0.01 -0.01 -0.00 -0.00(-1.71) (-1.54) (-0.38) (-0.22) (-0.22)
Leverage -0.07*** -0.09*** -0.15* -0.10*** -0.10***(-3.78) (-3.36) (-2.07) (-2.45) (-2.45)
Ratings 0.01 0.01 0.02 0.03 0.03(0.39) (0.27) (0.46) (1.72) (1.72)
LoanSpread 0.02 0.03(1.28) (1.10)
LoanMaturity 0.01 0.01(0.99) (0.78)
LoanSize 0.01 0.01 0.01(1.05) (1.52) (0.36)
LoanType -0.01 -0.01(-0.94) (-1.70)
Observations 1,957 1,957 570 871 871R-squared 0.40 0.40 0.59 0.21 0.21
First Stage F -test 13.09 13.05 9.25 21.34 21.34p-value 0.00 0.00 0.01 0.00 0.00Kleibergen-Paap LM Stat 5.97 5.67 6.7 7.33 7.33p-value 0.01 0.02 0.01 0.00 0.00
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
48
Table 7Second-Stage Results for Customer ConcentrationThis table shows the second-stage results of the relation between a firm’s customer concentration and bothits borrowing terms and future bank relationships. Customer concentration measures are instrumentedwith the acquisition levels in the customer’s industries (CustomerM&A). The dependent variables includeloan spreads (LoanSpread), number of loan covenants (LoanCovenants), loan maturity (LoanMaturity),and future bank relationships as measured by the future amount of loans extended by the same bank,scaled by the current loan amount (FutureLoans) and the duration of banking relationships in the future(FutureDuration). CustomerConcentration is the projected customer concentration from the first-stageregressions. Customer concentration is measured by CustomerSales. All regressions use industry-year-fixed effects and bank-fixed effects. Industry is classified as 2-digit SIC industry and lending banks areclassified by their ultimate parents. Robust-clustered z -statistics are shown in parentheses.
(1) (2) (3) (4) (5)Dep. Var.: LoanSpread LoanCovenants LoanMaturity FutureLoans FutureDuration
CustomerConcentration 235.82*** 4.21** -57.08*** -20.63** -83.92***(2.77) (2.41) (-3.05) (-2.18) (-3.64)
Size -17.13*** -0.06 -2.71* -0.04 2.60(-3.74) (-0.99) (-1.81) (-0.18) (1.42)
Age -0.59** 0.01** -0.24** -0.03 -0.17(-2.05) (2.15) (-1.97) (-0.79) (-1.61)
Profitability -259.51*** 1.22* 41.66*** 4.36 19.49(-8.28) (1.77) (3.59) (0.92) (1.27)
Tangibility -21.64 -0.13 -8.35 -0.35 -3.51(-0.89) (-0.39) (-0.89) (-0.14) (-0.29)
M/B -4.28* -0.04 -2.49** 0.52** -0.45(-1.94) (-0.87) (-2.16) (1.99) (-0.24)
Leverage 154.93*** -0.09 -13.16* -1.84 -6.30(12.68) (-0.25) (-1.66) (-0.80) (-0.61)
Ratings 36.25*** -0.05 14.49*** 0.56 8.95(3.64) (-0.38) (2.58) (0.68) (1.57)
LoanSpread 0.55*** 3.16(8.50) (1.11)
LoanMaturity 6.40 0.25***(1.31) (3.75)
LoanSize -15.12*** -0.00 3.06**(-4.80) (-0.07) (1.99)
LoanType 34.92*** 0.06(9.47) (1.00)
Observations 1,957 1,957 570 871 871R-squared 0.64 0.40 0.41 0.25 0.61
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
49
Table 8Customer Concentration and Loan Spreads following Deregulation EventsThis table shows how customer concentration and loan spreads change following customer industries’deregulation events. The dependent variable in columns (1) and (2) is customer concentration, measuredby CustomerSales. The dependent variable in columns (3) and (4) is loan spreads (LoanSpread). Treatedis a dummy variable indicating whether a firm’s customer industry experienced a Federally-mandatedAct or Order affecting its competitive environment. Post is a dummy variable covering the periodfollowing the regulatory event. All regressions use industry-fixed effects and bank-fixed effects. Industryis classified as 3-digit SIC industry and lending banks are classified by their ultimate parents. Robust-clustered z -statistics are shown in parentheses.
Dep. Var.: CustomerConcentration LoanSpread
(1) (2) (3) (4)
Treated 0.00 0.00 -9.98 -7.10(0.17) (0.20) (-1.97) (-1.37)
Post -0.03*** -0.03*** 5.08 6.68**(-6.17) (-4.35) (1.14) (3.01)
Treated×Post 0.03*** 0.04*** 21.39** 12.55***(4.09) (6.01) (3.72) (4.60)
Size -0.05** -0.07*** -6.48 -16.57(-2.97) (-6.19) (-0.43) (-0.75)
Age -0.00 0.00 -0.54 -0.99(-0.29) (0.04) (-0.81) (-1.74)
Profitability -0.14*** -0.11* 16.45 22.12(-5.12) (-2.48) (0.22) (0.26)
Tangibility 0.30** 0.26*** -112.16 -136.75(3.76) (4.29) (-0.84) (-1.06)
M/B 0.02 0.01 -17.01** -9.57***(0.87) (0.52) (-3.79) (-4.36)
Leverage -0.10* -0.06 173.40** 136.35(-2.06) (-1.80) (2.60) (1.86)
Ratings -0.03 -0.04 -31.72 -13.84(-0.41) (-0.57) (-1.32) (-0.66)
CreditSpread -0.31*** -0.29*** 83.59 93.50*(-34.78) (-10.16) (1.68) (2.07)
TermSpread -0.02 -0.03 -8.49 -7.08(-0.92) (-1.64) (-0.86) (-1.22)
GDPGrowth -0.01 -0.01 -2.66 -4.76(-0.99) (-1.07) (-0.28) (-0.46)
LoanMaturity -0.00*** -0.05(-5.28) (-0.07)
LoanSize 0.04 -0.55(1.94) (-0.05)
LoanType 0.02 55.59**(1.01) (3.80)
Observations 216 211 212 207R-squared 0.77 0.78 0.69 0.75
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
50
Table 9Loan Terms and Relationship-Specific InvestmentsThis table shows subsample results of the relation between loan terms and customer concentration, todemonstrate how this relation can be affected by firms’ relationship-specific investments. The dependentvariable are loan spread, covenants, and maturity. Customer concentration is measured by Customer-Sales. All regressions use industry-fixed effects and bank-fixed effects. Industry is classified as 2-digit SICindustry and the lending banks are classified by their ultimate parents. Robust-clustered t-statistics areshown in parentheses. Columns (1) through (3) show the regression results for the subsample contain-ing industries with high relationship-specific investments. Columns (4) through (6) show the regressionresults for for the subsample containing industries with low relationship-specific investments.
High-RSI Industries Low-RSI Industries
(1) (2) (3) (4) (5) (6)Dep. Var.: LoanSpread LoanCovenants LoanMaturity LoanSpread LoanCovenants LoanMaturity
CustomerConcentration 65.55*** 0.56 -9.56** 12.83 0.72 -4.01(5.46) (1.53) (-3.48) (0.72) (1.84) (-0.93)
Size -19.40*** -0.11 -0.05 -17.23*** -0.22* -2.27(-5.20) (-2.06) (-0.05) (-7.86) (-2.02) (-0.84)
Age -0.37 0.00 -0.14** -1.02** 0.01*** -0.04(-1.69) (1.06) (-3.67) (-2.86) (4.06) (-0.31)
Profitability -335.01 1.72 38.06* -174.97*** 1.74** 59.55(-1.79) (1.85) (2.26) (-6.42) (3.12) (2.12)
Tangibility 58.13** -0.51 -5.08 -24.89 0.29 -35.77**(3.03) (-0.52) (-1.28) (-0.86) (0.46) (-3.47)
M/B -6.40 -0.03 -1.59 -1.64 -0.03 -3.51(-1.15) (-0.73) (-1.38) (-0.53) (-0.37) (-1.73)
Leverage 134.51*** -0.28 6.19 132.03*** -0.26 7.35(6.86) (-1.15) (0.37) (15.50) (-0.70) (1.02)
Ratings 40.19*** -0.03 3.47 49.55*** 0.28* 2.55(9.38) (-0.24) (0.75) (5.27) (2.01) (1.25)
CreditSpread 59.61** 0.44 -1.45 96.52** 1.13** -0.84(3.95) (1.46) (-0.30) (3.55) (2.86) (-0.10)
TermSpread 3.30 -0.05 0.21 0.61 -0.26*** -1.61(1.67) (-0.96) (0.68) (0.43) (-7.30) (-1.64)
GDPGrowth -6.15* 0.13* 3.38** -0.20 0.22** 2.48(-2.62) (2.57) (3.18) (-0.07) (3.56) (1.90)
LoanSpread 0.69*** 1.58 0.74*** 1.10(12.36) (0.35) (11.80) (0.42)
LoanMaturity 1.62 0.33** 4.59 0.22***(0.26) (3.13) (0.87) (4.78)
LoanSize -17.24** 0.12 3.33* -15.61*** 0.05 2.81(-4.49) (1.67) (2.50) (-5.67) (0.54) (1.59)
LoanType 56.30*** 0.08 46.98*** 0.08(5.43) (1.32) (8.87) (1.53)
Observations 1,141 1,141 390 1,101 1,101 315R-squared 0.55 0.28 0.25 0.60 0.35 0.29
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
51
Table 10Customer Concentration and Account ReceivablesThis table shows the relation between a firm’s customer concentration and its receivables. Customerconcentration is measured by CustomerSales. The dependent variables are account receivables, scaledby cost of goods sold (COGS ), and total sales. All regressions use industry-fixed effects. The sampleincludes all manufacturing firms with available customer information and firm fundamental information.Industry is classified as 2-digit SIC industry. Robust-clustered t-statistics are shown in parentheses.
Dep. Var.: Receivables/COGS Receivables/Sales
(1) (2) (3) (4) (5) (6)
CustomerConcentration -0.05*** -0.06*** -0.06*** -0.02*** -0.03*** -0.03***(-4.18) (-5.61) (-5.70) (-3.71) (-5.51) (-5.30)
Size 0.01 0.01 -0.00 -0.00(1.52) (1.52) (-1.13) (-0.79)
Age -0.00 -0.00 -0.00* -0.00*(-1.46) (-1.71) (-2.00) (-2.02)
Profitability 0.05 0.05 -0.05*** -0.05***(0.64) (0.67) (-3.33) (-3.38)
Tangibility -0.22*** -0.22*** -0.06*** -0.06***(-7.74) (-7.99) (-6.06) (-6.50)
M/B 0.01** 0.01** -0.00*** -0.00***(2.55) (2.52) (-5.43) (-5.14)
Leverage -0.01 -0.00 0.01* 0.01*(-0.12) (-0.12) (1.84) (1.83)
Ratings -0.01 -0.01 0.00 0.00(-0.68) (-0.64) (0.01) (0.03)
CreditSpread 0.02 0.01***(1.61) (2.90)
TermSpread 0.00*** 0.00(4.53) (0.75)
GDPGrowth -0.00 0.00***(-1.12) (3.58)
Observations 17,225 15,858 15,858 17,225 15,858 15,858R-squared 0.12 0.16 0.16 0.12 0.16 0.16
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
52
Table 11Loan Terms and Customer Financial ConditionThis table shows the relation between loan terms and customer financial condition. The dependentvariable are loan spread, covenants, and maturity. All regressions use industry-fixed effects and bank-fixed effects. Industry is classified as 2-digit SIC industry and the lending banks are classified by theirultimate parents. Robust-clustered t-statistics are shown in parentheses. Columns (1), (3), and (5)show the regression results for CustomerLeverage: the leverage of major customers, weighted by thefirm’s percentage sales to these customers. Columns (2), (4), and (6) show the regression results forCustomerDefault : the inverse of the distance to default of major customers predicted by the KMV-Merton’s model, weighted by the firm’s percentage sales to these customers.
Dep. Var.: LoanSpread LoanCovenants LoanMaturity
(1) (2) (3) (4) (5) (6)
CustomerLeverage 151.43*** 1.43*** -23.55*(3.07) (3.45) (-1.73)
CustomerDefault 47.90*** 0.70*** -6.56(4.15) (3.71) (-1.29)
Size -17.45*** -18.39*** -0.17*** -0.16*** -2.44* -2.59*(-7.68) (-8.63) (-5.58) (-3.93) (-1.77) (-2.04)
Age -0.80*** -0.84*** 0.00** 0.01** -0.14** -0.15***(-3.63) (-3.94) (2.34) (2.30) (-2.72) (-3.17)
Profitability -279.22*** -276.11*** 1.53** 1.31** 50.05*** 52.04***(-4.11) (-3.83) (2.41) (2.16) (4.05) (4.32)
Tangibility -7.60 -2.07 -0.18 -0.27 -19.44** -18.48**(-0.40) (-0.10) (-0.37) (-0.55) (-2.46) (-2.39)
M/B -2.49 -4.00 -0.03 -0.04 -3.47** -3.49**(-0.58) (-1.01) (-0.58) (-0.74) (-2.51) (-2.51)
Leverage 122.86*** 121.88*** -0.25 -0.08 4.65 4.64(10.58) (9.76) (-1.23) (-0.40) (0.58) (0.56)
Ratings 40.61*** 41.33*** 0.08 0.04 2.32 2.46(7.04) (6.97) (1.14) (0.55) (0.74) (0.90)
CreditSpread 83.57*** 81.27*** 1.07*** 0.99*** -4.72 -6.68(5.77) (5.30) (3.61) (3.26) (-0.77) (-1.12)
TermSpread 3.39* 3.95* -0.11* -0.10* -1.14 -1.11(1.77) (2.00) (-2.04) (-1.79) (-1.41) (-1.27)
GDPGrowth -4.30 -4.81* 0.23*** 0.23*** 2.38** 2.07**(-1.73) (-1.98) (5.36) (5.14) (2.60) (2.25)
LoanMaturity 9.35** 10.56** 0.28*** 0.31***(2.13) (2.30) (5.52) (6.67)
LoanSize -16.62*** -15.81*** 0.12** 0.09* 4.45*** 4.58***(-7.20) (-6.06) (2.34) (1.93) (3.35) (2.97)
LoanType 47.53*** 48.69*** 0.06 0.04(7.50) (7.35) (1.30) (0.83)
LoanSpread 0.72*** 0.71*** 4.82** 4.64*(23.97) (18.49) (2.30) (2.01)
Observations 2,903 2,770 2,903 2,770 870 833R-squared 0.58 0.59 0.28 0.29 0.30 0.31
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
53
Table 12Loan Failure Rate and Customer ConcentrationThis table shows the relation between bank loan failure rates and customer concentration. The dependentvariable is loan failure (LoanFailure), a dummy variable that equals 1 if a firm files for bankruptcy beforeloan maturity, 0 otherwise. All regressions use industry-fixed effects and bank-fixed effects. Industry isclassified as 2-digit SIC industry and the banks are classified by their ultimate parents. Robust-clusteredz -statistics are shown in parentheses. Columns (1) through (3) show the logit regression results forCustomerSales. Columns (4) through (6) show the logit regression results for CustomerSize.
Dep. Var.: LoanFailure CustomerConcentration is CustomerSales CustomerConcentration is CustomerSize
(1) (2) (3) (4) (5) (6)
CustomerConcentration 1.79*** 1.73*** 1.59** 0.21*** 0.21*** 0.19**(2.79) (2.70) (2.28) (3.25) (3.30) (2.53)
Size -0.08 -0.07 -0.18 -0.12 -0.12 -0.25(-0.35) (-0.32) (-0.67) (-0.62) (-0.58) (-1.11)
Age -0.00 -0.00 0.01 0.00 0.00 0.01(-0.07) (-0.06) (0.66) (0.17) (0.18) (0.73)
Profitability -7.87*** -8.34*** -10.85*** -5.70*** -6.02*** -8.21***(-5.38) (-5.42) (-6.17) (-3.06) (-2.66) (-4.47)
Tangibility 1.65 1.73 2.11 1.94 1.95 2.21(1.05) (1.13) (1.40) (1.35) (1.43) (1.57)
M/B -1.60*** -1.67*** -1.42*** -2.40*** -2.49*** -2.33***(-4.97) (-4.83) (-2.99) (-5.72) (-5.35) (-5.02)
Leverage 5.69*** 5.56*** 4.66*** 6.11*** 5.95*** 5.02***(6.35) (6.42) (7.97) (5.55) (5.55) (6.27)
Ratings 0.63* 0.78** 0.61 0.58* 0.73** 0.57(1.70) (2.10) (1.51) (1.68) (2.10) (1.48)
CreditSpread -0.84 -1.31 -0.96 -1.51*(-1.28) (-1.46) (-1.50) (-1.74)
TermSpread 0.04 0.15** 0.04 0.12*(0.51) (2.20) (0.49) (1.73)
GDPGrowth 0.15 0.01 0.14 -0.01(1.51) (0.06) (1.39) (-0.08)
LoanSpread 0.79*** 0.81***(3.39) (3.22)
LoanMaturity 2.43*** 2.35***(4.02) (4.62)
LoanSize 0.18** 0.20**(2.02) (2.30)
LoanType -0.02 -0.02(-0.14) (-0.10)
Observations 2,166 2,166 2,118 2,104 2,104 2,056
*** p-value<0.01, ** p-value<0.05, * p-value<0.10
54