Customer Concentration and Loan Contract Terms*

Transcription

1 Customer Concentration and Loan Contract Terms* Murillo Campello Cornell University & NBER Janet Gao Cornell University This Draft: May 27, 2014 Abstract Recent research shows that firms enjoy operating efficiencies when dealing with fewer, larger customers. It ignores, however, how firms creditworthiness is affected by their large exposure to these customers. We look at multiple contractual features of bank loans to gauge how the credit market evaluates a firm s customer base. We first model how interactions between firms and their major customers influence bank loan terms. Empirically, we find that higher customer concentration leads to increases in interest rate spreads and in the number of restrictive covenants featured in bank loans. Customer concentration also reduces the maturity of those loans. The duration and depth of the relationship between firms and their banks are further negatively affected by increased customer concentration. All of these effects are aggravated by a deterioration in customers financial conditions (controlling for firms own finances, their industry, and the identity of their banks). Our results show that in spite of the fact that customer concentration contributes to profitability, it ultimately bears negative consequences for corporate credit. The analysis provides insights about integration along the supply chain and the limits of the firm. Key words: Customer Concentration, Bank Loans, Contract Terms, Financial Distress, Instrumental Variables, Fixed Effects. JEL classification: G21, G30, G32. *We are thankful for Edward Fee and Erasmo Giambona for sharing their data with us. We also thank Jean-Noel Barrot, Sudipto Dasgupta, Tomislav Ladika, Rafael Matta, and Justin Murfin for their comments. Electronic copy available at:

2 Customer Concentration and Loan Contract Terms Abstract Recent research shows that firms enjoy operating efficiencies when dealing with fewer, larger customers. It ignores, however, how firms creditworthiness is affected by their large exposure to these customers. We look at multiple contractual features of bank loans to gauge how the credit market evaluates a firm s customer base. We first model how interactions between firms and their major customers influence bank loan terms. Empirically, we find that higher customer concentration leads to increases in interest rate spreads and in the number of restrictive covenants featured in bank loans. Customer concentration also reduces the maturity of those loans. The duration and depth of the relationship between firms and their banks are further negatively affected by increased customer concentration. All of these effects are aggravated by a deterioration in customers financial conditions (controlling for firms own finances, their industry, and the identity of their banks). Our results show that in spite of the fact that customer concentration contributes to profitability, it ultimately bears negative consequences for corporate credit. The analysis provides insights about integration along the supply chain and the limits of the firm. Key words: Customer Concentration, Bank Loans, Contract Terms, Financial Distress, Instrumental Variables, Fixed Effects. JEL classification: G21, G30, G32. Electronic copy available at:

3 1 Introduction U.S. manufacturers attribute, on average, over one-third of their sales figures to a few major customers, and the level of customer concentration is increasing in recent years. A concentrated customer base is 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, such arguments find support in recent academic research (e.g., Patatoukas (2012) and Irvine et al. (2013)). Relying on major customers has shortcomings, nonetheless. Major customers demand lower prices, purchase irregularly, and delay payments (Fee and Thomas (2004), Kelly et al. (2013), Murfin and Njoroge (2013), and Barrot (2014)). 1 While these problems are shown to be important, the literature has not examined whether a close association with fewer, larger customers expose firms to costs and risks that affect their access to credit. This paper examines how the credit market evaluates a firm s customer-base profile, characterizing how customer concentration and financial status affect the firm s access to funds. To do so, we look at multiple features of bank loan contracts and firm bank relationships. This approach 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 and supply-chain relations. We first model the interplay between customer concentration, firm investment choices, and loan contract terms in a simple theoretical framework. The model we develop is useful in providing clear predictions about relations that can be taken to the data. We then empirically examine the impact of customer concentration on several features of 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 to the literature in revealing contracting costs associated with increased reliance on few, large customers. We show that these costs are significant and manifest themselves along various dimensions, pointing to important limitations to deeper integration among firms along their supply chain. 1 These behaviors have attracted attention from the press, with reports that large, powerful firms such as Walmart and P&G abuse their suppliers by delaying payment on their products. See Wall Street Journal article: Small Firms Big Customers Are Slow to Pay (June 6, 2012). 1 Electronic copy available at:

4 In a nutshell, our model characterizes firms incentives to invest in projects that enhance relations with their suppliers, showing how this affects their credit. Relationship-specific investments have been described in the existing theoretical and empirical literatures (e.g., Klein et al. (1978), Hart (1995), Bolton and Scharfstein (1998), Kale and Shahrur (2007), and Banerjee et al. (2008)). These projects may involve investment in R&D, unique fixed assets, and modifications of standard production processes. Relationship-specific projects are less desirable from lenders perspective because their uniqueness engenders higher risks and lower resale values in liquidation. Our equilibrium analysis shows that the higher the importance of major customers, the greater the gains from relationship-specific investments, and the lower the credit quality of firms undertaking those investments. The model implies, among other things, that increases in firms customer concentration will cause their 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 firm customers from Compustat s Segment Database over the window. Our baseline results can be summarized as follows. A more concentrated customer base increases both the interest rates and the number of restrictive covenants featured in 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 interest spreads on new bank loans; this compared to an average spread of 173 basis points. The same shift leads to, on average, 0.2 additional loan covenants; compared 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 45 months. The magnitudes of these effects are significant given the high level of competition in the market for corporate lending. We also examine whether customer concentration affects the length and depth of a firm s banking relationships. We find that a one-standard-deviation increase in customer concentration leads to 0.3 fewer loans from the firm s current bank in the future, and 0.4 years shorter relationship with that bank. These magnitudes are significant compared to the sample average levels of 4.2 future loans and 2.9 future years of relationship. 2

5 Observed relations between customer concentration and borrowing terms can be biased due to omitted variables. In particular, it can be argued that unobserved characteristics might lead a firm s customer concentration to increase and its credit terms to deteriorate this, despite of a positive relation between customer concentration and profitability. To alleviate such concerns, we experiment with the use of data on M&A activity in customers industries (downstream mergers) as an instrument for customer concentration. Downstream M&A activity is a plausible instrument for two reasons. First, it is related to customers own growth prospects (Fee and Thomas (2004) and Erel et al. (2014)) and following merger deals in customer industries, suppliers are likely to face higher customer concentration (inclusion restriction). Second, that activity is unlikely to affect suppliers borrowing terms through channels other than customer supplier linkages (exclusion restriction). Bearing in mind concerns that industry-level, time-varying dynamics could influence customers M&A activity and firms credit terms, we further control for industry-year-fixed effects in our tests. Our IV estimations imply that following high levels of M&A activity in customer industries, supplier firms observe higher customer-base concentration, which then lead to costlier, stricter borrowing terms and shorter banking relationships. We dig deeper into the meaning of our results by examining whether the financial conditions of a firm s large customers affect its credit terms. Customers in worse financial shape may, for example, face difficulties in maintaining purchase agreements and paying on time, eventually burdening their suppliers. Confirming the logic of our argument, we find that loan spreads increase even more and the number of covenants is even higher when a firm s large customers are likely to be distressed (as proxied by measures such as distance-to-default). 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 intermediaries, banks are particularly concerned about loan failures. If higher customer concentration is associated with higher loan failure rates for supplier firms, banks will naturally impose stricter loan terms. To establish this link, we identify loan failures by matching our data with the LoPucki bankruptcy database, which has records of corporate failures. We find a positive, 3

6 significant relation between customer concentration and supplier loan failure rates. To wit, a one-standard-deviation increase in customer concentration is associated with a 2 percentage points increase in loan failure rate. The impact is sizable when compared to the sample average loan failure rate of 6.5 percent. Our results show direct evidence that customer concentration is an important concern for banks decisions to offer corporate credit. They identify the cause of that concern and gauge its consequences across various dimensions of loan contracting. Our paper is related to various 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 lowering 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, and that this is reflected in stock prices. Greene et al. (2013) show that when customers become more powerful they demand better trade terms. Our study contributes to this literature by showing the responses from credit markets to changes in customer concentration. Like previous papers, we show that customer concentration is indeed associated with higher firm profitability. Using lenders perspective, however, we show that concentrated customer bases ultimately have 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 Banerjee et al. (2008) argue that firms tend to procure more unique assets when they rely on major customers. These firms have lower leverage ratios because customer liquidation imposes high redeployment costs for their relationship-specific assets (see also Kale and Shahrur (2007)). Hennessy and Livdan (2009) model a firm s leverage based on trade-offs between gains from bargaining power against suppliers and costs associated with lower input quality. We add to this research by showing how different features of debt contracting e.g., interest rates, maturity, and covenants relate to firms customer base. Our paper is also related to the literature on credit contagion along the supply chain. Existing studies show that a firm s financial distress can impact its suppliers and customers (e.g., Kolay et al. (2012)). In that vein, Cohen and Frazzini (2008) report that customers earnings 4

7 surprises are incorporated into suppliers stock prices. Consistent with these studies, our findings show that financially-distressed customers can generate severe negative externalities for their suppliers; in particular, have detrimental consequences for their borrowing. Finally, our study is related to the literature on the determinants of bank loan terms (examples are Graham et al. (2008), Roberts and Sufi (2009), Lin et al. (2011), 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 relations may reduce informational asymmetries between banks and firms in the long run. None of these papers consider the effect of customer concentration or distress on loan terms. The paper proceeds as follows. Section 2 contains our theoretical motivation for the interplay between customer concentration and firm credit terms. Section 3 describes our data and methodology. Section 4 reports univariate analyses. Section 5 describes our baseline results. Section 6 reports our instrumental variable analysis. Section 7 explores cross-sectional differences in the effect of customers financial health on suppliers borrowing terms. Section 8 describes a case study of credit contagion inside the supply chain. Section 9 reports a direct test of the effect of customer concentration on bank loan failure rates. Section 10 concludes. 2 A Model of Customer Concentration and Bank Credit We analyze the relation between a firm s customer concentration and bank credit using a simple theoretical framework. In it, we model the interplay between the customer, the supplier firm, and its bank, keeping the focus on the dynamics we want to study empirically. We do not model industry dynamics. In turn, we implicitly take customer concentration as given, reflecting the crux of our supply-chain-based story for a potentially deep association between a firm and its major customer. A major customer in our model may develop a deep relationship with a particular supplier to the point of shaping that firm s investment choices. This differentiates that firm from other producers in its industry. The model delivers several testable implications. 5

8 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 firm has to make investment decisions that maximize its profits, choosing between projects with different profiles. The model has two periods and at t = 0, the firm faces two mutually exclusive projects. Both projects require initial investment I and have a payoff at t = 1. The firm has no funds, so it borrows capital from the bank. Project A is risky. It pays αi with probability p, and 0 with probability 1 p. Project B is safe and pays βi with probability 1 (α < 1 < β). To make the problem interesting, project choice may have different impacts on the firm s relations with its major customer and bank. Project B gives a higher expected return; that is, β > pα. That project 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. The relationship-specific investment can involve expenditures with R&D, unique fixed assets, customization, and modifications to standard production processes. Project A is riskier for the firm because it has lower success rate and lower resale value. At the same time, it creates synergistic benefits for the firm s major customer that are ultimately shared by the firm. 2 The major customer derives value V A from project A and V B from project B, and we assume that it prefers the relationship-specific project pαv A > βv B. At t = 1, the firm sells a proportion µ of its output to the major customer; 1 µ is sold to a set of small customers. The major customer can observe the firm s project choice. To motivate the firm to take project A, the major customer offers different prices for different project outputs. For simplicity, we assume the non-major customers pay a price 1 per unit of output for either projects, while the major customer offers δ A per unit for project A and δ B per unit for project B; where δ A > δ B. 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, speeds of payment, and can also reflect variations in overhead costs during the production process. 3 2 Neither the major customer nor the supplier necessarily belongs to a perfectly competitive industry, and neither has absolute bargaining power against the other. 3 We assume that the relationship-specific investment does not impact the sales price to firm s non-major cus- 6

9 2.2 Base Analysis To ease the exposition, we momentarily assume that there is no asymmetry of information between the bank and the firm. The bank can observe the amount the firm borrows I, its 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 rate R = 1 p if the firm chooses the risky project and 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(µδ A + 1 µ)i RI, β(µδ B + 1 µ)i ri} (1) The firm will choose to invest in project A if pα(µδ A +1 µ) R > β(µδ B +1 µ) r. Simplifying this condition, the firm chooses project A if the customer s offer satisfies the following: pαδ A βδ B > (1 µ)(β pα) + R r. (2) µ The major customer will also benefit from project A if the following holds: pα(v A δ A ) > β(v B δ B ). (3) Therefore, the firm and its major customer will be in agreement and choose project A if: (1 µ)(β pα) + R r µ < pαδ A βδ B < pαv A βv B. (4) Conversely, project B will be chosen if: (1 µ)(β pα) + R r µ > pαδ A βδ B > pαv A βv B. (5) The conditions above make it clear that for high levels of customer concentration µ (that is, µ > β pα+r r β pα βv B +pαv A ), 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. tomers. Non-major customers do not have the ability to change the price or to delay their payments to the firm. 7

10 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 successfully invest in the relationship-specific (risky) project. Finally, we relax the assumption that banks have perfect information about firms. Instead, we only assume that banks can observe (ex-post) firms project choice, and that they know the general distribution of firm quality (ability to succeed in the relationship-specific project). We avoid 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]. The parameter 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 requirements and cannot diversify away the default risk of their loans. We assume a perfectly 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 customization and know their suppliers ability to successfully deliver the relationship-specific project A. This reflects the common assumption in the literature that major customers have information advantage over banks regarding real investment projects their suppliers implement, as customers can better understand input transactions, and trade credit works as a monitoring tool (Biais and Gollier (1997) and Burkart and Ellingsen (2004)). Major customers uniformly prefer project A; which is equivalent to: pαv A > βv B. 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. Accordingly, the bank charges a break-even rate R = 1 E[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 this equilibrium is that the better firms do not want to take the safe project and the worse firms do not want to take the risky project: pα(µδ A + 1 µ) R > β(µδ B + 1 µ) r, p > p (6) 8

11 Note that R decreases with p ( R p pα(µδ A + 1 µ) R < β(µδ B + 1 µ) r, p < p. (7) < 0), suggesting that when the bank knows only good firms undertake the risky project, it is less worried about default. The bank will thus charge a lower interest rate. The equilibrium threshold p will satisfy the following break-even condition: (1 µ)(β p α) + R r µ = p αv A βv B. (8) This expression describes the trade-off between projects for the marginal firm. The left-hand side presents the cost of undertaking the risky project. 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 the risky project, 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 αv A βv B )µ + (p α β)(1 µ) (R r). Given R p < 0, it follows that F p that p µ > 0 and F µ > 0. Using the Implicit Function Theorem, we have < 0, which implies that the quality of firms taking the risky project declines with customer concentration. Put differently, a higher level of customer concentration, µ, prompts more 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 Customer Financial Condition It is natural to consider a firm s major customer s financial condition as a concern to banks in this setting. We extend the model to shed light on how customers financial heath affects contracting. Assume at t = 1, there is a probability 1 λ that the major customer experiences financial difficulty and cannot pay the supplier firm. When a major customer is in financial distress, the supplier can still sell the output from its projects to other customers at price 1. This means that the firms who take project A receive α with probability 1 λ and the full output price δ A µα+(1 µ)α with probability λ. The firms that undertake project B receive δ B µβ +(1 µ)β 9

12 with probability λ and β with probability 1 λ. From the above setting, we can see that if the firm takes project A, it will default on its bank loan when the customer fails to pay. However, if the firm takes project B, it will not default, since β r 1 > α. To make the problem more realistic, we also allow for some degree of loan recovery when the firm is driven to default by its customer. We assume that the bank can recover the supplier s assets and resell them at a discount cost. Accordingly, we denote the recovery rate to be η, 0 < η 1. Therefore, the bank s break-even rate R given (p, λ, µ) is determined by pλr + ηpα(1 λ)(1 µ) = 1, or R = 1 ηpα(1 λ)(1 µ) pλ = 1 ηpα(1 µ) pλ + ηα(1 µ). (9) Similar to the case without customer distress, a separating equilibrium exists in this setting. In this equilibrium, the better firms (p > p ) take project A and the worse firms (p < p ) take project B. The threshold p satisfies the following condition: p αv A µ βλv B µ = p λr r + β((1 µ)λ + 1 λ) λp α(1 µ). (10) One can solve for p as follows: p = βλv Bµ + β((1 µ)λ + 1 λ). (11) αv A µ λr + αλ(1 µ) We prove in Appendix B that the bank will charge a fair rate R if it observes a firm undertake project A: R = 1 ηe[p p > p ]α(1 µ) E[p p > p ]λ + ηα(1 µ). (12) We also show in Appendix B that dp > 0, de[p p>p ] dλ dλ > 0, and R λ < 0. Our goal is to understand the relationship between R and customer s financial health λ. In this regard, the relation dp dλ > 0 is important as it shows that a higher customer distress risk causes lower quality firms to choose project A. This seemingly counter-intuitive result arises from firms increased risk-shifting incentives given the higher likelihood of default. With a 10

13 higher probability of financial distress, the major customer needs to pay more to induce the supplier to undertake project A. The supplier thus faces a compensation scheme that involves a large payment from the customer when it is in good financial shape, but no obligations otherwise (due to limited liability). This analysis helps us establish that the financial health of the customer influences the borrowing cost of the supplier. When the customer is in good financial condition, the supplier is offered more favorable contract terms. If the customer is in poor financial shape, however, the supplier has even higher incentives to risk-shift, and the bank imposes higher borrowing costs. 2.5 Empirical Predictions The model delivers very direct empirical implications and it is worth collecting them in a subsection. As µ increases, the firm s payoff depends more on its large customer. That large customer can therefore more easily induce the firm to undertake the relationship-specific, risky project. For larger µ, even firms with low ability will choose the risky project. It follows that the threshold p declines with customer concentration. A lower threshold 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 predictions as follows: Hypothesis 1 Banks will impose costlier, stricter loan contract terms on firms with higher customer-base concentration. Hypothesis 2 Firms with higher customer-base concentration experience higher loan failure rates. Hypothesis 3 Banks will impose costlier, stricter loan contract terms on firms that face customers in worse financial conditions. Our model reconciles the evidence that firms with more concentrated customer bases are more profitable (Patatoukas (2012)) with the observation that those relationships are inherently risky and may prompt default and bankruptcy along the supply chain (Hertzel et al. (2008) and Kolay et al. (2012)). Simply put, establishing deeper relationships with major 11

14 customers can be both profitable and risky. The model shows that the risk is passed on to the bank, which in turn responds by offering loan menus with costlier, stricter terms. 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, including 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 concentration affects derivative measures of the relationship between supplier firms and their banks: depth and duration. 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 more than 10% of a firm s total sales. The Segment database collects customer information including the names of the customers and their assigned sales figures. In identifying important customer relations, we focus on recurring customers and exclude customers that appear for fewer than three times for a firm in the sample period. We focus on manufacturers (SIC ) to ease comparisons across firms and because firms operating in this sector resemble our supply-chain story more closely. Notably, information from the U.S. input/output matrix suggest that supplier customer links in the manufacturing sector feature firms on both ends of the relationship. 4 We extract bank loan contract information from LPC Dealscan from 1985 through 2010, and link loan-level data to Compustat firm identifiers following Chava and Roberts (2008). We examine revolvers and term loans since both types of loans provide information on the pricing and the restrictiveness of bank credit. We construct our final sample by combining the customer and bank loan information. For a firm to be included in the sample, we require it to have available customer information, loan characteristics, and information on standard variables such as size, leverage, and market-to- 4 Over two-thirds of output in those industries is sold as intermediary goods to other manufacturers, the remainder goes to bulk retailers. 12

15 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. Notably, following prior literature (e.g., Campello et al. (2011), Lin et al. (2011), 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) Customer Concentration The unit of observation in the Segment database is a supplier customer pair. For each supplier, we aggregate all available customer information and define customer concentration in two ways. Our first measure of customer concentration is based on the percentage of sales that a firm assigns to its major customers (similar to Banerjee et al. (2008)). In particular, we define CustomerSales as the sum of the percentage sales to the set of customers the firm reports as major customers (i.e., those at least 10% of total sales). CustomerSales is computed as follows: n i CustomerSales i = %Sales ij, where n i is the number of firm i s major customers, and %Sales ij = j=1 Sales of i to j, is the per- T otal Sales of i centage sales from firm i to customer j over all i s sales. A high level of CustomerSales means a large proportion of a firm s sales go to its major customers. Accordingly, a small group of buyers may ultimately influence the firm s investment and profitability. Our second measure is the sales-weighted size of a firm s major customers. This measure is more nuanced than the first in that it gives more importance to larger customers that also happen to be larger firms, which presumably might have more bargaining power. We define CustomerSize as the size of major customers, weighted by the firm s percentage sales to these 5 We further identify firms whose customers also borrow from their same banks. These firms account for 4% of borrowers in our sample and excluding them does not change our results. 13

16 customers. CustomerSize is computed as follows: n i CustomerSize i = %Sales ij Size j, j=1 where Size j is the size (defined by log of total assets) of customer j. A high level of Customer- Size means that a firm relies more heavily on a few, large-sized customers. 3.2 Borrowing Terms Chava and Roberts (2008), Roberts and Sufi (2009), and Campello et al. (2011) describe 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 changing loan terms, banks can also react to a firm s customer concentration by terminating their relationships with the firm. If customer concentration is related to excessive credit risk or undesirable investment choices, banks can stop extending loans to the firm, terminating their relations. We design empirical measures of banking relationships to capture these dynamics. Each time a firm discloses its customer concentration, we look forward in the sample window searching for subsequent loan arrangements (renewed relations in the future) with its current banks. We measure these future banking relations using two methods. First, we measure the length of the future banking relationship as the number of years in which the bank continues to lend to the firm in the future (FutureDuration). For each bank loan contract, FutureDuration counts the number of years until the last occurrence of the firm receiving a loan from the current bank. Higher values of FutureDuration suggest that the bank and the firm maintain 14

17 relations for a long period after the disclosure of information about customer concentration. Our second measure of banking relationship is the additional loans extended by the bank to the firm after the information of customer concentration (FutureLoans). FutureLoans is defined as the number of loans issued by the same bank after the current loan. Similar to FutureDuration, FutureLoans measures a bank s commitment to the lending relationship. However, it emphasizes the intensity rather than the length of the relationship. Naturally, both measures of banking relationships suffer from attrition bias, in that we observe shorter future duration and fewer future loans as we approach the end of the sample. We therefore restrict our banking relationship tests to fiscal years prior to 2007, leaving at least 5 years, which is above the 85 th percentile of the length of future banking relationships in our sample Loan Failures To corroborate our argument that a more concentrated customer base is associated with worse creditworthiness, we 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 the LoPucki database, which provides detailed bankruptcy filings over the window. 7 We match the LoPucki 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 LoanFailure = 0. Finally, we match the loan failure variable with firms customer information. 3.5 Empirical Methodology We estimate panel regression models for our baseline tests. Our models regress loan term variables on customer concentration measures together with firm-level, loan-level, and macrolevel controls. The specifications also feature bank effects, capturing firm bank pairings. The 6 Nonetheless, our results are unaffected if we do not impose this time window constraint. 7 These data are provided free of charge by Professor Lynn LoPucki at UCLA. 15

18 model can be written as follows: LoanT erm i,k,t = β 0 +β 1 CustomerConcentration i,t +β 2 F irmcharacteristic i,t +β 3 MacroV ar t + β 4 LoanCharacteristic k,t + g Industry g + h Bank h + ɛ i,k,t, (13) where i indicates the supplier, k indicates newly initiated loans, t indicates the year of the loan initiation; LoanTerm {LoanSpread, LoanMaturity, LoanCovenants}, and CustomerConcentration {CustomerSales, CustomerSize}. Borrowing terms and customer concentration may vary significantly across industries due to industry-specific idiosyncrasies. We thus include an industry-fixed effect (Industry g ) for each 2-digit SIC industry. Differences of borrowing terms can also arise from banks screening technology. Some banks are able to better detect firms credit quality or to more closely monitor the firms. These banks can select firms with lower customer concentration and impose looser borrowing terms. We therefore include bank-fixed effects (Bank h ) to control for intrinsic differences across banks. We report heteroskedasticityrobust errors clustered by industry. Firm characteristics include standard proxies for profitability, size, age, tangibility, marketto-book, leverage, and credit ratings. Macroeconomic conditions are measured by credit spread, term spread, and GDP growth rate. Loan characteristics include logs of loan maturity, loan amount, 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, β 1, 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 banking relationship as follows: BankingRelation i,h,t = β 0 + β 1 CustomerConcentration i,t + β 2 F irmcharacteristic i,t + β 3 MacroV ar t + g Industry g + h Bank h + u i,h,t, (14) where h indicates the lending bank and BankingRelation {FutureLoans, FutureDuration}. 16

19 Figure 1. The frequency of firms distinct appearances in the sample. This figure shows the number of firms who appear in the sample for a certain number of distinct observations. The horizontal axis shows the number of distinct observations. The vertical axis shows the number (frequency) of firms. We expect customer concentration to hamper firms future relationship with their banks. Therefore, we expect the coefficient β 1 to be negative in both banking relationship regressions. 3.6 Data Structure Similar to prior studies on contracts features (Graham et al. (2008), Lin et al. (2011), Hertzel and Officer (2012), and Valta (2012)), the unit of observation in our baseline tests is a loan contract. As such, we only observe variation in a firm s customer concentration if the firm signs new contracts in different years. This results in relatively few recurrences for each firm. Figure 1 plots the histogram of a firm s distinct observations (entries) in the sample. The distribution is highly skewed, and there are very few firms that appear in the sample more than five years. Indeed, 45% of the firms appear in the sample only once. Because of this data structure, similar to prior studies in the area, we do not include firm-fixed effects in our regressions. Instead, we control for industry-fixed effects, a fixed-effect component that has been shown to capture important variation in firm credit terms. 17

20 3.7 Summary Statistics 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 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 sample have spreads of 173 bps over LIBOR, maturity of 45 months, and 1.8 covenants. Table 1 About Here 4 Univariate Analysis We start our investigation of the impact of customer concentration on borrowing terms by characterizing the very phenomenon of concentration, which is still understudied. Prior research points to significant benefits in concentrating sales to a small group of buyers. These benefits come 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 capacity that we document to the potentially negative effects of customer concentration to 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 important that we have a basic understanding of these relations. As we demonstrate below, customer concentration is related with fundamental characteristics such as firm size and age. 4.1 Customer Concentration and Firm Operating Performance We verify the positive relation between customer concentration and operating performance in Figure 2. Following Patatoukas (2012), we rank firms into deciles according to their customer 18

21 Figure 2. The relation between customer concentration and firm s operational performance. The left panel shows the relation between customer concentration and firms profitability; the right panel shows the relation between customer concentration and firms sales growth. Customer concentration is measured by the total percentage sales to all major customers, CustomerSales. The decile ranking of CustomerSales is shown on the horizontal axes. concentration measure CustomerSales and plot the average operating performance of firms in each decile. The right (left) panel shows the average profitability (sales growth) of firms in each customer concentration level. Profitability and sales growth increase with customer concentration. Firms in the lowest levels of customer concentration observe annual profitability of less than 15% and sales growth of less than 10% per year. Firms in the highest level of customer concentration observe over 16% profitability and 15% sales growth, on average, per year. The patterns we document in Figure 2 are consistent Patatoukas s argument that firms with concentrated customer bases enjoy improved operating performance (see also Irvine et al. (2013)). Important for our purposes, these patterns show that firms with high customer concentration are not necessarily poor 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 concentration with these firm characteristics, since they may also affect credit terms. Figure 3 shows that customer concentration is negatively correlated with firm size and age, indicating that smaller, younger firms tend to deal disproportionately more with major cus- 19

22 Figure 3. The relation of customer concentration with firm size and age. The left panel shows the relation between size and customer concentration. The decile ranking of firm size is shown on the horizontal axis. The right panel shows the relation between firm age and customer concentration. Firm age is shown on the horizontal axis. Customer concentration is measured by the total percentage sales to all major customers, CustomerSales. tomers. 8 This correlation can lead to spurious relation between customer concentration and loan terms, since smaller, younger firms also tend to face more informational problems, thus having higher borrowing costs. This analysis suggests that it is important to control for firm size and age effects in our empirical tests. Figure 4 provides further insights into firms that operate with higher levels of customer 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 loans (e.g., Graham et al. (2008), Campello et al. (2011), and Lin et al. (2011)). These findings further corroborate the argument that firms with major customers are not under-performing businesses naturally prone to receive costlier, stricter loan terms from their banks. 4.3 Customer Concentration and Relationship-Specific Investment The model analysis implies that higher customer concentration may prompt a higher level of relationship-specific investment. Eq. (8) shows that a higher µ induces a lower p, meaning that higher customer concentration will lead more firms to undertake the relationship-specific 8 To describe firms life cycle, we only include firms whose life spans exceed 10 years. Yet, including firms who exist in the sample for fewer than 10 years does not change our inferences. 20

23 Figure 4. The relation of customer concentration with firm leverage and market-to-book. The left panel shows the relation between firm leverage and customer concentration. The right panel shows the relation between firm market-to-book ratio and customer concentration. Customer concentration is measured by the total percentage sales to all major customers, CustomerSales. The decile ranking of CustomerSales is shown on the horizontal axes. project. This result is sensible: an important customer can more easily contract with the firm to invest in customized projects that are suitable to its particular needs. Firms who do not have to cater to a major customer, on the other hand, only need to supply standardized products. We study this implication by examining 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 in several different ways following the existing literature. Our first measure of relationship-specific investment is firms R&D intensity, measured by the ratio of firms R&D to total assets (this proxy is used by Kale and Shahrur (2007), among others). 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. Third, we trace the inputs of firms production and examine how specific are the inputs they employ in their production process. Giannetti et al. (2011) provide detailed information on industries use of differentiated goods as inputs. Firms that use more differentiated inputs are shown to offer more differentiated 21

24 Figure 5. The relation between customer concentration and firm s relationship-specific investment. The left panel shows the relation between customer concentration and firms R&D intensity; the middle panel shows the relation between customer concentration and the originality of firms patents; the right panel shows the relation between customer concentration and firms uniqueness of inputs, which is measured by the percentage of firms inputs from industries producing differentiated goods. The levels of differentiated inputs for each industry are as defined in Giannetti et al. (2011). Customer concentration is measured by the total percentage sales to all major customers, CustomerSales. The decile ranking of CustomerSales is shown on the horizontal axes. products. We follow Giannetti et al. s approach and assign a firm to a level of differentiated inputs according to the industry in which it operates. Figure 5 shows a positive relation between firms relationship-specific investments and customer concentration. Firms that have higher customer concentration conduct more R&D activity, produce more original patents, and use more specific inputs for production. These patterns are consistent with the economics of our model. They are also consistent with more general 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 (see, e.g., Titman (1984) and Hart (1995)). 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 18 such models. 22