The Role of Bank Reputation in Certifying Future Performance Implications of Borrowers Accounting Numbers

Transcription

1 DOI: /j X x Journal of Accounting Research Vol. 00 No. 00 xxxx 2012 Printed in U.S.A. The Role of Bank Reputation in Certifying Future Performance Implications of Borrowers Accounting Numbers ROBERT M. BUSHMAN AND REGINA WITTENBERG-MOERMAN Received 3 January 2011; accepted 14 February 2012 ABSTRACT We investigate the role played by the reputation of lead arrangers of syndicated loans in mitigating information asymmetries between borrowers and lenders. We hypothesize that syndications by more reputable arrangers are indicative of higher borrower quality at loan inception and more rigorous monitoring during the term of the loan. We investigate whether borrowers with more reputable lead arrangers realize superior performance subsequent to loan origination relative to borrowers with less reputable arrangers. We further examine whether certification by high-reputation lead banks extends to Kenan-Flagler Business School, The University of North Carolina at Chapel Hill; The University of Chicago Booth School of Business Accepted by Phil Berger. We thank the editor, an anonymous reviewer, Dan Amiram, Ray Ball, Ryan Ball, Douglas Diamond, Merle Erickson, Rich Frankel, Christian Leuz, Michael Minnis, David Ross, Florin Vasvari, Jieying Zhang, and participants at the 2010 Dopuch Conference at Washington University, the 2010 Duke/UNC Fall Camp, 2010 Financial Economics and Accounting Conference at the University of Maryland, the 2010 Stanford Summer Camp and seminar participants at Indiana University, Tsinghua University, and the University of Texas at Dallas for helpful comments. We thank the Thomson Reuters Loan Pricing Corporation for providing loan data. We gratefully acknowledge the financial support of the Kenan- Flagler Business School, The University of North Carolina at Chapel Hill, and the University of Chicago Booth School of Business. Regina Wittenberg-Moerman also gratefully acknowledges the financial support of the Neubauer Family Fellowship. 1 Copyright C, University of Chicago on behalf of the Accounting Research Center, 2012

2 2 ROBERT M. BUSHMAN AND REGINA WITTENBERG-MOERMAN the quality of borrowers reported accounting numbers. Controlling for endogenous matching of borrowers and lead banks, we find that higher bank reputation is associated with higher profitability and credit quality in the three years subsequent to loan initiation. We also show that bank reputation is associated with long-run sustainability of earnings via higher earnings persistence, and debt contracting value of accounting via a stronger connection between pre-loan profitability and future credit quality. We further document that the enhanced earnings sustainability associated with higher reputation lead banks reflects both superior fundamentals and accruals more closely linked with future cash flows. 1. Introduction Information asymmetries create frictions that can impact the costs of raising external funds. In this paper, we investigate the role played by the reputation of the lead arrangers of syndicated loans in mitigating information asymmetries between borrowers and lenders. By investing in rigorous pre-loan evaluations of borrowers quality and post-loan monitoring of borrowers performance, a bank can establish a strong reputation as a lead arranger via a track record of successful loans. We hypothesize that syndications by more reputable arrangers indicate higher borrower quality at loan inception and more rigorous monitoring during the term of the loan. To examine this hypothesis, we investigate whether borrowers with more reputable arrangers realize superior performance subsequent to loan origination, relative to borrowers with less reputable arrangers. We further examine whether the quality certification supplied by high-reputation lead banks extends to the quality of a borrower s reported accounting numbers. Syndicated lending, in which multiple lenders lend to a firm under a common contract, is an interesting setting for analyzing the role of reputation in certifying firm quality. Agency problems arise because lead banks privileged access to borrowers may reveal private information not available to other syndicate members, and a bank s due diligence and monitoring efforts are not observable (e.g., Lee and Mullineaux [2004], Sufi [2007], Ivashina [2009]). Further, the lead bank s formal contractual incentives to screen and monitor are limited, as the fraction of loans retained by lead banks is relatively small. The confluence of unobservable effort and insufficient contractual incentives with the repeated nature of the syndicated loan market creates a role for reputation. By observing outcomes of previously arranged loans, market participants can assess the extent to which a bank provides superior screening and monitoring on behalf of other investors. A recent literature supports the importance of lead arranger reputation in syndicated lending by examining the consequences for lead banks of a borrower s failure. Lin and Paravisini [2011] exploit large corporate frauds and find that, consistent with reputation loss, lead banks of loans to firms where fraud is discovered hold larger fractions of loans they syndicate following fraud discovery. Gopalan, Nanda, and Yerramilli [2011] focus on

3 ROLE OF BANK REPUTATION 3 Chapter 11 bankruptcies, showing that, following defaults, lead arrangers retain larger fractions of subsequent loans they syndicate, and are less likely to syndicate loans and attract participant lenders. 1 There is also an extensive literature documenting that loan announcements generate positive abnormal stock returns for borrowers. 2 It is also documented that these stock price responses are relatively more favorable for loans syndicated by higher reputation banks (Billett, Flannery, and Garfinkel [1995], Ross [2010]), suggesting that bank reputation certifies borrower quality. Ross [2010] further shows that the impact of a lead bank s reputation on abnormal returns is stronger for more opaque borrowers where certification is likely more important. This evidence is consistent with existing theories of financial intermediary s reputation. For example, Chemmanur and Fulghieri [1994a] show that higher reputation banks have incentives to perform more rigorous preloan evaluations of borrowers unobservable prospects than do lower reputation banks. Thus, in equilibrium, high-reputation banks are more likely to lend to borrowers with superior future prospects, reflecting favorably on these unobservable prospects. We extend prior literature by examining the relation between lead arranger reputation and two aspects of borrowers performance following loan origination: future profitability and future credit quality. Further, we investigate whether reputation is associated with long-run sustainability of earnings and enhanced debt contracting value of accounting information, as reflected in a stronger relation between borrowers accounting numbers reported just prior to loan origination and future performance outcomes. 3 Consistent with prior research (e.g., Sufi [2007], Ross [2010]), we measure bank reputation based on lead banks market share in the syndicated loan market. 4 To isolate the future performance implications of the certification role of reputation, we control for what are likely to be strong selection effects underpinning the matching of borrowers and lead banks. For example, borrowers with better future prospects or higher quality accounting may choose more reputable arrangers. Alternatively, reputable arrangers may choose to contract with borrowers likely to be lucrative consumers of the bank s services in the future. We employ a matching framework to explicitly control for observable and unobservable factors that 1 The results in Gopalan, Nanda, and Yerramilli [2011] and Lin and Paravisini [2011] suggest that lead bank reputation substitutes for contractual skin in the game. We discuss this in more detail in section 2. In section 6.2, we examine the impact of fraud on the certification role of reputation. 2 See James and Smith [2000] for a review of the empirical literature. Maskara and Mullineaux [2011] question these studies by examining the role of self-selection in determining which banks announce their loans. See also Boot [2000] and Saunders and Cornett [2004] for reviews of the special role of banks. 3 Ball, Bushman, and Vasvari [2008] define the debt contracting value of accounting as the ability of accounting numbers to capture deterioration in credit quality on a timely basis. 4 In section 2, we discuss in depth this measure of reputation and its potential limitations.

4 4 ROBERT M. BUSHMAN AND REGINA WITTENBERG-MOERMAN impact both the likelihood of a borrower dealing with a high-reputation bank and the borrower s future performance (Heckman and Navarro- Lozano [2004], Ross [2010]). 5 We control for observable factors that influence borrower-bank matching, including borrowers past performance, historical persistence of earnings, extreme accruals, earnings volatility, and the potential for cross-selling services to the borrower. Further, we include variables based on the location of borrowers relative to potential lenders that determine borrower arranger matching, but that are independent of borrowers future performance. To investigate whether certification by high-reputation lead banks extends to the quality of borrowers reported accounting numbers, we estimate endogenous switching regressions separately for high and low reputation partitions. Turning to our results, we document that higher lead arranger reputation is associated with higher profitability and credit quality in the three years subsequent to loan initiation. 6 With respect to accounting quality, we find that earnings persistence is higher for borrowers with high-reputation lead banks than for those with lower reputation lead banks, implying that borrower profitability reported at loan origination is more sustainable for high reputation banks. We also show that borrowers with high-reputation lead arrangers exhibit enhanced debt contracting value of accounting, as indicated by a stronger connection between pre-loan profitability and future credit quality. Finally, we examine whether the higher earnings sustainability associated with high-reputation lead banks is a consequence of better fundamentals, higher accrual quality, or both. We follow Minnis [2011] and regress oneyear-ahead cash flows from operations on the contemporaneous cash flow and accrual components of earnings separately for the high and low reputation partitions. We find that borrowers with high-reputation banks exhibit higher cash flow persistence and significantly greater accruals quality. Accruals of borrowers with high-reputation lead banks are more strongly related to future cash flows than are accruals of other borrowers. While we acknowledge that we may not have controlled for all possible alternative explanations, the totality of our results is consistent with the proposition that lead bank reputation certifies the quality of borrowers and that this certification extends to the quality of borrowers accounting numbers at loan inception. These results contribute to the literature across several dimensions. First, while a large literature examines relations 5 Private lending presents the possibility that private information, unobservable to the researcher, underpins both the decisions that determine bank-borrower matching and borrowers future performance. We also employ alternative specifications, finding that all results are robust to using OLS, propensity matching, and firm fixed effect specifications. 6 For example, in the third year following loan origination, borrowers with reputable arrangers report profitability that is 2.0 percentage points higher and credit ratings that are 2.5 notches lower (lower numerical ratings indicate higher credit quality) relative to borrowers with less reputable arrangers.

5 ROLE OF BANK REPUTATION 5 between auditor characteristics and accounting quality, less is known about the role of financial intermediaries in establishing the credibility of accounting numbers. 7 Lee and Masulis [2011] examine the link between underwriter reputation and earnings management and Agrawal and Cooper [2010] examine the effect of venture capitalist reputation on restatements. We extend these studies by documenting that lead bank reputation is associated with higher earnings and cash flow persistence, and with earnings that more strongly predict future credit quality. Our results suggest that the process of establishing the credibility of accounting reports operates through multiple channels, where intermediaries other than auditors use their reputation to certify firms accounting numbers. Second, we add to the literature on the role of intermediary reputation in mitigating financial contracting frictions. We extend Ross [2010] by establishing that the higher stock returns associated with announcements that a high-reputation bank syndicates a loan are consistent with reputation certifying borrowers quality, as reflected by their future performance. This evidence of superior economic and accounting quality of borrowers with high reputation lead banks also complements Lin and Paravisini [2011] and Gopalan, Nanda, and Yerramilli [2011], who, by establishing that a bank s reputation and skin in the game are substitutes in syndicated lending, show the importance of reputation for lead banks screening and monitoring incentives. Third, our result that pre-loan profitability better predicts future credit quality for borrowers of higher reputation lead banks contributes directly to the debt contracting literature, which posits that accounting quality is a determinant of a borrower s credit risk (e.g., Francis et al. [2005], Ashbaugh- Skaife, Collins, and LaFond [2006], Zhang [2007], Bharath, Sunder, and Sunder [2008]). Our analysis suggests that the debt contracting value of accounting also depends on the reputation of the lead bank arranging a loan. Our evidence also implies that borrowers endowed with favorable future prospects, but facing significant investor uncertainty over their accounting quality, can mitigate this uncertainty by having a high-reputation lead bank credibly certify its accounting quality. The rest of the paper is organized as follows. Section 2 presents the conceptual basis of reputation certification. Section 3 discusses our empirical design and econometric approach to self-selection issues. Section 4 describes the sample and data and presents descriptive statistics, while section 5 presents our main results. Section 6 discusses robustness issues and section 7 concludes. 7 Existing literature finds that auditor reputation is associated with fewer accounting errors and irregularities (Defond and Jiambalvo [1991]), less IPO underpricing (Beatty [1993]), higher earnings response coefficients (Teoh and Wong [1993]), lower abnormal accruals (Becker et al. [1998]), more predictive post-ipo delistings (Weber and Willenborg [2003]), and lower cost of public debt (Lou and Vasvari [2011]). See Francis [2004] for a review of the literature.

6 6 ROBERT M. BUSHMAN AND REGINA WITTENBERG-MOERMAN 2. Conceptual Basis of Reputation Certification It has long been posited that financial intermediaries play a key role as credible producers of information to market participants. One important mechanism for projecting credibility is reputation (e.g., Leland and Pyle [1977], Campbell and Keracaw [1980], Diamond [1984]). Klein and Leffler [1981], DeAngelo [1981], and Shapiro [1983], among others, demonstrate that reputable firms have incentives to supply high-quality goods to the market. In equilibrium, rents flow to firms as compensation for investments in reputational capital, and such rents bond the firm to good behavior since it is vulnerable to a loss of rents if shirking is detected. The literature applies this reputational capital paradigm to the certification role of financial intermediaries. Important theoretical contributions include Baron [1982] and Booth and Smith [1986], who examine the role of investment bankers in certifying the issuance of equity and risky debt securities. Numerous papers examine the role of financial intermediary reputation in equity underwriting (e.g., Beatty and Ritter [1986], Carter and Manaster [1990], Nanda and Yun [1997], Fernando, Gatchev, and Spindt [2005]), supporting the importance of reputation in certifying initial and secondary public offerings. Intermediary reputation has also been emphasized in additional financial market settings, such as bond underwriting (Fang [2005]), venture capital (e.g., Hsu [2004], Nahata [2008], Krishnan et al. [2011]) and auditing (e.g., DeAngelo [1981], Titman and Trueman [1986], Bachar [1989], Datar, Feltham, and Hughes [1991], Ball, Jayaraman, and Shivakumar [2012]). While reputation may be somewhat less important in the syndicated loan market relative to other settings, such as the equity market, where many players are individual investors rather than sophisticated institutions, the nature of syndicated lending and the role played by lead banks create a potentially important role for reputation. Lead arrangers establish relationships with borrowers, perform pre-loan due diligence, negotiate contract terms and monitor borrowers after loans are made. These actions provide them with access to borrowers private information not available to other syndicate participants. This privileged access to information, together with unobservability of due diligence and monitoring efforts, create agency problems (e.g., Lee and Mullineaux [2004], Sufi [2007], Ivashina [2009]). Further, while syndicate participants include large banks with independent lending arrangements with borrowers, many syndicate participants are smaller banks and nonbank institutional investors (collateralized loan obligations, hedge funds, pension funds, and insurance companies), who became major players in the syndicated loan market in the last decade (Bushman, Smith, and Wittenberg-Moerman [2010], Ivashina and Sun [2011]). Small banks and institutional investors typically do not have established relationships with borrowers and therefore rely heavily on lead banks monitoring and due diligence. In addition, the fraction of syndicated loans retained by the lead bank (i.e., the bank s skin in the game) is relatively small. Sufi [2007] shows that

7 ROLE OF BANK REPUTATION 7 the mean (median) fraction of loans retained by the lead bank is 28.5% (23.5%). Therefore, the lead bank s contractual incentives to screen and monitor are limited. Further, the syndicated loan market is characterized by repeated transactions between lead banks and participants. Unobservable screening and monitoring efforts, insufficient formal contractual incentives, and the repeated nature of the syndicated lending suggest that a bank s reputation may be a vital mechanism in the syndicated loan market. A key premise of certification is that an intermediary s reputation is enhanced by positive outcomes and damaged by negative outcomes. Lin and Paravisini [2011] examine the effect of frauds by Enron, WorldCom, and others on the lead arrangers of their loans outstanding during the fraud period, including some of the most reputable arrangers in the syndicated loan market, such as J.P. Morgan Chase, Bank of America, and Citigroup. They find that, consistent with reputation loss, these lead banks hold a larger fraction of the loans they arrange following the frauds discovery. Further, Gopalan, Nanda, and Yerramilli [2011] find that when lead banks borrowers are subject to Chapter 11 bankruptcy, the banks are less likely to syndicate subsequent loans and attract participant lenders, and retain larger fractions of loans they syndicate. Pertinent to our analysis, Chemmanur and Fulghieri [1994a] model the ex ante evaluation of borrowers by banks. In their model, firms with unobservable future prospects seek to raise external financing, while banks evaluate a firm s future prospects before agreeing to accept them as clients. More rigorous pre-deal screening increases the likelihood that borrowers with poor prospects are discovered and denied loans, but more rigorous evaluation is more costly to the bank. Borrowers who pass more rigorous evaluation will perform better on average relative to firms granted financing based on less rigorous evaluation. However, banks choice of rigor is not observable to outsiders, and so investors assess a bank s reputation for supplying rigorous evaluations by conditioning on observable outcomes of past transactions sponsored by the bank. In equilibrium, higher reputation banks adopt more rigorous evaluation standards than lower reputation banks do, and so are more likely to transact with firms with superior future prospects. Also, Chemmanur and Fulghieri [1994b] develop a model where reputation provides banks with endogenous incentives to devote more resources to information production about firms that subsequently become financially distressed. Investors believe that more reputable banks supply a greater ex post monitoring effort, and, in equilibrium, they do. Together, Chemmanur and Fulghieri [1994a,b] suggest that syndications by more reputable arrangers should indicate higher borrower quality at loan inception and more rigorous monitoring during the term of the loan. Consequently, if borrowers true quality is not publicly observable and high-reputation banks are more likely to transact with better borrowers, then the fact that a high-reputation bank contracts with a borrower reveals favorable information about the borrower s future prospects. However, it

8 8 ROBERT M. BUSHMAN AND REGINA WITTENBERG-MOERMAN is also possible that borrowers with better future prospects or higher quality accounting simply choose more reputable arrangers. Thus, empirically isolating the future performance implications of reputation is complicated by potential selection effects underpinning the matching of borrowers and lead banks. Ideally, we would like to examine a setting where borrowers are randomly assigned to high- and low-reputation banks, and banks then choose with which borrowers to contract. This would allow us to cleanly isolate whether the act of a high-reputation bank accepting a borrower reflects favorably on the borrower s quality. In the absence of such a setting, our empirical challenge is to control for the factors, observable and unobservable, that simultaneously impact the likelihood of a borrower dealing with a highreputation bank and the borrower s future performance. To address this challenge, we employ an endogenous matching framework (Heckman and Navarro-Lozano [2004], Fang [2005], Ross [2010]). We control for observable factors that may simultaneously influence borrower-bank matching and a borrower s future performance, including its past performance, historical earnings persistence, extreme accruals, earnings volatility, and the cross-selling potential of banking services. We also include variables based on the proximity of borrowers and potential lenders; these proximity measures are expected to significantly affect borrower arranger matching, but not borrowers future performance (we extend this discussion in section 3). Another empirical challenge is to proxy for lead bank reputation. We define a lead bank as reputable if its average market share of the syndicated loan market over our sample period ( ) is above 2% (see appendix A for details). We classify J.P. Morgan Chase, Bank of America, Citigroup, Wachovia, Credit Suisse First Boston, and Deutsche Bank as reputable arrangers. We also classify J.P. Morgan, Bank One, and Fleet Boston as reputable arrangers over the period prior to their merger/acquisition. Together, these banks syndicated over 65% of the loan issuances (by volume) over the sample period. In contrast, the remaining syndicated loans were arranged by more than 1,000 banks, the vast majority of which had a market share of less than 0.02%. The banks classified as reputable are among the largest and most sophisticated banks in the world, and are therefore likely to have a better feel for pricing conditions, better information on potential borrowers, and superior competence at screening and monitoring borrowers. This suggests that reputation is a mechanism by which market participants come to believe that a lead bank does indeed possess superior screening and monitoring technologies and, more importantly, that the bank consistently deploys these technologies on behalf of other loan investors. 8 8 Given that Bank of America and Citigroup had to be bailed out in the recent financial crisis, one may question our conjecture that these banks are reputable. However, Bank of America and Citigroup remained dominant arrangers over the period (with an

9 ROLE OF BANK REPUTATION 9 Our measure of bank reputation follows prior research (e.g., Sufi [2007], Ross [2010]). However, it is possible that dominant market share could be a consequence of factors other than reputation. First, dominant banks may have achieved their market share by offering more attractive terms than other lenders or from passing on an advantage of raising money at favorable rates. However, as argued by Ross [2010], these explanations are not plausible, given that lead banks with high market share syndicate most of the loan principal to other syndicate participants. Another alternative is that these banks have achieved oligopolistic market power through structural barriers that suppress competition. If so, they should be able to extract rents from borrowers, such as higher interest rates. However, Ross [2010] finds that lead banks with high market share charge lower interest rates, and are more likely to lend without the protection of a borrowing base that limits outstanding principal to a fraction of readily saleable assets. These results are consistent with such banks having a strong reputation for evaluating borrowers underlying business and true default risk, allowing them to offer borrowers more attractive terms while still inducing syndicate members to participate. Fang [2005] finds similar evidence in the public bond market: reputable underwriters charge higher fees, but a lower interest rate, with an overall favorable effect on issuers net proceeds. Third, reputable banks may achieve high market share because the large scale and scope of their operations allows them to offer borrowers lower interest rates based on expectations that the bank can later cross-sell significant services to the borrower (investment banking, derivatives, structured finance, etc.). Yasuda [2005] shows that serving as the loan s lead arranger helps banks to gain future underwriting of a borrower s public bonds. Drucker and Puri [2005] find that private lending increases the probability of receiving the current and future underwriting of a borrower s equity securities. In the context of syndicated loans supporting leveraged buyouts, Ivashina and Kovner [2011] find evidence that lead banks price loans to cross-sell future fee services. In our empirical tests, we explicitly control for the cross-selling potential of banking services to the borrower Empirical Design Section 3.1 describes the econometric framework. In section 3.2, we discuss our empirical strategy for isolating the implications of reputation certification for borrowers future performance and accounting quality. average market share of 18.3 and 14.4 percent, respectively), suggesting that their reputation continues to be strong in the syndicated loan market. J.P. Morgan Chase, Credit Suisse First Boston, Deutsche Bank, and Wachovia (until its merger with Wells Fargo) also continued to lead syndicated loan issuances over the crisis period (Bank One and Fleet Boston merged with J.P. Morgan Chase and Bank of America, respectively, in 2004). 9 We thank an anonymous reviewer for suggesting that we address the cross-selling potential in our analyses.

10 10 ROBERT M. BUSHMAN AND REGINA WITTENBERG-MOERMAN 3.1 ECONOMETRIC FRAMEWORK Consistent with recent research on intermediary reputation (Fang, [2005], Ross [2010]), we employ an endogenous matching framework designed to explicitly control for observable and unobservable factors associated both with the choice of lead arranger and the borrower s future performance (e.g., Heckman and Navarro-Lozano [2004]). The appropriateness of this framework rests on the presumption that private lending arrangements involve access to confidential information and that such private information may underpin the observed matching between banks and borrowers and also be associated with future performance. The basic framework consists of a binary outcome equation that models matching between borrowers and lead banks, and two regression equations, one for each reputation partition (see Maddala [1983], chapters 8 and 9). Formally, we have: Reputation i = Z i γ + ε i, (1) high rep FuturePerformancei low rep FuturePerformancei = x i β high rep high rep + μi, and (2) = x i β low rep low rep + μi. (3) Equation (1) is the borrower lead arranger matching equation. We implement equation (1) as a probit model where the dependent variable is an indicator variable set equal to one if a loan is arranged by a high-reputation bank, and zero otherwise. Equation (2) is the performance equation for reputable banks, and equation (3) is that for less reputable banks, where future performance is either profitability or credit quality. The vector x i includes observable borrower and loan characteristics posited to directly influence borrowers future performance. Equations (2) and (3) effectively allow interactions between arranger reputation and the explanatory variables x i in the borrower performance models. To control for unobservable factors affecting both bank-borrower matching and future performance, we allow the residuals in equations (2) and (3) to be correlated with the high rep low rep residual in equation (1). The three error terms (ε i,μi,μi )areassumed to have a trivariate normal distribution. We simultaneously estimate equations (1), (2), and (3) by maximum likelihood. 10 The vector Z i in (1) represents observable variables important to determining matching between borrowers and lead banks. This vector is comprised of two distinct subsets. The first subset contains factors that determine borrower lead arranger matching and at the same time may be associated with borrowers future performance and accounting quality. The 10 All results are robust to using a two-step procedure which: 1) estimates equation (1) as a probit model and derives the inverse Mills ratio from the fitted values of Reputation i,and then 2) includes the inverse Mills ratios in equations (2) and (3) to account for unobservable factors.

11 ROLE OF BANK REPUTATION 11 second subset includes two variables based on the location of borrowers relative to potential lenders that determine borrower arranger matching, but that are independent of borrowers future performance and are properly excluded from the future performance regressions. With respect to the first subset, recall that one of our main research objectives is to examine whether reputation certifies borrowers accounting quality. It is thus important to control for the possibility that borrowers with higher earnings quality simply select higher reputation lead banks. In addition to a borrower s prior earnings, we include a range of variables that capture observable earnings quality at a loan s issuance. In particular, we include two measures of historical earnings persistence (see appendix A for a detailed description of all variables). We estimate our first persistence measure Earnings-persistence-1 by the coefficient from a time series regression of earnings on prior year earnings (e.g., Ali and Zarowin [1992], Francis et al. [2004], Frankel and Litov [2009], Dechow, Ge, and Schrand [2010]). Alternatively, we follow Skinner and Soltes [2011], who find that reported earnings are significantly more persistent for dividendpaying firms and firms that make stock repurchases. We define Earningspersistence-2 as an indicator variable taking the value of one if a borrower paid cash dividends and performed stock repurchases in the majority of years over the five-year period preceding a loan s issuance, zero otherwise. To further control for historical earnings persistence, we include a measure of extreme positive accruals (Accruals-pos-extreme), as such accruals tend to quickly reverse and decrease future profitability (Sloan [1996]). We also control for whether a borrower reported losses in the year prior to the year of a loan s issuance (Loss), as losses are typically transitory and less persistent (Hyan [1995], Basu [1997], Frankel and Litov [2009]). In addition, we address the possibility that firms with a more stable performance select more reputable arrangers and include a measure of earnings volatility. We measure Earnings-volatility by the standard deviation of the ratio of EBITDA to total assets over the 10-year period prior to a loan s issuance (Dichev and Tang [2009]). In untabulated analysis, we substitute earnings volatility by cash flow volatility and find that the results are unchanged. The first subset of variables in the borrower-lead arranger matching equation also includes controls for a variety of firm- and loan-specific characteristics, such as credit risk; leverage; and a loan s size, maturity and purpose. In addition, we incorporate a measure of cross selling potential to control for the possibility that high-reputation lead banks cherry pick borrowers to whom they can cross-sell significant services in the future. We adopt the approach of Ivashina and Kovner [2011] and proxy for a lead bank s expectations of future fee business with a borrower s past business. We measure a borrower s past business by the amount of total equity and bond issuances in the five-year period preceding the year of the loan. We focus on equity and bond issuances as Drucker and Puri [2005] and Yasuda [2005] show that serving as the loan s lead arranger helps banks to gain future

12 12 ROBERT M. BUSHMAN AND REGINA WITTENBERG-MOERMAN underwriting of a borrower s equity and public bonds, respectively. 11 The value of equity and bond transactions should represent a good proxy for bank fee amounts because, in these transactions, fees are typically strongly related to the transaction value (Chen and Ritter [2000], Ivashina and Kovner [2011]). 12 We define Cross-selling-potential as an indicator variable taking the value of one if a borrower s cross-selling potential, estimated by the value of past equity and bond transactions, is above the sample median, zero otherwise. 13 For the second subset in the vector Z i, we follow Ross [2010] and include two variables based on the proximity of borrowers and potential lenders that impact borrower arranger matching but not borrowers future performance. The first variable derives from the idea that a borrower is more likely to borrow from a high-reputation bank if it is in close proximity to the bank. The importance of proximity is supported by evidence that a shorter distance between the borrower and the lender enhances lenders information gathering and monitoring abilities (Peterson and Rajan [2002], Berger et al. [2005], Sufi [2007], Dass and Massa [2011]). In particular, proximity to the borrower facilitates lenders collection of soft information (Peterson and Rajan [2002], Berger et al. [2005]) and access to private information (Dass and Massa [2011]). Further, the home bias literature also suggests that investors have better access to information about local firms (e.g., Coval and Moskowitz [1999] and [2001]). We define the variable Nearreputable-arranger as equal to one if the borrower s headquarters is within 60 km of any high reputation banks headquarters, zero otherwise. 14 We expect Near-reputable-arranger to be positively associated with the choice of a high-reputation lead bank. The second variable captures the idea that a borrower will be less likely to borrow from a high-reputation bank if there are credible local alternative lenders nearby. We consider a local bank to be a credible alternative if it arranges, on average, at least 30 syndicated loan deals per year over the sample period, suggesting that it has a considerable expertise in 11 The total amount of a borrower s equity and bond issuance over the five-year period preceding the loan issuance is 54%, correlated with the total amount of a borrower s equity and bond issuances over the five-year period following loan issuance, supporting the appropriateness of the cross-selling measure based on borrower past business. In unreported analysis, we base the cross-selling variable on the amount of future transactions. The results continue to hold. 12 The Thompson One Banker database, from which we obtain transactions values, reports fee amounts very sparsely; fees are not available for the vast majority of the sample borrowers past and future transactions. 13 In untabulated analyses, we find that our inferences are similar if we employ the continuous measure of cross- selling potential (the total value of a borrower s past equity and bond transactions) instead of the indicator variable. The economic and statistical significance of the cross-selling potential variables are somewhat smaller in this case. 14 Following Ross [2010], we conjecture that, while reputable banks have satellite offices, these offices rarely engage in the syndicated loan activities of the large, public borrowers that are the focus of this study.

13 ROLE OF BANK REPUTATION 13 syndicated lending. 15 This variable, Near-local-arranger, is set equal to one if a borrower s headquarters is within 60 km of a credible regional lead arranger s headquarters and zero otherwise. We expect Near-local-arranger to be negatively associated with the choice of a high reputation lead bank. 16 While location likely plays a large role in determining the propensity of lenders and borrowers to do business together, it is not plausible that a lender or borrower would change the location of its headquarters solely for the purpose of consummating a loan (Ross [2010]). Moreover, two otherwise observationally equivalent borrowers should not differ in their susceptibility to the loan certification effect and in post-loan performance merely because the borrowers headquarters are located in different regions of the United States. Therefore, we treat the Near-reputable-arranger and Near-local-arranger variables as valid instruments that are independent of borrowers future performance and are properly excluded from the future performance regressions. 3.2 ESTIMATING THE FUTURE PERFORMANCE IMPLICATIONS OF BANK REPUTATION We first investigate the basic question of whether firms with high reputation banks exhibit a superior future performance (section 5.2). Here, we utilize a treatment effects model that includes a reputation dummy variable. 17 The specification simplifies equations (1), (2), and (3) to a twoequation framework that retains equation (1), and replaces equations (2) and (3) with a single equation (4) that restricts β high rep = β low rep. The single equation takes the form: FuturePerformance i = β R Reputation i + x i β + μ i. (4) Reputation i is set equal to one if a loan is arranged by a reputable bank, zero otherwise. The error terms (ε i,μ i ) are assumed to be jointly normally distributed. The coefficient β R represents an unbiased estimate of the impact of reputation certification on future performance. FuturePerformance i is either a borrower s profitability or credit rating in each of the three years subsequent to loan initiation. The set of control variables includes the same variables as in the borrower-matching equation, including borrowers past profitability, historical earnings quality and volatility measures, cross-selling potential, and various firm and loan characteristics. In addition, we control 15 The results are not sensitive to the threshold number of syndicated loan deals. When we change the threshold to 40 or 60 deals per year, this instrument performs similarly in both the arranger borrower matching equation and the future performance equation. 16 Our results are robust when we define Near-reputable-arranger and Near-local-arranger based on the distance between the borrower s and the lender s headquarters within 150 km. 17 We use the treatment effects model primarily to facilitate simple comparisons of the results using the matching model to alternatives such as OLS, propensity matching, and firm fixed effects. We also compute treatment effects using counterfactuals constructed from the endogenous switching model estimated in section 5.3.

14 14 ROBERT M. BUSHMAN AND REGINA WITTENBERG-MOERMAN for loan interest rate and whether institutional investors participate in a loan syndicate. 18 When FuturePerformance is credit ratings, we augment the controls with measures of whether the borrower is on the S&P watch list or has an S&P outlook at the time of a loan s issuance. To investigate whether reputation operates through the quality of borrowers accounting numbers, we estimate the full switching regression framework represented by equations (1), (2), and (3) above (see section 5.3). We estimate separate performance mappings for high and low reputation partitions. The implication of reputation for accounting quality is captured by significant differences in the mapping between borrowers preloan profitability and future performance across high and low reputation partitions. With respect to future profitability, significant differences in the mapping between borrowers pre-loan profitability and future profitability capture differences in earnings persistence across reputation partitions. When future performance is credit quality, significant differences capture differences in the strength of the connection between pre-loan profitability and future credit ratings. 4. Sample, Data, and Descriptive Statistics 4.1 DATA SOURCES AND SAMPLE SELECTION We employ the DealScan database provided by the Thomson Reuters Loan Pricing Corporation (TRLPC). We obtain firm characteristics from Compustat. Firms senior debt ratings, watch list additions, and outlook changes (at the firm level) are retrieved from the S&P historical database. If the S&P historical database does not cover a particular firm, we retrieve the Moody s, Fitch, or DPR senior debt rating from the Mergent Fixed Income Securities Database (FISD). For borrowers missing ratings on S&P and FISD, we hand-collect ratings from the Internet-based version of TRLPC. CDS data is obtained from Markit. We employ Thompson One Banker to compile information on sample borrowers equity and bond transaction; this database covers all bond and equity issuances in the United States. Table 1 summarizes the sample selection process. For the period 1998 to 2006, DealScan reports 68,368 facilities outstanding to U.S. firms and issued in U.S. dollars. Merging this data with Compustat allows us to identify 25,518 facilities issued to public firms. Next, we exclude facilities with insufficient loan data, leaving 19,141 facilities. We also require sample borrowers 18 We require all the variables in the borrower lead arranger matching model to be observable prior to the matching process. We exclude the interest rate spread and institutional investor indicator variables from the model because these variables become known later in the loan syndication process, when the lead arranger assesses the market demand for the loan and recruits syndicate participants. At the same time, the matching model does comprise a number of loan characteristics, including size, maturity, and purpose. While the first two characteristics may be subject to some negotiation between the borrower and the syndicate, the borrower typically requires a particular loan size and duration when approaching the lead arranger. Loan purpose is also known prior to loan initiation.

15 TABLE 1 Sample Selection ROLE OF BANK REPUTATION 15 Filters Number of facilities Syndicated loans to U.S. borrowers, in U.S. dollars, issued over the 68,368 period from 1998 to 2006 Intersection with Compustat 25,518 After elimination of facilities with missing loan data 19,141 After elimination of facilities with insufficient firm data, including 9,857 past and future profitability After elimination of unrated facilities 6,675 This table presents the sample selection process. to have sufficient Compustat data for estimating a borrower s performance prior to and following a loan s issuance; this restricts our sample to 9,857 facilities. 19 Finally, we exclude loans of non rated borrowers. The remaining sample contains 6,675 facilities related to 1,272 firms. 4.2 DESCRIPTIVE STATISTICS Panel A of table 2 reports that 70% of sample loans are issued by reputable arranger. Sample loans have, on average, a size of $546M, a maturity of 43 months, and an interest spread of 156 basis points. Institutional loans represent 12% of the sample loans, 14% of the loans are issued for restructuring purposes, and 49% are syndicated by a relationship arranger. In terms of credit quality characteristics at loan origination, sample firms have a mean and median S&P senior debt rating of BBB. The mean CDS spread for the firms with traded CDS is 1.62%, and 8% (18%) of the loans relate to firms on the S&P negative watch list (Outlook). Sample firms are relatively large, with a mean and median value of total assets of 9,439M and 2,776M, respectively, and they have average profitability, as measured by the ratioofebitdatototalassets,of13%. In panels B and C of table 2, we report summary statistics for the reputable and less reputable arranger samples, respectively. The two samples differ along a number of dimensions. Loans syndicated by reputable arrangers are larger, have a shorter maturity, and are also more likely to be issued by arrangers that have a previous relationship with the borrower; loans with a prior lending relationship represent 52% of the reputable arranger sample, relative to 43% of the less reputable arranger sample. As suggested by the Interest-spread, Credit-rating, CDS-spread, andleverage variables, reputable arrangers loans are issued to less risky firms relative to the firms of less reputable arrangers. Reputable arrangers borrowers are significantly larger but do not differ in terms of profitability at the time of the loan s origination. 19 We require at least nine years of profitability-related data to estimate a borrower s past earnings persistence and volatility, and profitability following the loan issuance.

16 16 ROBERT M. BUSHMAN AND REGINA WITTENBERG-MOERMAN TABLE 2 Descriptive Statistics Panel A: Total sample Number of Distribution Loan and Firm Characteristics observations Mean SD 25% 50% 75% Loan characteristics: Reputation 6, Loan-size (in millions) 6, Maturity (in month) 6, Interest-spread (in bps) 6, Institutional 6, Restructuring-purpose 6, Prior-relationship 6, Credit risk characteristics: Credit-rating 6, CDS-spread (in %) 2, Watch-negative (Watch-positive) 6, (0.02) Outlook-negative (Outlook-positive) 6, (0.07) Additional firm characteristics: Firm-size (in millions) 6,675 9,439 25,573 1,056 2,776 9,533 Leverage 6, ROA 6, Interest-coverage 6, Earnings-persistence-1 6, Earnings-volatility 6, Earnings-persistence-2 6, Accruals-pos-extreme 6, Loss 6, Near-reputable-arranger 6, Near-local-arranger 6, Cross-selling-potential 6, (Continued)

17 ROLE OF BANK REPUTATION 17 TABLE 2 Continued Panel B: Reputable arranger sample Number of Distribution Loan and Firm Characteristics observations Mean SD 25% 50% 75% Loan characteristics: Loan-size (in millions) 4, Maturity (in month) 4, Interest-spread (in bps) 4, Institutional 4, Restructuring-purpose 4, Prior-relationship 4, Credit risk characteristics: Credit-rating 4, CDS-spread (in %) 1, Watch-negative (Watch-positive) 4, (0.02) Outlook-negative (Outlook-positive) 4, (0.07) Additional firm characteristics: Firm-size (in millions) 4,690 10,907 24,246 1,511 3,675 11,499 Leverage 4, ROA 4, Interest-coverage 4, Earnings-persistence-1 4, Earnings-volatility 4, Earnings-persistence-2 4, Accruals-pos-extreme 4, Loss 4, Near-reputable-arranger 4, Near-local-arranger 4, Cross-selling-potential 4, (Continued)

18 18 ROBERT M. BUSHMAN AND REGINA WITTENBERG-MOERMAN TABLE 2 Continued Panel C: Less reputable arranger sample Number of Distribution Loan and Firm Characteristics observations Mean SD 25% 50% 75% Loan characteristics: Loan-size (in millions) 1, Maturity (in month) 1, Interest-spread (in bps) 1, Institutional 1, Restructuring-purpose 1, Prior-relationship 1, Credit risk characteristics: Credit-rating 1, CDS-spread (in %) Watch-negative (Watch-positive) 1, (0.02) Outlook-negative (Outlook-positive) 1, (0.08) Additional firm characteristics: Firm-size (in millions) 1,985 5,975 28, ,358 3,806 Leverage 1, ROA 1, Interest-coverage 1, Earnings-persistence-1 1, Earnings-volatility 1, Earnings-persistence-2 1, Accruals-pos-extreme 1, Loss 1, Near-reputable-arranger 1, Near-local-arranger 1, Cross-selling-potential 1, This table provides descriptive statistics (see table 1 for the sample selection procedure). Panel A provides descriptive statistics for all facilities. Panels B and C provide descriptive statistics for the reputable arranger and less reputable arranger samples, respectively. and indicate whether a variable s mean value is significantly different for the reputable arranger sample than it is for the nonreputable arranger samples at the 1% and 5% levels, respectively. Variables are defined in appendix A.