The Effect of Banks Financial Reporting on Syndicated Loan Structures

Transcription

1 The Effect of Banks Financial Reporting on Syndicated Loan Structures Presented by Dr Scott Liao Associate Professor University of Toronto #2016/17-12 The views and opinions expressed in this working paper are those of the author(s) and not necessarily those of the School of Accountancy, Singapore Management University.

2 The Effect of Banks Financial Reporting on Syndicated Loan Structures Anne Beatty Fisher College of Business The Ohio State University 2100 Neil Avenue Columbus, OH Scott Liao* Rotman School of Management University of Toronto 105 St. George Street Toronto, ON M5S 3E Haiwen (Helen) Zhang Fisher College of Business The Ohio State University 2100 Neil Avenue Columbus, OH August 2016 Corresponding author. We would like to thank Brad Badertscher, Jeff Burks, Shane Heitzman, Chi-Chun Liu, Gus De Franco, Gord Richardson, Ewa Sletten, Jack Stecher, Kailong Wang, participants at the University of Houston Accounting Conference and 2015 FARS annual meeting, and seminar participants at SUNY Binghamton, BARC, Boston College, FDIC, National Taiwan University, Notre Dame University, University of New Mexico, the University of Toronto, McGill University, and University of Southern California.

3 The Effect of Banks Financial Reporting on Syndicated Loan Structures Abstract We examine whether lead lenders commercial and industrial loan (C&I) loss provision validity, measured as the extent of C&I provisions capturing subsequent net charge offs, affects the fraction of loans retained by lead lenders. Consistent with the argument that provision validity provides information about banks underlying screening and monitoring abilities, we first document positive associations between C&I provision validity and both ex post monitoring outcomes and cross-sectional variation in equity market reactions to borrowers loan announcements. We then find that the fraction of syndicated loans retained by lead lenders decreases with C&I provision validity but not with non-c&i provision validity, suggesting that C&I provision validity does not simply capture banks overall operation efficiency or risk taking. We further find that the importance of lead lenders C&I provision validity on syndication structures is attenuated by the existence of the borrower s credit rating, lead lenders previous syndicating relationships with participating banks, and participants previous lending relationships with the borrower, suggesting that our measure captures an information effect, rather than merely the lead lenders attributes. Our study contributes to the literature by exploring how lead lenders accounting information affects information asymmetry between lead lenders and participating banks and thereby influences syndication structures.

4 1. Introduction The growth of U.S. syndicated loan originations to more than $2 trillion annually is consistent with the diversification benefits of these multiple lender loans. Participant lenders delegation of loan screening and monitoring to lead lenders avoids the duplication of efforts by multiple lenders, but creates a potentially costly additional layer of agency problems between the participants and lead lenders (Leland and Pyle, 1997; Diamond, 1984). Asymmetric information about the lead lenders screening and monitoring efforts could give rise to adverse selection and moral hazard problems necessitating larger lead lender loan shares than required for optimal diversification. We examine the importance of the lead lenders reported accounting numbers in addressing this information asymmetry problem. Controlling for lead lenders past reputation and other non-accounting information related to lead lenders screening and monitoring ability, 1 we argue that participants may use lead lenders accounting information to assess their monitoring and screening ability, reducing the need for lenders to sub-optimally hold larger shares of the syndicated loans. 2 Specifically, we focus on lead bank s loan loss provisions because the OCC Loan Loss Handbook (2012) emphasizes that an effective loan review system and controls that identify, monitor, and manage asset quality problems in an accurate and timely manner are essential to generate high quality loan loss estimate. 3 Thus, lead lenders loan loss provision quality likely mitigates the agency problems between lead banks and participants by providing 1 Examples include lead lenders prior syndication experience and borrowers default rate. We examine whether the lead lender s accounting information quality is among the information sources participants use. We do not claim that it is the most important or only source, and acknowledge that information about borrowers creditworthiness is fundamental to lead bank-participant information asymmetry. 2 For example, participants may consider the lead banks loss recognition incentives and policy when assessing potential agency conflicts. See as an example from practitioners. 3 FDIC (2014) expresses similar opinions. 1

5 information about lead lenders screening and monitoring effectiveness, beyond the level of provisions and charge offs. 4 Our primary measure of the provision quality is the extent of loan loss provisions mapping into subsequent net charge offs (provision validity). We contend that if provision validity provides useful information in mitigating agency conflicts between lead lenders and participants, then the fraction of loans retained by lead lenders should decrease with this measure. To separately identify the information effect of the provision validity from other bank attributes such as banks overall risk and operation efficiency and to mitigate concerns of omitted correlated variable problems, we employ two identification strategies. First, we examine whether the loan share retained by a lead lender is differentially negatively associated with the provision validity of commercial and industrial (C&I) loans compared to the provision validity of non-commercial loans (non-c&i). Second, we examine how the association between the C&I provision validity and lead lender syndicate share differs based on alternative information sources. Specifically, we consider borrowers credit ratings, and prior lending relationships between participants and lead lenders, and between participants and borrowers as alternative information sources in this identification strategy. We contend that, if the provision validity measures merely capture bank-level risk or efficiency, we should observe that the fraction of loans retained by lead lenders decreases with provision validity measured using both commercial loans and noncommercial loans. In contrast, if C&I provision quality specifically captures lead lenders 4 Note that a bank s screening and monitoring ability and its asset quality or riskiness are two distinct concepts. Given the risk reward trade-off that banks face when making loans, a bank with a higher screening and monitoring ability may choose to assume higher risks to earn higher rewards even though the bank will exhibit higher levels of non-performing loans, loan loss provisions, or charge offs. 2

6 ability to screen and monitor borrowers of commercial loans (beyond omitted bank attributes), then we should find C&I provision validity to be more important than non- C&I provision validity in affecting loan shares retained by lead lenders. We further argue that the importance of the C&I provision validity in providing information to participants about lead lenders screening and monitoring ability should decline in the presence of alternative information sources. In contrast, if bank fundamental attributes drive the association between commercial provision validity and lead lender loan shares, then we would not observe such a decline. Among these alternative information sources, credit rating agencies provide additional information about borrowers that disciplines borrowers risk taking and help participant lenders evaluate the borrowers. Thus, we contend that loan participants face less information asymmetry with lead lenders about the borrowers and tend to rely less on lead lenders screening and monitoring efforts when borrowers are rated. Further, we argue that when participants and lead lenders have previous syndicating relationships then participants have less uncertainty about the lead lender s screening and monitoring effectiveness so the lead lender s accounting information is less important. Similarly, lead lender s information advantage regarding the borrower s creditworthiness should be lower when participants have previously lent to the same borrower, so we expect the importance of lead lender s accounting information to be lower. A necessary condition for C&I provision validity to provide information to participants is that it indeed reflects lead banks screening and monitoring abilities. Thus, we first examine the relation between C&I provision validity and ex post monitoring outcomes and cross-sectional variation in equity market reactions to loan announcements 3

7 made by borrowers. We find that banks future loss recovery rate for C&I loans and market reactions to loan announcements both increase with C&I provision validity, but not with non-c&i provision validity. While this validation would indicate that high C&I provision validity likely signals high screening and monitoring ability, low provision validity might reflect either low ability or a poor signal of the lead lenders abilities perhaps due to earnings management. Therefore, C&I provision validity provides useful but imperfect information about lead lenders screening and monitoring abilities. Using 4,371 facility-lender pairs found in 3,660 loan facilities in 2,946 syndicated loan packages syndicated by U.S. commercial banks from 1999 to 2010, we find results consistent with our predictions. First, we find that, holding bank risk and other bank and borrower attributes constant, the proportion of loans retained by the lead lenders decreases with lead lenders C&I provision validity but not with non-c&i provision validity, with the difference being significant. These on average findings are consistent with the notion that participating banks rely on lead lenders accounting information to infer banks abilities in monitoring and screening industrial borrowers when deciding to take part in the syndicates. The difference in the effects between C&I and non-c&i provision validity suggests that the C&I provision validity measure does not simply capture bank-specific correlated omitted variables such as overall risk or efficiency. Despite this falsification analysis, the on average findings do not distinguish between the possibilities that 1) the lead lenders provisioning validity informs participants of their underwriting and monitoring quality and 2) the mere association between provision quality and monitoring and screening ability. Our cross-sectional tests based on 4

8 alternative agency conflict-mitigating information sources help distinguish between these two possibilities in addition to addressing potential omitted correlated variables concerns. We find that the negative association between C&I provision validity and lead lender loan share is attenuated when borrowers are rated, when the participants have strong syndicating relationships with the same lead lender, and when participants have previously lent to the same borrowers. These results together suggest that the C&I provision validity is informative to the participants about lead lenders screening and monitoring ability rather than capturing the mere association between provision validity and the underlying lender ability that can be learned from other non-accounting information sources. More importantly, these results are supportive of the notion that C&I provision information is used to mitigate information asymmetry between lead lenders and participants and therefore affects the loan structure. To further address the possibility that the above findings are caused by bank or borrower heterogeneity that happens to be reflected in the partition based on alternative information sources, we employ a propensity score matching approach. Specifically, for facility-lender observations in the top quintile of C&I provision validity, we find a matched sample among the rest of the facility-lender observations with a similar propensity of having high C&I provision validity based on various lender and borrower attributes. We continue to find that 1) while lead banks C&I provision validity is negatively associated with the share of loans retained by lead banks, non-c&i provision validity is not significantly associated with lead banks loan shares and 2) the negative association with C&I provision validity is attenuated when agency conflicts between lead 5

9 banks and participants are lower, i.e., when borrowers are rated and when participants have previous relationships with the lead banks or borrowers. This study makes several contributions to the literature. Our study expands the literature associating accounting quality with debt contracting and syndication structures. This literature mostly focuses on the information asymmetry between lenders as a group and borrowers (e.g., Zhang, 2008; Beatty et al., 2008; Bharath et al., 2008). We instead focus on another important agency problem arising from asymmetric information among lenders. While limited prior research studies the information asymmetry mitigating effect of borrower accounting information on the proportion of the loan held by lead lenders (see Sufi, 2007, and Ball et al., 2008), few studies have directly examined how lead lenders financial reporting affects this information asymmetry among lenders. Our study differs from prior research by providing evidence that participating banks not only depend on borrowers accounting information in assessing differences in information between themselves and lead lenders, they also use lead lenders accounting information to evaluate lead lenders screening and monitoring effectiveness. Further, this study broadens our understanding of the economic consequences of the informativeness of banks loan loss provisions. The literature on loan loss provisions has focused on two potential roles of provisions. Provisions are likely used to convey management s private information to mitigate information asymmetry with external investors (e.g., Beaver and Engel, 1996; Wahlen, 1994) or are used opportunistically for capital or earnings management (e.g., Liu and Ryan, 2006; Collins et al., 1995; Beatty et al., 1995). Our study extends this literature by documenting that syndicate participants use the lead lenders provision validity to mitigate information problems and that 6

10 provisions have real economic effects on syndicate structures. Given the importance of banks role in providing capital to other sectors (Beatty and Liao, 2014), it is important to understand the effect of banks financial reporting on the capital provision process. Finally, using data collected from 10-K footnote disclosures, our study is among the first that separately examines loan loss provisions by loan type and documents that C&I provisions provide distinct information from non-c&i provisions for capital market participants. The rest of the paper is organized as follows. Section 2 provides background for our study and prior literature. We motivate our hypotheses in Section 3. We describe our sample and research design in Section 4. We discuss our empirical results in Section 5 and conclude in Section Background and Literature Review 2.1 Information Asymmetry in Syndicated Loans A syndicated loan involves multiple banks jointly offering funds to a borrower. The importance of syndicated loans in providing capital has increased drastically in the past several decades (Sufi, 2007). The lead lender is the bank that develops a relationship with the borrower, negotiates terms of the contract, and guarantees a loan amount. The lead lender then finds other syndicate members or participating banks to fund part of the loan (Taylor and Sansone, 2007). Lead lenders form syndications to avoid the regulatory lending restrictions and limit the exposure to individual borrowers (Simons, 1993; Ball et al., 2008). Specifically, loans to a single borrower cannot exceed 15% of a bank s capital for uncollateralized loans or 25% for collateralized loans (Ivashina, 2009; Beatty et al., 2012). Lead lenders screen the borrowers and monitor the 7

11 borrower s compliance with contractual terms on behalf of the syndicate. Lead lenders also act as administrative agents in collecting payments and renegotiating debt terms. In the process of the due diligence, lead lenders acquire public and private information about the borrower on an on-going basis and choose appropriate information to share with syndicate members (Taylor and Sansone, 2007). Participating banks face two types of information asymmetry in a syndicated loan: information asymmetry between borrowers and lenders and information asymmetry between lead lenders and syndicate members. Because of the information asymmetry between lead lenders and participating banks and because the monitoring efforts by lead lenders are not directly observable, agency problems arise when lead lenders screening and monitoring efforts are not aligned with syndicate member banks interests. Morris (2002) notes that participating banks often believe the Agent is too close to the borrower, too tolerant of the borrower s continuing inability to meet projections, and less sensitive to the needs of other syndicate participants. Ball et al. (2008) argue that the agency problems between lead lenders and participants can be separated into those that are ex ante (before contract signing) and ex post (after contract signing). Ex ante, lead lenders may have private information about the borrower, which leads to adverse selection problems and an incentive to shirk on their due diligence role in screening monitors due to a moral hazard (hidden effort) problem. Ex post, lead lenders may have an incentive to shirk on their monitoring role or to engage in self-serving activities at participants costs. These agency problems can be mitigated either by increasing lenders screening and monitoring incentives by requiring lead lenders to hold a significant 8

12 proportion of the loans (Sufi, 2007) or by reducing the information asymmetry among lenders. Prior debt contracting studies have mostly focused on how information asymmetry between borrowers and lenders as a group affects debt contracting. For example, Bharath et al. (2008) and Francis et al. (2005) consider the effect of borrower s accounting quality on information asymmetry between borrowers and lenders and thus on interest rates. Zhang (2008) and Beatty et al. (2008) examine the relation between accounting conservatism and loan terms such as interest rates and debt covenants. Further, Sufi (2007) reports that lead lenders share of loans increases with borrowers credit risk and information opacity, arguing that participants require lead lenders to have more skin in the game when the information problem between the borrower and lenders is greater. Because information asymmetry about the borrower s credit quality is the root of the agency conflicts between participants and lead lenders, the importance of borrowers accounting information in mitigating information problems among lenders has also been considered. Specifically, Ball et al. (2008) argue that participants may use borrowers accounting information to assess lead lenders screening efforts to mitigate adverse selection, a concern that privately informed lead lenders may attempt to sell them low quality loans while keeping good loans for themselves. In addition, borrowers accounting information also helps participating banks to gauge lead lenders monitoring effectiveness to overcome potential shirking by lead lenders. Fewer studies have focused on the importance of lender attributes in measuring monitoring and screening ability and in alleviating information asymmetry among lenders. Previous studies examining lenders screening and monitoring ability have used banks 9

13 credit ratings (Billet et al., 1995), ratio of loan loss provision to loans (Johnson, 1997), market share (Sufi, 2007), and reputation (Ross, 2010, and Goplan et al., 2011). 5 In addition to these information sources, we further consider how lead lenders financial reporting addresses information asymmetry between the lead lender and other syndicate members by providing information that participating banks may use to evaluate lead lenders screening and monitoring effectiveness. 2.2 Validation of Loan Loss Provisions The current accounting standards governing loan loss provision are FAS 114 and FAS 5, which provide specific guidance for loans deemed to be impaired when it is probable that not all interest and principal payments will be made as scheduled. The purpose of loan loss recognition is to reflect changes in management s expectations of future loan losses. Bank regulators require bank mangers to incorporate all relevant information and adopt and adhere to appropriate written policies. For example, FDIC (2014) emphasizes that at a minimum, these policies and procedures should ensure that the institution has an effective loan review system and controls (including an effective loan classification or credit grading system) that identify, monitor, and address asset quality problems in an accurate and timely manner and that the loan loss allowance estimate should take into consideration all available information including environmental factors such as industry, geographical, economic, and political factors. 6 The 1997 OCC Advisory Letter (OCC97-8, 1997) points out that banks must recognize losses in accordance with regulatory charge-off criteria, suggesting that 5 These studies typically validate their screening and monitoring measures using the borrower s stock price reaction to the loan announcement. 6 Our conversations with bank examiners at the FDIC confirm the idea that the quality of the loan loss provision is affected by the effectiveness of a bank s loan review and control system. 10

14 provisions should be verified by the subsequent charge-offs and recoveries. The OCC (1997) further indicates that many banks generally consider coverage of one year s losses an appropriate benchmark for most pools of loans because the probable loss on any given pool should ordinarily become apparent in that time frame, suggesting that banks tend to recognize loan loss provisions based on the estimated losses likely to materialize within a year. Specifically, collectively impaired loans governed by FAS 5 are usually charged off based on the numbers of days past due. Based on Federal Financial Institutions Examination Council s (1999) guidance on charge offs for consumer loans, charge offs practice ranges from 120 to 240 days past due depending on the type of loan. 7 In contrast, individually impaired loans governed by FAS 114 are usually charged off based on management s judgment. More importantly, the OCC (1997) argues that bankers and examiners should verify the reasonableness and accuracy of loss estimation methodologies. Back testing should be considered to evaluate accuracy of loss estimates from prior periods. The SEC (2001) provides similar guidance to banks on loan loss provision validation and documentation in SAB 102, which states: The staff believes that a registrant s loan loss allowance methodology is considered valid when it. Include(s) procedures that adjust loan loss estimation methods to reduce differences between estimated losses and actual subsequent charge-offs. The SEC further argues that to validate the reasonableness of the loan loss allowance methodology, banks should review the trends in loan volume, delinquencies, 7 For example, in Bank of America, for non-bankrupt credit card loans, real estate secured loans, and openend unsecured consumer loans are charged off no later than 180 days past due. Personal property secured loans are charged off no later than 120 days past due. 11

15 restructurings, concentrations and previous charge-off and recovery history, including an evaluation of the timeliness to record both the charge-offs and the recoveries. 8 Finally, the SEC argues that validation of banks provisions depends on internal control over the provisioning process. Consistent with this argument, Altamuro and Beatty (2010) find evidence that provision validity increases after FDICIA internal control provisions take effect, suggesting that provision validity is reflective of a bank s internal control effectiveness. 3. Hypothesis Development Syndicated loan participation relies on lead lenders who can and will provide appropriate screening and monitoring of borrowers. Because these screening and monitoring efforts are not directly observable to potential participants, both adverse selection and moral hazard problems arise that may affect the level of participation. To address these problems, potential participants may use lead lenders financial reporting and disclosure to assess their underwriting and monitoring ability among other accounting and non-accounting information. While the level of the loan loss provisions, allowances, charge offs, nonperforming loans, and other credit risk disclosures provide information about overall risk taking, separately these numbers do not capture the risk reward trade-off or the lenders screening and monitoring abilities. We argue that the validity or the quality of accounting information provides additional information important in assessing bank s screening and monitoring abilities. We focus on C&I provision validity, which relates directly to the type of loans (i.e., commercial and industrial loans) included in syndicated lending. 8 Based on a KPMG (2013) survey of banks, a majority of banks (76 percent) have performed validation by back testing their allowance methodology although some of them only conduct partial model validation. 12

16 The OCC s handbook for Allowance for Loans and Lease Losses (2012) suggests that the quality of banks loan loss provision information depends on whether the bank maintains effective systems and controls for identifying, monitoring, and addressing asset quality problems. This suggests that loan loss provisioning reflects lenders screening and monitoring effectiveness, consistent with Dou et al. (2014). The SEC also argues that a bank's loan loss provision is influenced by bank organizational structure, business environment and strategy, management style, loan portfolio attributes, loan administration procedures, and management information systems, suggesting that banks underwriting and monitoring behaviour impacts loan loss provisioning. The importance of provision or loan loss allowance information to investors is further reinforced by the OCC s statement (2012) that the SEC is concerned about the accuracy of the loan loss provision and allowance because of investors and analysts reliance on this information in assessing a bank s operation and credit risk. Finally, based on the argument that lead banks internal controls also affect their screening and monitoring effectiveness, provision validity reflective of internal control effectiveness may capture screening and monitoring ability (Altamuro and Beatty, 2010). Based on these arguments and those in Section 2.2, we contend that C&I provision validity reflects a bank s ability in screening and monitoring borrowers beyond the level of provisions and charge offs. Our first hypothesis is: H1: The validity of lead lenders C&I provision is positively associated with its screening and monitoring ability. While high provision validity likely signals high screening and monitoring ability, low provision validity may reflect either low ability or a poor signal of the lead lenders abilities. If participants are unable to assess lead lenders monitoring and screening 13

17 ability due to low quality of lead lenders provision information, they face higher information asymmetry. As a result, lead lenders may be required to hold a larger fraction of loans to ensure that lead lenders interests are aligned with participating banks interests and have adequate incentives to monitor the borrower. If C&I provision validity provides information about lead lenders monitoring and screening ability that participants use to evaluate lead lenders, then we expect that the fraction of loans retained by lead lenders decreases with C&I provision validity. To address the concern that provision validity captures omitted correlated variables including bank overall risk appetite and operation efficiency that participants may have learnt from other information sources, we employ two identification strategies and hypothesize based on these two strategies. First, we argue that while provision validity for C&I loans is reflective of lead banks ability to screen and monitor industrial borrowing firms, we do not expect that provision validity for non-c&i loans to capture the lead banks screening and monitoring of industrial borrowers. We thereby expect the relation between the fraction of loans retained by lead lenders and non-c&i provision validity to be significantly attenuated. In contrast, if the negative relation between lead lenders loan share and C&I provision validity simply captures the overall bank efficiency or risk taking, then we expect the fraction of loans retained by lead lenders to be similar for C&I provision validity and non-c&i provision validity. Our second hypothesis, which is stated in two parts, is: H2A: The lead lenders loan share decreases with lead lenders C&I provision validity. H2B: The association between the lead lenders loan share and provision validity is more negative for C&I loans than for non-c&i loans. 14

18 As our second identification strategy, we expect the importance of lead lenders C&I provision validity in addressing information asymmetry depends on the existence of alternative information sources about borrowers and lead lenders. We expect that the importance of the C&I provision validity, but not of the non-c&i provision validity in reducing information asymmetry between the lead lender and syndicate members to be lower when the participants have previous syndicating relationships with the lead lender. These participants should know more about the lead lender s ability from the prior experiences and therefore rely less on lead lenders accounting information to mitigate information asymmetry. We also expect the importance of provision information in addressing lead lender-participant information problems to be lower when participating banks have independent sources of information about the borrower. If syndicate members have acquired knowledge about the borrowers from previous lending relationships or if there are other sources of information about the borrower such as credit ratings, then participants can better assess whether the lead lender is performing screening and monitoring appropriately. Based on these arguments, our third hypothesis is: H3: Lead lenders C&I provision validity is more negatively associated with the fraction of loans retained by lead lenders when loan participants lack alternative sources of information about borrowers and lead lenders. 4. Research design 4.1 Sample We use the Loan Pricing Corporation s Dealscan database, COMPUSTAT, Bank Holding Companies quarterly Y9-C reports, and banks annual reports filed with the SEC to gather necessary information and construct our sample. We obtain comprehensive information about syndicated loan contracts from Dealscan for the period from 1999 to 15

19 In addition to the attributes of loan contracts, Dealscan also provides lenderspecific and syndication-specific information including lender names, locations, lender roles, and percentage of loans retained by each lender within the syndications. We follow Sufi (2007) by using the lead lender credit information provided by Dealscan to identify lead lenders. 10 We link the lead lender information from Dealscan with bank holding companies Y9-C reports to obtain financial reporting information for lead lenders that are also U.S. banks. 11 To ensure the link between these databases is accurate, we rely on the historical information for financial institutions provided by National Information Center to account for the bank merger and acquisition activities during our sample period. Further, we hand collect loan loss provision information by loan type from banks 10K filings with the SEC. Finally, we obtain borrower attributes from COMPUSTAT. Our final sample consists of 4,371 facility-lender pairs found in 3,660 loan facilities in 2,946 syndicated loan packages for 1,486 borrowers syndicated by 33 bank holding companies as lead lenders with all available information. 12 In our main analysis, we view each facility-lead lender pair as an observation. That is, in a loan facility with multiple lead lenders, we treat each lead lender as a separate observation. 13 In additional 9 We exclude data prior to 1999 because of the adoption of market-flex language in loan commitment letters after 1998, which allowed lead lenders to change the pricing of the loan based on investor demand (S&P, 2011), results in lead lender retained shares that better reflect lenders economic incentives. 10 That is, a bank is classified as a lead lender in our sample if its lead lender credit is Yes. 11 U.S. banks arranged 84% of total facilities for our sample period. 12 We start from 46 bank holding companies serving as lead lenders across our sample period with necessary information available from the bank regulatory reports. We hand collected provision information by loan type from the SEC 10-K filings for 43 publicly traded bank holding companies that serve as lead lenders for publicly traded U.S. borrowers for at least 5 facilities during a given year. We further require that the difference between the total disclosed loan loss allowance in the 10-K filings and that in the regulatory reports is less than 5%. After requiring all necessary control variables and lead-lag information, our final sample includes 33 bank holding companies. Using the larger sample based on regulatory provision disclosures that do not distinguish loan types produces similar results, although it does not allow us to compare C&I and non-c&i provision validity. 13 Consistent with the 31% reported in Sufi (2007), 26% of our sample facilities have more than 1 lead lender. In addition, a single package may contain multiple loan facilities. For our sample loan packages, 16

20 analyses, we pick the one lead lender retaining the largest proportion of loan shares as the main lead lender and allow each facility to have only one observation or only include the facility with the largest loan amount in each package in the regressions. Results are similar. 4.2 Research Design Screening and Monitoring Quality Our main test variable is a measure of how well a bank s current year loan loss provision predicts future net charge offs, i.e., provision validity. We measure C&I provision validity, C&I_VALID, as -1* past three-year average of the absolute value of ((C&I_NCOt+1/C&I_PROVt) - 1), where C&I_NCOt+1 is net charge off for C&I loans for year t+1 and C&I_PROVt is the provision of C&I loans for year t. We construct this validity measure based on the OCC s (1997) argument that one year is an appropriate window over which the loan loss provision reflects future losses and the SEC s (2001) argument that valid provisions reduce differences between estimated losses and actual subsequent charge-offs. In Section 5.2, we validate that the threshold of value 1 as a legitimate better cut-off value based on the ex post monitoring ability; however, we also examine alternative threshold values ranging from 0 to 1.5 as the robustness checks. 14 This parsimonious measure of provision validity differs from that in Altamuro and Beatty (2010) because their approach uses a time series of quarterly data and the loan loss provision by loan type disclosures are only available annually in the 10-K filings. Higher validity suggests that current period provisions map into future charge offs to a higher extent and that the bank assesses the credit quality of its existing loan portfolios more 80% have only one facility and 17% have two facilities. Our main analysis is at the facility - lead lender level because a lead lender may retain different portions for different facilities. 14 Please see section 5.3 for details. 17

21 accurately, reflecting the bank s effective systems and controls for identifying, monitoring, and addressing asset quality problems. Provision validity for non-c&i loans, Non-C&I_VALID, is measured similarly. It is measured as -1* past three-year average of the absolute value of (non-c&i_ncot+1/non-c&i_provt) - 1), where non-c&i_ncot+1 is net charge off for non-c&i loans for year t+1 and non-c&i_provt is the year t provision of non-c&i loans. We perform two tests to examine whether C&I_VALID, measured with a range of thresholds, captures the lead lenders monitoring and screening ability. The first test examines whether C&I_VALID is positively associated with a direct measure of lenders ex-post monitoring quality. Specifically, we examine whether C&I_VALID is associated with a measure of future loss given default. Following the approach taken by Ferguson and Stevenson (2007) and Banerjee and Canals-Cerda (2012), we use the ratio of recoveries to charge-offs to capture loss given default. Ferguson and Stevenson (2007) argue that the best monitors are able to recover the greatest proportion of previously charged off loans (controlling for bank risk measured by the ratio of past due loans to total assets). We use the following model (1) to estimate the extent to which C&I provision validity captures the ex-post monitoring quality, where we regress one-year-ahead commercial loan recovery ratio (C&I_REC_RATIO), calculated as the ratio of C&I loan recoveries measured at year t+1 to C&I charge-offs measured at year t, on C&I_VALIDt. We expect the coefficient on C&I_VALID to be positive if provision validity is informative about the banks monitoring ability. In addition, we also examine the association between the C&I loan recovery ratio on Non-C&I_VALID as a falsification 18

22 test, and expect a significantly lower coefficient on non-c&i_valid than on C&I_VALID. C&I_REC_RATIOi,t+1 = δ0 + δ1c&i_validi,t + δ2non-c&i_validi,t +δ3past_c&i_reci,t+δ4past_non-c&i_reci,t + δk Lender attributesi,t + υi,t (1) We control for the average commercial and non-commercial loan recovery ratio from the past 3 years (PAST_C&I_REC and PAST_Non-C&I_REC), lender attributes that may affect both loan recovery and provision validity, including lender size (Size_L), nonaccruing commercial loans (C&I_NONACC_L), ratio of commercial loans to total assets (C&I_LOAN), leverage (LEV_L), and profitability (ROA_L), along with year and bank fixed effects. We also include the level (C&I_NCO) and standard deviation of commercial loan net charge offs (C&I_NCO_STD) in an alternative specification to control for bank risk taking among commercial loans. SIZE _L is measured as the natural log of the lead lender s total assets at the beginning of the year. We use the ratio of commercial loans relative to total assets C&I_LOAN to control for lead lenders C&I loan concentration and the ratio of non-accrual commercial loans to total commercial loans C&I _NONACC to control for overall commercial loan quality. ROA_L is earnings before extraordinary items divided by beginning balance of total assets. Detailed definitions of these variables are provided in the Appendix. Our second test of whether C&I_VALID captures the lead lenders monitoring and screening ability relies on an indirect measure, specifically the equity markets reaction to borrowers announcements of new lending agreements. The advantage of this measure is that it has been used extensively to test lenders monitoring and screening ability, although this measure captures both the actual and perceived association between 19

23 monitoring and screening ability and our C&I_VALID measure. Specifically, we use the following model (2) to test whether loan loss provision validity captures lenders perceived or actual screening and monitoring ability. ABRETi,j= δ0 + δ1c&i_validj + δ2 Non-C&I_VALIDj+ δklender attributesj + δmborrower attributesi + δnloan attributesi + δsprior_ret +υi,j (2) where ABRET is the 2-day [0, 1] market-adjusted abnormal return around the loan announcement dates. Based on the notion that banks with effective systems in underwriting, and identifying and addressing loan problems are likely to have a higher provision quality, we expect the coefficient on C&I_VALID but not Non-C&I_VALID to be positive. In addition to lender attributes mentioned above, we also control for lead lenders average charge offs in the past three years (PAST_C&I_NCO), commercial loan loss allowance (C&I_ALLL), % change in total assets (ASSG_L), and market share of the lead arranger in syndicated loans market based on loan amount constructed following Sufi (2007) (MKT_SHARE). We also control for the standard deviation of quarterly charge offs (C&I_NCO_STD) to account for bank operating risk. The second set of control variables includes borrower attributes that the previous literature (e.g., Ball et al., 2008) finds important in affecting a borrower s information environment. Specifically, we control for borrower size (SIZE_B), leverage (LEV_B), profitability (ROA_B), and growth opportunities (MTB_B). The third set of control variables includes various loan attributes. For example, we control for the loan amount relative to the borrower s total assets (FAMT), loan maturity (MATURE), whether the facility is a term loan (TERM), whether the loan has collateral (SECURE), and the number of financial covenants (NCOV). We use the number of lenders to control for the syndicate size (NLENDER). We also include the natural log of the loan spread above 20

24 LIBOR to control for the overall credit risk of the loan (LOG_SPREAD). Finally, we control for the abnormal return before the loan active dates [-20, -1] to account for the potential information leakage (PRIOR_RET). Detailed definitions of these variables are provided in the Appendix Impact of Provision Validity on Syndication Structures After validating whether C&I provision validity reflects lead lenders screening and monitoring ability, we use the following model (3) to examine the association between lead lenders loan loss provision validity and the fraction of the loan retained by the lead lenders. SHARE_LEADi,j=δ0 + δ1c&i_validj + δ2non-c&i_validj + δklender attributesj + δmborrower attributesi + δnloan attributesi + υi,j (3) where SHARE_LEADi,j refers to the fraction of loan facility i retained by lead lender j. Based on previous research, lead lenders retain a higher proportion of loans when information asymmetries between participants and lead lenders are more severe and when lead lenders cannot credibly commit to perform due diligence because their monitoring effort is unobservable (Sufi 2007; Ball et al. 2008). Based on H2 that C&I provision validity can be used to infer lead lenders screening and monitoring ability, thereby reducing information asymmetry among lenders, we expect the coefficient on C&I_VALID to be significantly negative and to be significantly more negative than the coefficient on Non-C&I_VALID. Further, the use of this information is likely to depend on the extent of alternative sources of information about the borrower and about the lead lenders monitoring and screening ability. To conduct the cross-sectional analyses, we partition the sample based on alternative information sources including existence of borrower credit rating and various lending relationships. Consistent with H3, we expect 21

25 C&I_VALID, but not Non-C&I_VALID, to be more negative in model (3) in the absence of these alternative information sources. We predict that lead arrangers screening and monitoring ability is less important when borrowers have credit ratings both because loan participants can rely on the additional information provided in the ratings and because the rating agencies can serve as alternative monitors. We consider two attenuating lending relationship that may affect the importance of lead lender s accounting information. The first is the relationship between lead lenders and participating lenders. For each pair of lead lender and participating lender within a loan package, we count the total number of unique loan packages originated during the year before the current loan is initiated involving the two parties. 15 We then add up the number of previous pairings across all participating lenders and divide it by the number of lenders within the syndication to measure the average previous lending relationships between a lead lender and participating lenders. PART_LEAD is an indicator that equals 1 if the average lending relationship between lead lenders and participating lenders is above the sample median and 0 otherwise. Based on H3, we predict the estimated coefficient on C&I_VALID, (but not Non-C&I_VALID,) is more negative when a lead lender has fewer prior syndicating relationships with the participants (PART_LEAD = 0). The second attenuating lending relationship we investigate is between a borrower and the lending participants in a given loan. BORROWER_PART is measured as an indicator variable that equals 1 if at least one of the loan participants in the current deal has participated in loans for the same borrower during the past three years. Based on the argument that participants can reduce information asymmetry with lead lenders regarding 15 Our results are robust if we measure this relationship using information from the past two or three years. 22

26 borrower credit quality via prior lending relationships with the borrower, we expect to find C&I_VALID (but not Non-C&I_VALID) to be more negative when borrowerparticipants relationship is weak (BORROWER_PART=0) than when the borrowerparticipants relationship is strong (BORROWER_PART=1). 5. Empirical results 5.1 Descriptive Statistics Table 1 reports descriptive statistics for our main variables. We find that the sample average lead lenders retain 19.8% of loans. The average (median) ratio of future net charge off to current period provision for commercial loans is 0.93 (0.92) with a standard deviation of The observed distribution is consistent with the OCC s and the SEC s arguments that loan loss estimates should on average predict next year s realized loan losses, although there is substantial cross-bank/year variation. The average (median) value of lead lender C&I_VALID is (-0.66) with a standard deviation of On the other hand, the average (median) value of Non-C&I_VALID is (-0.45) with a standard deviation of We also find that in 61.3% of sample loans, participating banks have formed previous lending relationships with the same borrower during the past three years. We report Pearson correlations in Table 2. Consistent with the existing literature, we document that lead lenders retain lower proportions of loans when they have a high reputation as proxied by the high market share in the syndicated loan market. We do not find significant correlations between C&I_VALID and shares retained by lead lenders, although inferences should be made with controls for bank and borrower attributes. Finally, we document lead lenders retain lower shares when the borrowers are larger, and 23

27 when loans are less risky as reflected in lower spreads. The correlations among other control variables are largely consistent with the existing literature and our expectations. 5.2 Main Results Table 3 reports the first validation of the relation between C&I_VALID and our proxies for lender monitoring ability. Controlling for the past history of the recovery ratio, we find that C&I provision validity is positively associated with next year s recovery ratio, suggesting that banks with higher provision validity derived from better systems and controls for identifying and addressing loan problems are more effective in recovering future loan losses. When the C&I_VALID threshold is changed from 1 to either 0.75 or 1.25 the significance level is reduced to 10% versus 1% when the threshold is set to 1. This provides some evidence that a threshold of one better captures lender s monitoring ability. As documented in Table 3 Column (2), this result is robust after controlling for the current period commercial loan net charge offs and the standard deviation of past quarterly commercial loan net charge offs. The above finding confirms that provision validity is indeed informative of lead lenders monitoring effectiveness. In contrast, we do not find the same relation between the commercial loan recovery ratio and Non-C&I_VALID, suggesting that our C&I provision validity is not likely to capture omitted variables such as bank overall risk taking or operation efficiency. To assess the association between provision validity and loan announcement returns in Table 4, we first partition the samples based on whether the borrower is rated by S&P to provide additional identification. The average 2-day abnormal returns around loan announcements for rated borrowers are versus for unrated borrowers (untabulated). In Table 4, we find that for unrated firms C&I_VALID is positively 24

28 associated with the abnormal returns around loan announcements while Non-C&I_VALID is not associated with the abnormal returns, suggesting that banks with higher C&I provision validity are associated with higher screening and monitoring abilities that equity investors value. When the C&I VALID threshold is changed to either 0.75 or 1.25 similar significance levels are found but the coefficients on this variable becomes insignificant when the threshold is set further away from 1. This further supports using a threshold of 1 in our primary analyses. For rated firms where information problems are less serious and bank screening and monitoring is relatively less important we do not find a significant association with C&I_VALID. In addition, the coefficients on C&I_VALID are significantly different across the two subsamples. These results together suggest that equity investors perceive provision validity as an indication of banks screening and monitoring effectiveness when the information problems are the most serious. Given the validation test results suggesting that C&I_VALID is positively associated with banks monitoring and screening ability, we next examine how this variable affects loan syndicate structures. Specifically, we consider whether syndicate participants use this provision validity information to address information asymmetry with the lead lenders. In the first column of Table 5, where we do not control for bank fixed effects, we find that the fraction of loans retained by lead lenders decreases with C&I_VALID (at the 5% level). In addition, we find that, in this specification, lead lenders loan share also decreases with the overall market shares held by the lead lender (MKT_SHARE), suggesting that lead bank s reputation mitigates agency conflicts 25