Relationship bank behavior during borrower distress and bankruptcy

Transcription

1 Relationship bank behavior during borrower distress and bankruptcy Yan Li Anand Srinivasan March 14, 2010 ABSTRACT This paper provides a comprehensive examination of differences between relationship bank behavior for a set of borrowers that either underwent distress or filed for bankruptcy relative to normal times. Prior to distress, banks offer preferential contract terms in the form of lower interest rates and less collateral requirement to their relationship borrowers. After the onset of distress, banks offer identical loan contract terms to their relationship borrowers and outside borrowers. Further, loan availability from relationship lenders (relative to outside lenders) is significantly lower after the onset of distress. However, after filing for bankruptcy, banks again offer preferential terms to their relationship borrowers in terms of collateral requirement. Further, availability of loans from relationship lenders is comparable to that in normal times. Our findings contribute to both the literature on soft budget constraints as well as that on DIP financing. We would like to thank Yakov Amihud, Miao Bin, Edmund Cheung, Allaudeen Hameed, Michael Lemmon, Srinivasan Sankaraguruswamy and seminar participants at Korea University, Financial Management Association and CEPR-UBC-UA conference on competition in banking markets for their comments. Srinivasan thanks NUS Faculty Research Grants R and R for financial support for this project. Korea University Business School, Anam-Dong, Seongbuk-Gu, Seoul, , Korea, Tel: liyan09@korea.ac.kr Dept of Finance, NUS Business School and Risk Management Institute, National University of Singapore, 1 Business Link, BIZ1 Building 04-34, Singapore , Tel: (65) , bizas@nus.edu.sg

2 Relationship bank behavior during borrower distress and bankruptcy ABSTRACT This paper provides a comprehensive examination of differences between relationship bank behavior for a set of borrowers that either underwent distress or filed for bankruptcy relative to normal times. Prior to distress, banks offer preferential contract terms in the form of lower interest rates and less collateral requirement to their relationship borrowers. After the onset of distress, banks offer identical loan contract terms to their relationship borrowers and outside borrowers. Further, loan availability from relationship lenders (relative to outside lenders) is significantly lower after the onset of distress. However, after filing for bankruptcy, banks again offer preferential terms to their relationship borrowers in terms of collateral requirement. Further, availability of loans from relationship lenders is comparable to that in normal times. Our findings contribute to both the literature on soft budget constraints as well as that on DIP financing. 2

3 1. Introduction Most of the empirical work on relationship lending has focused on banks and borrowers when borrowers are performing well financially. 1 Little work has focused on the impact of lending relationships when borrowing firms undergo financial distress. 2 This is an important gap in the literature as the benefits of bank lending arise because of the (assumed) better screening ability or better refinancing decisions of banks relative to capital markets during borrower distress. 3 This paper seeks to fill this gap in the literature by exploring the behavior of relationship banks when their borrowers undergo financial distress and bankruptcy and comparing this to relationship bank behavior during normal times. In particular, we compare loan availability as well as loan rates and collateral to borrowers in normal conditions, distress and bankruptcy. Further, we also study the impact of maintaining relationships on the likelihood of distress and bankruptcy. On one hand, the incentive of a relationship lender to help the borrower is reduced if a borrower enters distress or bankruptcy. Theoretically, a reduced incentive may arise because a lender would want to maintain a reputation as being tough to protect its outstanding loans with other borrowers (Chemmanur and Fulghieri (1994)), because the borrower becomes more locked in during distress (Sharpe (1990), Rajan (1992)) or because of the reduced likelihood of future business from a distressed firm. 4 A reduced incentive to help may manifest in less preferential loan contract terms as well a reduced loan availability. On the other hand, the borrower may want to continue helping a distressed or bankrupt firm due to soft budget constraints. Such constraints may arise from a de- 1 See for example, Petersen and Rajan (1994) and Berger and Udell (1995). 2 See Elsas and Krahnen (1998) for evidence from Germany and Peek and Rosengren (2005) for evidence from Japan. It should be noted that a number of studies study the impact of bank debt during borrower distress. For example, Gilson, John, and Lang (1990) examine the impact of bank debt on the decision to enter a formal bankruptcy process versus informal out of court restructuring. Likewise, James (1995) examines the determinants of bank concessions in out-of-court restructurings. However, these studies do not typically study relationship bank versus outside bank finance during distress which will be a key focus of this study. 3 See, for example, Diamond (1984) and Diamond (1991). 4 Bharath, Dahiya, Saunders, and Srinivasan (2007) document that an important benefit of relationship banking from the lender s perspective is the likelihood of repeat business from the same borrower. 3

4 sire of the relationship bank to protect its reputation with other borrowing firms (Boot, Greenbaum, and Thakor (1993)), to protect its outstanding loans (Dewatripont and Maskin (1995)) and the borrower s threat of strategic default (Anderson and Sundaresan (1996)). Thus, a relationship lender may not reduce, in fact, may even increase its loan availability to its distressed borrower and may continue to offer preferential loan contract terms. While many of the above arguments are equally applicable to distress as well as bankruptcy, the provision for the super priority claims in the form of DIP financing after bankruptcy may change incentives of relationship lenders. The super priority feature may give the relationship lender additional incentives to help the relationship borrower in the DIP financing. Thus, it is interesting to contrast distress to bankruptcy to study if the possibility of making DIP loans changes relationship bank behavior. The sample for studying the above hypotheses requires the use of loan data, financial data on companies as well as bankruptcy data. We use the Dealscan database maintained by the Loan Pricing Corporation (henceforth, LPC) to obtain loan data. The LPC database is currently the most comprehensive data source on loans made to publicly traded companies and as such has been used in several papers that test the impact of relationships for large publicly traded borrowers. 5 Borrowers in the LPC database are manually matched with the merged CRSP and Compustat database. The base set of the sample consists of borrowing firms that can be matched to the CRSP and Compustat databases. We obtain the bankruptcy data information from New Generation Research s bankruptcy data. The empirical methodology uses the above borrowing firms and loan sample in two ways. First, we construct a firm year sample for all borrowers in the matched LPC- CRSP-Compustat sample to examine the impact of maintaining a strong prior relationship on the likelihood of distress or bankruptcy of the borrowing firm. Second, we use the loan sample to investigate differences in the loan contract terms (particularly focusing on the differences in the behavior of relationship lenders) when the firm is in a normal financial condition, in distress or in bankruptcy. Using both the loan sample and firm 5 See Carey, Post, and Sharpe (1998), Drucker and Puri (2005) and Sufi (2007) among others. 4

5 year sample, we construct proxies of the relative importance of relationship lending and examine the variation in these proxies as a borrowers financial condition deteriorates. Since a focal point of this paper is documenting the evolution of relationship bank behavior as the firm undergoes distress and bankruptcy, we provide brief definitions of relationships, distress and bankruptcy here. 6 For the firm year sample, relationship is computed based on beginning of the year, while for the loan sample, it is based on the date of the loan. We define a firm as having a relationship with a lender if the borrowing firm retained the bank as a lead lender in any of its prior loans in the past five years. At a given point of time, a firm and bank are said to have maintained a strong lending relationship if the firm had retained the given bank in more than 50% of its past loans. To classify a firm as being distressed in year t, we use the KMV-Merton model to compute the expected default frequency of the firm in each month. If a firm falls in the top 10% of the unconditional EDF distribution for all firm years for at least 6 months in an year, we classify the firm as distressed in that year. A loan made in a distressed year is defined as a distressed loan. A loan with a purpose of Debtor-in-possession is classified as a loan in bankruptcy. Using the firm year sample, we find that strong prior relationship does not impact the likelihood of distress or the likelihood of bankruptcy. This is also true if the likelihood of bankruptcy is examined in the distress sub-sample. This suggests that the positive wealth effects that have been associated with bank loans (James (1987),Lummer and McConnell (1989)) principally benefit equity holders of the firm. Default risk does not appear to be reduced at least in a statistically significant manner. Next, we examine the relative importance of relationship lending (loan availability from relationship banks) after the onset of distress. Using four different proxies, that are based on the presence of a relationship loan, and the relative size of the relationship loans to the total loans in a given year, we find that the relative importance of relationship lending is much lower after the onset of distress relative to normal times. The reduction in magnitude is around 6-10% which is an economically large change as well. Further, after the onset of distress, loans made by relationship banks and outside banks are indistinguishable from each other in terms of interest rate or collateral requirement. 6 More detailed definitions are provided in section 3. 5

6 In the last stage of analysis, we examine the behavior of relationship lenders after the borrower has filed for bankruptcy, again in terms of loan contract terms and loan availability from relationship banks. Using a sub sample of firms that file for bankruptcy and get DIP loans, we find that majority of DIP loans (67%) are from the relationship lenders, which is consistent with earlier results in Dahiya, John, Puri, and Ramirez (2003). Using a sample of DIP loans, we find that DIP loans from the relationship lenders require less collateral. However, the fees from relationship and outside lenders are comparable as is the case for distress. Thus, there is no benefit in terms of lower fees for relationship DIP loans. Further, the relative importance of relationship lending after filing for bankruptcy is comparable to that in normal times. Thus, for the bankruptcy sub-sample, there is a reversal of results relative to distress. In particular, relationship banks again appear to help their borrowers, at least in terms of collateral and loan availability. Finally, we test the impact of DIP loans on the likelihood of emerging from bankruptcy. Consistent with the results from Dahiya, John, Puri, and Ramirez (2003), we find that DIP loans reduce the likelihood that a firm is liquidated. One problem for the tests conducted so far is that the observed lending relationship could be endogenous. The observed pattern of lending relationship could be determined by some observable or unobservable characteristics of the borrowing firm, which in turn may affect the results that we obtain. Two methods have been commonly used to alleviate the endogeneity viz. Propensity Score Matching (PSM) and instrumental variables (IV). We find that our original results continue to hold, after controlling for potential endogeneity by both of these methods. This paper is related to several strands of literature. First, in contrast to existing literature on relationship lending, we examine the impact of relationships when it is most needed - when the borrower is in distress or bankruptcy. In contrast to the papers by Elsas and Krahnen (1998) and Peek and Rosengren (2005), who find that German and Japanese banks tend to help their borrowers in distress, we find that US banks do not appear to help their borrowers in distress. Institutional differences between the US versus Germany or Japan may account for this difference. In particular, the German status as a Hausbank probably imposes a much larger obligation on the bank than being a lead lender in the US context. Likewise, explicit government pressure on banks identified in Peek and Rosengren (2005) may play an important role in the decision to 6

7 help distressed borrowers in the Japanese context. Overall, the findings in this study suggest that soft budget constraints do not play a large role in the lending decision, at least for loans to public companies in the US. Our results are also consistent with James (1995) who finds that banks give relatively few concessions in distressed restructurings especially in the absence on large concessions by public debt holders. Second, by comparing the changes in the behavior of relationship lenders during distress and bankruptcy, we add to this literature in that DIP financing papers do not study the financing choice prior to bankruptcy. By showing that relationship banks give no benefits (price or non-price) to borrowers in distress but some non-price benefits to borrowers in bankruptcy, we are able to demonstrate that existing lenders would not lend if the loans were not super-priority as demonstrated by the results in distress prior to bankruptcy. The paper proceeds as follows. In Section 2, we develop the hypotheses that we will test in more detail. In Section 3, we describe the construction of the data set and various variables used for empirical tests. In Section 4, we present summary statistics and conduct univariate tests of the hypotheses in this paper. In Section 5, we conduct multivariate tests and analyze possible biases. In Section 6, we conclude with directions for future research. 2. Hypothesis This section develops the hypotheses to be tested in this paper. Since a focal point of this paper lies in documenting the changing nature of relationship bank behavior as the firm enters distress or bankruptcy, the hypotheses will also be developed in this order Relationships and future likelihood of distress Theoretically, Diamond (1984) and Diamond (1991) posit that banks have a better ability to screen borrowers as well as monitor borrowers. Thus, firms that have well established relationships have been screened by their banks and are subject to continual monitoring. Empirically, benefits from granting or renewal of a loan by a relationship 7

8 lender has resulted in positive certification effect (James (1987), Lummer and McConnell (1989), Billett, Flannery, and Garfinkel (1995), Puri (1996)). Further, benefits of relationship lending in the form of lower interest rates and/or less collateral have been documented in Petersen and Rajan (1994), Berger and Udell (1995) and Bharath, Dahiya, Saunders, and Srinivasan (2009). Further, the relationship lender may continually monitor the borrower, which would result in lower likelihood of risk shifting. All of these suggest that maintaining strong lending relationships in normal times is likely to lower the likelihood of value reducing actions and consequently reduce the likelihood of distress (that would have resulted from such value reducing actions). This is formalized in the following hypothesis: Hypothesis 1 (H1: Future likelihood of distress) A borrowing firm that maintains a strong prior relationship with its lender is less likely to enter distress relative to an equivalent borrower that does not maintain a strong prior relationship with its lender Loan contract terms during distress The second hypothesis involves contract terms for loans made after a firm enters distress. Even though the relationship bank does have an incentive to offer preferential terms in normal times, after a firm has entered distress, it may not have the same incentives. In particular, the time of distress is one where external financing is not easily available. This implies that there is a greater ability for the bank to hold up its relationship borrower (Sharpe (1990), Rajan (1992), Von-Thadden (2004)). Consequently, the incentive to offer preferential terms is reduced relative to normal times. 7 Secondly, the relationship bank may want to develop a reputation for being tough in dealing with distressed borrowers (Chemmanur and Fulghieri (1994)). Lastly, a reduced likelihood 7 The maintained assumption in this paper will be that in normal circumstances, borrowing firms derive some price or non-price benefits from their lending relationships. Several event studies such as James (1987) and follow on studies establish this. Direct evidence on the relative unimportance of hold up costs for publicly listed Norwegian firms is also provided in Ongena and Smith (2001). Further, Bharath, Dahiya, Saunders, and Srinivasan (2009) show that lending firms derive overall benefits from lending relationships. Lastly, publicly listed firms that are likely to be subjected to hold up costs can mitigate this by having multiple bank relationships as documented by Houston and James (1996). Of course, under some circumstances, even publicly listed firms can be subjected to hold up costs such as documented in Santos and Winton (2008) and Hale and Santos (2009). 8

9 of repeat business due to the onset of distress may reduce the incentive of the bank to given preferential terms to its relationship borrower. We will henceforth refer to all of these potential reasons that banks may not help a borrower in distress as hard budget incentives. However, the relationship bank also has incentives to help its distressed borrowers. This may arise because of soft budget constraints that make it ex-post optimal to help the distressed borrower. Such constraints may arise because the bank s desire to protect its reputation with other borrowers (Boot, Greenbaum, and Thakor (1993)), to protect its outstanding loans (Dewatripont and Maskin (1995), the borrower s threat to strategically default (Anderson and Sundaresan (1996)), governmental pressure to help distressed borrowers (Peek and Rosengren (2005)) and loan originating officers being reluctant to recognize losses (Hertzberg, Liberti, and Paravisini (2008)). Thus, the relationship lender may continue to offer the same or even more preferential terms to its borrowers in distress as in normal times, due to soft budget constraints. At worst, the relationship lender would charge the same terms as that offered by an outside lender, due to hard budget incentives. This argument is formalized in the second hypothesis. Hypothesis 2 (H2: Loan contract terms during distress) If soft budget constraints dominate, the bank will continue to give the same level or greater level of preferential treatment to its relationship borrowers in distress relative in normal times. If hard budget incentives dominate, the degree of preferential treatment given to a relationship borrower in distress will reduce relative to normal times. At the extreme, the bank may treat the relationship borrower at par with the outside borrower Loan Availability: Relative importance of relationship lending during distress The above hypothesis focuses on the contract terms offered by a relationship lender. Another dimension of relationship lending is availability of loans from a relationship lender. As discussed above, the model by Boot, Greenbaum, and Thakor (1993) implies that a relationship bank may have an incentive to continue making loans to a borrower 9

10 in distress. Outside lenders have no such incentive. This implies the relative importance of relationship lending should be greater after a firm enters distress. On the other hand, the relationship lender may also be better at screening its relationship borrowers relative to outside lenders and make the efficient liquidation or continuation decisions (Diamond (1991), Chemmanur and Fulghieri (1994)). Screening would result in the relationship bank rejecting a fraction of its inside borrowers, some of who may obtain loans from outside lenders due to the lower precision of outside lenders in evaluating the credit risk of the borrowing firm. This implies a lower relative importance of relationship lending after the onset of distress. Likewise, most of the arguments made in section 2.2 are equally applicable for the availability of relationship loans relative to outside loans. The above discussion (one of which implies lower likelihood of relationship lending after distress and one of which implies a higher likelihood of relationship lending after distress) leads to the following hypothesis. Hypothesis 3 (H3: Loan availability: Relative importance of relationship lending during distress) If soft budget constraints dominate, the relative importance of relationship lending should be comparable to or even greater in distress relative to normal times. If screening and hard budget incentives dominate, the relative importance of relationship lending should be lower in distress relative to normal times Relationships and bankruptcy While the above arguments are formulated for distress, the same arguments would be equally applicable for a firm that has filed for bankruptcy. The first hypothesis (which corresponds to H1 in the case of distress) is as follows: Hypothesis 4 (H4: Future likelihood of bankruptcy) A borrowing firm that maintains a strong prior relationship with its lender is less likely to enter bankruptcy relative to an equivalent borrowing firm that does not maintain a strong prior relationship with its lender. 10

11 However, after filing for bankruptcy, it may be the case that the incentives of relationship lenders change relative to their incentives during distress. After a firm has filed for bankruptcy, the possibility of making super-priority loans in the form of DIP financing may give relationship lenders the incentive to identify the safer borrowers using their information advantage and extend DIP loans to them. Consistent with this view, Dahiya, John, Puri, and Ramirez (2003) find that borrowers obtaining DIP loans tend to have a higher likelihood of recovering from bankruptcy. This would imply that firms in bankruptcy should get a larger discount in terms of fees and/or collateral requirement relative to distressed times. This also implies that the relative importance of relationship lending should increase in bankruptcy relative to distress. On the other hand, after a firm files for bankruptcy, the chances of recovery and the likelihood of repeat business, which is one of the important determinants of the relationship bank offering a discount may be even lower. This implies that any discount (if given) in bankruptcy should be even smaller relative to that in distress. This also implies a lower relative importance of relationship lending post bankruptcy. The above arguments lead to the next two hypotheses (which correspond to H2 and H3 in the case of distress). These hypotheses incorporate the idea that after filing for bankruptcy, in addition to soft budget constraints, the possibility of making DIP loans provides an additional incentive for relationship lenders to offer preferential terms to their borrowers and increase the relative importance of relationship lending. Hypothesis 5 (H5: Loan contract terms during bankruptcy) If soft budget constraints and the possibility of making DIP loans dominate, the bank will continue to give the same or greater level of preferential treatment in loan contract terms to its relationship borrowers relative to normal times. If hard budget incentive dominate, the degree of preferential treatment in loan contract terms to a relationship borrower will reduce relative to normal times. At the extreme, the bank may treat the relationship borrower at par with an outside borrower. Hypothesis 6 (H6: Loan availability: Relative importance of relationship lending during bankruptcy) If soft budget constraints and/or the possibility of making DIP loans dominate, the relative importance of relationship lending should be greater than or comparable to that in normal times. If screening incentives and hard budget incentives 11

12 dominate, the relative importance of relationship lending should be lower in bankruptcy relative to normal times. To summarize, our hypotheses imply that if soft budget constraints dominate in distress, there should be some net benefit of relationship lending to the borrower in terms of the loan rate and/or collateral and/or availability. If hard budget incentives dominate, then there should be no benefit to the borrower after the onset of distress. The only difference between the distress and bankruptcy is that there would be an additional incentive to lend due in bankruptcy due to the superpriority of the DIP loans which should at least increase loan availability. It may also induce the lender to share some of its informational advantage in the form of lower fees and/or collateral. 3. Data Sample Construction 3.1. Data source The data for firm year sample comes from CRSP/COMPUSTAT Merged Database and the Dealscan database maintained by the Loan Pricing Corporation (henceforth, LPC). LPC has been collecting information on loans to large U.S. corporations primarily through self-reporting by lenders, SEC filings, and its staff reporters. While the LPC database provides comprehensive information on loan contract terms (LIBOR spread, maturity, collateral, etc.), it does not provide much information on borrowers. Borrowers in the LPC database are manually matched with the merged CRSP and Compustat database, after excluding financial service companies and real estate companies. The version of data we have starts in 1986 and ends in For several empirical tests, we need accounting information from the Compustat database. To ensure that only accounting information that is publicly available at the time of a loan is used, the following procedure is adopted: For those loans made in calendar year t, if the loan activation date is 6 months or later than the fiscal year ending month in calendar year t, we use the data of that fiscal year. If the loan activation date is less than 6 months after the fiscal year ending month, the data from the fiscal year ending in calendar year t-1 is used. The accounting and stock price information are 12

13 used in the construction of the distress measures, as well as to control for firm level heterogeneity that may impact variables that we study, such as the loan rate, collateral requirement as well as the likelihood of distress and bankruptcy. A second panel data set of firm years is constructed based on the firms in the above loan sample. For each firm in the loan sample, the first available loan is identified and that year is used as the starting year for the firm to be included in the firm year sample. The rationale for such a restriction is as follows: The key focus of this study is on determining the impact of lending relationships. To determine the identity of the relationship bank or banks, we need at least one past loan taken by the borrower. Consequently, we can define the relationship measures (defined later in Section 3.3) only on or after the firm year when we have the firm s loan history. In addition, the last year when any firm can be in the firm year sample would be either 2003 or The reason for having two ending dates is as follows: For tests that link relationships to loan contract terms or the relative importance of relationship lending during distress or bankruptcy, we need to have contemporaneous measures of loan contract terms and the relative importance of relationship lending during distress or bankruptcy. This could be done on the firm year sample till On the other hand, when evaluating the impact of relationships on the future likelihood of distress or bankruptcy, we can use an additional firm year of data that ends in Of course, if a firm delists earlier than this date due to any reason including bankruptcy, the last year that the given firm will be present in the firm year sample will be the year of bankruptcy or delisting. Based on an ending date of 2004, the total sample size is firm years and a loan sample size of Accounting data for the firm year is constructed along the same lines as that for the loan sample. Lastly, a list of public firms that filed for bankruptcy is obtained from the New Generation Research s bankruptcy database. This database provides information on the dates of filing, outcome of the bankruptcy as well as the final date of emergence or liquidation. Loans made to bankrupt firms are obtained from the LPC database identified as those with primary purpose being DIP financing. 13

14 3.2. Construction of the distress measure Our principal measure of distress is based on the option pricing model developed by Merton (1974). This method is being used by the KMV corporation (a subsidiary of Moody s) and forms the basis of the market price based measures of bankruptcy prediction. 8 We elaborate on the classification of the firm year sample into distress and normal years. For each year and each month, we compute the expected default frequency (EDF, henceforth) as implied by the KMV-Merton model for all firms in the merged CRSP-Compustat database. Subsequently, for each calender year, we sum up the months where the EDF of each borrowing firm in LPC database lies in the top 10% of the unconditional EDF distribution for all firms for all years and all months. 9 If this sum is equal to or greater than 6, we classify the given firm year as one where the borrowing firm is distressed. 10 At the end of this process, each firm year when the firm has sufficient trading and accounting data available is either classified as distressed (Distress=1) or not distressed (Distress=0). Using the filing date for bankruptcy, a Bankruptcy dummy variable is constructed which takes a value of 1 if the firm filed for bankruptcy in the given year and 0 otherwise. Note that with the above definitions, a given firm year could be classified both as a bankruptcy as well as a distress year. Once a firm has filed for bankruptcy in a given year, the firm is no longer included in the sample of firm years, i.e., a bankruptcy event is included only once in the firm year sample, even though the bankruptcy process may last longer than one firm year. All loans made after the filing of bankruptcy and before the resolution of bankruptcy are considered to be loans made during bankruptcy firm year Shumway (2001), Hillegeist, Keating, Cram, and Lundstedt (2004) and Bharath and Shumway (2008) provide evidence that market based measures of financial distress provide better prediction of bankruptcy than the earlier accounting based measures such as the Altman score and the Zmijewski score. The exact methodology for computation is detailed in Appendix 1. 9 Note that the distribution of EDF s for the entire universe of CRSP-Compustat merged firms is used in computation of this percentile. 10 The results are robust to alternate values of the cut off percentile using distress defined by the top 20% and top 30%. 11 While one could easily include the actual number of bankruptcy firm years as independent observations, firm specific accounting or stock information is rarely, if ever, available after bankruptcy filing. Hence, no data analysis is possible with these additional observations. 14

15 Given the above classification of firm years, the classification of loans into normal times, distress and bankruptcy is relatively straightforward. A loan facility with starting date in a normal year is classified as a normal loan, and one made during a distress year is classified as a distressed loan. However, unlike for the firm year sample, a loan that is made in a year where the firm is in distress as well as files for bankruptcy is classified as a distressed loan unless it is explicitly classified as a DIP loan. Thus, in the loan sample, there is no overlap between the distress and bankruptcy sub-samples Construction of relationship measures Several loans in the LPC database are syndicated loans where many banks are retained in several different roles. Hence, before defining the relationship measures, it is important to identify the banks that are playing an lead role. we follow the methods used in Sufi (2007) and Bharath, Dahiya, Saunders, and Srinivasan (2009) to classify banks into the lead role. In particular, a bank is defined as playing a lead role in a given loan facility if any one of the following conditions were met. (1) The bank is given a lead arranger credit for the given loan facility or (2) the bank was retained in any of the following roles: (a) Agent, (b) Arranger, (3) Administrative Agent, (4) Lead bank, and (5) Sole lender. The rationale for this selection is that banks retained in these roles typically retained a large fraction of syndicated loans (over 25%) on average, and for the last role, the given loan is not syndicated at all. Consequently, it is reasonable to assume that banks retained in these roles are truly one of the lead lenders in the given loan facility. All measures of relationship lending are constructed only using lenders retained in a lead role as defined above. Next, we elaborate on the construction of relationship measures for the loan sample. For each loan, we have a look back period of 5 years starting on the date of the loan. A given loan is classified as a relationship loan (RELLOAN =1) if any of the lead lenders retained in the given loan facility was retained as the lead lender in any loan taken by the same borrower over the last 5 years. As an additional measure of lending relationships, we define a firm and a lender as maintaining a strong relationship if more than 50% loans (using the number of loans) in 15

16 the last 5 years came from the same bank. 12 A dummy variable (STRONGRELLOAN ) which takes a value of 1 if a strong relationship lender is retained for the current loan and 0 otherwise. Thus, the strong relationship measure can also be thought of as capturing the incremental effect of the strength of a relationship (among the set of relationship borrowers) on loan contract terms, whereas the relationship measure reflects the impact of outside versus inside lenders. For borrowers where there was no loan in the past 5 years, neither of these variables are defined. Along similar lines, we construct relationship measures for the firm year sample. For each firm year, we identify relationship lenders by searching all the previous loans (over a 5-year window excluding the current year) of that borrower as recorded in the LPC database. If at least one loan in the given year comes from a bank which has extended loans to the firm in the past five years, the given year is classified as one where a relationship lender made a loan. If none of the loans in the current year were made by any relationship lender, the given year is classified as one where relationship banks did not lend to the given firm. Based on the above classification, we construct a dummy variable (RELLOAN ) that takes a value of 1 for firm years with relationship loans and 0 for firm years with no relationship loan. One can also define years when a strong relationship lender was used (STRONGRELLOAN =1) and those where a strong relationship lender was not used (STRONGRELLOAN =0), conditional on the borrower taking at least one loan in the given year. For years where the borrowing firm did not take any loan, or years when the borrowing firm did not take any loans in the past 5 years, neither RELLOAN or STRONGRELLOAN is defined. We also construct a dummy variable STRONGPRIORELATION along the lines of Burch, Nanda, and Warther (2005). At the beginning of a firm year, if a given borrowing firm has retained any bank in more than 50% of its loans over the last 5 years, it is classified as having a strong prior relationship (STRONGPRIORELATION =1). If not, it is classified as not having a strong prior relationship. Therefore, borrowing firms that are very loyal to their lead banks will be classified as having a strong prior relationship 12 This is similar to the loyalty measure proposed by Burch, Nanda, and Warther (2005). 16

17 and borrowing firms that split their business across multiple banks will be classified as not having a strong prior relationship. 13 This measure will be critical in examining the impact of relationships on the future likelihood of distress or bankruptcy. The reason is that the firm year sample, by construction, comprise of firm years that have at least one loan in the past 5 years. Therefore, to evaluate differences in likelihood of distress or bankruptcy, we cannot use the presence or absence of a lending relationship because the entire sample by construction consists of firms that have a lending relationship with some bank. For these tests, the strong prior relationship measure as defined above can be used. 14 Construction of relationship measures is complicated by the fact that the sample period was one where several banks merged with one another. we collect data on such mergers using the SDC merger database and news searches on the bank mergers in our sample. In case of a merger, we assume that all the lending relationships of both the merging banks carry over to the new merged bank. To the extent that this procedure is classifying related banks as unrelated (for example, in case our search procedure misses the merger among two banks in our sample), we may be classifying relationship loans as non-relationship loans. This biases against finding any significant results for our measures of lending relationships, and consequently, any results obtained here are likely to underestimate the true impact of relationship lending Measures of relationship benefits to borrowing firms The hypotheses developed in Section 2 have implications for three sets of variables: (1) Interest rate charged on the loan, (2) Whether or not it is collateralized, and (3) The relative importance of relationship lending. we now elaborate on the variables used to measure each one of these. Following Drucker and Puri (2005), we use the LPC reported All-in-Spread-Drawn (hereafter Fee) as the measure of interest rate for a loan. Fee is the coupon spread over 13 As defined, the set of firm years where STRONGRELLOAN takes a value of 1 will be a subset of the observations where STRONGPRIORRELATION is More detailed explanations and examples on the construction of relationship measures for the firm year sample and loan sample are shown in Appendix 2. 17

18 LIBOR assuming the loan is fully drawn plus the annual fee. For collateral, we use a dummy variable in LPC that indicates whether or not a given loan is secured. As is evident from the definitions, both of these measures can only be defined for the loan sample. Lastly, to measure loan availability from relationship lenders, we use the relative importance of relationship lending. We construct four different proxies the relative importance of relationship lending relative to outside lending, all of which are defined for the firm year sample. The first measure is simply the RELLOAN for the given firm year. Recall from the previous subsection that this variable takes a value of 1 if the given year was one where a relationship lender extended at least one loan to the given borrower and 0 otherwise. The next three measures look at the fraction of loans and the average loan size coming from relationship banks relative to outside banks. They are defined below. Relloanratio1 : This is the ratio of the sum of loan facility amounts of all relationship loans taken by a given borrower in a given year to the sum of facility amounts of all loans taken by the same borrower in the given year. Relloanratio2 : This is the ratio of average facility amount of all relationship loans taken by a given borrower in a given year to average facility amount of all loans taken by the same borrower in the given year. Relloanratio3 : This is the ratio of number of relationship loans taken by a given borrower in a given year to the total number of loans taken by the same borrower in the given year. Similar to the relationship measures for the firm year sample, all these measures are set to be missing if the given borrower has no loans in the given year. These three ratios capture the relative importance of relationship lenders and outside lenders in terms of overall bank lending to the borrowing firm in the given year. 18

19 4. Univariate Analysis In this section, we present summary statistics on the data sample and perform univariate tests of the hypotheses developed in Section Summary statistics Table 1, Panel A reports the summary statistics for total number of observations of the firm year sample and the loan sample, where both are sub-divided by the financial condition of the firm (normal times, distress and bankruptcy). The statistics are presented for firm years till Panel B of Table 1 provides firm characteristics for firms classified as in normal times, distressed or bankrupt. The differences in firm characteristics across these three different conditions provide an independent justification for the distress measure, as it is constructed based solely on the price dynamics of the firm s stock price and the only accounting inputs used are the total assets and total debt of the firm. For example, EBITDA the coverage ratio (defined as natural log of ratio (1+ ) is 2.68 for the Interest Expenses distressed sample whereas it is during normal times. Likewise the profitability of firms in normal times is 16%, while it is 9% in distressed times. Further, firms classified as distressed using our measure also have a lower current ratio. This suggests that our measure of distress is reasonable when evaluated using the firm s accounting variables that measure firm performance or liquidity. Panel C of this table reports differences in loan characteristics (fee, collateral, maturity and size) across firms in these three sub-samples. As expected, there is a large increase in the fee as a firm goes from normal times to distress and/or bankruptcy. The mean fee during distress is 338 basis point spread, relative to a mean value of 174 basis point spread in normal times. Likewise, the percentage of collateralized loans is 43% in normal times while it is 72% in distressed times. The size of the loan and maturity also 15 As mentioned earlier, the total number of firm years will be less than the sum of firm years in normal times, distress and bankruptcy due to some overlap in firm years that are classified as both distressed and bankrupt. In particular, there are 44 firm years that are classified as both distressed and bankrupt. 19

20 decrease. Thus, loan contract terms also reflect the onset of distress. Further, in most cases, the values for fee and collateral in distress lie in between the corresponding values for normal times and bankruptcy. This provides a further confirmation that the distress measure does indeed capture distress and this is reflected in the loan terms. Table 2 presents summary statistics of the relationship measures, again stratified into normal, distressed and bankrupt conditions. From panel A, we can see that the median number of relationship banks is 1, for all three conditions, which indicates that on average, firms maintain a single lending relationship. Panel B reports relationships by firm year sample. Out of the 8382 firm years where firms take a loan, 5492 come from relationship banks. Out of these, firms get loans from strong relationship banks for 4275 firm years. In percentage terms (relative to the 8382 firm years where firms take a loan), the likelihood of a relationship loan in normal times/distress/bankruptcy is 66%/54%/72%. Thus, in distressed times, the likelihood of a relationship loan or strong relationship loan is much lower, both relative to normal times and relative to bankruptcy. In panel C, using loan facilities instead of firm years, we find a similar pattern with the likelihood of a relationship loan in normal times/distress/bankruptcy being 71%/62%/67% respectively. Both of these findings provide preliminary evidence that relationship lending is less likely in distress whereas it is more likely in bankruptcy. However, in terms of relationship loans obtained from strong relationship lenders, the pattern differs somewhat. In particular, using the loan sample, there is a monotonic decrease in the likelihood of a loan from a strong relationship bank as a firm goes from a normal condition to distress and then to bankruptcy Univariate tests In this subsection, we present univariate comparisons of the likelihood of distress, fees charged, percentage of collateralized loans, as well as the relative importance of relationship lending. These will serve as preliminary tests of the hypotheses in section 2. To test H1 (strong prior relationship helps to reduce the future likelihood of distress), we stratify the firm year sample into those where the borrowers maintained a strong prior 20

21 relationship with their bank and those where this was not the case. We compute distress probabilities for each set of firms one year subsequent to the computation of the strong prior relationship measure. Thus, if the strong prior relationship measure was computed as of the beginning of year t, the distress measure was computed as of end of year t. Table 3 Panel A presents the results of this univariate comparison. Firms that maintain a strong prior relationship in a given year have a distress probability of 8.2% in the next year while firms that did not maintain a strong prior relationship in a given year had a distress likelihood of 9.5% in the next year. 16 The difference is also significant at the 1% level providing support for H1. Similarly, the results in Table 3, Panel B suggest that firms that maintain a strong prior relationship in a given year have a significantly lower probability of filing for bankruptcy in the future. Thus, H4 that suggests that a strong prior relationship should reduce the likelihood of bankruptcy is supported. However, if the likelihood of bankruptcy is measured for the subsample of firms that are in distress (Panel C), the differences in the future likelihood of filing for bankruptcy is not significant for firms with and firms without a strong prior relationship. Thus, the above univariate tests suggest that borrowing firms that maintain a strong prior relationship have a lower likelihood of distress and bankruptcy but not a lower likelihood of bankruptcy when firms are already in distress. In table 4, we conduct univariate tests of H2 (loan contract terms in distress versus normal times) and H5 (loan contract terms during bankruptcy versus normal times). To test these hypotheses, we examine the differences in fees and percentage of collateralized loans for relationship and non-relationship loans made in normal times, distress and bankruptcy. In panel A, we use the relationship loan dummy to stratify the sample, and in Panel B, we use the strong relationship loan dummy to stratify the sample. In both cases, relationship loans have much lower fees and collateral requirement, both in normal times and in bankruptcy. The magnitudes of the differences are quite significant. For example, a relationship loan in bankruptcy has a 92 basis point lower fee relative to 16 Recall that the criterion for being included in the firm year sample is only after a given borrowing firm has taken their first loan. Therefore, all the firm year sample observations are those where the borrowing firm has at least one relationship bank. Hence, it is not possible to conduct a test of the difference in distress likelihood of lenders with lending relationships and those without one, as all borrowers in the sample have to have at least one lending relationship to enter the sample. 21

22 a non-relationship loan in bankruptcy. Likewise, the probability that a non-relationship loan in bankruptcy is collateralized is 94% whereas the probability that a relationship loan in bankruptcy is collateralized is 68%. In contrast, for the distress sub-sample, the only statistically significant difference in loan contract terms between relationship and non-relationship loans is in the collateral requirement where the difference is around 12%. However, the difference in the collateral requirement between relationship and non-relationship loans in distress is still lower than the difference in collateral requirement between relationship and non-relationship loans in normal times (around 14% difference based on Table 4, Panel A). Thus, the difference in collateral requirement between the relationship and non-relationship loans has narrowed during distress. The difference in fees is insignificant. The pattern is similar if one were to examine differences between loans made by lenders with a strong relationship and those made by lenders without a strong relationship (Table 4, Panel B). Next, in Table 5, we examine the relative importance of relationship lending after the borrowing firm enters distress or bankruptcy relative to that in normal times (H3 for distress and H6 for bankruptcy). In normal times, we find that the likelihood of a relationship loan is around 74% whereas in distress, it is around 62%, with the difference being highly significant (Table 5, Panel A). Other measures of the relative importance of relationship lending that measure the ratio of relationship loans to total loans in a given year also show a similar pattern of reduction in distress relative to normal times. The measures of the relative importance of relationship lending in bankruptcy relative to normal times shows a similar pattern (Table 5, Panel B). While measures of the relative importance of relationship lending in bankruptcy appear to be much lower than those measures in distress, if one were to compare the relative importance of relationship lending in bankruptcy for the sub-sample of distressed firms, there is no significant difference (Table 5, Panel C). The results in Tables 4 provide support for the reduced incentives for banks to help their borrower in distress in terms of less preferential loan terms. If the true reason for the reduction in help during distress (as reflected in the univariate tests) were a reduction in repeat business, then the relationship lenders should be even more reluctant to offer 22