The Relation between Borrower Risk and Loan Maturity in Small Business Lending

Transcription

1 The Relation between Borrower Risk and Loan Maturity in Small Business Lending KAROLIN KIRSCHENMANN and LARS NORDEN * First version: August 24, 2007; this version: December 8, 2007 ABSTRACT We empirically test a set of hypotheses on the relation between borrower risk and loan maturity in small business lending. Analyzing data on new loan approvals and renewals made by a German bank in 2005, we find a robust, significantly positive and monotonic riskmaturity relation. This relation is more pronounced in case of loans to commercially active individuals (relative to SMEs), relatively high asymmetric information, and low borrower bargaining power. Our results are consistent with both theoretical models on adverse selection and the view that relationship lenders provide liquidity insurance to certain borrowers. JEL classification: G20, G21, G32 Key words: Relationship lending; Loan maturity; Asymmetric information; Bargaining power; Liquidity insurance * Karolin Kirschenmann is Doctoral Student and Research Assistant at the Department of Economics, University of Heidelberg, Germany, karolin.kirschenmann@awi.uni-heidelberg.de. Lars Norden (corresponding author) is Assistant Professor at the Department of Banking and Finance, University of Mannheim, Germany, e- mail: norden@bank.bwl.uni-mannheim.de, and currently visiting the Finance Department, Kelley School of Business, Indiana University, 1309 East Tenth Street, Bloomington, IN 47405, USA, Phone: (812) , Fax: (812) , lnorden@indiana.edu. We are grateful to Daniel Foos, Daniel Sungyeon Kim, Eva Terberger, Gregory F. Udell, Xiaoyun Yu and participants at the Doctoral Seminar in Finance at Indiana University. We also thank the management and three anonymous employees of the bank who gratefully provided us with the data. 1

2 In financial contracting, lenders face two potential problems arising from an asymmetric distribution of information: adverse selection (hidden characteristics, uncertainty about the borrower quality) in case of new debt issues and moral hazard (hidden action, uncertainty about borrower behavior) in case of existing financing relationships. Interestingly, the relative importance of the information problems, the type of debt (bonds, loans) and the scope of financing relationships (arm s length vs. relationship lending) varies considerably across firms (e.g., Boot and Thakor (2000)). Small businesses are usually considered as informationally opaque because there is little publicly available information that can be used for screening. Since the risk of adverse selection is relatively high, most of these firms cannot access capital markets but have to rely on bank loans. In addition, especially small firms can benefit from relationship lending (e.g., Petersen and Rajan (1995)). Note that for these firms the risk of moral hazard (risk shifting) in existing bank-firm relationships is relatively less important than the issue of adverse selection because management and ownership of SMEs are frequently in one hand, the scope for accounting manipulation is restricted, as well as the organizational structure and business activity are of low complexity. In contrast, big firms are subject to a variety of disclosure rules, ranging from accounting standards to legal requirements. In addition, many of these firms are listed at stock exchanges and have been assigned a credit rating by at least one of the major rating agencies. Hence, screening these companies on the basis of public information should yield a relatively fair evaluation which allows the firms to issue marketable debt securities like bonds and commercial papers. Although important, private information about big firms seems to be of relatively lower importance in comparison to small firms. As a consequence, the dominant form of debt financing by big firms is borrowing from arm s length lenders (i.e. capital markets, lending syndicates or single arm s length bank lenders). In opposite, the risk of moral hazard associated with big firms can be considered as relatively high because of the 2

3 separation of management and ownership, accounting rules, feedback effects from capital markets as well as the high organizational and operational complexity (multiple products, multiple countries, etc.). Banks can overcome these problems by collecting private information to reduce the risk of adverse selection and restrict problems due to moral hazard by screening new and monitoring extant borrowers. In addition, banks can use debt contract terms (e.g. maturity, collateral, covenants) to deal with risks arising from asymmetric information. One important contractual element in bank lending is the maturity of a loan. The latter has a strong impact on, for example, default risk, liquidity, financial flexibility and financing costs which matter for both lenders and borrowers. In particular, although the relation between borrower risk and loan maturity has been addressed in some theoretical models (e.g., Flannery (1986), Diamond (1991)), there is very little empirical research that tests these theoretical models in the context of bank lending to small firms (e.g., Berger, Espinosa-Vega, Frame and Miller (2005)). In this paper, we analyze the role of loan maturity and its relation to borrower risk in small business lending. 1 This is interesting for several reasons. Small firms play a significant role in many economies, promoting growth, invention and employment. Moreover, as discussed beforehand, lending to small firms (in comparison to big firms) represents a particularly interesting opportunity to test theoretical models on adverse selection. In addition, note that small firms are also more risky in the sense that they exhibit considerably higher average default rates than big firms. Analyzing data on incremental financing decisions, i.e. all new loan approvals and renewals of commercial and industrial loans from a German universal bank in 2005, we find a robust, significantly positive and monotonic risk-maturity relation which is consistent with both theoretical models on adverse selection (e.g., Flannery (1986)) 1 There are two standard definitions for micro and small businesses: Either firms with less than 10 employees and annual sales of less than 1 million Euros (Institut für Mittelstandsforschung Bonn (2004)) or firms with less than 50 employees and annual sales of less than 10 million Euros (Kommission der Europäischen Gemeinschaften (2003)). 3

4 and the view that relationship lenders provide liquidity insurance to certain borrowers. The risk-maturity relation is more pronounced for loans to commercially active individuals than for loans to SMEs. Furthermore, the relation also holds for unsecured and secured loans but it is stronger for secured loans. Moreover, our analysis reveals that loans made under high asymmetric information exhibit a significantly shorter maturity and the risk-maturity relation is stronger for these loans. We also find that loans granted to borrowers with low bargaining power display a shorter maturity and the risk-maturity relation is stronger for loans to these borrowers. Based on the results from several model specifications we can conclude that our key findings are not changed by potential problems of endogeneity. Finally, repeating all analyses at the borrower level, with alternative risk measures and different regression estimation techniques confirm previous findings. Our study contributes in several ways. First, we can explain the positive risk-maturity link found for loans made in different situations by two complementary theoretical rationales that explicitly fit these situations (signaling to overcome adverse selection, relationship lending and liquidity insurance ). Second, Germany, the prototype of a bank-based financial system and the world s third largest banking sector in terms of total assets (as of year-end 2006), represents a particularly interesting case for the following reasons. Most important, German banks typically do not write covenants into the loan contract (except in syndicated lending to big firms). Therefore, the relation between borrower risk and maturity may be stronger than in the U.S. since the maturity of a loan can be interpreted as a (restrictive) substitute for covenants. In addition, the huge majority of firms are micro and small businesses that heavily depend on borrowing from banks. In particular, strong relationships with Hausbanks play an important role for credit availability and lending terms (e.g., Elsas and Krahnen (1998), Machauer and Weber (1998)). To the best of our knowledge this is the first study analyzing the relation between borrower risk and loan maturity in micro and small business lending in 4

5 Germany. 2 Third, we propose a multi-attributive index of asymmetric information, based on several empirical proxies, to study the impact on the risk-maturity relation. Fourth, we also condition the analysis on three new measures of borrower bargaining power which has not been done in any related paper. This is particularly challenging because of potential problems of endogeneity in loan contracting. Finally, we go beyond related studies by differentiating the analysis between different types of small businesses (SMEs, commercially active individuals) and loan types (new loans, renewals). The remainder of this paper is organized as follows. In Section 1 we review related theoretical and empirical literature. In Section 2 we briefly describe the data and propose a set of empirically testable hypotheses. Section 3 includes the empirical analysis and reports main findings on the relation between borrower risk and loan maturity. We also investigate the influence of collateral, asymmetric information and borrower bargaining power on the riskmaturity relation. Section 4 summarizes results from various tests of robustness. Section 5 concludes and offers suggestions for further research. 1. The risk-maturity relation in the literature In this section, we briefly outline theoretical models on the relation between borrower risk and debt maturity under asymmetric information (and the problem of adverse selection 3 ) and then turn to related empirical studies focusing on small business lending. Flannery (1986) considers a situation in which firm insiders are better informed about the project they want to carry out than the market. In such a setting, the choice of debt maturity 2 In contrast to our study, Fedorenko, Schäfer and Talavera (2007) analyze whether the loan maturity is above or below five years for a sample of medium-sized and large borrowers from two German banks (mean loan size of 13 millions Euros). They find a non-monotonic relation for corporate borrowers and a negative relation for sole proprietors. 3 As discussed above, we do not consider potential problems due to moral hazard since these are relatively less important (in comparison to adverse selection) in small business lending. Interestingly, there is evidence for a negative relation between borrower risk and loan maturity in lending to big firms (e.g., Strahan (1999)) where the scale and scope of moral hazard can be substantial. 5

6 may be used as a quality signal to the market under certain conditions. In detail, firms have two-period positive net present value (NPV) projects which they may finance either long-term with a debt maturity of two periods or short-term with one-period borrowing that has to be refinanced at an ex-ante unknown interest rate. There are good and bad projects, whereby project quality is the firms private information. However, creditors can observe project performance at the end of the first period, which gives them probabilistic information on the quality of the projects since good projects have a higher probability to increase in value than bad projects. If transactions costs to roll over debt are high enough to prevent bad firms (those with unfavorable private information) from imitating good firms (those with favorable private information), a separating equilibrium may occur with good firms borrowing short-term and rolling over debt at a relatively low interest rate and bad firms borrowing long-term at a higher rate. Bad firms choose to pay this relatively higher rate to avoid transactions costs and a very probably high interest rate when having to roll over short-term debt. Good firms, in contrast, benefit from transactions costs because they may signal their good quality and thus their low risk to the market by choosing short debt maturity. Summarizing, Flannery (1986) predicts a positive and monotonic relation between borrower risk and debt maturity. Diamond (1991) extends the previously described model by adding liquidity risk as well as a third risk category (medium risk). In this setting, the firms debt maturity choice is based on a trade-off between the preference for short-term debt due to an expected better credit rating in the future and the risk of liquidation, i. e. the inability to roll over short-term debt. This model also considers a two-period setting in which firms may finance a two-period project either short-term or long-term. However, in contrast to Flannery (1986), firms are distinguishable in the beginning and there are projects with negative NPV. In detail, there are good borrowers with favorable private information and positive NPV projects and bad borrowers with unfavorable private information and negative NPV projects. Lenders may distinguish firms in the beginning by their credit ratings, which reflect a firm s previous credit 6

7 reputation but they do not know whether firms have positive or negative NPV projects. After one period, lenders receive new non-verifiable information about borrowers and either upgrade or downgrade their credit ratings. Accordingly, the terms for refinancing short-term debt depend on this new information. As a result, low-risk good borrowers choose short-term debt because their probability of a downgrade is low and they thus can refinance at favorable terms when good news arrives. At the same time, medium-risk good borrowers prefer longterm debt at a higher interest rate as they must fear liquidation after the first period. Bad low and medium risk borrowers will imitate these strategies because otherwise they would be identified as having projects with negative NPVs and be unable to receive any financing. Finally, some high risk borrowers cannot obtain long-term finance due to their high probability of having projects with negative NPVs. However, they may get short-term debt if the lender receives sufficiently high returns from liquidation after one period in the case of bad news. Consequently, Diamond (1991) predicts a nonmonotonic, inversely U-shaped relation between borrower risk and debt maturity. While low risk and the very risky borrowers have short maturities, the medium risk borrowers choose long-term finance. Related empirical studies on the relation between borrower risk and debt maturity at SMEs are of two types. One group analyzes the maturity structure of debt at a particular point in time. For example, Scherr and Hulburt (2001) analyze data of small U.S. firms and find evidence for the nonmonotonic relation between risk and the debt maturity structure of firms. In contrast, Heyman, Deloof and Ooghe (2003) detect a negative relation between risk and the debt maturity structure of Belgian firms. Note that both studies proxy risk with an accounting measure (Altman s Z-score and an adapted version for Belgium). However, this approach does not allow for a rigorous test of Flannery (1986) because risk ratings in that model are meant to include originally private information while accounting measures typically represent public information. Moreover, theory focuses on incremental financing decisions while these 7

8 studies analyze the debt maturity structure. Therefore, we do not follow this strand of the literature. The second group of studies analyzes incremental financing decisions, i.e. the relation between borrower risk and the maturity of new loans. This approach has the advantage that contract terms are more easily identified and the problem of averaging all outstanding debt financing decisions over time and across contract types is avoided (see Dennis, Nandy and Sharpe (2000)). Unfortunately, this literature does not provide a clear picture on the relation between borrower risk and loan maturity because the studies are based on different data sets and different methodologies. For example, Berger, Espinosa-Vega, Frame and Miller (2005) analyze the risk-maturity relation by means of commercial and industrial loans granted to small U.S. firms in They consider bank risk ratings to proxy for borrower risk, which allows to jointly test the positive and monotonic versus nonmonotonic relation in one empirical model. They find evidence in favor of a positive relation between risk and loan maturity, which is in line with Flannery (1986) and partially consistent with Diamond (1991) for low and medium risk borrowers. Moreover, Ortiz-Molina and Penas (2007) analyze new credit lines of small U.S. firms. They find a negative and monotonic relation between borrower risk and loan maturity. However, they rely on an accounting measure (firm and owner delinquency) to proxy for firm risk, which is not entirely consistent with the theoretical models. Some of these empirical studies employ proxies to take into account the degree of asymmetric information between the borrower and lender (e.g. firm age, firm size, duration of the bank-firm relationship). For example, Ortiz-Molina and Penas (2007) control for firm age and firm size and detect a positive relation between these proxies for asymmetric information and loan maturity. They also investigate the impact of the duration of the bank-firm relationship on loan maturity but cannot find a statistically significant effect. Using survey data from several European countries, Hernández-Cánovas and Koëter-Kant (2006) examine the number of bank relationships and the provision of soft information to proxy asymmetric 8

9 information. They find that the number of bank relationships influences loan maturity on average but the results heavily depend on country characteristics. Finally, Berger, Espinosa- Vega, Frame and Miller (2005) test how the risk-maturity relation is influenced by different levels of asymmetric information between banks. They use the fact whether a bank utilizes Small Business Credit Scoring (SBCS) as part of its lending technology as a proxy for asymmetric information. It is assumed that banks which do not use SBCS face higher informational asymmetries in comparison to banks that use SBCS. It turns out that maturities become significantly longer for low risk borrowers when informational asymmetries are smaller (i.e. at banks using the SBCS), which is consistent with the implications of Flannery (1986) and Diamond (1991). 2. Data and hypotheses Our data set consists of all new commercial and industrial loans and renewals to borrowers from a German universal bank 4 during The total volume of these loans amounts to 86.1 million Euros which corresponds to approximately 10% of the bank s entire commercial lending portfolio. The data set includes information on the borrower risk, further borrower characteristics, and loan contract terms. We exclude all consumer loans and all observations with missing data for relevant variables. This procedure leads to a final sample of 668 loans, hereof 201 new and 65 renewed loans to SMEs, as well as 276 new and 126 renewed loans to commercially active private customers such as craftsmen or persons from liberal professions (subsequently referred to as individuals). Our sample is comparable in terms of size to Brick and Palia (2007) who analyze data on 766 credit lines from the 1993 National Survey of Small Business Finances (NSSBF). However, we do not analyze credit 4 The bank, which requests to stay anonymous, is among the largest 5% by total assets in the category of comparable banks, as defined by the Deutsche Bundesbank. The credit portfolio can be seen as representative for small business lending in Germany. 9

10 lines and loan commitments because their nominal maturity is typically either short term (e.g., 6 or 12 months) or not specified whereas the effective maturity may be relatively long since credit lines are frequently rolled over. Loan maturity MAT, the dependent variable in our analysis, is the nominal maturity of each single loan measured in months as stipulated in the loan contract. The main explanatory variable in our study is borrower risk. Thanks to the structure of our data set, we are able to measure borrower risk by means of the bank s internal credit ratings (RAT) which is consistent with the theoretical models outlined above. The bank s internal credit rating system consists of rating categories from grade 1 being the highest creditworthiness to grade 5 including borrowers in financial distress (the borrower is 90 days past due on any of his obligations, the bank has established a specific loan loss provision or the borrower has filed for bankruptcy). Taking the internal credit rating as a measure of borrower risk has several advantages. Most important, the rating is not part of the loan contract and it is assigned before the bank negotiates the loan maturity and the amount of collateral with the customer. Accordingly, the rating can be seen as a predetermined (or exogenous) variable in comparison to loan maturity. This interpretation is consistent with literature on the determinants of lending terms (e.g., Dennis, Nandy, and Sharpe (2000), using Altman s Z-Score 5 as a predetermined variable). Second, the bank includes all relevant public and private (hard and soft) information into the rating and considers this measure systematically for all loan approval decisions, loan pricing and loan loss provisioning (e.g., Machauer and Weber (1998), Treacy and Carey (2000), Grunert, Norden, and Weber (2005) for studies on internal credit rating systems of U.S. and German banks). Although rating grade 5 refers to borrowers in financial distress, we include these observations in our analysis for the following reason. The fact that these firms have been granted new loans or renewals indicates that the bank is 5 This approach is more conservative because internal ratings are based on public and private information whereas Altman s Z-Score is exclusively based on publicly available accounting variables. 10

11 engaged in a restructuring process and that it expects recovery from distress in the mediumterm. In the subsequent empirical analysis we investigate whether there is a significantly positive and monotonic relation between internal credit ratings RAT and loan maturity MAT as predicted by Flannery (1986) or, alternatively, a nonmonotonic relation as suggested by Diamond (1991). For this purpose, we state the following hypothesis: H1: The higher the borrower risk measured by the internal credit rating, the longer the maturity of the loan. Moreover, we also consider additional loan contract terms in some of our analyses as they may influence the choice of loan maturity. However, it is possible that these additional contract terms are determined simultaneously with loan maturity. They may be traded off during the negotiation process between the bank and the borrower, which, in turn, renders these variables endogenous. Alternatively, it seems plausible to assume that the loan amount is already defined at the beginning of the negotiations, which makes it a predetermined variable that may be treated exogenously. Studies examining the determinants of borrowing costs argue that collateral may also be considered as exogenous because the amount of collateral pledged is frequently negotiated before the final loan rate is set (see Berger and Udell (1995), and Degryse and Ongena (2005)). Ortiz-Molina and Penas (2007) follow these arguments in their study of the determinants of loan maturity. In addition, Berger, Espinosa- Vega, Frame and Miller (2005) also include other contract terms to check the robustness of their results. We adopt this research strategy and first examine the risk-maturity relation univariately, then turn to a multivariate analysis with borrower characteristics as control variables and finally check whether our findings are biased due to the endogeneity of some variables. To study the influence of collateral on the risk-maturity relation we use COL, measuring the collateral relative to the loan amount (the secured percentage of the loan). Note that only very few studies have explicitly considered the amount of collateral pledged by the borrower while the vast majority has distinguished between secured and unsecured loans by 11

12 means of a dummy variable. If a borrower pledges more collateral, the credit risk of the loan may be reduced considerably. The loan becomes safer for the bank. Consequently, the bank does not need anymore to consider loan maturity as a contracting device to discriminate between high and low risk borrowers. This mechanism can also be derived from the agency literature (e.g., Myers (1977), Smith and Warner (1979), Stulz and Johnson (1985)). Either a short maturity or collateral pledges help to mitigate agency problems between equity holders and debtors. Considering these arguments, we propose the following hypothesis on the role of collateral: H2: The higher the amount of collateral, the weaker the relation between borrower risk and loan maturity. In addition, we analyze the influence of asymmetric information on the risk-maturity relation. One possible proxy for asymmetric information is the age of the borrower since there may be more information available about older firms or individuals. This variable has already been used in earlier studies (e.g., Berger and Udell (1995), Blackwell and Winters (1997)). As our dataset includes the age of the firm for SMEs in contrast to the age of individual borrowers, we use AGE_LOW. This dummy variable indicates whether the age of the firm or the age of the borrower is below the respective median age in the two subsamples of SMEs and individuals. A proxy that perhaps even better captures the extent of asymmetric information is the duration of the bank-borrower relationship (DUR). The longer the bankcustomer relationship, the more information the bank can gather on its borrowers and the better it can assess their credit risk (see Boot (2000)). We also consider the variables NEAR (=1 if a borrower s domicile is within a radius of 10 kilometers around the bank s head office, see Degryse and Ongena (2005)), CHECK (=1 if the borrower has a checking account with the bank, see Mester, Nakamura, and Renault (2007), Norden and Weber (2007)), CLUSTER (=1 if the borrower is among the TOP 10% borrowers in the credit portfolio of the bank based 12

13 on the net loan exposure) and NEW (=1 if the loan is new and zero for renewals). 6 We predict that the informational asymmetry is lower in the case of borrowers (i) with a long bank relationship, (ii) with a checking account at the bank, and (iii) whose loans have been renewed. If the informational asymmetry is relatively high, it is more likely that loan maturity matters as a contracting device. Recall that this is especially important in Germany because banks do not make use of covenants in small business lending. In view of these arguments, we state the following hypothesis on the influence of asymmetric information on the riskmaturity relation: H3: The higher the informational asymmetry, the stronger the relation between borrower risk and loan maturity. Moreover, the general or deal-specific bargaining power of the bank and the borrower may be a further factor influencing the strength of the risk-maturity relation. If the borrower has a relatively low bargaining power in comparison to the bank, it is likely that there is a strong relation between borrower risk and loan maturity. In contrast, if the borrower has a relatively high bargaining power, this relation should be mitigated or disappear completely. Hence, we propose the following hypothesis: H4: The higher the borrower bargaining power, the weaker the relation between borrower risk and loan maturity. Finally, we consider the following additional loan contract terms. AMOUNT measures the loan amount of a single loan in Euros. SPREAD is the maturity adjusted credit spread for each loan. Moreover, FIX is a dummy variable that equals one if the loan s interest rate is fixed and zero if it is floating; BUL is a dummy variable indicating bullet loans; TRANS is a dummy variable which indicates whether the loan is a start up, development or special 6 The variables NEAR and CHECK exhibit a Spearman rank correlation coefficient of 0.83 (p<0.001), i.e. borrowers with a domicile near to the bank are most likely to have a checking account with the bank (and vice versa). Therefore, we consider only one of these variables (CHECK) in the subsequent multivariate tests. 13

14 purpose loan initiated under a federal development bank program (e.g., Kreditanstalt für Wiederaufbau (KfW)). BUILD is a dummy variable indicating whether the loan is used for building and construction purposes. We do not include macro-economic control variables (like the level and term structure of interest rates) because all loans were granted within the same year. Table 1 reports summary statistics. Insert Table 1 here Panel A indicates that the mean internal credit rating is The mean duration of the bank-borrower relationship DUR is 5.4 years, 52% of the loans are granted to borrowers having a checking account with the bank, 8% of the loans are granted to borrowers that are among the TOP 10% largest borrowers of the bank, and 39% of the loans in our sample are granted to SMEs. The mean loan maturity MAT amounts to 83 months (6.9 years) and the median is 55 months. Moreover, 71% are new loans. The mean loan amount AMOUNT is 129,044 Euros (maximum of 3.1 million Euros) which is close to other empirical studies on small business lending. For comparison, the mean loan size in the U.S. data set analyzed by Berger, Espinosa-Varga, Frame, and Miller (2005) is 43,580 USD (for loans < 100,000 USD) and 183,720 USD (for loans < 250,000 USD) and 21,814 Euros (maximum 1.9 million Euros) in the Belgian sample used by Degryse and Ongena (2005). On average, 46% of a loan amount is secured with collateral (the maximum is slightly above 100% because in five cases the bank has collateral that exceeds the loan amount). Since MAT, DUR, and AMOUNT exhibit a strongly skewed distribution, we take the natural logarithm of these variables in all subsequent analyses. Finally, the mean loan spread is 2.48 percentage points above the bank s same-maturity refinancing costs (while six new loans and six renewals exhibit negative 14

15 spreads). 7 Panel B reports the average loan maturity (in months), differentiated by rating, borrower type, loan type and loan size. It can be seen that the relation between borrower risk and loan maturity is positive and monotonic for the entire sample. A non-parametric Wilcoxon rank sum test shows that differences between adjacent rating grades are significantly different at the 0.01-level when comparing maturities for RAT1 vs. RAT2 and RAT3 vs. RAT4, and at the 0.10-level when comparing maturities for RAT4 vs. RAT5. Maturities between RAT2 and RAT3 are not significantly different (p-value = ). The positive and monotonic relation also holds for SMEs, individuals, new loans, small loans and big loans. Only for renewals we observe a small deviation from a positive and monotonic relation. In summary, these descriptive statistics provide a first indication that higher borrower risk is associated with longer loan maturities. 3. Empirical analysis 3.1 Basic results on the risk-maturity relation Subsequently, we analyze the empirical relation between borrower risk and loan maturity by means of multivariate regression models. We are mainly interested in answering the question whether loan maturity is an upward-sloping function of borrower risk as predicted by Flannery (1986) or a nonmonotonic function as suggested by Diamond (1991). Specifically, we proceed as follows. First, we estimate cross-sectional ordinary least squares regressions to investigate the basic link between borrower risk (measured by dummy variables RAT = {RAT2,, RAT5} for the internal credit ratings, RAT1 serves as a reference category) and the natural logarithm of loan maturity MAT (Model I). Second, we re-estimate Model I augmented by control variables for borrower characteristics. Third, we extend Model II by 7 Micro and small businesses are usually considered as non-investment grade borrowers. The corresponding average credit spreads on loans granted to borrowers with a S&P credit rating of BB (B) in 2005 are 1.84 (2.60) percentage points over Libor which is relatively close to the spreads we observe in our data set. 15

16 also including further loan contract variables (Model III). The stepwise regression model has the following specification: Ln (MAT) β + β RAT + β Borrower Characteristics + β Loan Contract Terms + ε (1) i = 0 1 i 2 i 3 i We are well aware of potential endogeneity problems and its consequences for the estimated coefficients and standard errors. However, note that Model I and II show the unbiased effects of the risk variable RAT (because we do not include other loan contract terms) while all estimates in Model III may be biased because of a potential endogeneity problem. Consequently, the specification in Model III can be seen as a robustness check of Model II. We refrain from estimating a simultaneous equation model because in the case of loan contracting it is very difficult to find adequate and reliable instrumental variables. Table 2 reports the basic results for the entire sample. Insert Table 2 here Model I clearly indicates a significantly positive and monotonic relation between borrower risk, measured by RAT, and loan maturity. Note that the coefficients of the dummy variables for all rating grades are highly significant and increase when moving to lower grades. Model II which additionally includes the borrower characteristics DUR, AGE_LOW, CHECK, CLUSTER, and SME as control variables confirms findings from Model I. Moreover, if we also include further loan contract terms, we observe that the significantly positive risk-maturity relation persists. Note that the coefficients of the control variables SME and CHECK are statistically significant in Models II and III. SMEs exhibit a shorter maturity, and, as expected, the existence of a checking account with the bank allows the borrower to obtain loans of a relatively long maturity. The strong positive influence of duration detected 16

17 in Model II disappears in Model III due to the correlation with further loan contract terms. Overall, these results on the risk-maturity relation represent clear support for Hypothesis H1 and are consistent with the theoretical model of Flannery (1986) but do not provide evidence for the model of Diamond (1991). In a next step, we deepen the previous analysis to gain insights on factors influencing the relation between borrower risk and loan maturity. For this purpose, we re-estimate the Models II and III from Table 2 on various subsamples. Specifically, we differentiate by borrower type (individuals vs. SMEs) and loan size (small vs. big). Table 3 presents the results. Insert Table 3 here Panel A reveals that there is a significantly positive relation between borrower risk and loan maturity in case of both borrower types but the effects are considerably stronger for loans to individuals (in terms of the magnitude and statistical significance of the coefficients RAT2 to RAT5 as well as in terms of the adjusted R 2 ). Moreover, the inclusion of further loan contract terms (Model II) does not substantially change the basic results from the reduced Model I in case of loans to individuals. For SMEs, we observe that there appears to be a stronger impact of endogeneity between the risk measure, borrower characteristics and loan contract terms. For example, the coefficients of RAT2 and RAT3 become highly significant in Model IV whereas they are not significant at all in Model III. This finding is not surprising since loans to firms are typically more complex financial products than loans to individuals, making effects from endogeneity more likely (that is exactly what explains the difference between Model III and IV). Most important, for the lowest rating grade (RAT5) we obtain a significantly positive coefficient in Model III. Summarizing, we find evidence for Hypothesis H1 for loans to both types of borrowers but the effects are stronger for individuals. 17

18 Panel B presents the results differentiated by loan size. We detect a significantly positive and monotonic relation between borrower risk and loan maturity for small and big loans, again supporting Hypothesis H1. However, the relation is substantially more pronounced for big loans and this is not because big loans are mainly granted to SMEs. The relative share of big loans to all loans granted to the same borrower type is higher among individuals (53%) than among SMEs (45%) which relates these findings to those of Panel A. Summarizing, differentiating the analysis by borrower type and loan size confirms the basic results obtained for the entire sample. Empirical evidence is consistent with Hypothesis H1, stating a positive and monotonic relation between borrower risk and loan maturity. 3.2 Collateral and the risk-maturity relation We now investigate how the risk-maturity relation is affected by the amount of collateral pledged by borrowers. As stated in H2, we expect the risk-maturity relation to become weaker in the presence of collateral. Recall that collateral is an important device for banks to mitigate default risk and problems arising from asymmetric information (for an excellent overview see Berger, Espinosa-Vega, Frame, and Miller (2007)). Furthermore, it represents a major determinant of the recovery rate of bank loans (see Basel Committee on Banking Supervision (2006), Grunert and Weber (2007)). In the following, we analyze the risk-maturity relation for unsecured (COL = 0%) and secured loans (COL > 0%). With respect to loan maturity, we observe that it is considerably higher for secured loans than for unsecured loans (means: vs months, medians: 61 vs. 47 months). The difference between the two groups is significantly different from zero at the 0.01-level, using a non-parametric Wilcoxon rank sum test. Moreover, the mean credit rating is 2.54 for secured loans and 2.63 for unsecured loans. Again, this difference is 18

19 significant (p-val. = 0.058). We now turn to the multivariate analysis, including control variables for borrower characteristics. 8 Table 4 summarizes the results. Insert Table 4 here The multivariate analysis yields two main findings. First, it can be seen that the riskmaturity link is weaker for unsecured loans (I) than for secured loans (II). This interpretation is based on the magnitude and statistical significance of the coefficients as well as on the explanatory power of the regression models (adjusted R 2 ). Note that maturity may be an important signaling device for low-risk borrowers since these loans exhibit relatively short maturities (see Table 1, Panel B) regardless whether the loans are unsecured or secured. 9 Moreover, re-estimating the regressions on the two subsamples for SMEs and individuals separately confirms the above results (but the effects become considerably stronger for individuals). Second, we find a significantly positive relation for both subsamples (except in case of RAT2 and RAT3 for unsecured loans). This result confirms that our basic findings from the previous section also hold if we differentiate by the amount of collateral. Nevertheless, we do not find evidence for Hypothesis H2 since the risk-maturity relation is stronger for secured loans. To further investigate the role of collateral and its relation to risk and maturity, we consider an alternative way to measure risk in the remainder of this section. We calculate the expected loss (which includes collateral) and test its relation to loan maturity. The expected loss is an important variable for credit risk management and represents the key input for loan pricing, i.e. calculating risk-based loan spreads. Moreover, it is widely considered in the new 8 We do not include further loan contract terms since they may be endogenous to both maturity and collateral. 9 Collateral may also be used as a signaling device in an environment with asymmetric information (e.g., Chan and Kanatas (1985) and Chan and Thakor (1987)). 19

20 capital adequacy regulation for banks (see Basel Committee on Banking Supervision (2006)). It is defined as the product of the probability of default, the loss given default (in our case: 100 secured fraction of a loan), and the exposure at default (in Euros). The absolute expected loss (AEL) is calculated as follows: AEL = PD x ((100 COL)/100) x AMOUNT (2) In the subsequent regressions we include the natural logarithm of the absolute expected loss. In addition, we also consider the relative expected loss per Euro loan amount. The relative expected loss (REL) is calculated as follows: REL = PD x ((100 COL)/100) (3) We are aware of the problem that collateral (as an input factor to calculate the expected loss) may be endogenous to loan maturity. However, given the practical relevance of the expected loss for banks and banking supervisors, we believe that it is useful to study the relation between this risk measure and loan maturity as well. Table 5 reports the main findings on the basic regression model. Insert Table 5 here Both models based on the expected loss (I and II) reveal a highly significant and positive link between risk and loan maturity. This is in line with our findings from Section 3.1 and with results from this Section. Interestingly, we observe the positive risk-maturity relation in both the subsamples of secured and unsecured loans. However, in contrast to unsecured loans for which signaling à la Flannery (1986) may represent a good explanation it is not clear why 20

21 signaling (by choosing short maturities) should play a role in the case of (fully) secured loans. We investigate the question whether additional factors are driving the positive risk-maturity relation in the remainder by examining the influence of asymmetric information and borrower bargaining power. 3.3 Asymmetric information and the risk-maturity relation We now go beyond previous analyses in the sense that we investigate the influence of a more fundamental issue, the extent of informational asymmetries between a bank and the borrower, on the risk-maturity relation. The main goal here is to disentangle the effects on maturity due to borrower risk (internal credit rating) and asymmetric information. Unlike Berger, Espinosa-Vega, Frame and Miller (2005) who measure differences in asymmetric information between banks we investigate how varying informational asymmetries of different borrowers of the same bank influence the loan maturity and the risk-maturity relation. According to Hypothesis H3, we expect the risk-maturity relation to be stronger the higher the information asymmetries. To test how the relation between borrower risk and loan maturity varies across high and low levels of informational asymmetry, we proceed as follows. We consider the duration of the bank-borrower relationship (DUR), the existence of a checking account (CHECK), and the type of the loan (NEW) as potential indicators for the extent of asymmetric information. Although NEW refers to the loan level and is not a borrower characteristic as the other indicators, it captures some details on the extent of informational asymmetries. In the case of a loan renewal there has already been interaction between the bank and the borrower with respect to the financing of a specific project. Moreover, cash flows from the project have been realized and the bank may draw better conclusions on the project quality. Comparing the relative importance of these variables, we believe that the duration of the bank relationship is the key variable while the other two indicators mainly matter in combination with duration. 21

22 For example, a long duration may be useful for the bank to accumulate information of the borrower (e.g., Boot (2000)), even without access to checking account information. However, the usefulness of duration may be considerably increased if the borrower has a checking account with the bank since there is theory and empirical evidence that this information is valuable for monitoring existing borrowers (e.g., Mester, Nakamura, and Renault (2007), Norden and Weber (2007)). In addition, making a new loan to a new borrower implies a higher extent of asymmetric information than making a new loan to a standing, well-known customer. We do not consider AGE_LOW in this context since the duration of bank-firm relationship is a more precise proxy. Based on this argumentation we construct a multiattributive index to differentiate between loans made under relatively high and low asymmetric information. The main advantage of this approach is that we can condition our analysis on a compact measure that is based on several input factors instead of reporting univariate relationships. Note that considering indices as a proxy for complex constructs is not uncommon in financial research (e.g., the credit rights index used by La Porta, Lopez-de Silanes, Shleifer, and Vishny (1998)). More specifically, taking into account the importance of DUR, we decide to multiply an indicator variable for short duration (SHORTDUR) with the sum of indicators for no checking accounts and new loans to obtain INDEX. 10 INDEX = SHORTDUR x [(1 CHECK) + NEW] (4) INDEX can take values of 0, 1, 2, and higher values indicate a higher extent of asymmetric information. The rank correlation between the three input variables ranges from 0.2 ((1 CHECK vs. NEW) to 0.5 (SHORTDUR vs. NEW), indicating a positive but not a perfect correlation. Multiplying by SHORTDUR implies that this variable serves as a 10 Note that INDEX is based on variables that are not included in the bank s internal credit ratings. 22

23 knockout criterion, i.e. in case of long durations (SHORTDUR = 0) we always assume a low informational asymmetry. 11 Furthermore, lacking quantitative information on the relative importance of these factors, we apply equal weights. Before turning to the multivariate test of Hypothesis H3, we calculate some descriptive statistics differentiated by INDEX. For example, we find that the loan maturity is shorter in case of loans made under high asymmetric information (means: 51 vs. 115 months, medians: 36 vs. 68 months) and that this result is highly significant (p<0.01, Wilcoxon rank sum test) for the entire sample and for rating grades 1 to 3. In addition, the Spearman rank correlation coefficient between the internal credit rating and loan maturity amounts to 0.32 for INDEX=0 and 0.47 for INDEX>0, indicating a stronger association for loans made under high asymmetric information. In a next step, we re-estimate the basic regression models from Table 2 separately on samples of loans made under relatively low (INDEX = 0) and high (INDEX > 0) informational asymmetries. Table 6 reports the regression results. Insert Table 6 here The differentiation by INDEX reveals several interesting findings. Most important, it turns out that the relation between borrower risk and loan maturity is more pronounced for loans made under high informational asymmetries, which is support for Hypothesis H3. Note that this finding is relatively strong because all coefficients for the ratings (RAT2-RAT5) are considerably higher in case of high informational asymmetries, the goodness-of-fit (R 2 ) of the regression model is more than twice as large (Model I vs. III), and including further loan contract terms does not change the conclusion. Moreover, also note that for both levels of 11 Alternatively, we have also calculated a simpler, additive index by summing up the dummy variables SHORTDUR and NEW. Conditioning the risk-maturity relation on this index leads to similar results as INDEX and does not change our conclusions. 23

24 asymmetric information we observe a significantly positive and monotonic risk-maturity relation which is consistent with previous results. In summary, we find a relatively shorter maturity and, in line with Hypothesis H3, a stronger risk-maturity relation for loans made under high asymmetric information. Considering both results jointly, we conclude that borrowers can clearly benefit from a reduction of asymmetric information. Moreover, as in case of secured loans we find evidence that signaling might not be the driving force behind the uncovered positive risk-maturity relation. In the next section, we further investigate this issue and provide an additional explanation for our findings. 3.4 Borrower bargaining power and the risk-maturity relation The outcome of the loan contracting process depends, among other factors, on the bargaining power of the bank and the borrower. As discussed above, borrowers with strong bargaining power may be able to obtain more favorable lending terms than others. Note that analyzing effects from bargaining power in the context of loan contracting is not easy and therefore relatively rare in the literature (for the decision between private and public debt, e.g. Rajan (1992); for effects due to market power in relationship lending, e.g. Petersen and Rajan (1995)). In this section, we investigate the influence of borrower bargaining power on the risk-maturity relation as a test of Hypotheses H4. Measuring bargaining power in loan contracting is highly challenging. As a starting point, we have conducted informal interviews with loan officers from the bank that provided the data and with other bankers to learn more about the subsequent steps in loan contracting. We learned that loan negotiations work generally as follows. A borrower approaches a bank, having in mind the purpose of the loan, asking for a specific loan amount and maturity (both amount and maturity should be highly dependent on the purpose of the loan). Moreover, some borrowers also reveal their preferences for the level and type of loan rates and collateral, 24