EVIDENCE OF JOINTNESS IN THE TERMS OF RELATIONSHIP LENDING

Transcription

1 EVIDENCE OF JOINTNESS IN THE TERMS OF RELATIONSHIP LENDING Ivan E. Brick, Rutgers Business School a Edward J. Kane, Boston College b Darius Palia, Rutgers Business School c Abstract This paper examines the joint impact of borrower-lender relationships on loan interest rates, collateral and fees. Using a simultaneous system of equations with well-identified instruments, we find that firms that have a shorter relationship with their lenders face a higher likelihood of posting collateral and paying higher fees. Endogenizing collateral and fees eliminates any significant correlation between contract interest rates and the length of the borrower-lender relationship, but uncovers a positive correlation between contract interest rates, collateral and fees. These results show jointness in the terms of relationship lending and suggest that simultaneous-equation bias may explain the variation in the results obtained in single-equation studies. We also find that fees are higher for fixed-rate loans, suggesting that fees help compensate the lender for the interest-rate put option customers receive from fixed-rate credit lines. a Professor of Finance and Economics at Rutgers Business School, 111 Washington Street, MEC 136, Newark, NJ , , ibrick@andromeda.rutgers.edu, b James F. Cleary Professor in Finance at Boston College, Fulton Hall 330A, Chestnut Hill, MA 02467, (617) , edward.kane@bc.edu, and c Associate Professor of Finance and Economics at Rutgers Business School, 109 Janice Levine Building, Rockafeller Road, Piscataway, NJ, 08854, , dpalia@rci.rutgers.edu. We thank N. Chiddambaram, William Greene, Kose John, Michael Long, Michael Pagano, Bob Porter, Matt Richardson, Ben Sopranzetti, Robert Whitelaw, and seminar participants at University of Connecticut, Indiana University-Purdue University at Indianapolis, NYU, and Yale University for helpful comments and discussions. All errors remain our responsibility. Corresponding author: Edward Kane. 1

2 1. Introduction Contemporary banking firms may be conceived as organizations that produce funds-management, informational, and transactional products for a targeted base of customers with whom bank managers maintain a durable relationship. This definition emphasizes the repeat-business and multi-product nature of bank-customer relationships. It also clarifies that the bargaining process that determines the explicit and implicit prices on the financial product can entail joint adjustments of product terms (Hodgman, 1963). The modern theory of financial intermediation stresses that the repeated points of contact a bank has with regular customers results in the transmission of inside information. Relationship banks can capture information about the borrower that an unrelated lender cannot (Kane and Malkiel, 1965; Diamond, 1984 and 1991; Ramkrishnan and Thakor, 1984; Fama, 1985; Lummer and McConnell, 1989; Sharpe, 1990; Rajan, 1992; Padilla and Pagano, 1997). Information flows that accrue over the relationship include: fuller knowledge of the borrower s repayment history, better understanding of the customer s managerial style, revelation of subtle strengths and weaknesses through negotiations, and other kinds of soft information. If this information is important, a relationship customer should be granted credit on more favorable terms than a borrower with the same observable prospects and financial condition. Researchers have, with mixed success, tested the narrower hypothesis that the length of a customer s relationship ( relationship length ) significantly reduces the explicit interest rate charged on its loans. Although studies conducted by Berger and Udell (1995), Harhoff and Korting (1998), and Degryse and Cayseele (2000) support this 2

3 hypothesis, research by Petersen and Rajan (1994) and Elsas and Krahnen (1998) rejects it, finding a positive but statistically insignificant association between the length of the relationship and loan interest rates. These results may be reconciled by noting that theory about relationships between financial intermediaries and their customers does not require that the length of their relationship affect the explicit loan interest rate only or at all. It suggests instead that relationship value affects the sum of a borrower s implicit and explicit loan interest rates. Implicit interest rates include, for example, additional charges or restrictions imposed on borrowers such as fees, collateral, covenanted duties, and documentation and reportingfrequency requirements. Kane and Malkiel (1965) portray the allocation of relationship benefits as a bilateral bargaining problem, so that benefits should be divided between banks and customers according to their relative bargaining power. A borrower with a longer customer relationship offers a better risk-return trade-off to the lender and therefore such a borrower deserves better terms. At the same time, borrowers may find loan terms from non-relationship banks to be more costly because these banks lack the flow of soft information that an extended relationship generates. This paper uses a simultaneous-equation model to test whether firms with a shorter relationship with their banks incur higher implicit and/or explicit loan interest rates. In particular, we examine how the borrower-lender relationship affects two implicit components of total interest cost: whether the borrower posts collateral and pays upfront fees. The theoretical literature on collateral identifies two distinct functions. When inside information is important, collateral serves as a positive signal of the insiders view of unobservable elements of borrower quality (see, e.g., Besanko and Thakor, 1987 a&b; 3

4 Chan and Kanatas, 1985; Bester, 1985). Whether or not inside information concerning borrower quality is important, a moral hazard rationale, as developed by Boot, Thakor and Udell (1991), Boot and Thakor (1994), and John, Lynch, and Puri (2002), assumes that managerial intentions are unobservable whereas the current quality of the firm is observable. In these circumstances, collateral serves as a performance bond against postloan activity -- managerial shirking and unanticipated risk-taking activities -- that would allow the borrower to shift uncompensated risk to the lender. If longer relationships build trust and lessen information asymmetries, then relationship length should be negatively associated with posting collateral, which is in fact what Berger and Udell (1994, 1995) find. However, these authors treat collateral as an exogenous regressor in their loanrate regressions, and not as a simultaneously determined one. Moreover, we know of no empirical study that models the tradeoffs that might exist between upfront fees and the use of collateral and the level of explicit contract rates. The model that we estimate treats fees, collateral, and explicit loan rates as simultaneous variables, and then examines the effect that the length of the lender-borrower relationship has on these focal variables. We find that firms with longer lending relationships are charged lower fees. As in Berger and Udell (1995), we also find that firms with longer lending relationships have a lower probability of having to post collateral. After endogenizing fees and collateral in a system of equations, we find a positive correlation between contract interest rates, collateral, and fees, but do not find a significant correlation between contract interest rates and the length of the borrower-lender relationship.. These results indicate jointness in the terms of relationship lending and the existence of simultaneous-equation bias in 4

5 single-equation studies. We also find that fees are higher for fixed-rate loans, suggesting that fees help compensate the lender for the interest-rate put option customers receive from fixed-rate credit lines. The evidence supports the dominance of the moral-hazard rationale for collateral, in that a positive relationship emerges between collateral use and the riskiness of the firm. We introduce a new proxy to measure the bank s confidence in managers character and ability: a dummy variable that is set to unity when either the principal owner or the firm itself has previously defaulted. This variable proves more strongly related to the probability of having to post collateral than the proxies others have used (for example, CEO ownership, etc). Borrowers and lenders might prefer jointness in relationship lending for two reasons. First, lenders might insist that borrowers with a shorter relationship length pay implicit interest upfront as a way to reduce their interim loss exposure. The lower are the periodic interest payments, the lower the probability that bank might have to declare a prematurity default. 1 Second, it might be legally and reputationally advantageous for lending institutions to mask their efforts to shade interest rates to relationship borrowers by exchanging value in implicit ways those other customers, regulators, and litigious parties cannot easily observe. The paper proceeds as follows. Section 1 describes the data, the simultaneousequation model that is estimated, and the various proxy and control variables that the tests employ. Section 2 presents and analyzes the empirical findings. Finally, Section 3 summarizes our results. 1 To test these possibilities, one would require a time series of loan transactions for a specific borrowerlender relationship. This study and the extant empirical literature do not have access to such data. 5

6 2. Explanation of Data Sources, Estimation Methods, and Included Variables Data on borrowers and loan terms are drawn from the 1993 and 1998 National Survey of Business Finances, a cooperative effort of the Board of Governors of the Federal Reserve System and the Small Business Administration. We follow Berger and Udell (1995) in restricting the sample to firms who negotiate a credit line 2 : a total of 1109 firms from the 1993 Survey and 258 firms from the 1998 Survey. After deleting observations with missing data, our final sample consists of 1125 firms. Three variables are chosen to control for differences in the economic environment at the time the line was opened: the prime rate, the slope of the term structure, and a default spread. These variables were downloaded from the Federal Reserve Bank of St. Louis website ( 2.1 Limitation of Survey Data One observes that the two surveys yield different sample sizes. This is because the 1993 Survey includes all lines of credit, namely, whether they are new lines of credit or the renewal of an existing line, and the 1998 Survey includes new lines of credit only. Unfortunately, the 1993 Survey does not provide data as to which lines of credit are loan renewals and which are new loan initiations. While James (1987) found that all bank loans generate positive abnormal returns, Lummer and McConell (1989) found that only loan renewals generate positive abnormal returns and loan initiations do not. However, Slovin, Johnson, and Glascock (1992) show that differentiating between loan initiations and loan renewals is unnecessary, because both type of loans generate abnormal returns only in the case of small firms and not in the case of large firms. Additionally, the Survey 2 Berger and Udell justify focusing on credit lines on the ground that this excludes from the data set most loans that are transaction driven rather than relationship driven (1995, p.353). 6

7 provides only one year of borrower characteristics. This limits our ability to test hypotheses described in footnotes 2 and 9. Finally, Kayshyap, Stein and Wilcox (1993), Kayshyap and Stein (1997), and Hubbard, Kuttner, and Palia (2002) among others argue that characteristics of banks have an independent impact on loan contract rates. Although it would be desirable to have bank characteristic variables in our study, the Survey data do not provide the name of the lender. This also limits us in examining the impact of financial institution mergers on small business lending as documented by Berger, et.al. (1998) and Sapienza (2002). 2.2 Regression Model Our research compares single-equation estimates of the sensitivity of explicit loan rates to relationship benefits with the sensitivity of simultaneous-equation ones. We begin by regressing each borrower s contract Loan Rate on a collection of potential regressors whose relevance is supported by prior studies. We then interpose simultaneous equations for Collateral and Fees. The specifications we estimate differentiate between a vector of common control variables X and a vector of specific instruments Z for each endogenous variable. The instruments are chosen to satisfy rank and order conditions that identify the following system: Loan Rate = α LR + β LR Collateral + γ LR Fees + Ω LR X + λ LR Z LR + ε LR (1) Fees = α PF + β PF Collateral + γ PF Loan Rate + Ω PF X + λ PF Z PF + ε PF (2) Collateral = α CO +β CO Fees +γ CO Loan Rate + Ω CO X + λ CO Z CO + ε CO. (3) 2.3 Definitions and Estimators Used for Endogenous Variables Loan Rate is the contractual coupon interest set for the credit line. Fees is stated as the ratio of fees collected by the lending institution to the total amount that can be borrowed. Collateral is a binary dummy variable that equals one if collateral is required and is zero 7

8 otherwise. Table 1 lists these measures along with the control and instrumental variables we employ. 2.4 Table 1 Control variables are chosen to meld the explanatory variables used by Berger and Udell (1995) with those used by Petersen and Rajan (1994, 1995). 3 While the Loan Rate regression can be estimated by the usual least-squares techniques, the other two variables require the use of limited dependent-variable methods. It turns out the 527 of the 1,125 borrowing firms (almost half of the sample) paid zero Fees. The large number of datapoints concentrated at zero and the lack of negative observations for Fees call for Tobit estimation. As a binary variable, the range of Collateral is restricted to only two values and is therefore estimated by the logistic regression. 2.5 Definition of Control Variables Control variables cover four broad dimensions of the negotiation process: borrower characteristics, loan characteristics, macroeconomic variables, and observation year. Borrower characteristics are measures of the profitability and risk exposure of the firm: a) Debt (i.e., leverage), defined as the ratio of outstanding debt to annual sales; b) Profit, defined as the ratio of earnings before interest and taxes to annual sales; c) Cash (i.e., liquidity), defined as the ratio of cash holdings to the annual sales of the firm; d) Size, defined as the logarithm of annual sales; e) Company, which equals unity if the firm s owners enjoy limited-liability protection and equals zero if the firm is a sole 3 Although Berger and Udell test a broad range of potential determinants for borrowers loan rates, they exclude two macroeconomic variables that Petersen and Rajan find to be significant: macroeconomic interest-rate variables and bank concentration ratios. Petersen and Rajan examine a wider variety of loan arrangements than either Berger and Udell or us, but their models suppress the role of collateral and fees. 8

9 proprietorship or partnership; 4 f) Number, which proxies borrower bargaining power by the number of lending sources available to the firm at the contracting date; g) Relation, which measures the number of years the firm has dealt with the lending institution; and h) industry dummies which proxy industry characteristics by the borrower s two-digit SIC code. 5 Loan Rate should increase with characteristics that increase the bank s exposure to loss and decrease with characteristics that increase the firm s bargaining power. On these grounds, we expect the Loan Rate equation to show positive coefficients for Debt and Company and negative coefficients for Profit, Cash, Size, Relation, and Number. Two loan characteristics and one lender characteristic are also examined. Comp Bal takes on the value of one when the credit line must be supported by a compensating balance and is zero otherwise. Maturity expresses how long the credit line remains open. To consider whether deposit institutions might pass economies in monitoring or safetynet subsidies through to their borrowers, we also include a dummy variable, Bank, that equals unity of the lender is a commercial bank or savings institution, and is zero otherwise. Macroeconomic variables represent economic conditions in the month and year the deal was completed. Three financial macroeconomic variables are employed: a) Prime, the prime rate; b) Term, the yield spread between the five-year Treasury note and the yield of a monthly three-month Treasury Bill; c) Default, the spread between the yields of Baa and Aaa bonds. We expect cyclical forces to affect both the risk-free rate 4 For example, Company is set to unity for firms that are classified as an S-Corporation, C-Corporation, Limited Liability Company or Limited Liability Partnership. 5 To conserve space, we do not report coefficients for the industry controls. 9

10 and market wide risk premiums. Consequently, we expect the Loan Rate to increase with all three variables. A fourth macroeconomic control represents the survey date: a dummy variable, Year, which is set to unity for observations from the 1998 survey, and is zero for observations obtained from the 1993 Survey. Although all borrower, lender, and loan characteristics are potentially endogenous, the number of endogenous variables one can handle is constrained by the difficulties of identifying a larger system. Because the case for endogenizing maturity is particularly strong (see, for example, Brick and Ravid 1985, 1991, and Barclay, Marx and Smith 2002), it can nicely illustrate the problem. It is difficult to conceive of an instrument for loan maturity that would not also be related to Loan Rate, Collateral, and Fees. As a robustness test, we show that our estimates of the impact of other variables do not change significantly when Maturity is moved in and out of each equation. 2.6 Instrumental Variables Identification is achieved by drawing on prior knowledge about particular model coefficients or about the covariance matrix of the error terms. Ideally, the instrumental variables inserted into any equation should be correlated with that equation s endogenous variable, but uncorrelated with the error term (Bowden and Turkington 1984). To better compare our results with the existing literature that focuses on the impact of lending relationship on loan rate, our main concerns are to identify and obtain consistent estimates of the Loan Rate equation (1). As a result, we introduce a fairly large number of instruments (see Chao and Swanson 2003). We also use an F-test to make sure that the residuals from the Loan Rate equation are uncorrelated with the vector of instrumental variables we deploy. 10

11 The instrumental variables inserted into the Loan Rate equation are measures of market power and the age of the borrowing firm. The importance of these variables is emphasized by Petersen and Rajan (1995). They argue that bankers in concentrated markets use their explicit loan rate as a loss leader to secure long-term rents on other relationship business. Petersen and Rajan believe that this strategy of loss-leadership is directed particularly at young firms. In estimating our models, Firm Age is measured in years and we proxy market power by a zero-one indicator variable, HHI Conc. HHI Conc equals one if the Herfindahl index for deposits in the borrowers Standard Metropolitan Statistical Area exceeds If Loan Rate functions more strongly as a loss leader in concentrated markets, HHI Conc should receive a negative sign. If a firm s bargaining power increases with Firm Age, relationship rents would decline with Firm Age so that it could receive either a positive or a negative sign. However, if loss leadership is directed especially at young firms, the interacted variable, Inter 1 defined as the product of Firm Age and HHI Conc should receive a positive sign. Financial theory suggests that the required return lenders demand from borrowers should be positively related to the loan s default risk. However, borrowers can compensate lenders for risk by various combinations of loan rates and lending fees. For a given borrower risk class, if fees and rates are perfect substitutes, the contract interest rate should fall with loan fees. 6 On the other hand, loan fees and contract rates might be positively related because fees can better compensate lenders for the refunding risk embedded in the prepayment option conveyed by a fixed-rate line of credit. Two 6 Bruekner (1994), Dunn and McConnell (1981), Dunn and Spatt (1985), and Leroy (1996) demonstrate theoretically that points can be used by banks to differentiate those mortgage borrowers with high probabilities of prepayment because they are more mobile from those borrowers with low probabilities of prepayment. Borrowers who tend not to prepay would, in equilibrium, select low interest rate and high loan fee contracts. 11

12 possible explanations underlie this positive relationship. The first relates to possible changes in macroeconomic interest rates. As interest rates fall, borrowers have the incentive to refund the loan. The lender can then be forced to reprice its existing loan at a lower contract rate. Upfront fees more reliably compensate the lending institution for refunding risk because for a given coupon rate and fee, the ex post return the fee generates is inversely related to the amount of time the loan remains outstanding. 7 Under this argument, one would expect to find no relationship between borrower credit quality and the probability of borrowing at fixed rates. The second reason is that low-quality firms might prefer to borrow at fixed interest rates because the impact of a credit-rating change on their contract rate is greater than that for high-quality firms. More specifically, low-quality firms that borrow at a fixed rate buy protection against the impact of future credit-risk deterioration on their interest expense. If their credit risk improves, they can exercise the implicit embedded option to refinance. Under this argument, one expects to find a positive relationship between borrower credit quality and the probability of borrowing at fixed rates. 8 Consequently, for the Fees equation (2), we use the instrumental variable, Fixed, a dummy variable that equals unity if the line of credit has a fixed coupon rate, and is zero otherwise. As we have explained, fees and loan rates might be substitutes or complements. If fees were simply a substitute for contract interest rates, one would Borrowers who tend to prepay would select low loan fees and high interest rate contracts. Note that these papers argue for substitution that is unrelated to the default risk factor of the firm. 7 As shown in Table 2, the average (median) maturity for the line of credit is 2 years (1 year). But a significant number of our sample firms have a line of credit greater than three years. Thus, refunding risk is definitely present in the longer-term contracts. However, it should be noted that refunding risk is still present for short-term contracts (one year or less) since firms can always switch to other lenders should they get better rates. 8 We ran a logistic regression of Fixed against all of our variables. (Results are available from the authors upon request.) Only the size of the firms proves significant and it is negative. This suggests that small firms 12

13 expect no relationship between fees charged and whether the loan is fixed, and a negative relationship between fees and contract interest rates. If fees are complements to contract rates because fees are used to compensate lenders for refunding risk, one would expect a positive relationship between fees charged and whether the loan is fixed, and a positive relationship between fees and the loan interest rates. Refunding risk is greater in a fixedrate loan than in a variable-rate loan and the probability of refunding increases with the contract rate. Whatever other functions upfront fees might perform, they can assure a return when future interest rates decline. Because such declines can render a fixed rate credit line too expensive to draw down, it is reasonable for lenders to charge fixed-rate borrowers for the interest-rate option conveyed to them. The previous literature on Collateral indicates that it serves two conceptually distinct functions. On the one hand, collateral may serve to mitigate asymmetric information about borrower quality. In this case collateral serves as a positive signal about borrower quality that is known to firm managers, but is unobservable to lenders (see, e.g., Besanko and Thakor, 1987 a&b; Chan and Kanatas, 1985; Bester, 1985). On the other hand, collateral can also mitigate against moral hazard, in that it serves as a performance bond against post-loan managerial shirking and risk taking activities that might shift uncompensated risk to the lender (see, e.g., Boot, Thakor and Udell, 1991; Boot and Thakor 1994; John, Lynch, and Puri 2002). To account for both functions, we introduce five instrumental variables into the Collateral equation, (3). These variables are: a) CEO age; b) CEO share of the ownership of the firm; c) CEO exp, measuring the number of years the CEO has headed (who tend to be riskier) tend to borrow at fixed rates. However, to test this argument sharply would require time series data on the borrower s financial history which is unavailable from the Surveys. 13

14 the firm; d) Delinq, a zero-one dummy that equals unity if either the principal owner or the firm has previously defaulted; and e) NFA, the ratio of long-term tangible assets to sales. We suppose that the bonding element loses importance to the lender for older and more-experienced CEOs and gains importance if either the firm or its principal owner has a history of default. Finally, we suppose that the signaling element becomes less important to the lender, the more importantly that independently assessable tangible assets figure into the borrower s assets. On the other hand, the ease of posting and monitoring collateral might increase with NFA as well. Table 2 presents summary statistics for the variables included in this study. Borrower firms paid a mean explicit interest rate of 8.46 percent and a median rate of 8.0 percent. Loan fees increased the effective rate by a mean of 68 basis points, but the median effect is only 3 basis points. This is because only 598 of the 1,125 sample borrowers paid fees at all. At these firms, the net contribution by fees averaged 129 basis points, with a median of 50 basis points. Fully 60 percent of sample lines included collateral. Borrower sales averaged $8.84 million, while median sales were only $2.88 million. Debt averaged 36.1 percent of sales. Profits averaged 5.3 percent of sales, but the median ratio was only 3.2 percent. Firms held an average of 7.6 percent of their sales in cash. The mean value of net fixed tangible long-term assets was percent of sales, but the median value was a very much smaller 8.62 percent. 2.7 Table 2 Across sample dates, the mean prime rate was 6.9 percent, while the median rate was a slightly lower 6.45 percent. The mean yield spread (Term) between the five-year Treasury note and the monthly Treasury bill averaged 2.38 percent, with a median value 14

15 of 2.71 percent. The yield difference between the Baa and Aaa bonds (interpreted as a market wide default premium) averaged 0.71 percent. Relationship length (Relation) showed a mean of 7.6 years and a median of five years. The average age of the firms is years. 9 On average, 92 percent of the lines of credit were written by deposit institutions and 10.5 percent of the lines carried a compensating balance. The mean maturity of sample lines was 24 months, with a median maturity of only one year. Although 40.4 percent of the firms borrowed from institutions in highly concentrated markets, the median firm borrowed in a competitive market. The mean age of sample firms is 16.5 years. Nearly 29 percent of the lines carried a fixed interest rate, which implies that the median firm accepted a variable-rate contract. Firms averaged only 1.15 sources of lines of credit and the median firm showed only one source. Mean CEO is 50 years old, with 20.1 years in experience. The mean (median) share of equity owned by the CEO was 65.8 percent (58 percent) and approximately 21.4 percent of the sample showed a record of prior default. 3. Empirical Evidence of the Interaction of Implicit and Explicit Loan Terms As a benchmark against which to evaluate simultaneous-equation results, Table 3 reports single-equation estimates of two models for the Loan Rate. The models differ only in whether or not Maturity is included among the regressors. The table shows that estimates of the other coefficients and overall model performance are not much affected by how Maturity is handled. 9 Now that this average age is close to the average age of 14.1 years in the Berger and Udell (1995) sample. However, the average length of relationship in our sample is only 7.6 years, which is well below the Berger-Udell mean of 11.4 years. 15

16 3.1 Table 3 In these runs, implicit-interest variables (i.e., Fees and Collateral) receive insignificant coefficients. This single-equation model generates no evidence that explicit interest rates and implicit interest rates are correlated, and accordingly no indication that they are jointly determined. Cash, Size, Prime, Bank, and Year prove significant at onepercent, while relationship variables (Relation and Number), Term, Default, CEO share, and NFA are significant at weaker levels. These estimates indicate, as in the previous literature, that smaller borrower firms are charged a higher contract rate, whereas the limited-liability status of the borrower firm has no significant effect on this rate. Borrowers who get their lines of credit from banks are charged a lower contract rate. As in Petersen and Rajan (1994, 1995), contract rates vary significantly with macroeconomic variables. Specifically, contract rates respond positively to the prime rate, negatively to the slope of the term structure, and positively to the default premium. As in Berger and Udell (1995), Harhoff and Korting (1998), and Degryse and van Cayseele (2000), the length of the bank-borrower relationship relates negatively (Ed wants reduces the loan rate) to the loan rate. This result is inconsistent with Petersen and Rajan (1994) and Elsas and and Krahnen (1998), who find no significant association between relationship length and explicit interest rates. We next investigate the effects of treating Collateral and Fees as determined simultaneously with the Loan Rate. We begin by assessing the strength of our instrumental variables in Table 4, 5, and 6. Each table reports results for three specifications. The first specification measures the explanatory power of the instrumental variables alone. The second specification incorporates the contribution of the 16

17 instrumental variables and control variables taken together, while the third specification shows the effect of excluding Maturity from the second model. Table 4 summarizes the Tobit regression for Fees. The first specification establishes a significantly positive relationship between Fees and Fixed. The experiment achieves a significant likelihood ratio of In the table s second specification, all control variables except Maturity and Relation prove insignificant. The positive sign assigned to the Fixed dummy supports the hypothesis that Fees help to price the interestrate option held by the borrower. Dropping Maturity does not render any other variables significant and has little effect on the magnitude or significance of the coefficient for Fixed. These results indicate that Fixed is an acceptable instrument for Fees. 3.2 Table 4 In parallel fashion, Table 5 assesses the performance of the instruments included in the Collateral equation. This table reports on experiments using logistic regressions and incorporates the various regressors employed by Berger and Udell (1990, 1995). Depending on its sign, each significant coefficient increases or decreases the probability that the borrower is required to post collateral. In the first specification, the instruments CEO Share and Delinq prove to be significant. However, only Delinq retains its significance when the control variables are added. Nevertheless, a Chi-Square test rejects the hypothesis that instrumental-variable coefficients jointly equal zero at the one-percent level. Among the controls, Debt, Relation, Maturity, and Year are significant in the second and third experiment. The logistic regression shows a significant likelihood ratio in all three specifications. In the third specification, coefficient signs and significance are 17

18 unaffected by dropping Maturity. This suggests that the impact of Maturity on collateral is orthogonal to that of the other regressors. 3.3 Table 5 In Table 6, the instrumental variables for Loan Rate prove strongly significant in the first specification, but their significance weakens when control variables are introduced. Again, an F-test rejects the hypothesis that instrumental-variable coefficients jointly equal zero at one percent. Many of the control variables prove significant in the second and third runs. However, as in the single-equation runs of Table 3, Maturity is insignificant and dropping it has little effect on the magnitude and significance of other coefficients. 3.4 Table 6 In these single-equation experiments, coefficient signs are only partly consistent with the results of Petersen and Rajan (1995). The negative relation for HHI continues to indicate that banks in more concentrated markets may charge a lower explicit rate, but the negative sign for Firm Age implies -- in line with the bargaining-power hypothesis-- that older firms receive the lowest explicit rates. Still, the positive sign for the interaction term Inter 1 supports the Petersen-Rajan hypothesis that, in concentrated markets, banks may charge younger firms a lower explicit rate on the grounds they can confidently plan to extract rents from these customers later. Table 7 reports two-stage least-squares (2SLS) estimates of the Loan Rate equation using the three-equation system. In the first-stage, we regress Fees (Tobit regression) and Collateral (logistic regression) on the instrumental variables. Fitted 18

19 values from this stage are used as regressors in Loan Rate specifications that alternately include and exclude Maturity Table 7 To ensure that each simultaneous equation is well-identified in our sample, it is useful to examine whether the residuals from the loan-rate equation prove appropriately uncorrelated with the instrumental variables Z. In the last line of Table 7, we test the hypothesis, E(ε LR, Z )=0. Low F-statistics emerge in both specifications. This allows us to maintain the null hypothesis that the error terms are uncorrelated with the instruments. Although the magnitude of the adjusted R 2 and the signs and significance of all exogenous variables are the same between the OLS regressions and the 2SLS regressions, the endogenous variables Fees and Collateral become significant only in the 2SLS experiments. The significantly positive signs that simultaneous modeling assigns to these variables support two hypotheses: (1) that Fees help compensate the lender for the interest-rate option customers receive from fixed-rate credit lines; and (2) that posting Collateral controls and implicitly prices some (but not all) of the loss exposure lenders face in high-risk loans. As compared to the instrumental-variable runs of Table 6, the 2SLS estimates of Table 7 support the Petersen-Rajan hypothesis by reversing the sign of Firm Age (older firms now pay higher rates), but completely eliminate the significance of the marketconcentration and interaction terms. Simultaneous estimation also raises the p-value of the negative impact found for relationship length slightly above the 10-percent bound for statistical significance. Coefficients for Size, Bank, Number, and Cash continue to 10 As before, results for the Loan Rate prove robust to the exclusion of this variable. 19

20 support the hypothesis that banks have informational advantages and that they shift more of the benefits generated by these advantages to customers that evidence greater bargaining power. 4. Conclusions Drawing on banking theory, Berger and Udell (1995) and Petersen and Rajan (1994), among others, have estimated empirically the determinants of a firm s borrowing cost, and (focusing on the impact of lender-borrower relationships on explicit loan rates) obtained mixed results. However, these empirical studies do not recognize that banking theory does not require that the length of the borrower-lender relationship affect the explicit loan interest rate. It requires instead that these benefits affect the sum of a borrower s implicit and explicit loan rate. We examine the impact of the borrower-lender relationship on two implicit elements of a loan s interest cost, namely, whether the loan requires collateral and whether it carries any fees. Using a simultaneous system of equations with well-identified instruments, we find that firms that have a shorter relationship with their lenders face a higher likelihood of posting collateral and are charged higher fees. After endogenizing fees and collateral in a system of equations, we no longer find a significant correlation between explicit loan interest rates and the length of the borrower-lender relationship, but do find a positive correlation between loan interest rates, collateral and fees. These results show that explicit and implicit terms of relationship loans are set jointly, which suggests that disagreement in the findings of single-equation studies may be a product of simultaneousequation bias. We also find that fees are higher for fixed-rate loans, suggesting that fees 20

21 help compensate the lender for the interest-rate put option customers receive from fixedrate credit lines. 21

22 References Barclay, M, Marx, L. M., and Smith, C. (2002) The Joint Determination of Leverage and Maturity, Journal of Corporate Finance, forthcoming. Berger, A., Saunders, A., Scalise, J.M., and Udell, G. (1998) The effect of bank mergers on small business lending, Journal of Financial Economics, 50, Berger, A & Udell, G. (1995) Relationship lending and lines of credit in small firm finance, Journal of Business, 68, Berger, A. & Udell, G. (1990) Collateral, loan quality, and bank risk, Journal of Monetary Economics, 25, Besanko, D. & Thakor, A. (1987)a Collateral and rationing: Sorting Equilibria in monopolistic and competitive credit markets, International Economic Review, 28, Besanko, D. & Thakor, A. (1987)b Competitive equilibrium in the credit market under asymmetric information, Journal of Economic Theory, 71, Bester, H. (1985) Screening vs. rationing in credit markets with imperfect information, American Economic Review, 75, Boot, A., Thakor, A., and Udell, G. Secured lending and default risk: Equilibrium analysis and policy implications and empirical results, Economic Journal, 101, Boot A. & Thakor, A. (1994) Moral hazard and secured lending in an infinitely repeated credit market game, International Economic Review, 35, Bowden, R., & Turkington, D. (1984) Instrumental Variables, Econometric Society Monograph in Quantitative Economics, Publication No. 8. Cambridge: Cambridge University Press. Brick, I. & Ravid, A. (1985) The Relevance of Debt Maturity Structure, Journal of Finance, 40, Brick, I. & Ravid, A. (1991) Interest Rate Uncertainty and the Optimal Debt Maturity, Journal of Financial and Quantitative Analysis, 26, Brueckner, J. (1994) Borrower mobility, adverse selection, and mortgage points, Journal of Financial Intermediation, 3, Chan, Y. & Kanatas, G. (1985) Asymmetric valuation and the role of collateral in loan agreements, Journal of Money, Credit and Banking, 17,

23 Chao, J. & Swanson, N. (2003) Consistent Estimation with a Large Number of Weak Instruments, Yale University, Cowles Foundation Discussion Paper No Degryse, H. & Cayseele, P. (2000) Relationship lending within a bank-based system: evidence from European small business data, Journal of Financial Intermediation, 9, Diamond, D., & Rajan, R. (2001) Liquidity risk, liquidity creation, and financial fragility: a theory of banking, Journal of Political Economy, 109, Diamond, D. (1991) Monitoring and reputation: the choice between bank loans and directly placed debt, Journal of Political Economy, 99, Diamond, D. (1984) Financial intermediations and delegated monitoring, Review of Economics Studies, 51, Dunn, K. & McConnell, J. (1981) A Comparison of alternative models for pricing GNMA mortgage-backed securities, Journal of Finance, 36, Dunn, K. & Spatt, C. (1985) An analysis of mortgage contracting: Implications for mortgage contract terms and pricing, Journal of Finance, 40, Elsas, R., & Krahnen, J. (1998) Is relationship lending special? Evidence from creditfile data in Germany, Journal of Banking and Finance, 22, Fama, E. (1985) What s different about banks? Journal of Monetary Economics,15, Harhoff, D, & Korting, T. (1998) Lending relationships in Germany: empirical results from survey data, Journal of Banking and Finance, 22, Hodgman, D. (1963) Commercial Bank Loan and Investment Policy. Champaign, IL: Bureau of Economic and Business Research, University of Illinois. Hubbard, R.G., Kuttner, K., and Palia, D. (2002) Are they bank effects in borrower cost of funds? Evidence from matched sample of borrowers and banks, Journal of Business, 75, James, C. (1987) Some evidence on the uniqueness of bank loans, Journal of Financial Economics, 19, Kane, E. J. & Malkiel, B.G. (1965) Bank portfolio allocation, deposit variability, and the availability doctrine, Quarterly Journal of Economics 79, Kashyap, A. & Stein, J. C. (2000) What do a million banks have to say about the transmission of monetary policy? American Economic Review 90,

24 Kashyap, A., J., Stein, C., and Wilcox, D. W. (1993) Monetary policy and credit conditions: Evidence from the composition of external finance, American Economic Review 83, LeRoy, S. (1996) Mortgage valuation under optimal prepayment, Review of Financial Studies 9, Lummer, S. & McConnell, J. (1989) Further Evidence on the Bank Lending Process and the Capital-Market Response to Bank Loan Agreements, Journal of Financial Economics, 25, Padilla, A. & Pagano, M. (1997) Endogenous communication among lenders and entreprenuerial incentives, Review of Financial Studies, 10, Petersen, M. & Rajan, R. (1995) The effect of credit market competition on lending relationships, Quarterly Journal of Economics, 110, Petersen, M. & Rajan, R. (1994) The benefits of lending relationships: evidence from small business data, Journal of Finance, 49, Rajan, R. (1992) Insiders and outsiders: the choice between informed and arm s length debt, Journal of Finance, 47, Ramakrishnan, S. & Thakor, A. (1984) Information reliability and a theory of financial intermediation, Review of Economics Studies, 51, Sapienza, P. (2002) The effect of banking mergers on loan contracts, Journal of Finance, 57, Sharpe, S. (1990) Asymmetric information, bank lending and implicit contracts: a stylized model of customer relationships, Journal of Finance, 45, Slovin, M.B., Shushka, M.F., and Polonchek, J.A. (1993) The value of bank durability: borrowers and stakeholders, Journal of Finance, 48,

25 Table 1 Variable definitions Variables Definitions Dependent variables: Loan rate Fees Collateral Control variables: Debt Profit Cash Size Company Prime Term Default Relation Bank Comp bal Number Maturity Year contractual coupon rate on line of credit fees collected by the lending institution as a fraction of the total amount borrowed set to unity if collateral is required by the loan, and zero otherwise ratio of total debt outstanding to level of sales ratio of earnings before interest and taxes to sales ratio of level of cash holdings to sales natural logarithm of firm s sales set to unity if the firm s owners enjoy limited liability protection, and zero otherwise prime rate yield spread between the five-year Treasury note and the monthly three-month Treasury bill difference between the yield of a Baa bond and a Aaa bond number of years the firm has had a relationship with the lending institution set to unity if the lender is a bank, savings or thrift institution, and zero otherwise set to unity if the line of credit agreement requires a compensating balance, and zero otherwise number of lending sources available to the firm number of months the line of credit is outstanding set to unity if the firm observation is obtained from the 1998 Survey, and zero if the firm observation is obtained from the 1993 Survey Table 1 (continued) 25

26 Table 1 (continued) Variables Definitions Instrumental Variables: For Loan rate: HHI Firm age Inter1 For Fees: Fixed For Collateral: CEO age CEO share CEO exp Delinq NFA set to unity if the Herfindahl Index for deposits in the MSA of the firm is greater than 1800, and zero otherwise age of the firm interaction of HHI and Firm age (HHI x Firm age) set to unity if the line of credit has a fixed coupon rate, and zero if the coupon rate is a variable rate age of CEO percentage ownership of firm held by CEO number of years experience as CEO of the firm set to unity if either the principal owner or the firm has defaulted, and zero otherwise ratio of tangible long-term assets to sales 26

27 Table 2 Descriptive statistics Variable Unit # Mean Median Standard Deviation Dependent Variables: Loan rate Percent 1, Fees (all L/Cs) Fraction 1, Fees (L/Cs where fees Fraction charged) Collateral Dummy 1, Control Variables: Debt Fraction 1, Profit Fraction 1, Cash Fraction 1, Size Log 1, Company Dummy 1, Prime Percent 1, Term Basis Pts 1, Default Basis Pts 1, Relation Years 1, Bank Dummy 1, Comp bal Dummy 1, Number Quantity 1, Maturity Months Year Dummy 1, Instrumental Variables: For Loan Rate: HHI Dummy 1, Firm age Years 1, Inter1 Interaction 1, For Fees: Fixed Dummy 1, For Collateral: CEO age Years 1, CEO share Percent 1, CEO exp Years 1, Delinq Dummy 1, NFA Fraction 1,

28 Table 3 Single-equation models of the explicit loan rate (all regressors treated as exogenous) Ordinary-least-squares regression of Loan rate on all exogenous variables specified in the previous empirical literature. The dependent variable is Loan rate, defined as the contractual coupon on the line of credit. The independent variables are defined as follows: Fees, fees collected by the lending institution as a fraction of the total amount borrowed; Collateral, set to unity if collateral is required by the loan, and zero otherwise; Debt, ratio of total debt outstanding to level of sales; Profit, ratio of earnings before interest and taxes to sales; Cash, ratio of the level of cash holdings to sales; Size, natural logarithm of firm s sales; Company, set to unity if the firm s owners enjoy limited liability protection, and zero otherwise; Prime, prime rate; Term, yield spread between the five-year Treasury note and the monthly three-month Treasury bill; Default, difference between the yield of a Baa bond and a Aaa bond; Relation, number of years the firm has had a relationship with the lending institution; Bank, set to unity if the lender is a bank, savings or thrift institution, and zero otherwise; Comp bal, set to unity if the line of credit requires a compensating balance, and zero otherwise; Maturity, number of months the line of credit is outstanding; Year, set to unity if the firm observation is obtained from the 1998 survey, and zero if the firm observation is obtained from the 1993 survey; Fixed, set to unity if the line of credit has a fixed coupon rate, and zero if the coupon rate is a variable rate; CEO age, age of the CEO; CEO share, percentage ownership of firm held by CEO; CEO exp, number of years experience as CEO of the firm; Delinq, set to unity if either the principal owner or the firm has defaulted, and zero otherwise; NFA is the ratio of tangible long-term assets to sales; HHI, set to unity if the Herfindahl Index for deposits in the MSA of the firm is greater than 1800, and zero otherwise; Firm age, age of the firm; and Inter 1, interaction of HHI and Firm age (HHI x Firm age). (p-values are given in parentheses). Two digit industry dummies are included, the results of which are not reported. with maturity without maturity Variable parameter (p-value) parameter (p-value) Intercept a (0.003) a (0.004) Fees (0.187) (0.208) Collateral (0.656) (0.578) Debt (0.578) (0.548) Profit (0.113) (0.105) Cash c (0.007) c (0.007) Size a (0.000) a (0.000) Company (0.346) (0.371) Prime a (0.000) a (0.000) Term c (0.084) c (0.088) Default c (0.060) c (0.064) Relation c (0.068) c (0.071) Bank a (0.000) a (0.000) Comp bal (0.865) (0.865) Number c (0.052) b (0.058) Maturity (0.396) Year a (0.002) a (0.003) Fixed a (0.000) a (0.000) CEO age (0.613) (0.649) CEO share b (0.014) b (0.013) CEO exp (0.464) (0.407) Delinq (0.136) (0.120) NFA b (0.035) b (0.034) HHI (0.485) (0.477) Firm age (0.969) (0.969) Inter (0.209) (0.208) Adj. R F-value (p-value) a (0.000) a (0.000) a statistically significant at 1% level, b statistically significant at 5% level, and c statistically significant at 10% level, respectively. 28