Firm Size Dependence in the Determinants of Bank Term Loan Maturity

Transcription

1 Journal of Business Finance & Accounting, 32(1) & (2), January/March 2005, X Firm Size Dependence in the Determinants of Bank Term Loan Maturity STEVEN A. DENNIS AND IAN G. SHARPE* Abstract: We examine the hypothesis that firm size affects the sensitivity of bank term loan maturity to its underlying determinants. As borrower size increases, negotiating power with the lender and information transparency increase, while the lender is able to spread the fixed costs of loan production across a larger dollar value of the loan. We find strong evidence of firm size dependency in the determinants of bank term loan maturity and show that this is unrelated to syndication. Only large borrowers can manipulate bank loan contract terms so as to increase firm value. Keywords: firm size, bank, loan, maturity 1. INTRODUCTION Recent empirical research examining the maturity of debt contracts has focused on four hypotheses relating debt maturity to taxes, to signaling issues, to contracting (or agency) problems, * The authors are respectively John H. Poteat Chair of Banking and Director, Center for Banking, East Tennessee State University; and Professor of Finance, School of Banking and Finance, University of New South Wales. This paper has benefited from comments by Christopher Anderson, John Erickson, Neil Esho, Mark Flannery, Paul Kofman, Don Mullineaux, Li-Anne Woo, an anonymous referee, and participants at the Australasian Finance and Banking Conference, the Financial Management Association Meeting, the European Financial Management Conference, and the Southern Finance Association Meeting. The authors gratefully acknowledge the research support provided by the Australian Research Council and by the George A. Ball Distinguished Research Fellowship. They thank both Chris Bradley and Debarshi Nandy for research assistance. (Paper received January 2002, revised and accepted November 2003) Address for correspondence: Ian Sharpe, Professor of Finance, School of Banking and Finance, University of New South Wales, Sydney, NSW 2052, Australia. i.sharpe@unsw.edu.au # Blackwell Publishing Ltd. 2005, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 31

2 32 DENNIS AND SHARPE and to the risk of not being able to obtain funding at maturity (or liquidity risk ). A potential shortcoming of many of these studies is that, while they include firm size as a direct determinant of debt maturity, they implicitly assume that the tax, signaling, contracting, and liquidity risk hypotheses are independent of firm size. However, several empirical studies in this area have provided cursory evidence that firm size dependence exists in the determinants of debt maturity structure. Barclay and Smith (1995, p. 625), who first empirically examined debt maturity structure in this context, conclude that the effect on the firm-size coefficient indicates a nonmonotonic relationship when they exclude firms with less than $100 million of debt from their sample. More recently, Demirguc- Kunt and Maksimovic s (1999) study of the effects of crosscountry differences in financial and legal institutions on debt maturity reports separate regressions for large and small firms and concludes that different policies would be necessary to lengthen the debt maturity of large and small firms (p. 334). However, existing debt maturity studies have not provided a rigorous statistical analysis of the firm size dependency of debt maturity, nor have they offered any detailed explanation for its existence. In this paper, we investigate the hypothesis that the behavioral relationships developed to explain debt maturity are firm size dependent. Firm size may influence the relationships affecting debt maturity in several ways. Where the borrower has little bargaining power in negotiations with the lender, maturity will be less sensitive to the borrower s preferences. Additionally, where the lender can more easily distinguish creditworthiness, the lender may be able to offer a richer menu of contract characteristics. When information concerning the borrower is opaque, the lender may impose a pooling equilibrium on the borrower in order to avoid being selected against if a menu of maturity choices were provided. Moreover, where the amount of borrowings is small, the lender s production function may not allow it to negotiate the maturity of smaller loans so precisely. Given that each of these influences (bargaining power, information transparency, and issue size) is positively correlated with firm size, we would expect the maturity of large firms debt contracts to be more sensitive to the underlying determinants of debt maturity.

3 FIRM SIZE DEPENDENCE IN THE DETERMINANTS OF LOAN MATURITY 33 If information, bargaining, and cost economies problems are significant impediments to maturity choices for smaller firms, these problems are likely to be most prevalent in bank loan contracts because firms experiencing these problems may be confined to bank debt and may not borrow in other debt markets. Consequently, we examine the hypothesis of firm size dependency in the determinants of debt maturity within a sample of 1,236 term loan contracts between a large cross-section of banks and borrowers. Within a sample of incremental bank loans, we suggest that the small firms, because of their information, bargaining, and cost economies problems, will have fewer loan maturity choices. Moreover, an examination of the debt maturity of individual loans allows us to control for other important characteristics of the loan that may influence maturity, such as secured status and pricing. We find strong support for the hypothesis that behavioral relations concerning loan maturity are firm size dependent and show that this size dependence is not related to syndication activity. Only large firms alter term loan maturity so as to reduce contracting and liquidity problems and to optimize the tax advantages of debt, and only large firms tradeoff higher spreads to obtain longer maturity loans. Moreover, the firm size dependency does not appear to be driven by information problems that force the bank to impose a pooling equilibrium with respect to debt maturity. Rather, the evidence is consistent with both the bargaining power and cost economies arguments. Our results are important because they imply an advantage to large firms. Whereas previous empirical studies have found that firms manipulate debt maturity to maximize firm value, we find that only large firms have the ability to manipulate loan maturity. Small firms have little influence in the setting of loan maturity. That is, the maturity is set independently of the characteristics of the borrower. It may well be that small firms take whatever maturity is offered to them. This suggests a substantial incentive to be a large firm, either through growth, merger, or acquisition. As the firm grows, it has more bargaining power with lenders and has a larger dollar volume over which to spread the bank s costs involved with contracting. In addition to any other advantages of firm size, such as diversification, increased firm size allows the firm another avenue to enhance shareholder value via the manipulation of debt maturity.

4 34 DENNIS AND SHARPE The remainder of the paper is organized as follows. Section 2 outlines the hypotheses to be tested while Section 3 describes the data and estimation techniques. Section 4 then provides the estimation results and Section 5 concludes the discussion. 2. BANK TERM LOAN MATURITY STRUCTURE In this section, we initially examine the contracting, signaling, liquidity risk, tax, and interrelated contract terms hypotheses concerning the maturity of bank term loans. We conclude by developing a framework within which we test if these behavioral relationships are firm size dependent. (i) The Contracting Hypothesis There are at least two ways that contracting problems between debt holders and equity holders can influence loan contract features. After contracting risky debt, managers can transfer wealth from bondholders to equity holders by increasing the risk of projects undertaken (asset substitution). Smith and Warner (1979) suggest covenants and security provisions are important in limiting this incentive. Also, Myers (1977) describes a problem of passing over low-risk projects when the firm has risky debt outstanding. The acceptance of low risk, positive NPV projects transfers wealth to existing bondholders, reducing the incentive for equity holders to undertake these projects. These contracting problems can be reduced by employing short-maturity debt, by lowering leverage, by lowering risk, and/or by issuing claims with high priority (see Myers, 1977; and Stultz and Johnson, 1985). If contracting problems are important, then the maturity of new term loans will be negatively related to the extent of the firm s contracting problems. The most prevalent proxy for the extent of the firm s contracting problems has been the Market/Book ratio. 1 If the firm s assets are recorded at historical value for accounting purposes and if markets are reasonably efficient in pricing future earnings, then the market-to-book ratio captures the degree to 1 The definitions of all variables appear in the Data Appendix.

5 FIRM SIZE DEPENDENCE IN THE DETERMINANTS OF LOAN MATURITY 35 which the firm has many opportunities to invest. More growth opportunities (higher market-to-book ratios) represent a higher degree of contracting problems. (ii) The Signaling Hypothesis Flannery (1986), Robbins and Schatzberg (1986) and Kale and Noe (1990) argue that, given asymmetric information, the firm will determine its debt maturity structure based upon the mispricing of the firm s debt in the capital markets. High-quality firms choose short-term debt, which allows the interest rate to adjust after revelation of favorable private information. Low-quality firms, in contrast, issue long-term debt to avoid a revelation of unfavorable news and to lock in the borrowing cost on the overpriced debt. The signaling literature, therefore, suggests the maturity of the firm s debt will be inversely related to firm quality. Although firm quality is difficult to measure, studies have typically proxied it with some ex-post measure of performance, such as earnings surprises or a stock price response. Following Barclay and Smith (1995) and Stohs and Mauer (1996), we employ the abnormal (or unexpected) future earnings of the firm, Abnormal Earnings, which is defined in the Data Appendix. The signaling hypothesis suggests Abnormal Earnings will be inversely related to term loan maturity. (iii) The Liquidity Risk Hypothesis Diamond (1991b and 1993) develops a model in which borrowers prefer short-term financing, but face liquidity (or refinancing) risk if bad news concerning the project arrives before the project matures. 2 Lenders will be prone to liquidate the project too often at the rollover date because lenders cannot be assigned future cash flows and control rents. In this model, highly-rated firms issue short-term debt because their exposure to the risk of not being able to obtain refinancing when the debt matures is relatively low. Lower-rated firms, however, issue long-term debt to avoid the refinancing (liquidity) risk. 2 See also Sharpe (1991).

6 36 DENNIS AND SHARPE Diamond (1991b), though, notes an important limitation for very low rated firms. These firms are unable to borrow long term because the high cost of borrowing will induce risky behavior. Therefore, the maturity of term loans should be related to the firm s credit quality in a nonlinear fashion. High-rated and lowrated firms borrow short term for different reasons and intermediate-rated firms borrow long term. We proxy credit quality by computing a Z-score, similar to that in Altman (1993). 3 High Z-scores represent low probabilities of default. Diamond s (1991b and 1993) liquidity risk hypothesis suggests a positive sign for the linear term Z-Score and a negative sign for the quadratic term Z-Score Squared in the maturity equation. (iv) The Tax Hypothesis There are several theoretical articles relating taxes to debt maturity. Leland and Toft (1996) present a model in which firms tradeoff taxes, bankruptcy costs, and the agency costs of debt issuance. 4 At any given tax rate for the firm, long-term debt will be preferred to short-term debt because long-term debt reduces the endogenously-determined bankruptcy point for the firm. If bankruptcy occurs at lower asset values as debt maturity increases, the firm is less likely to lose these tax benefits in bankruptcy. As the firm s marginal tax rate decreases, the firm will shorten the maturity of its debt to avoid the agency problems of debt, including asset substitution. 5 Following Guedes and Opler (1996) we proxy the firm s marginal tax rate by Tax/Assets and expect a direct relationship with term loan maturity. 6 (v) Interrelated Contract Terms Melnik and Plaut (1986) characterize bank loan commitment contracts as providing a package of n contract terms that 3 The Z-Score we compute in this study is not exactly that of Altman (1993) because leverage is excluded in our calculation. 4 See also Kane et al. (1985). 5 Leland and Toft (1996) also show that the term structure of credit spreads may be upward sloping, which increases the tax benefit of long-term debt. 6 For robustness, we substituted Tax/Sales for Tax/Assets. This had no effect on the conclusions of the study.

7 FIRM SIZE DEPENDENCE IN THE DETERMINANTS OF LOAN MATURITY 37 cannot be split and traded separately. Banks offer borrowers an n dimensional array of bundles from which to choose their desired contract features, which allows for tradeoffs in the design of the loan. For example, borrowers may offer more collateral or higher pricing to receive a longer maturity loan. If borrowers tradeoff these loan characteristics, then the All-in- Spread (pricing) and the Secured status of the loan will each be directly related to loan maturity. 7 Given the interrelated nature of these contract terms, we discuss the necessary econometric techniques in Section 2(ii). (vi) Controls While testing the above hypotheses, we control for other possible influences, such as the firm s size, risk, leverage, and asset maturity. Though these variables can be connected with the hypotheses described above, we discuss them as controls because we cannot separate the influences of several competing hypotheses. For example, we control for the firm s Leverage as potentially a substitute for maturity in reducing the underinvestment problem (Myers, 1977). However, leverage also influences the firm s susceptibility to liquidity risk (Diamond, 1993) and may have tax consequences (Modigliani and Miller, 1963) and signaling implications (Ross, 1977). Along similar lines, we have difficulty differentiating competing theories concerning both the firm s risk and its asset maturity. Concerning Firm Risk, Kane et al. (1985), Wiggins (1990) and Leland and Toft (1996) all suggest a link with debt maturity through expected bankruptcy. However, the underinvestment problem is also more pervasive for higher-risk firms. 8 Also, firms may attempt to match the maturity of their debt contracts with Asset Maturity to control contracting problems (Myers, 1977; and Chang, 1992) or to control liquidity risk (Diamond, 1993). 7 Dennis et al. (2000) report strong evidence that these contract terms are interrelated. 8 We proxy firm risk by the standard deviation of its earnings before interest, taxes and depreciation for the five year period preceding the deal date and scale it by average total assets for that period. Guedes and Opler (1996) use a similar proxy but based on the firm s industry. For robustness, we also substituted the firm s Beta for the standard deviation of earnings. This had no effect on the conclusions of the study.

8 38 DENNIS AND SHARPE Finally, Firm Size has traditionally been included in empirical studies of debt contract design, though its inclusion is not explicitly rationalized in most cases. Using measures of (real) total asset size Barclay and Smith (1995), Stohs and Mauer (1996) and Dennis et al. (2000) all report a significantly positive relation between firm size and debt maturity. However, when Guedes and Opler (1996) employ sales as a measure of firm size, they find a negative relationship. The model we propose testing, then, is of the following form: 0 1 Constant; Market=Book; AbnormalEarnings; Z-Score; Z-Score Squared; Tax=Assets; Maturity ¼ fb All-in-Spread; Secured; Leverage; FirmRisk; A : ð1þ Asset Maturity; Firm Size (vii) Firm Size Dependence of the Behavioral Parameters In equation (1) we model firm size as a direct determinant of the maturity of bank term loans. However, in this section, we also argue that the behavioral relationships in equation (1) are dependent on firm size (i.e., that firm size affects the sensitivity of loan maturity to the hypotheses developed above). This dependency arises from several sources including differences in borrower bargaining power, in borrower information transparency, and in characteristics of the lender s loan production function. Rajan (1992) has shown that bank (or inside ) debt differs from public (or arm s length ) debt in that the bank has substantial knowledge of the borrower s activities. The informed position of the bank results in efficient monitoring of the borrower, but also results in a hold-up problem whereby the bank extracts monopoly rents from its superior information. In essence, this information allows the bank to price the loan higher than the true price for the borrower s quality. In such an environment, the other negotiated contract terms, including loan maturity, are also likely to reflect the borrower s bargaining power with the lender. Rajan (1992) and Detragiache et al. (2000) employ this analysis to suggest that borrowers will enhance their bargaining power by accessing multiple sources

9 FIRM SIZE DEPENDENCE IN THE DETERMINANTS OF LOAN MATURITY 39 of funds to reduce the hold-up problem. Houston and James (1996) report results consistent with borrower access to multiple borrowing sources reducing the information monopoly problem. Along similar lines, Diamond (1989 and 1991a) describes a model in which firms gain a reputation (or track record ) for performance, which lowers the agency costs associated with trusting the borrower. This produces a life-cycle effect of borrowing whereby the firm initially borrows from a bank that monitors the firm s actions closely. Over time, however, the firm gains a favorable reputation enabling it to access the public debt markets. Hence, a firm s access to multiple borrowing sources, and therefore its bargaining power with lenders, will be directly related to its reputation which, in turn, is directly related to firm size. Where the borrower has little bargaining power with the bank, we suggest the maturity of the loan will be less sensitive to the borrower s desires. That is, small borrowers (with little bargaining power) are not able to manipulate bank loan maturity as easily as large borrowers. Therefore, the maturity structure of large firms bank loans should be more responsive to the determinants of maturity than should small firms. Moreover, as the firm becomes larger, information concerning the firm becomes more transparent (Klein and Bawa, 1977; and Zeghal, 1984). Coverage of the firm by security analysts intensifies and the firm is more likely to be subject to disclosure rules from the SEC and listed stock exchanges. Together, these factors increase the information transparency of the firm. Stiglitz and Weiss (1981) argue that where lenders cannot easily distinguish the credit quality of borrowers, lenders impose a pooling equilibrium concerning loan pricing, resulting in credit rationing in debt markets. Charging a higher rate of interest results in adverse selection among borrowers. Along similar lines, lenders may offer a limited menu of maturity choices for information problematic borrowers. Hence, loan maturity should be more responsive to its underlying determinants for information-transparent (larger) firms than for informationopaque (smaller) firms. Finally, given that there are fixed costs in the origination, funding, and monitoring of borrowers, small loans provide less volume over which lenders can spread these fixed costs. If

10 40 DENNIS AND SHARPE lenders cannot achieve cost economies for smaller loans, the lender s production function could make it prohibitively expensive to offer a menu of maturity choices on smaller loans. Any or all of these three factors (bargaining power, information transparency, and loan production costs) tend to limit maturity choices for smaller firms. If maturity choices are limited for smaller firms, we would expect less variance in the maturity of smaller firms term loans. Therefore, we would expect there to be differences in the coefficients of estimated behavioral relations in a regression analysis involving bank term loans to firms of varying size. That is, proxies for contracting problems, liquidity risk, signaling, and taxes should have a greater effect (in absolute value terms) on the maturity of larger firm s term loans. (i) Data 3. DATA AND ESTIMATION The sample of bank term loans is obtained from the Loan Pricing Corporation s (LPC) Dealscan database, which provides detailed market information on more than $3 trillion of large corporate and middle-market commercial and industrial loans. Dealscan defines a term loan as an installment loan where amounts repaid may not be reborrowed. The funds are typically drawn down all at once, though the loan may have a series of takedowns or a delayed takedown period. Much of the data gathered by LPC comes directly from commitment letters or actual credit agreements contained in public filings with the SEC, including 13-Ds, 14-Ds, 13-Es, 10-Ks, 10-Qs, 8-Ks, and S-series filings. LPC also reports a small quantity of deals (5 10%), which are based on direct research at the lending banks. All loan data are confirmed through senior management or reported as unconfirmed or partially confirmed. Extraction of the data involved identifying all confirmed deals with at least one term loan facility and which have a deal date between January 1987 and December 1995 inclusive. 9 This 9 To reduce measurement error, we ignore unconfirmed and partially confirmed data.

11 FIRM SIZE DEPENDENCE IN THE DETERMINANTS OF LOAN MATURITY 41 produced a sample of 7,819 term loan facilities (6,885 loan deals) originated by 1,958 different lending entities. The data were then filtered to remove term loans where the facility size, maturity, or pricing details were not specified, where they were part of a 144A or non-144a private placement, and where we could not identify the lender as being a bank. As a final filter, we required borrower data availability on Computstat. A sample of 1,236 bank term loans survived these criteria. While the sample is smaller than in Barclay and Smith (1995) and Guedes and Opler (1996), it has more crosssectional variation than in Stohs and Mauer (1996). Considering the availability of detailed characteristics concerning the loan contract for a cross-section of borrowers, the sample size seems reasonable. A feature of the data requiring attention is that only 718 of the 1,236 observations include details of the secured status of the loan. The 518 missing secured status observations were treated using regression imputation, as in the SPSS treatment of missing values. This involved running a logit regression for Secured on non-missing observations of the exogenous and other endogenous variables in the model. For missing observations of secured status, the predicted probability of secured status is obtained. If this probability is 0.5, the missing secured observation takes the value of unity (¼secured). Otherwise it takes the value of zero (¼unsecured). This approach maintains the efficiency of the estimates by allowing the regressions to be estimated on the full sample of 1,236 observations. 10 Table 1 contains descriptive statistics for the sample. 11 The median term loan is $20 million in size, has a maturity of 4.8 years, and is priced at an all-in-spread of 200 basis points. Average loan size is significantly higher than the median at $140 million, suggesting some very large loans in the data. 10 As an alternative, the model was also estimated using the listwise deletion approach whereby the regression analysis is undertaken only on the N ¼ 718 non-missing observations. Apart from the larger standard errors of the estimates on the listwise deletion results, reflecting the loss of estimation efficiency, the results mirrored those of the full sample estimates. Moreover, the results do not differ substantially when employing the naïve approach of assuming all zero observations are unsecured. 11 To reduce the distorting effects of extreme values in the transformed data, extreme values are truncated at six standard deviations from the mean.

12 42 DENNIS AND SHARPE Table 1 Descriptive Statistics for Variables Variable Median Mean Standard Deviation Minimum Maximum Loan Size ($million) Firm Size ($million) Maturity (yrs.) Secured (proportion) All-in-Spread (basis pts.) Market/Book Leverage Tax/Assets Abnormal Earnings Z-Score Firm Risk Asset Maturity (log) Loan Concentration LIBOR (%) Term Premium (%) Int. Rate Volatility (%) Fixed Asset Ratio Revolver (proportion) Notes: Total observations equal 1,236. All variables as defined in the Data Appendix, except Firm Size. The maximum loan size is $5.9 billion and the minimum is $50,000. The average loan all-in-spread is marginally higher than the median at 208 basis points. For the non-missing secured status observations, 88% of the loans are secured. After regression imputation for the missing observations, 89% of the loans are classified as secured, suggesting that the missing observations had a similar proportion of secured loans as in the non-missing data. Loan purpose is described as being for working capital and general corporate purposes in 46% of the sample, debt repayment or recapitalization in 29%, and acquisition or leveraged buy-out financing in 20%. The average loan size in the sample is consistent with the average loan size in the samples of James (1987), Lummer and McConnell (1989), Slovin et al. (1992), Preece and Mullineaux (1994) and Billet et al. (1995). The median loan size in our

13 FIRM SIZE DEPENDENCE IN THE DETERMINANTS OF LOAN MATURITY 43 sample is smaller than in all of these studies, but the minimum loan size in our sample is also much smaller. Median loan maturity is consistent with Billet et al. (1995) and Preece and Mullineaux (1989), but slightly shorter than the sample medians reported in the other studies. The sample appears to represent the population of bank lending at least as well as in these studies. (ii) Estimation There are two major econometric concerns associated with the estimation of the maturity structural equation. Dennis et al. (2000) argue that loan contract terms, such as maturity, secured status, and pricing, are determined simultaneously and they find strong evidence of interrelated contract terms in bank revolving credit agreements. In the presence of such simultaneity, application of OLS to equation (1) would result in biased and inconsistent estimates. The second concern relates to the treatment of leverage. Where leverage is assumed to be exogenous, as in Stohs and Mauer s (1996) study of debt maturity, the implicit assumption is that leverage and contract terms are determined recursively with the firm initially determining its desired leverage and then selecting debt contract terms conditional on the leverage decision. An alternative specification, with leverage assumed endogenous, is suggested by several theories in which leverage and contract terms are viewed as alternative or substitute mechanisms for limiting agency or contracting problems (see Myers, 1977; and Leeth and Scott, 1989). Moreover, the theory of optimal capital structure suggests that leverage is related to other exogenous variables in the model of loan contract terms including the proxies for growth opportunities, asymmetric information, taxes, and credit risk. Indeed, Barclay et al. (1997) and Stohs (1998) present evidence that leverage and maturity may be interrelated. In light of this evidence, we assume that leverage is also endogenously determined. Thus, equation (1) is estimated using Two-Stage Least Squares where the instruments for the 2SLS estimates are

14 44 DENNIS AND SHARPE obtained from equation (1) and the following specifications for Secured, All-in-Spread, and Leverage: Constant;Maturity; All-in-Spread;Market=Book; AbnormalEarnings; Z-Score;Leverage; FirmRisk; Secured¼hB FirmSize; Revolver; LoanConcentration; A ð2þ Fixed Asset Ratio 0 1 Constant; Maturity; Secured; Market=Book; AbnormalEarnings; Z-Score;Leverage; All-in-Spread¼jB Firm Risk; Firm Size; Revolver; Loan Concentration; A LIBOR; Term Premium; Int: Rate Volatility 0 1 Constant; Maturity; Secured; All-in-Spread B C Leverage ¼ k@ Market=Book; Abnormal Earnings; Z-Score; A: ð4þ Leverage; Firm Risk; Firm Size Variables not previously defined include Revolver, abinary variable for whether the deal containing the term loan facility also includes a revolving credit facility, Loan Concentration, a proxy for borrower reliance on the lender, and the firm s Fixed Asset Ratio. WealsoincludeLIBOR, Int. Rate Volatility, andtheterm Premium, all of which are defined in the Data Appendix. The instruments for the 2SLS estimates of equation (1) are then the exogenous variables included in the specifications of equations (1) to (4). They are as follows: Constant, Market/Book, Abnormal Earnings, Z-Score, Z-Score Squared, Tax/Assets, Firm Risk, Asset Maturity, Firm Size, Term Premium, Loan Concentration, LIBOR, Fixed Asset Ratio, Revolver, and Int. Rate Volatility. ð3þ 12 The specifications for Secured and All-in-Spread are consistent with those in Dennis et al. (2000). The primary difference is that the Secured equation now allows for a feedback effect from the All-in-Spread while Revolver, Firm Risk and the Fixed Asset Ratio represent additional controls that assist in identification. The Leverage specification is broadly consistent with that in Barclay et al. (1997) and Stohs (1998).

15 FIRM SIZE DEPENDENCE IN THE DETERMINANTS OF LOAN MATURITY 45 (iii) Firm Size Dependency In testing whether the behavioral parameters of the model are firm size dependent, we allow each of the parameters to have a fixed and variable component, k ¼ f k þ v k Size where the f and v subscripts indicate fixed and variable, respectively, and Size is a measure of firm size. In the regressions, we consider two firm size measures. The first is a dichotomous [1,0] indicator variable, Sizedum, which takes the value of unity if the firm is in the upper half of the distribution of the Firm Size variable and zero otherwise. We refer to firms in the upper and lower halves of the distribution as large firms and middle-market firms, respectively. This seems a reasonable representation given that the median real firm size of the middle-market sub-sample is $60.91 million relative to that of the large firm sub-sample of $ million. When Sizedum is used in equation (5), each explanatory variable, X, is included in the regression in linear as well as in multiplicative form, X*Sizedum. The fixed coefficients are those on the linear variables and are interpreted as those applicable to the middle-market sample when Sizedum ¼ 0. On the other hand, the variable coefficient on the multiplicative term provides a test of whether the effect of the X variable on term loan maturity is different for large vis-a-vis middle-market firms. The coefficients for large firms, when Sizedum ¼ 1, are obtained by summing the fixed and variable coefficients. The coefficients for middle-market and large firms obtained in this manner are identical to those obtained when the model is estimated on the respective sub-samples of 618 observations. Table 2 contains descriptive statistics of the middle-market and large firm sub-samples. Term loans to middle-market firms have a significantly shorter mean maturity, a higher mean all-in-spread, and a higher probability of being secured. Moreover, middle-market firms have fewer growth opportunities (proxied by market/book), shorter asset maturity, lower leverage and tax/asset ratios, and higher loan concentration and abnormal earnings than do large firms. Interestingly, the two risk measures are conflicting with the Z-score suggesting ð5þ

16 46 DENNIS AND SHARPE Table 2 Descriptive Statistics for Middle-Market and Large Firm Sub-Samples Variable Middle-Market Sample N ¼ 618 Mean Standard Deviation Large Firm Sample N ¼ 618 Mean Standard Deviation t -statistic for Difference in Means Firm Size (in $million) *** Loan Size (in $million) *** Maturity (in yrs.) *** Secured (proportion) *** All-in-Spread *** (basis points) Market/Book ** Abnormal Earnings *** Z-Score ** Tax/Assets *** Term Premium Int. Rate Volatility *** Loan Concentration *** Revolver (proportion) Asset Maturity (log) *** Firm Risk *** LIBOR Leverage *** Fixed Asset Ratio *** Notes: Total observations equal 1,236. All variables as defined in the Data Appendix, except Firm Size. *, ** and *** indicate significance at the 10%, 5% and 1% levels respectively. that middle-market firms have significantly lower risk while the firm risk (volatility of earnings) measure suggests they have higher risk. A referee has noted that the disparity in the two measures is not surprising as the Z-score captures credit quality and liquidity risk factors while firm risk captures inherent business risk. The volatility of earnings measure may be higher for middle-market firms due to lack of diversification. However, they may compensate for the increased volatility of earnings by increasing liquidity, which increases the Z-score (suggesting lower risk). The second measure of size we employ is the Firm Size variable itself, as defined in the Data Appendix. In this case, the fixed parameter is interpreted as the coefficient of the X

17 FIRM SIZE DEPENDENCE IN THE DETERMINANTS OF LOAN MATURITY 47 variable when Firm Size ¼ For a firm of a given size, its coefficient is calculated by substituting the appropriate value of Firm Size in equation (5). 4. ESTIMATION RESULTS The 2SLS estimates of the structural model, with maturity in years 14 as the dependent variable, are summarized in Table 3. The first set of results relate to the size invariant or fixed coefficient model while the second set of results relate to the varying coefficient model based on Sizedum. Although the explanatory power of the regressions is low, with adjusted R 2 statistics ranging from 5% to 7%, this is a general characteristic of incremental debt maturity studies (see Guedes and Opler, 1996). (i) Fixed Coefficient Results Generally, the fixed coefficient results are consistent with the contracting, liquidity risk, and tax hypotheses. Firms with a higher degree of contracting problems, proxied by the Market/Book ratio, limit asset substitution and underinvestment problems by shortening the maturity of term loans as in Myers (1977) and Stulz and Johnson (1985). Moreover, there is a nonlinear relationship between term loan maturity and the borrower s credit quality, proxied by Z-Score, with intermediate-rated firms borrowing for longer terms than high-rated and low-rated firms. Also, the significantly positive relationship between Tax/Assets and loan maturity is consistent with the Leland and Toft (1996) model in which lengthening the firm s debt maturity lowers the endogenously-determined bankruptcy point for the firm and thereby creates a higher expected value of the tax advantage of debt. There is also some evidence of interrelated debt contract terms, consistent with Dennis et al. (2000), with loan pricing 13 That is, a real market value of only $1 million, which is outside the range of Firm Size within the sample. The smallest firm in the sample has a real total asset size of $2.4 million. 14 Regressions were also run using log of maturity as the dependent variable but this had little effect on the results.

18 48 DENNIS AND SHARPE Table 3 Two Stage Least Squares Estimates of Maturity of Bank Term Loans: Size Invariant Model and Size Varying Model Based on Sizedum Size Varying Model Based on Sizedum Independent Variables Size Invariant Model Mid-Market Firms Larger Firms Constant (1.9491) Market/Book *** (0.1986) Abnormal Earnings (0.2670) Z-Score * (0.2393) Z-Score Squared ** (0.0354) Tax/Assets *** (4.0064) Firm Size *** (0.1829) Asset Maturity *** (0.1733) Firm Risk ** (2.3187) Fitted Leverage *** (1.4284) Fitted Secured (0.1560) Fitted All-in-Spread *** (0.5616) (2.6854) (0.2846) (0.3286) (0.3201) (0.0479) (5.5716) (0.2820) (0.2465) (3.1140) (2.0069) (0.2310) (0.7792) R Model F-stat 6.94*** 4.97*** Size Sub-Sample F-stat 3.03*** *** (2.8811)### *** (0.2797)### (0.4650) * (0.3647) * (0.0534) *** (5.8054)### *** (0.2628)### ** (0.2412) (4.0109) *** (2.0260) (0.2101) *** (0.8023)### Notes: The size invariant model is as summarized in equation (1). Definitions of variables appear in the Data Appendix. Standard errors are in parentheses. *, ** and *** indicate coefficient significance at the 10%, 5% and 1% level while #, ## and ### indicate that the large firm and mid-market firm coefficients are significantly different at the 10%, 5% and 1% levels, respectively. Sample size is 1,236 observations. and maturity significantly and positively related. Borrowing firms appear willing to tradeoff a higher borrowing rate in order to increase the maturity of their term loans. In contrast

19 FIRM SIZE DEPENDENCE IN THE DETERMINANTS OF LOAN MATURITY 49 with the loan pricing interdependence, the results suggest that there is little tradeoff in terms of providing collateral for a more favorable loan maturity. The four control variables in the structural model, Leverage, Firm Risk, Asset Maturity and Firm Size, are significant determinants of term loan maturity. Moreover, the signs of the coefficients are consistent with contracting cost explanations of debt maturity with: (i) the negative coefficient on Leverage consistent with leverage being a substitute for debt maturity in controlling contracting problems; (ii) the positive coefficient on Asset Maturity being consistent with firms matching debt to asset maturities to control contracting problems; and (iii) the negative coefficient on Firm Risk and positive coefficient on Firm Size being consistent with higher risk firms having greater contracting problems and shortening debt maturity to alleviate these problems. In contrast with these relatively strong results supporting the contracting, liquidity risk, and tax hypotheses, there is little support for the signaling hypothesis. Although the coefficient of Abnormal Earnings has the correct negative sign, it is not statistically significant. The size-invariant results in Table 3 are important in that they are consistent with previous empirical studies in this area, including those who measure total debt maturity structure rather than the maturity of individual debt issues. The results bolster our belief that the maturity choices in our sample of bank loans are representative of debt maturity choices in general. We now turn our attention to the firm size varying coefficients in Table 3. The fixed-coefficient structural estimates were examined for stability across firm size using the Chow F -test. Allowing each coefficient in the model to have a fixed and variable component, as in equation (5), and representing size by either the dichotomous Sizedum variable or the continuous Firm Size variable, we then test whether the size varying parameters in the model are jointly zero. The relevant F-statistic is shown in the final row of Tables 3 and 4. For both measures of firm size, the Chow test rejects the null hypothesis that the parameter estimates are stable across firm size. Thus, in the following section we examine the results of firm size varying coefficient models.

20 50 DENNIS AND SHARPE Table 4 Two Stage Least Squares Estimates of Maturity of Bank Term Loans: Size Varying Model Based on Firm Size Regression Result Implied Coefficient Independent Variables Fixed Coefficient Firm Size Varying Coefficient At Mean Firm Size of Mid- Market Firms At Mean Firm Size of Large Firms Constant *** (5.9031) Market/Book *** (0.6330) (0.1122) Abnormal Earnings (0.9556) (0.1889) Z-Score ** (0.7300) (0.1364) Z-Score Squared * (0.1049) (0.0198) Tax/Assets ** ( ) (2.1918) Firm Size *** *** (1.5566) (0.0960) Asset Maturity (0.5701) (0.1012) Firm Risk ** * (6.9918) (1.3382) Fitted Leverage (4.5931) (0.8092) Fitted Secured (0.5389) (0.0924) Fitted All-in-Spread * *** (1.7605) (0.3106) R Model F-stat 5.12*** Varying Coef. F-stat 3.15*** *** (5.9031) (0.2523) (0.3149) (0.2815) (0.0407) (4.9121) *** (1.1952) (0.2227) ** (2.8022) ** (1.8304) (0.2148) (0.7085) *** (5.9031) *** (0.2607) (0.4612) *** (0.3431) *** (0.0514) *** (5.2860) *** (0.9454) *** (0.2283) (3.6641) *** (1.8480) (0.1939) *** (0.7239) Notes: Definitions of variables appear in the Data Appendix. Standard errors are in parentheses. Implied coefficients are those applying to the mean Firm Size of mid-market and large firms where mid-market (large) firms are those in the lower (upper) halves of the distribution of firms by Firm Size. *, ** and *** indicate coefficient significance at the 10%, 5% and 1% levels, respectively. Sample size is 1,236 observations. (ii) Firm Size Varying Coefficient Results The 2SLS estimates of the firm size varying coefficient model based on the discrete Sizedum variable (corresponding to middle-market and large firms respectively) are summarized

21 FIRM SIZE DEPENDENCE IN THE DETERMINANTS OF LOAN MATURITY 51 in the two right-hand columns of Table 3 while those based on the continuous variable, Firm Size, are shown in Table 4. The results in Table 3 are presented so as to emphasize the estimates for each of the size sub-samples. The middle-market estimates are the fixed parameters, f k, in equation (5) whereas the large firm coefficients are the sum of the fixed and variable coefficients, f k þ v k. Where the large and middle-market estimates are significantly different, this is shown by an # on the standard errors of the large firm coefficients. For the results based on the continuous firm size variable in Table 4, the fixed and firm size varying coefficients are reported. 15 While the significance of specific varying coefficients is important in identifying the source of the parameter size dependence, the sensitivity of loan maturity to an independent variable requires the substitution of a particular firm size into equation (5). To facilitate comparisons between the estimates in Table 4 with those in Table 3, in Table 4 we use equation (5) to compute the implicit regression coefficients of the firm size varying coefficient model at the mean Firm Size of the middlemarket and large firm sub-samples, respectively. The similarity of the estimates in Table 3 with those in Table 4 confirms that the size dependence results are robust across the alternative measures, Sizedum and Firm Size. 16 The results in Table 3 suggest that the fixed coefficient results are largely driven by firms in the upper half of the distribution of firms by firm size. Although most of the coefficients of the middle-market firms have the same sign as those of large firms, their coefficients are all smaller in absolute size and are not significantly different from zero. This is consistent with the maturity of bank term loans to mid-market firms being less responsive than that of large firms to the underlying determinants of debt maturity choice. Moreover, the statistically 15 For this model, because of singularity it is not possible to estimate a firm size varying parameter for the constant term. 16 The exceptions are the coefficients of the firm size and constant terms which, for large firms, switch signs between Tables 3 and 4. The negative implied coefficients for firm size in the size varying model of Table 4 are attributable to the positive component of the firm size effect being captured in the coefficient of the constant term. It is important to note however, that the results are consistent across the models with the firm size coefficient increasing with firm size (in Table 4 the implied coefficient becomes less negative as firm size increases).

22 52 DENNIS AND SHARPE significant difference at the 99% confidence level in the coefficient of the Fitted All-in-Spread variable, across large and mid-market firms, reinforces the view that mid-market term loan maturity is less responsive. Large firms are able to tradeoff loan maturity and pricing by purchasing longer maturities at higher spreads, a tradeoff option not evident in the middle-market sub-sample. Other behavioral parameters in Table 3 that are significantly different across the sub-samples are the proxy for contracting problems, Market/Book, the Tax/Asset ratio, and Firm Size. Only firms in the large firm sub-sample shorten loan maturity to alleviate the contracting problems associated with growth opportunities or lengthen maturity to reduce the endogenously determined bankruptcy point for the firm so as to maximize the tax benefit of debt. As the significant difference in the parameter of the Firm Size variable across the sub-samples may reflect an incorrect functional form, the Chow F stability tests were re-run to test the null that all size varying coefficients other than that on the Firm Size variable were jointly equal to zero. In all cases the null was strongly rejected, indicating that the size dependency property of the model is not due to an incorrect functional form of the firm size variable. In addition to the four variables that have significantly different parameter estimates across the sub-samples, the liquidity risk proxies, Z-Score and Z-Score Squared, and the Asset Maturity and Fitted Leverage variables are statistically significant for only the large firm sub-sample, while not being significantly different from the mid-market sub-sample parameters. When the varying coefficient model is based on the continuous variable, Firm Size, as in Table 4, some further interesting results emerge. In addition to the four significant size dependent parameters observed in Table 3, the liquidity risk parameters, Z-Score and Z-Score Squared, and the control variable, Firm Risk, now exhibit firm size dependence. Moreover, an examination of the implied coefficients at the mean firm size of mid-market and large firms, respectively, indicates that support for the contracting, liquidity risk, and tax hypotheses, and for interrelated pricing and maturity decisions, is largely confined to large firms in the sample. Among the control variables, the significantly positive relationship between asset

23 FIRM SIZE DEPENDENCE IN THE DETERMINANTS OF LOAN MATURITY 53 maturity and loan maturity is only evident for large firms while the negative relationship with firm risk is only evident for midmarket firms. On the other hand, the inverse relationships of loan maturity with leverage and firm size are observed across both mid-market and large firms. (iii) Loan Syndication and Firm Size Dependence Dennis and Mullineaux (2000) have shown that syndicated loans have longer maturity and less incidence of collateral as compared to non-syndicated loans. In our sample, half the term loans are syndicated while Syndication (a binary variable equaling unity for syndicated loans) is positively correlated with loan maturity and inversely correlated with secured status as in Dennis and Mullineaux (2000), and inversely correlated with the All-in-Spread. Importantly, there is a very high positive correlation of 0.55 between Firm Size and Syndication. These significant correlations suggest the possibility that the firm size dependence results may be a syndication effect rather than a firm size effect. To examine this possibility, the syndication variable was added as a control in equations (1) to (3) and substituted for firm size in equation (5) to produce a syndication varying coefficient model. Because of singularity, it is not possible to estimate a syndication varying parameter for the constant or the syndication control variable. The results, as reported in Table 5, reveal that the syndication varying coefficients are both individually and jointly statistically insignificant. The insignificance of the syndication varying coefficients in Table 5 may at first seem contrary to our firm size hypothesis. Syndication, by its very nature, involves multiple banks. However, as noted in Dennis and Mullineaux (2000), the lead lender usually negotiates the deal with the borrower and then offers the loan for syndication. The significant agency considerations between the lead lender and participating banks, as described in Dennis and Mullineaux (2000), suggest that the lead lender has significant information advantages vis-a-vis the other participant banks. Hence, it is not surprising that the sensitivity of term loan maturity to its underlying determinants does not differ between syndicated and non-syndicated loans.