Debt maturity and the cost of bank loans

ARTICLE IN PRESS

JID: JBF

[m5G;November 2, 2017;7:1]

Journal of Banking and Finance 0 0 0 (2017) 1–22

Contents lists available at ScienceDirect

Journal of Banking and Finance journal homepage: www.elsevier.com/locate/jbf

Debt maturity and the cost of bank loans Chih-Wei Wang a, Wan-Chien Chiu b,∗, Tao-Hsien Dolly King c a

National Sun Yat-sen University, Department of Finance, College of Management, 70. Lianhai Rd., Kaohsiung 80424, Taiwan National Tsing Hua University, Department of Quantitative Finance, College of Technology Management, No. 101, Section 2, Kuang-Fu Road, Hsinchu, Taiwan 30013 c University of North Carolina at Charlotte, Department of Finance, Belk College of Business, Charlotte, NC 28223, USA b

a r t i c l e

i n f o

Article history: Received 2 September 2016 Accepted 13 October 2017 Available online xxx JEL classification: G12 G21 G31 G32 Keywords: Bank loan Short-term debt Debt maturity Rollover risk Asset substitution

a b s t r a c t This study explores the extent to which a firm’s debt maturity structure affects the cost of bank loans. By examining the U.S. syndicated loans from 1990 and 2014, we find that debt maturity structure is a major determinant of loan spreads, after accounting for firm- and loan-specific variables and firm and year fixed effects. A one standard deviation increase in the ratio of short-term debt to total assets is associated with an increase of 11.44 basis points in loan spread, representing an additional $0.644 million in interest expenses. The results support the rollover risk hypothesis, which predicts that short-term debts intensify the shareholder and bondholder conflicts and lead to greater credit risk. In addition, high-growth firms experience significantly smaller increases in their loan spreads than low-growth firms when the shortterm debt ratio increases. This finding is consistent with the asset substitution theory that short-term debt mitigates the managerial/shareholders’ risk-taking incentives, leading to a decrease in firm risk. Our results remain strong when we use alternative short-term debt proxies, address endogeneity concerns, and perform various robustness tests. © 2017 Elsevier B.V. All rights reserved.

1. Introduction A simple debt–equity choice cannot fully reflect a firm’s capital structure. Debt maturity is an important attribute of the debt structure that has received substantial attention. He and Xiong (2012) suggest that the rollover risk associated with shortterm debt intensifies the shareholder–debtholder conflict, in which shareholders are motivated to default earlier, leading to a higher probability of default. In contrast with the rollover risk explanation, short-term debt alleviates the asset substitution problem, as firms with more short-term debt are subject to more frequent renegotiations and closer scrutiny (Jensen and Meckling, 1976). Motivated by these theoretical predictions, we examine whether and how a firm’s debt maturity structure affects the cost of bank loans. We are particularly interested in bank loans based on the following reasons. First, most prior studies on debt maturity structure and cost of debt focus on public debt markets (e.g., Gopalan et al., 2014). Second, literature on debt mix suggests that private debt is a critical source of debt financing for corporations (e.g., see empir-

ical works by Denis and Mihov, 2003; Arena, 2011; Dhaliwal et al., 2004). For example, Khang et al. (2016) report that bank debt accounts for 56% of total debt outstanding for a typical U.S. company. Given the generally shorter maturity of bank loans, we conjecture that private debt issues are especially important to investigate, as the effect of short-term debt on the cost of debt should be more prominent for bank loans. If creditors consider the rollover risk, they are likely to demand a higher risk premium on loans to compensate for both credit and rollover risks (i.e., rollover risk hypothesis). If creditors take into account that short-term debt restrains managerial risk-seeking incentives, interest rates on loans are likely to be adjusted downward (i.e., asset substitution hypothesis). We test these two primary hypotheses using a large sample of syndicated loans in the U.S. market from 1990 to 2014. We first find strong evidence supporting the rollover risk hypothesis: a shorter-term maturity structure increases the cost of bank loans, and the impact is both economically and statistically significant. On average, a one-standard-deviation increase in the ratio of shortterm debt to total assets increases the loan spread by 11.44 basis points, which is equivalent to 5.66% of the average loan spread and $0.644 million in interest expenses.1 The result is robust

∗

Corresponding author. E-mail addresses: [email protected] (C.-W. Wang), [email protected] (W.-C. Chiu), [email protected] (T.-H. Dolly King).

1 Consistent with prior literature, we use all-in-drawn loan spreads to capture the overall cost of bank loans.

https://doi.org/10.1016/j.jbankfin.2017.10.008 0378-4266/© 2017 Elsevier B.V. All rights reserved.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF 2

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

after we control for various firm- and loan-specific variables, and for firm and time fixed effects.2 Our findings suggest that banks demand a premium on loan rates to compensate for the rollover risk in short-term debt that may intensify agency conflicts between shareholders and debtholders. In addition, we find support for the asset substitution hypothesis. Prior literature suggests that the incentives to shift investments toward risky assets are contingent on a firm’s growth opportunities. Because high-growth firms have the strongest incentives to engage in asset substitution behavior (e.g., Billett et al., 2007), the mitigating effect of short-term debt on loan costs should be most prominent for these firms. We find by increasing the short-term debt ratio by one standard deviation, the mean loan spread will go up by 2.05% for a high-growth firm but 8.62% for a low-growth firm. This result implies a strong tradeoff especially for high-growth firms in choosing a debt maturity structure: the rate-reducing effect on loan spread due to the alleviation of asset substitution problem significantly counteracts with the rate-increasing effect associated with the rollover risk. Lowgrowth firms may be better off choosing a longer debt structure, because the rate-increasing effect of short-term debt (rollover risk) dominates the rate-reducing effect (asset substitution). Additional tests offer a deeper understanding of the impact of the debt maturity structure on loan contract rates. Specifically, this effect is more pronounced in firms with higher firm risk, greater dependence on bank financing, speculative grade, and a greater use of credit lines. Moreover, short-term debts exert an amplifying effect on the credit cost to corporations, as well as the price paid by corporations to guarantee access to liquidity (i.e., all-in-undrawn fees). Relative to other studies, the endogeneity concern due to the simultaneity between short-term debt and the cost of bank loans is minimized in our analysis because loan spreads are set by a firm’s creditors based on the competitive forces in the market (i.e., observed outcomes rather than firms’ choices). Yet it is difficult to rule out the possibility that a set of unobserved factors affect both the short-term debt and the cost of bank loans. Our proposed causal relationship from short-term debt to bank loans may be also subject to reverse causality concerns. In particular, firms with poor credit quality are likely to face higher borrowing costs and thus are restricted from longer-term financing, resulting in a shorter debt maturity structure. To address these concerns, we first use the ratio of long-term debt maturing in one year to total assets as a proxy for short-term debt. This proxy substantially reduces the endogeneity concern because unlike short-term debt, long-term debt is predetermined and less likely to correlate with the firm’s current risk characteristics Almeida et al., 2012). The results based on this proxy are qualitatively similar to our main findings. In addition, we use a panel data model with firm and year fixed effects with clustered standard errors at the firm level to account for possible estimation biases, further reducing the endogeneity concern.3 Finally, we adopt a system of simultaneous equations to address the concern that the short-term debt ratio, loan spread, and leverage are simultaneously determined. The results support our conclusion that banks consider the effects of rollover risk and alleviation of asset substitution when determining loan rates. The primary contribution of this study is to provide new insights into loan pricing literature by highlighting that a firm’s debt maturity structure is a major determinant of bank loan spreads. 2 Our sample comes from the Loan Pricing Corporation’s Dealscan database, which focuses on large loans and firms that presumably suffer lower rollover risk compared with smaller firms. Thus, the database should bias against emphasizing the prevalence of monopolistic loan pricing behavior. 3 We consider two additional model specifications: ((1) ordinary least squares (OLS) regressions with standard errors adjusted for heteroskedasticity and withinfirm clustering and (2) random fixed effect models, in which we include industry dummies and clustered standard errors at the firm level. The results confirm our main findings.

Our findings complement recent empirical studies that document the amplifying mechanism of rollover risk on financing costs in public debt markets (Chen et al., 2012; Gopalan et al., 2014; Valenzuela, 2016). In addition, we show that banks recognize that shortterm debt helps mitigate asset substitution problems and charge lower interest rates. This finding advances the understanding of how short-term debt can alleviate the debt overhang problem (Johnson, 2003). Finally, previous studies focusing the association between the duration of incremental debt issues and bank loan costs show mixed results (see Appendix A for a summary of related literature). Our study offers new evidence that a firm’s overall debt maturity structure is more informative for predicting loan spreads than the duration of loan contracts.4 The rest of the paper is structured as follows. Section 2 presents the theoretical arguments for how a firm’s debt maturity structure affects the cost of debt. After detailing the data and variables in Section 3, we discuss the empirical results regarding the interplay of debt maturity and loan spreads in Section 4. Section 5 addresses endogeneity issues and presents the results of the robustness tests; Section 6 concludes. 2. Theoretical background and hypotheses To examine how a firm’s debt maturity structure affects the cost of bank loans, we present prior theories on the link between debt maturity structure and bank loan pricing. The rollover risk of short-term debt prompts banks to charge a higher premium in addition to the required credit premium. On the other hand, agency theory postulates that short-maturity debt reduces a firm’s riskseeking behaviors, so banks should charge lower interest rates on corporate loans. 2.1. Rollover risk He and Xiong (2012) argue that rollover risk can be a source of credit risk, because it intensifies the shareholder–debtholder conflict, in which shareholders bear refinancing costs, resulting in insolvency, even if the value of the firm’s assets is greater than the insolvency threshold without rollover risk. Gopalan et al. (2014) provide empirical evidence that firms with greater exposure to rollover risk are more likely to be downgraded than firms with similar risk characteristics. According to Wang et al. (2016), exposure to rollover risk increases the expected default probabilities of a company. Therefore, if creditors recognize the effect of rollover risk on a borrower’s credit worthiness, they would demand a higher risk premium to compensate for the increased credit risk. Prior studies support this conjecture through examinations of the pricing of credit default swaps (Chen et al., 2012) and corporate bonds (Gopalan et al., 2014; Valenzuela, 2016). Most studies in this literature focus on public debt and swap markets. We argue that the rollover risk, through which creditors demand a credit premium, is crucial for the pricing of private debt. Private debt issues are typically structured with shorter maturity than public debt issues, so the effect of rollover risk should be more prominent in the private debt market than it would be in the public debt market. Furthermore, the majority of private debt takes the form of a syndicated bank loan, which motivates us to consider the impact of debt maturity on the cost of bank loans. A firm with shorter debt maturity has a higher likelihood of refinancing, resulting in greater exposure to rollover risk and stronger interdependence between rollover risk and credit risk. This leads to a higher level of credit risk. If banks recognize this increase in 4 The incremental approach provides only noisy tests of agency theories of maturity choice (central to this study) that depend on gradually changing characteristics, such as asset lives and the investment opportunity set.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

credit risk and price the loans accordingly, firms with a shorter debt structure may incur a higher bank loan rate. On the basis of the foregoing discussion, we propose: Hypothesis 1. Firms with a shorter debt maturity structure pay a higher premium when obtaining bank loans. 2.2. Asset substitution In contrast, the asset substitution problem in agency theory suggests that short-term debt can reduce the cost of bank loans. Jensen and Meckling (1976) argue that shareholders prefer investments in risky projects, because their payoffs increase when a firm’s volatility increases. However, debtholders (i.e., fixed income claimants) may be negatively affected by riskier investments that increase the likelihood of default. The asset substitution problem emerges when shareholders have incentives to exploit bondholder wealth by replacing low-risk with high-risk investments. Shortterm debts may alleviate this asset substitution problem. The investment policies of firms with more short-term debts are closely monitored as they are subject to more frequent renegotiations and borrowers’ scrutiny (Jensen and Meckling, 1976). Substantial evidence confirms that banks demand higher loan interest rates in anticipation of the potential risks they might encounter in debt contracting (Bharath et al., 2008; Graham et al., 2008; Hasan et al., 2014). If banks recognize that short-term debts restrain managerial risk-seeking incentives, they price loans rationally, demanding lower interest rates on loans extended to firms with a shorter debt maturity in their capital structure. In addition, incentives to shift investments toward risky assets vary across firms, such that firms with more growth options have stronger incentives to engage in asset substitution behaviors (Johnson, 2003; Billett et al., 2007; Eisdorfer, 2008). Therefore, the mitigating effect of short-term debt on loan costs tend to be most prominent in high-growth firms, and we propose the following hypothesis: Hypothesis 2. High-growth firms pay lower loan spreads than low-growth firms, given a similar increase in short-maturity debts. 3. Sample data and variable construction 3.1. Sample construction We collect information on all U.S. syndicated loans, for the sample period 1990–2014, from the Dealscan LPC database,5 according to several screening processes. First, we exclude firms from highly regulated industries, including financial firms (standard industrial classification [SIC] codes 60 0 0–6999), utilities (SIC 490 0– 4999), and quasi-public firms (SIC greater than 8999). Second, we exclude privately held firms, because we require accounting and equity information to measure debt maturity and other firm characteristics. Thereafter, we merge the loan sample with data from Compustat and the Center for Research on Securities Prices (CRSP), using the conversion table provided by Chava and Roberts (2008). This process generates the initial sample of bank loans. For the cost of bank loans, we rely on all-in-drawn spreads (Spread) as the overall cost of the loan (Santos, 2011). We require nonmissing values of the main variables of interest, namely, Spread, short-term debt, and long-term debt that matures in a year. We also exclude observations with missing values for firm- and loanspecific variables. Our particular focus on unrated firms—which

5 The data in the Dealscan LPC database are more comprehensive after 1990, as Santos and Winton (2008) suggest, Dealscan’s coverage of the loan market improved markedly into the early 1990s, and loans from the 1980s may not be representative.

struggle to access public debt markets, rendering them very dependent on bank borrowing—leads us to exclude rated firms from our main analysis. A major premise of the rollover risk theory is that banks determine the costs of loans to borrowers, so we conjecture that there should be solid evidence for rollover risk theory among unrated firms.6 To minimize the effects of outliers, we winsorize Spread and all explanatory variables at the 1st and 99th percentiles. Our final sample contains 9941 loan facilities and 2754 unique firms. Fig. 1 presents Spread (solid line scaled to the left of the Y axis) and the total number of loans (dotted line scaled to the right of the Y axis). Spread exhibits a significant increase during the 2007–2010 financial crisis, as expected. The number of loans has increased steadily since 1992, except for the significant drop during the financial crisis. 3.2. Variable construction 3.2.1. Debt maturity structure This study investigates whether a firm’s debt maturity structure is associated with costs when the firm borrows from banks. Our hypotheses highlight the importance of short-term debt. We use the ratio of firms’ short-term debt to total assets (ST) as the main measure of their debt maturity structure.7 However, a concern with ST is that the level of short-term debt may be determined simultaneously with the cost of bank loans, or there may be unobserved risks or factors that affect both (ratio and cost), resulting in an endogeneity problem. Therefore, we consider the ratio of long-term debt that matures in a year to total assets (LT1AT) as a second proxy for debt maturity. Evidence suggests that LT1AT is appropriate for testing rollover risk theory (e.g., Almeida et al., 2012; Gopalan et al., 2014), because unlike short-term debt, long-term debt is predetermined and less likely to correlate with the firm’s current risk characteristics. Finally, we use the ratio of short-term debt to total debt as a third debt maturity proxy (STDEBT). 3.2.2. Control variables Consistent with loan contracting literature, we consider several firm- and loan-specific variables as determinants of loan spreads (Santos and Winton, 2008; Santos, 2011). The firm-specific variables include age and size (logarithm of total sales). Both variables should be negatively associated with loan spreads, because older and larger firms typically are well established and more diversified and therefore considered less risky. In addition, we use leverage as a control variable, because a higher level of firm leverage is associated with greater default risk and expected to have a positive effect on loan spreads. To measure a firm’s capability to service debt, we include profit margin and interest coverage. Improved profitability and a higher interest coverage ratio both indicate lower credit risk, so they should have negative effects on loan spreads. To examine the impact of credit risk, we control for the size and quality of the assets that debt holders can draw on, in the case of default. Tangible assets (tangibility) provide more satisfactory protection for debtholder wealth in the event of default, so they should have a negative effect on loan spreads. However, R&D and advertising proxy for a firm’s brand equity, which is less likely to protect debtholders from default loss, and they thus might relate positively to loan spreads. We include net working capital to reflect liquid assets, which help reduce value losses during default events, 6 In contrast, rated firms are less dependent on bank borrowing, because they have a broader access to funds; therefore, we expect weaker support for the rollover risk theory for these firms. We repeat the baseline analysis with rated firms and present the findings as additional evidence. 7 Short-term debt comprises all current liabilities, including loans, trade credits, and other current liabilities, with maturities of less than one year.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

3

ARTICLE IN PRESS

JID: JBF 4

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

400

800

350

700

300

600

250

500

200

400

150

300

100

200

50

100

0

0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

(Basis pionts)

Spreads

Number of loans

Fig. 1. Syndicated loans in the U.S. market during 1990–2014 The all-in-drawn spreads are indicated by the solid line scaled to the left on the Y axis, and the total number of loans is indicated by the dotted line scaled to the right on the Y axis, for the U.S. syndicated loan market from 1990 to 2014.

and we expect it to negatively affect loan spreads. The market-tobook (MTB) ratio is a proxy for firm growth, which we expect to relate negatively to loan spreads. In addition to these accounting-based measures, we control for two market-based risk indicators. Excess stock return represents a firm’s financial performance relative to the market, with an anticipated negative impact on loan spreads. Stock volatility measures stock return volatility, which is positively linked to default risk and therefore likely to affect loan spreads positively. Finally, we include a widely used forward-looking default risk indicator, distance-todefault, based on the KMV model,8 to proxy for the likelihood of a borrower’s default. A higher value indicates that a greater difference exists before the firm will reach its default threshold. We expect it to be negatively related to loan spreads. Finally, for the loan characteristics, we include log loan size and log loan duration. The effects of the two variables on loan spreads are ambiguous. A larger loan may lead to greater credit risk, but it also may facilitate processing and monitoring economies of scale. Similarly, loans with longer maturity are characterized by greater credit and term risks, but they also are more likely to be granted to borrowers with more favorable credit scores. We include dummy variables to indicate dividend restrictions, seniority, and security. Because the purpose of a loan may affect its spreads, additional dummy variables distinguish among loans that are for general corporate purposes, for repaying existing debt, or working capital. To include the type of the loan contract, we determine whether a loan is a term loan, bridge loan, or line of credit.9 In line with Santos and Winton’s (2008) proposal that the logarithm number of lenders can proxy for the hold-up, we include it and expect a negative effect on loan spreads. To account for the effects of any intertemporal economic shocks (Acharya et al., 2013), we include LIBOR (the London interbank offered rate). Detailed descriptions of all of the variables are in Appendix B.

3.3. Descriptive statistics Table 1 contains the descriptive statistics. The mean value of ST is 0.051; for an average firm, the amount of short-term debt is 5.1% of the total assets. As expected, LT1AT is lower than ST, with a 8 A detailed description of the KMV-Merton model is provided by Vassalou and Xing (2004). 9 Chava et al. (2008) suggest that the pricing of term loans may be very different from that of revolving loans, so we include dummy variables for each loan type.

mean value of 0.027, which indicates that maturing long-term debt accounts for nearly half of the short-term debt. The mean value of STDEBT is 0.25, so on average, one-quarter of debt matures in a year. Table 1 also reports the summary statistics for the loan characteristics. On average, bank loans have an issue size of $140.67 million, a spread of 202 basis points over the LIBOR rate, and a duration of 4 years. 59.7% of loan contracts are secured with collaterals and almost all loans are senior. Finally, term loans and credit lines together represent the majority loan contracts in the sample with about 25.7% and 65.3% respectively. Table 1 also reports summary statistics on other firm-level control variables used in the regression analysis. On average, firms have a leverage ratio of 26.8%, indicating that about a quarter of capital comes from debt financing. Profit margin and excess stock return are positive, and stock volatility is 47.6%.

3.4. Variable correlation Table 2 presents the correlations among loan spreads, firmspecific variables, and loan duration. The positive correlations between short-term debt ratios and loan spreads provide preliminary support for the rollover risk hypothesis. In particular, the correlations between Spread and short-term debt ratios range from 0.03 to 0.17, whereas the correlation between Spread and loan duration is 0.01. Despite extensive evidence of a strong link between loan duration and spreads, these preliminary results suggest that the balance sheet debt maturity structure may be more significant for determining the cost of bank loans than the incremental debt maturity reflected in the loan duration. Therefore, examining the impact of this balance sheet debt maturity structure on loan spreads is imperative. The correlations of leverage and the short-term debt variables also are relatively weak, suggesting that a firm’s capital structure does not completely explain its debt maturity structure. In the following sections, we thus investigate in-depth the effect of a firm’s debt maturity structure on the cost of its bank loans, after controlling for leverage and other risk factors. We also examine the correlations between the independent variables used in the regressions to ensure that our results are not subject to collinearity problems. Table 2 shows that most variables are not highly correlated with each other and in general, the signs of the correlations between Spread and other firm-specific variables are consistent with expectations.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

5

Table 1 Summary statistics. This table presents the summary statistics for the short-term debt variables and firm- and loan-level control variables. The short-term debt variables are ST, the ratio of short-term debt to total asset values; LT1AT, the ratio of long-term debts that mature within a year to total asset values; and STDEBT, the ratio of short-term debt to total debts. The variable Spread is the all-in-drawn spreads, representing banks’ overall borrowing costs over LIBOR. The detailed construction of the other variables is provided in Appendix B.

Short-term debt variables ST LT1AT STDEBT Firm characteristics Log age Log sales Leverage MTB (market-to-book) Profit margin Interest coverage Tangibility Net working capital R&D Advertising Excess stock return Stock volatility Distance-to-default Loan characteristics Spread (all-in-drawn spread) Log loan spread (all-in-drawn spread) All-in-undrawn spread Log loan size ($million) Loan size ($million) Log loan duration (months) Loan duration (years) Secure Senior Dividend rest Corporate purposes Debt repay Working capital Term loan Bridge loan Credit line Log number of lenders Macro controls LIBOR (%)

Obs.

Mean

Std. Dev.

p25

Median

p75

9941 9941 9941

0.051 0.027 0.253

0.086 0.053 0.311

0.003 0.001 0.023

0.020 0.009 0.116

0.058 0.031 0.361

9941 9941 9941 9941 9941 9941 9941 9941 9941 9941 9941 9941 8888

2.516 5.866 0.268 1.743 0.009 23.854 0.296 11.911 0.019 0.009 0.083 0.476 6.401

0.773 1.286 0.206 1.002 0.223 57.083 0.226 55.99 0.052 0.024 0.625 0.259 5.009

1.946 4.999 0.111 1.123 0.005 3.524 0.120 0.22 0 0 –0.264 0.304 2.934

2.485 5.884 0.237 1.451 0.035 7.221 0.233 0.834 0 0 0.058 0.414 5.330

3.091 6.733 0.377 2.001 0.070 17.481 0.417 2.147 0.013 0.005 0.389 0.575 8.556

9941 9941 6347 9941 9941 9941 9941 9941 9941 9941 9941 9941 9941 9941 9941 9941 9935

202.045 5.123 33.035 18.022 140.672 3.735 3.998 0.597 0.998 0.609 0.295 0.204 0.218 0.257 0.012 0.653 1.321

120.322 0.645 16.700 1.282 235.847 0.596 1.735 0.491 0.043 0.488 0.456 0.403 0.413 0.437 0.107 0.476 0.906

115 4.745 22.500 17.217 30 3.584 3 0 1 0 0 0 0 0 0 0 0.693

175 5.165 30 18.133 75 3.970 4.417 1 1 1 0 0 0 0 0 1 1.386

275 5.617 47.500 18.859 155 4.094 5 1 1 1 1 0 0 1 0 1 1.946

9941

3.791

2.247

1.559

4.765

5.623

4. Empirical results This section contains the results of univariate and multivariate analyses, as well as additional tests that we conduct to gain a more thorough understanding of the relation between the debt maturity structure and the costs of bank loans. 4.1. Univariate analysis of loan spreads and debt maturity structure Using a univariate analysis, we examine Spread across quartiles of short-maturity debt proxies (i.e., ST, LT1AT, and STDEBT). In a given year, firms can be classified into one of four quartiles; we report the mean and median Spread by quartile. Panel A of Table 3 contains the results for the entire sample. Spread monotonically increases with an increase in ST from the lowest to the highest quartile. The mean (median) comparison between the lowest and highest quartiles indicates a difference in loan spread of 43 basis points (62 basis points), which is significant at the 1% level. We observe similar patterns when using LT1AT or STDBET as alternative debt maturity proxies.10 These preliminary results support Hypothesis 10 Debt maturity literature considers relatively longer debt ratios as a proxy for short-term debts, such as ST3 (percentage of total debt that matures in less than 3 years) and ST5 (ratio of debts within 5 years to total debt) (see Datta et al., 2005; Billett et al., 2007; Brockman et al., 2010). However, short-term debts that mature

1, such that banks charge a higher loan rate for firms with shortermaturity debts due to their greater rollover risk. For loan duration, we observe a U-shaped pattern, suggesting that firms pay a lower interest rate when they obtain loans with an intermediate duration but pay a higher rate for loans with the shortest or longest durations. This result may explain why prior literature reveals ambiguous effects of loan durations on loan spreads. With regard to Hypothesis 2, we perform the same analysis on low-growth (Panel B) and high-growth (Panel C) firms. For lowgrowth firms, we continue to observe a strong, positive relationship between the short-term debt ratios and loan spreads for all proxies. However, for high-growth firms, this positive relation disappears for STDEBT and becomes weaker for ST, depending on

within a year are more appropriate proxies for our study, because we focus on unrated firms, whereas 3-year (or longer) debt proxies are probably more suitable for rated firms. The difference arises because loans are the main financing sources for unrated firms, and their duration typically is shorter than that of corporate bonds. Nevertheless, we acknowledge that no perfect debt maturity proxy exists. Therefore, we use ST3, ST5, and MAT (book-value weighted numerical estimate of debt maturity; see Appendix B for a detailed definition) as complementary measures and analyze the findings in Table 3 using these alternative proxies. The results are generally consistent with our benchmark debt maturity proxies, if somewhat weaker. For our study, the short-maturity proxies dominate the relatively longer short-term debt proxies. These results are not included in the main text but are summarized in Table OA1 of our Online Appendix.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)

(14)

(15)

(16)

(17)

(18)

1.00 0.67 0.59 −0.01 −0.09 0.31 −0.09 −0.09 −0.14 −0.12 −0.09 −0.05 −0.01 0.18 −0.03 −0.22 −0.11

1.00 0.32 0.00 −0.11 0.30 −0.07 −0.10 −0.13 −0.10 −0.01 −0.02 −0.01 0.15 −0.02 −0.18 −0.05

1.00 0.02 −0.05 −0.25 0.09 −0.02 0.16 0.21 −0.22 0.07 0.04 0.11 −0.01 0.03 −0.14

1.00 0.25 −0.11 −0.12 0.08 0.05 0.01 −0.05 0.02 −0.04 −0.16 0.01 0.18 0.02

1.00 −0.08 0.01 0.21 0.09 0.02 −0.15 −0.10 0.03 −0.23 −0.05 0.26 0.02

1.00 −0.14 −0.17 −0.35 −0.26 0.17 −0.15 0.00 0.13 −0.02 −0.38 0.05

1.00 0.03 0.29 0.14 −0.10 0.24 0.04 −0.01 0.10 0.28 0.00

1.00 0.14 0.02 −0.03 −0.29 −0.03 −0.24 0.02 0.18 0.06

1.00 0.40 −0.06 0.06 0.00 −0.10 −0.01 0.31 0.00

1.00 −0.12 0.14 0.03 −0.01 0.01 0.21 −0.01

1.00 −0.19 −0.07 −0.01 −0.02 −0.08 0.04

1.00 −0.01 0.09 0.04 0.07 −0.06

1.00 0.00 −0.02 0.00 0.00

1.00 0.24 −0.51 −0.14

1.00 0.18 0.04

1.00 0.05

1.00

ARTICLE IN PRESS

(1) 1.00 0.17 0.17 0.03 −0.09 −0.24 0.26 −0.19 −0.20 −0.13 −0.02 0.02 −0.01 0.02 0.34 0.01 −0.38 0.01

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

(1) Spread (2) ST (3) LT1AT (4) STDEBT (5) Log age (6) Log sales (7) Leverage (8) MTB (9) Profit margin (10) Interest coverage (11) Net working capital (12) Tangibility (13) R&D (14) Advertising (15) Stock volatility (16) Excess stock return (17) Distance-to-default (18) Log loan duration

JID: JBF

6

[m5G;November 2, 2017;7:1]

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017),

https://doi.org/10.1016/j.jbankfin.2017.10.008

Table 2 Correlation matrix. This table presents Pearson correlations of the variables used. All variables are defined in Appendix B.

ARTICLE IN PRESS

JID: JBF

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

7

Table 3 Mean and median loan spreads, categorized by short-term debt proxies. This table presents the Spread (basis points) across quartiles of short-maturity debt proxies (i.e., ST, LT1AT, and STDEBT) and new issuance loan duration (log loan duration). For each year, firms are classified into one of four groups. The means are reported, with the medians in brackets, among firms classified to these quartiles. Panel A presents results based on the full sample; Panels B and C present results for low- and high-growth firms, respectively. The lowgrowth (high-growth) firms are identified when firms’ MTB value is less than (greater than) the median of MTB ratios among all firms for a given year. We test the difference in means and medians between high and low quartile groups based on the Wilcoxon one-way sample t-test. ∗∗∗, ∗∗, and ∗ denote significance of the t-tests at the 1%, 5%, and 10% levels, respectively. Panel A: All firms Debt maturity variable quartiles

ST

LT1AT

STDEBT

Log loan duration

1 = Low

183.74 (162.5) 192.61 (175) 204.93 (200) 226.84 (225) 43.11∗ ∗ ∗ 62.5∗ ∗ ∗

177.92 (150) 193.54 (175) 207.64 (200) 228.95 (225) 51.02∗ ∗ ∗ 75∗ ∗ ∗

197.84 (175) 206.11 (200) 200.73 (182.5) 203.51 (185) 5.66∗ 10∗ ∗ ∗

210.54 (200) 196.18 (187.5) 179.99 (150) 229.12 (225) 18.58∗ ∗ ∗ 25∗ ∗ ∗

Debt maturity variable quartiles

ST

LT1AT

STDEBT

Log loan duration

1 = Low

206.34 (187.5) 209.31 (200) 226.25 (225) 247.49 (250)

202.62 (187.5) 215.03 (200) 225.29 (225) 246.4 (250)

215.16 (200) 221.39 (225) 220.61 (200) 232.2 (225)

232.5 (225) 212.52 (200) 207.89 (200) 242.54 (225)

41.15∗ ∗ ∗ 62.5∗ ∗ ∗

43.78∗ ∗ ∗ 62.5∗ ∗ ∗

17.04∗ ∗ ∗ 25∗ ∗ ∗

10.03∗ ∗ ∗ 0∗ ∗ ∗

Debt maturity variable quintiles

ST

LT1AT

STDEBT

Log loan duration

1 = Low

164.96 (150) 182.12 (150) 183.49 (152.5) 197.27 (175)

160.44 (125) 171.11 (150) 189.59 (175) 206.75 (192.08)

180.45 (150) 191.7 (175) 180.36 (165.63) 175.49 (150)

186.34 (175) 181.52 (175) 157.52 (137.5) 227.8 (200)

32.31∗ ∗ ∗ 25∗ ∗ ∗

46.31∗ ∗ ∗ 67.08∗ ∗ ∗

−4.97 0

41.46∗ ∗ ∗ 25∗ ∗ ∗

2 3 4 = High Two-sample differences tests High – Low (Mean) High – Low (Median) Panel B: Low-growth firms

2 3 4 = High Two-sample differences tests High – Low (Mean) High – Low (Median) Panel C: High-growth firms

2 3 4 = High Two-sample differences tests High – Low (Mean) High – Low (Median)

the magnitude of the difference in the mean (or median) spread between the highest and lowest quartiles. The non-significant or weaker relations between short-term debt ratios and loan spreads for high-growth firms implies the contrary influences of the negative effect based on an asset substitution argument and the positive effect based on the rollover risk explanation, in preliminary support of Hypothesis 2. The univariate results regarding the link between short-term debt ratios and loan spreads thus strongly support our hypotheses. Firms with more short-term debt pay higher loan spreads to banks. Furthermore, the results indicate that short-term debt reduces loan spreads by mitigating the asset substitution problem, particularly for high-growth firms. 4.2. Multivariate analysis of loan spreads and debt maturity structure We explore the relation between loan spreads and debt maturity structure in a multivariate framework, in which we regress

loan spreads on short-maturity debt proxies after controlling for firm- and loan-specific variables that represent crucial determinants of loan spreads. 4.2.1. Empirical methodology To investigate the impact of short-term debts on loan spreads, we estimate the following model:

Spreadi, j,t,d = c + β × STi,t−1 + Xi,t−1 + Yi, j,t,d + LIBORd , +F irm F ixed F f f ects + Year F ixed E f f ects + εi, j,t,d (1) where i, j, t, and d denote the ith firm and jth loan for year t and day d. Spread is the loan interest payment over LIBOR (i.e., all-indrawn spread) for a loan facility j of firm i on date d in year t. Then ST is our variable of interest. We also use two alternative short-term debt proxies: LT1AT (ratio of long-term debts maturing within a year) and STDEBT (ratio of short-term debt to total debt). Consistent with Hypothesis 1, a firm with a shorter debt maturity

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF 8

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

structure should be charged a higher loan spread, so we expect that β > 0. Furthermore, X represents a vector of firm-level control variables, and Y represents a vector of contemporaneous loan-level control variables that likely affect the loan spreads. All firm-level variables are measured at the fiscal year-end, immediately before the origination of the loan contract. We follow Santos (2011) and include firm fixed effects. The loan spread also might be affected by time fixed effects; certain unobserved factors systematically influence loan spreads across firms, at a specific time. To address this concern, we estimate the models by including year fixed effects. The LIBOR is used to control for macroeconomic conditions.11 Finally, we estimate all models with clustered standard errors at the firm level, as suggested by Petersen (2009). 4.2.2. Relation between short-term debt and the cost of bank loans Table 4 presents the regression results of the loan spreads on the short-term debt ratio and control variables. We estimate six models, featuring different combinations of short-term debt ratios and control variables. All the models include the firm and year fixed effects. We thus find that ST is positive and highly significant at the 1% level for the first four models; firms with a higher level of short-maturity debt pay a higher loan rate, after controlling for firm- and loan-level characteristics, in strong support of Hypothesis 1. The effect of ST on the cost of bank loans is both statistically and economically significant. To illustrate the economic impacts, we consider the results of Model 4, in which ST is the variable of interest and all control variables are included. According to the estimates in Model 4, a one-standard-deviation increase in ST leads to an increase in the loan spread by 11.44 basis points, which is 5.66% of the overall average loan spread of 202 basis points.12 In dollars, this effect is also substantial: A one-standard-deviation increase in ST results in an increase of $0.644 million in interest expenses, using an average loan size of $140.67 million and time to maturity of 4 years (i.e., $140.67 million × 0.001144 × 4). We also observe similar results with the two alternative shortterm debt proxies, LT1AT and STDEBT. The estimated coefficients are positive and highly significant (Models 5 and 6). The impact of LT1AT or STDEBT on Spread also is economically significant. A one-standard-deviation increase in LT1AT (STDEBT) increases loan spreads by 6.32 (4.41) basis points.13 Collectively, firms with a higher level of short-maturity debt pay a much higher loan spread when borrowing from banks.14 The short-term debt ratio ST thus appears to have a stronger impact on loan spreads than other elements identified in prior literature. As noted, increasing ST by one standard deviation prompts an estimated increase of 11.44 basis points in the loan spread. In contrast, the collective evidence gathered by Bharath et al. (2008), Francis et al. (2012), and Hasan et al. (2014, 2016) reveal that onestandard-deviation increases in accounting quality, board indepen-

11 With the LIBOR data, we consider the level of LIBOR in the month the firm initiates the loans. 12 The detailed calculation is as follows: The standard deviation of ST is 0.086 (see Table 1), and the estimated coefficient of ST in Model 4 of Table 4 is 133, so a one-standard-deviation increase in ST leads to an increase of Spread by 0.086 × 133 = 11.44 basis points. The mean value of Spread is 202 basis points (Table 1), so the percentage increase is 11.44/202 = 5.66%. 13 We reexamine the impact of short-term debts on loan spreads by replacing ST with the alternative short-maturity proxies ST3, ST5, and MAT in the baseline regression (Model 4, Table 4). The results are generally consistent with the main analysis; the coefficients of ST5 and MAT are significant and support the hypothesis. Detailed results are available in Table OA2 of the Online Appendix. 14 We also test our central hypotheses using a sample of rated firms. In Table OA3 of the Online Appendix, only the LT1AT short-maturity measure is significant, thereby confirming the weak relation as we conjecture for rated borrowers.

dence, cash effective tax rate, and social capital reduce bank loan spreads by 6.65, 5.50, 4.87, and 4.33 basis points, respectively. Among the firm-specific variables, the results are generally significant and consistent with expectations. First, firms with higher stock return volatility have greater default risk, leading to a positive effect on spreads. Firms that outperform the market and those with asset values above the default barrier instead are expected to pay lower loan spreads. Second, the firm size, leverage, profitability, and growth results are consistent with those of Santos (2011): Larger, less leveraged, more profitable, and high-growth firms pay significantly lower loan spreads. In certain models, firm age is positively related to loan spreads.15 The coefficients for the loan characteristics (loan amount, loan type, purpose dummies, and number of leaders) are generally significant. We focus here on loan duration, for two reasons. First, the duration of newly issued loans contributes to a firm’s debt maturity structure and represents the concept of incremental debt maturity. Second, literature on loan contracting generally accepts that loan duration is an essential determinant of loan spreads; however, its impact on the loan spread is ambiguous.16 The estimated coefficient of log loan duration is not significant across our models, implying that a firm’s overall debt maturity (measured by shortterm debt ratio proxies) is more informative than the incremental debt maturity (i.e., loan duration) for explaining loan spreads. Furthermore, LIBOR is negatively associated with spreads, consistent with Acharya et al. (2013) results. Overall, we thus find strong support for Hypothesis 1: The rollover risk that accrues to firms with a short-maturity debt structure is recognized and accounted for in corporate loan rates. 4.2.3. Effects of short-term debt on loan spreads, conditional on growth opportunities To elucidate how asset substitution theory can explain the relation between the debt maturity structure and loan spreads, we perform a regression analysis and examine whether the effect of rollover risk on loan spreads varies systematically with growth opportunities. If short-term debt mitigates the asset substitution problem, its net effect on spreads may result from the combination of the positive effect based on rollover risk and the negative effect based on alleviation of asset substitution. Asset substitution is most severe for high-growth firms, so we expect to observe significantly smaller effects of short-term debt on loan spreads for such firms. In contrast, the effect of short-term debt on spreads likely is due mainly to the rollover risk for low-growth firms, so we anticipate a strongly positive effect. To test these predictions, we use the MTB ratio as a proxy for a firm’s growth options. A dummy variable, High_MTB, identifies firms with a MTB ratio greater than the median value of all firms in a given year. To test Hypothesis 2, we modify the baseline model in Eq. (1) by adding High_MTB and STi , t − 1 × High_MTB. The interaction variable can test the possibility that the effect of short-term debt on loan spreads is conditioned on a firm’s growth opportunities. The model is thus as follows:

Spreadi, j,t,d = c + β1 × STi,t−1 + β2 × High_MT B + β3 ×(STi,t−1 × High_MT B ) + Xi,t−1 + Yi, j,t,d + LIBORd . +F irm F ixed F f f ects + Year F ixed E f f ects + εi, j,t,d (2) As previously noted, we expect β 3 to be negative, because for high-growth firms, the net effect of short-term debt on loan 15

Santos (2011) indicates that the log age variable is also positive. Loans with longer durations instead may face greater credit risk, and banks may charge higher spreads. Alternatively, banks may grant loans to firms that appear creditworthy, which produces a negative relationship (e.g., Santos, 2011; Goss and Roberts, 2011). 16

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

9

Table 4 Short-term debt and loan spreads. This table presents the results of regressing loan spreads on short-term debt ratios (ST, LT1AT, and STDEBT). The sample contains syndicated loans in the U.S. market from 1990 to 2014. We estimate models with or without control variables, and all models include firm and year fixed effects. In terms of ST, Model 1 is estimated without including any firm- and loan-level control; Model 2 includes only loan-level controls; Model 3 includes only firm-level controls; and Model 4 includes both loan- and firm-level controls and is our benchmark model. We replace ST in Model 4 with LT1AT and STDEBT and report the estimation results in Models 5 and 6, respectively. The p-values, reported in parenthesis, are obtained after considering clustered standard errors at the firm level. Indicator variables for year and firm fixed effects are not reported. ∗, ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. Model Short-term debt variables ST

(1)

(2)

(3)

(4)

235.85∗ ∗ ∗ (0.00)

211.24∗ ∗ ∗ (0.00)

149.26∗ ∗ ∗ (0.00)

133.10∗ ∗ ∗ (0.00)

(5)

134.12∗ ∗ ∗ (0.00)

LT1AT

19.25∗ ∗ ∗ (0.00)

STDEBT Firm-level characteristics Log age Log sales Leverage MTB Profit margin

−36.19∗ ∗

Interest coverage

0.01

Net working capital Tangibility

22.30

R&D Advertising

63.55

Stock volatility

48.35∗ ∗ ∗

Excess stock return

−1.31

Distance-to-default Loan-level characteristics Log loan size Log loan duration

5.46 (0.50) −30.18∗ ∗ ∗ (0.00) 55.03∗ ∗ ∗ (0.00) −11.98∗ ∗ ∗ (0.00) −36.77∗ ∗ (0.04) −0.01 (0.70) 0.07∗ ∗ (0.02) 23.51 (0.36) −77.91 (0.45) 39.82 (0.72) 50.09∗ ∗ ∗ (0.00) −3.72 (0.69) −3.23∗ ∗ ∗ (0.00)

−9.81∗ ∗ ∗ (0.00) −1.95

−1.28 (0.48) 34.60∗ ∗ ∗ (0.00) −180.91∗ ∗ ∗ (0.00)

Secure Senior Dividend rest

(6)

−9.42∗ ∗ ∗

−6.03∗ (0.01)

Corporate purposes

−32.05∗ ∗ ∗

Debt repay

−14.71∗ ∗ ∗

Working capital

−31.30∗ ∗ ∗

Term loan

45.03∗ ∗ ∗

Bridge loan

93.71∗ ∗ ∗

Credit line

5.92

Log number of lenders

−9.18∗ ∗ ∗

−27.14∗ ∗ ∗ (0.00) −16.69∗ ∗ ∗ (0.00) −26.20∗ ∗ ∗ (0.00) 39.23∗ ∗ ∗ (0.00) 82.52∗ ∗ ∗ (0.00) 3.76 (0.16) −8.10∗ ∗ ∗ (0.00) −5.53∗ (0.07)

LIBOR

4.43 (0.55) −19.39∗ ∗ ∗ (0.00) 62.64∗ ∗ ∗ (0.00) −9.92∗ ∗ ∗ (0.00) −37.87∗ ∗ (0.02) −0.01 (0.68) 0.06∗ ∗ (0.03) 18.46 (0.29) −33.49 (0.73) 28.21 (0.82) 51.77∗ ∗ ∗ (0.00) −4.29 (0.23) −2.38∗ ∗ ∗ (0.00)

4.66 (0.53) −20.13∗ ∗ ∗ (0.00) 71.53∗ ∗ ∗ (0.00) −9.77∗ ∗ ∗ (0.00) −37.71∗ ∗ (0.02) −0.02 (0.77) 0.06∗ ∗ (0.05) 18.01 (0.40) −39.36 (0.69) 44.13 (0.87) 50.58∗ ∗ ∗ (0.00) −4.11 (0.16) −2.38∗ ∗ ∗ (0.00)

−6.80∗ ∗ ∗ (0.00) −1.41 (0.62) 28.35∗ ∗ ∗ (0.00) −152.15∗ ∗ ∗ (0.01) −6.09∗ (0.08) −27.65∗ ∗ ∗ (0.00) −16.88∗ ∗ ∗ (0.00) −26.88∗ ∗ ∗ (0.00) 38.95∗ ∗ ∗ (0.00) 82.85∗ ∗ ∗ (0.00) 3.58 (0.37) −8.00∗ ∗ ∗ (0.00) −8.18∗ ∗ ∗ (0.00)

−6.78∗ ∗ ∗ (0.00) −1.30 (0.59) 28.48∗ ∗ ∗ (0.00) −153.56∗ ∗ ∗ (0.01) −6.14∗ ∗ ∗ (0.08) −27.50∗ ∗ ∗ (0.00) −16.74∗ ∗ ∗ (0.00) −26.53∗ ∗ ∗ (0.00) 39.28∗ ∗ ∗ (0.00) 82.06∗ ∗ ∗ (0.00) 3.88 (0.39) −8.25∗ ∗ ∗ (0.00) −7.95∗ ∗ ∗ (0.00)

4.39 (0.55) −20.15∗ ∗ ∗ (0.00) 88.66∗ ∗ ∗ (0.00) −9.88∗ ∗ ∗ (0.00) (0.02) (0.57) 0.04 (0.14) (0.41) −32.42 (0.74) (0.80) (0.00) (0.18) −2.40∗ ∗ ∗ (0.00) −6.82∗ ∗ ∗ (0.00) (0.62) 28.54∗ ∗ ∗ (0.00) −156.33∗ ∗ ∗ (0.01) (0.08) (0.00) (0.00) (0.00) (0.00) (0.00) (0.36) (0.00) −7.82∗ ∗ ∗ (0.00)

CONSTANT

132.31∗ ∗ ∗ (0.00)

534.48∗ ∗ ∗ (0.00)

264.82∗ ∗ ∗ (0.00)

546.67∗ ∗ ∗ (0.00)

553.24∗ ∗ ∗ (0.00)

547.56∗ ∗ ∗ (0.00)

Year fixed effects Firm fixed effects Observations R-squared

Yes Yes 9941 0.13

Yes Yes 9935 0.34

Yes Yes 8888 0.36

Yes Yes 8882 0.46

Yes Yes 8882 0.46

Yes Yes 8882 0.46

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF 10

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

Table 5 Effect of short-maturity debts on loan spreads by growth opportunities. This table presents the regression results for the effect of short-maturity debt variables on loan spreads, conditional on growth opportunities. The High_MTB dummy variable identifies firms as high-growth opportunity firms if their MTB values are greater than the median value of that variable for all firms in a given year. Control variables on firm- and loan-specific variables, firm and year fixed effects, and LIBOR in the month of the loan are included in all regressions, but these coefficients are not reported. The p-values, reported in parenthesis, are obtained after considering clustered standard errors at the firm level. ∗, ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. Model

(1)

(2)

(3)

ST

20 0.0 0∗ ∗ ∗ (0.00) −2.23 (0.61) −149.27∗ ∗ ∗ (0.00)

−3.17 (0.45)

−6.23 (0.17)

High_MTB ST × High_MTB

4.3. Additional tests

222.39∗ ∗ ∗ (0.00) −209.05∗ ∗ ∗ (0.00)

LT1AT LT1AT × High_MTB STDEBT STDEBT × High_MTB CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects # of observations R-squared

Yes Yes Yes Yes Yes 8882 0.46

facility by $0.92 million (= 120.8 × 0.001914 × 4). For high-growth firms, the same increase in ST raises the total interest expense per loan facility by a substantially lower amount of $0.24 million (= 160.3 × 0.0 0 0374 × 4). These results strongly support Hypothesis 2: High-growth firms experience two contradictory effects of short-maturity debt, and the net effect on loan spreads becomes nonsignificant. Yet for lowgrowth firms, the rollover risk effect outweighs the attenuation of the asset substitution problem, so the net effect of short-term debt on loan spreads is significantly positive.

Yes Yes Yes Yes Yes 8882 0.46

26.43∗ ∗ ∗ (0.01) −13.10 (0.25) Yes Yes Yes Yes Yes 8882 0.46

spreads is determined by the decreasing effect from the mitigation of the asset substitution problem. Conversely, β 1 should be significantly positive, as predicted in Hypothesis 1, mainly due to rollover risk, because low-growth firms have little or no incentive to use short-term debt to mitigate the asset substitution problem. In the regressions, we control for the direct effect of MTB (i.e. one of firm-level control variables represented in X), and thus we expect β 2 to be negative. We present the regression results in Table 5. For ST (Model 1), the coefficient of STi , t − 1 × High_MTB is negative and highly significant at the 1% level, whereas the coefficient of STi , t − 1 is positive and highly significant at the 1% level. Model 2 also suggests similar results when we use LT1AT. Model 3 that uses STDEBT as the short-term debt proxy is largely consistent with the results as in Models 1 and 2, although the coefficient of STi , t − 1 × High_MTB is not significant. The economic impact of ST on Spread is substantial for low-growth firms. Model 1 suggests that a one-standard-deviation increase in ST increases the loan spread by 19.14 basis points (= 0.0957 × 200), which is approximately 8.62% (= 19.14/222) of the average Spread.17 For a highgrowth firm, the same increase in ST is associated with an increase in Spread by 3.74 basis points (= 0.0738 × 50.73), which is approximately 2.05% (= 3.74/182) of the average value.18 In addition, the economic impacts of the high- and low-growth firms differ significantly. For low-growth firms, increasing ST by one standard deviation increases the total interest expense per loan

17 To compute the economic impacts in Eq. (2), instead of using the summary statistics for the entire sample (as in Table 1), we use separate summary statistics for low- and high-growth firms. The standard deviation of ST for low-growth (high-growth) firms is 0.0957 (0.0738), and the average Spread is 222 basis points (182 basis points). The loan size in the sample, on average, is $120.8 million for low-growth firms and $160.3 million for high-growth firms, and the time to loan maturity is 4 years for both firm types. 18 The total effect of STi , t − 1 on loan spreads is the summation of the estimated coefficient STi , t − 1 and the estimated coefficient STi , t − 1 ×High_MTB. Therefore, in this case, it is 200 + (–149.27) = 50.73.

Additional tests establish a more thorough understanding of the impact of the debt maturity structure on loan contract rates. In particular, we examine whether this impact is more pronounced among firms with higher risk, greater dependence on bank financing, speculative grades, and more committed credit lines. 4.3.1. Debt maturity, loan spreads, and firm risk In this section, we investigate how firm risk affects the link between short-term debt and the cost of bank loans. He and Xiong (2012) highlight that when a firm is sensitive to negative shocks and short-term debt accounts for a significant portion of its capital structure, an unfavorable event may lead to a large drop in liquid reserves, such that the firm suffers considerable refinancing losses in rolling over its short-term debt. With the same debt maturity structure, a high-risk firm likely faces greater rollover (and credit) risk than a low-risk firm. In addition, compared with less risky borrowers, riskier borrowers have stronger incentives to engage in asset substitution behaviors (Campbell and Kracaw, 1990).19 Thus, the use of short-term debt to reduce risktaking behaviors should be more effective for riskier firms, and banks in turn should charge lower loan rates. According to this argument, our baseline results should be more pronounced for highrisk than for low-risk firms. To test this prediction, we distinguish high-risk from low-risk firms according to a set of risk indicators. The first, STOCKVOLA50, equals 1 if a firm’s equity volatility is higher than the median of the sample firms in a given year and 0 otherwise. We also create three additional risk indicators using the Altman Zscore, distance-to-default, and interest coverage. In contrast with stock volatility, these variables are inversely related to the level of risk.20 For each indicator, we create a dummy variable (ZSCOREB50, DTD-B50, or INTCOVERAGE-B50) that assumes a value of 1 for firms for which the variable is lower than the median of all sample firms in a given year, to indicate high risk, and 0 otherwise. We replace ST in the baseline regression (Eq. 1) with two interaction terms: ST × Risk_dummy and ST × (1 − Risk_dummy), where the Risk_dummy variable equals 1 if the firm is identified as high-risk and 0 otherwise. The coefficient of ST × Risk_dummy should be positive and more significant than the coefficient of ST × (1 − Risk_dummy). Table 6 contains the estimation results for the three short-term debt ratios (ST, LT1AT, and STDEBT, in Panels A, B, and C, respectively). The coefficient of the interaction of the short-term debt

19 Campbell and Kracaw (1990) demonstrate that manager–equityholders’ incentives to substitute riskier assets relates to the level of observable risk in the firm. When observable and unobservable risks are positively correlated, increases (decreases) in observable risk generate incentives for manager–equityholders to increase (decrease) unobservable risk. Risker firms thus have more incentives to engage in risky asset substitution. 20 The interest coverage ratio indicates a firm’s capability to pay interest; a lower value of this ratio renders the firm’s debt more risky.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

11

Table 6 Effect of short-maturity debts on loan spreads by firm risk. This table presents the results of regressing loan spreads on short-term debt ratios (ST, LT1AT, and STDEBT), conditional on the firm risk level. The main variables of interest are ST × Risk_dummy and ST × (1 − Risk_dummy), in which Risk_dummy is a dummy variable, and a value of 1 indicates high-risk firms. We consider four risk indicators: STOCKVOL-A50, ZSCORE-B50, DTD-B50, and INTCOVERAGE-B50. Results of the tests of the differences between coefficients on the interaction terms are presented in the row titled Coef. We estimate models with firm- and loan-specific variables, firm and year fixed effects, and LIBOR. We only report the results of our main explanatory variables. The p-values, reported in parenthesis, were obtained by considering clustered standard errors at the firm level. ∗, ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. Panel A: ST Risk_dummy

ST × Risk_dummy ST × (1– Risk_dummy)

Coef. CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects #of observations R-squared

STOCKVOL-A50 (1)

ZSCORE-B50 (2)

DTD-B50 (3)

INTCOVERAGE-B50 (4)

171.92∗ ∗ ∗ (0.00) 81.09∗ ∗ ∗ (0.00) 90.83∗ ∗ (0.03) Yes Yes Yes Yes Yes 8882 0.46

161.79∗ ∗ ∗ (0.00) 18.82 (0.61) 142.97∗ ∗ ∗ (0.00) Yes Yes Yes Yes Yes 8882 0.46

148.39∗ ∗ ∗ (0.00) 70.94∗ ∗ (0.04) 77.45∗ ∗ (0.04) Yes Yes Yes Yes Yes 8882 0.46

168.71∗ ∗ ∗ (0.00) −25.93 (0.48) 194.64∗ ∗ ∗ (0.00) Yes Yes Yes Yes Yes 8882 0.47

STOCKVOL-A50 (1)

ZSCORE-B50 (2)

DTD-B50 (3)

INTCOVERAGE-B50 (4)

163.1∗ ∗ ∗ (0.01) 95.62∗ ∗ ∗ (0.01) 67.48 (0.28) Yes Yes Yes Yes Yes 8882 0.46

153.55∗ ∗ ∗ (0.00) 56.95 (0.22) 96.6∗ (0.10) Yes Yes Yes Yes Yes 8882 0.46

139.31∗ ∗ ∗ (0.00) 114.97∗ ∗ ∗ (0.01) 24.34 (0.64) Yes Yes Yes Yes Yes 8882 0.46

160.24∗ ∗ ∗ (0.00) 7.31 (0.90) 152.93∗ ∗ (0.02) Yes Yes Yes Yes Yes 8882 0.46

STOCKVOL-A50 (1)

ZSCORE-B50 (2)

DTD-B50 (3)

INTCOVERAGE-B50 (4)

24.73∗ ∗ ∗ (0.01) 14.79∗ ∗ (0.02) 9.94 (0.30) Yes Yes Yes Yes Yes 8882 0.46

44.35∗ ∗ ∗ (0.00) 0.96 (0.88) 43.39∗ ∗ ∗ (0.00) Yes Yes Yes Yes Yes 8882 0.47

27.51∗ ∗ ∗ (0.00) 11.79∗ (0.06) 15.72∗ (0.09) Yes Yes Yes Yes Yes 8882 0.46

48.64∗ ∗ ∗ (0.00) −2.70 (0.66) 51.34∗ ∗ ∗ (0.00) Yes Yes Yes Yes Yes 8882 0.47

Panel B: LT1AT Risk_dummy

LT1AT × Risk_dummy LT1AT × (1– Risk_dummy)

Coef. CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects #of observations R-squared Panel C: STDEBT

Risk_dummy

STDEBT × Risk_dummy STDEBT × (1– Risk_dummy)

Coef. CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects #of observations R-squared

ratio and high-risk dummy is greater than that of the low-risk dummy across all model specifications. The coefficients on the two interaction terms differ significantly from each other at the 10% (or lower) level in 9 of the 12 models. On the basis of the risk indicators, we divide the firms in our sample into high-risk and low-risk subsamples. We rerun individual regressions according to the model in Eq. (2); the results

are in Table 7. Across all risk indicators in the high-risk subsample, the coefficient of the interaction between short-term debt and High_MTB (i.e., additional effect of short-term debt on loan spreads for high-growth firms) is systematically negative and significant, except for the case when STDEBT is used as a proxy. However, the coefficient is not significant in majority of the models for the lowrisk sample.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF 12

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22 Table 7 Effect of short-maturity debt on loan spreads, depending on growth opportunity for high- and low-risk firms. This table presents the results of regressing loan spreads on short-term debt ratios (ST, LT1AT, and STDEBT), conditional on the growth opportunity and firm risk levels. The main variables of interest are ST × High_MTB and ST × (1 − High_MTB). The High_MTB is the dummy variable, identifying firms as high-growth opportunity firms when their MTB values are greater than the median value of the variable for all firms in a given year. We consider four risk indicators: STOCKVOL-A50, ZSCORE-B50, DTD-B50, and INTCOVERAGE-B50. We estimate models with firm- and loan-specific variables, firm and year fixed effects, and LIBOR. We only report the results of our main explanatory variables. The p-values in parenthesis were obtained by considering clustered standard errors at the firm level. ∗, ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. Panel A: ST STOCKVOL-A50

ST High_MTB ST × High_MTB CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects # of observations R-squared

ZSCORE-B50

DTD-B50

INTCOVERAGE-B50

High risk (1)

Low risk (2)

High risk (3)

Low risk (4)

High risk (5)

Low risk (6)

High risk (7)

Low risk (8)

279.83∗ ∗ ∗ (0.00) −3.38 (0.69) −230.52∗ ∗ ∗ (0.00) Yes Yes Yes Yes Yes 4399 0.37

−10.29 (0.80) −1.56 (0.75) 25.29 (0.59) Yes Yes Yes Yes Yes 4483 0.47

236.91∗ ∗ ∗ (0.00) 6.96 (0.43) −139.65∗ ∗ (0.02) Yes Yes Yes Yes Yes 4303 0.39

65.06 (0.20) 0.34 (0.95) −26.94 (0.63) Yes Yes Yes Yes Yes 4579 0.44

220.66∗ ∗ ∗ (0.00) −3.00 (0.75) −185.06∗ ∗ ∗ (0.00) Yes Yes Yes Yes Yes 4446 0.33

56.60 (0.27) −0.71 (0.86) −17.93 (0.78) Yes Yes Yes Yes Yes 4436 0.47

187.70∗ ∗ ∗ (0.00) −1.10 (0.90) −148.56∗ ∗ (0.01) Yes Yes Yes Yes Yes 4395 0.41

−49.78 (0.49) −5.48 (0.22) 94.06 (0.23) Yes Yes Yes Yes Yes 4487 0.44

Panel B: LT1AT STOCKVOL-A50

LT1AT High_MTB LT1AT × High_MTB CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects # of observations R-squared

ZSCORE-B50

DTD-B50

INTCOVERAGE-B50

High risk (1)

Low risk (2)

High risk (3)

Low risk (4)

High risk (5)

Low risk (6)

High risk (7)

Low risk (8)

246.15∗ ∗ (0.02) −5.69 (0.49) −326.32∗ ∗ ∗ (0.01) Yes Yes Yes Yes Yes 4399 0.38

54.03 (0.40) −0.46 (0.93) −4.83 (0.95) Yes Yes Yes Yes Yes 4483 0.47

198.10∗ ∗ ∗ (0.01) 6.86 (0.44) −162.83∗ ∗ (0.04) Yes Yes Yes Yes Yes 4303 0.40

252.78 (0.00) 1.86 (0.69) −152.64 (0.12) Yes Yes Yes Yes Yes 4579 0.44

210.42∗ ∗ ∗ (0.00) −4.02 (0.66) −260.01∗ ∗ ∗ (0.00) Yes Yes Yes Yes Yes 4446 0.34

204.38 (0.01) 0.31 (0.94) −78.98 (0.40) Yes Yes Yes Yes Yes 4436 0.46

183.44∗ ∗ ∗ (0.01) −4.30 (0.62) −144.68∗ (0.07) Yes Yes Yes Yes Yes 4395 0.42

99.56 (0.34) −1.87 (0.67) −46.39 (0.70) Yes Yes Yes Yes Yes 4487 0.43

Panel C: STDEBT STOCKVOL-A50

STDEBT High_MTB STDEBT × High_MTB CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects # of observations R-squared

ZSCORE-B50

DTD-B50

INTCOVERAGE-B50

High risk (1)

Low risk (2)

High risk (3)

Low risk (4)

High risk (5)

Low risk (6)

High risk (7)

Low risk (8)

43.68∗ ∗ (0.02) −9.04 (0.31) −26.11 (0.20) Yes Yes Yes Yes Yes 4399 0.37

−9.23 (0.31) −4.30 (0.40) 20.99∗ (0.05) Yes Yes Yes Yes Yes 4483 0.47

50.95∗ ∗ ∗ (0.00) −1.31 (0.89) 3.93 (0.86) Yes Yes Yes Yes Yes 4303 0.40

1.44 (0.87) −0.85 (0.87) 1.14 (0.91) Yes Yes Yes Yes Yes 4579 0.44

39.29∗ ∗ (0.02) −7.72∗ ∗ (0.41) –31.65∗ (0.09) Yes Yes Yes Yes Yes 4446 0.34

7.34 (0.46) −1.75 (0.68) 2.19 (0.85) Yes Yes Yes Yes Yes 4436 0.47

44.01∗ ∗ (0.01) −4.28 (0.63) –31.93 (0.14) Yes Yes Yes Yes Yes 4395 0.42

−7.40 (0.47) −6.42 (0.18) 17.11 (0.15) Yes Yes Yes Yes Yes 4487 0.44

Overall, these results support our predictions: Given the same increase in the short-term debt ratio, the increase in loan spread is greater for high-risk firms than for low-risk firms. The mitigating effect of short-term debts on the costs of bank loans for highgrowth firms is more pronounced among high-risk firms. 4.3.2. Bank dependence According to Hypothesis 1, banks perceive borrowers’ debt maturity structure and decide which interest rates to charge, driven

mainly by supply side effects. Hale and Santos (2009) argue that banks are able to hold up borrowers by charging higher interest rates due to private information the banks possess. Therefore, the amplifying effect of short-term debt on loan spreads should be more pronounced for firms that depend strongly on bank debt financing. To identify bank-dependent firms, we collect information from the Capital IQ database. For each firm, we compute the bank debt to total assets ratio, then classify a firm as bank dependent if its ratio is greater than the median ratio for all firms in a given

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

13

Table 8 Bank dependence. This table presents the regression results for the Capital IQ-based sample, over the sample period 2002 to 2014; the sample size is 3557 unrated firms and 3669 speculative grade firms at the loan level. The dummy variable Bank_Dep equals 1 if the firm’s ratio of bank debt to total assets exceeds the median value of the ratio, and 0 otherwise. This dummy variable is annually updated. We examine unrated firms in Columns 1–3 and speculative grade firms in Columns 4–6. Results of the tests of the differences between coefficients on the interaction terms are in the row titled Coef. ∗, ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively.

ST × Bank_Dep ST × (1–Bank_Dep)

Unrated firms

Unrated firms

Unrated firms

Speculative grade firms

Speculative grade firms

Speculative grade firms

(1)

(2)

(3)

(4)

(5)

(6)

293.31∗ ∗ ∗ (0.00) 175.89 (0.18)

125.97∗ (0.05) −113.79∗ (0.09) 300.06∗ ∗ ∗ (0.00) 86.66 (0.45)

LT1AT × Bank_Dep LT1AT × (1–Bank_Dep) STDEBT × Bank_Dep STDEBT × (1–Bank_Dep)

Coef. CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects # of observations R-squared

117.42 (0.41) Yes Yes Yes Yes Yes 3557 0.41

213.40 (0.14) Yes Yes Yes Yes Yes 3557 0.42

155.10∗ (0.09) –94.10 (0.30) 53.98∗ ∗ ∗ (0.00) 4.20 (0.73) 49.78∗ ∗ ∗ (0.01) Yes Yes Yes Yes Yes 3557 0.42

year. The dummy variable Bank_Dep equals 1 if a firm is bank dependent and 0 otherwise. The Capital IQ database provides reliable information only since 2002; the sample period for this analysis therefore is 2002–2014 (hereafter, the CIQ-based sample), which features 3557 observations. Columns 1–3 of Table 8 contain the regression results related to the rollover risk effect, conditional on bank dependence. The coefficients of the interaction variables between short-term debt and Bank_Dep are highly significant at the 1% level; the coefficients of the interaction variables between short-term debt and (1 – Bank_Dep) are not significant. These results clearly indicate that the rollover risk effect is more pronounced for bank-dependent firms. Given the same increase in short-term debt, bank-dependent borrowers pay higher interests to banks than borrowers that are less bank dependent, in further support of Hypothesis 1.21 4.3.3. Speculative grade firms Although we focus on unrated firms in our main analysis, rated firms are also likely to require bank financing. Thus rollover risk may also affect the loan spreads for rated firms. We compute the

21

We rerun our baseline regressions using the CIQ-based sample (starting from 2002) and confirm that the results are consistent with the main findings. Remarkably, the amplification effect of short-term debts on loan spreads is strongly positive for all short-term debt proxies and more prominent than the baseline regression results in Table 4. For example, in the CIQ-based sample, the coefficients of ST, LT1AT, and STDEBT are 252, 223, and 24, whereas in the main sample, the values are 133, 134, and 19, respectively (Model 4, Table 4). These results also might reflect the increasingly significant amplification effect of the short-term debt structure on the cost of bank loans in recent years (2002–2014). The economic impact is more sizable than that revealed by the results of the main analysis too. A one-standarddeviation increase of the short-term debt proxy leads to an increase in loan spreads by 21, 12, and 8 basis points when we use ST, LT1AT, and STDEBT, respectively; in the main analysis, it increases loan spreads by 11, 6, and 4 basis points, respectively. In addition, the results confirm that the reduction effect of short-term debt on the loan rate due to the alleviation of asset substitution is more significant for highgrowth firms. The difference in the coefficient between ST × High_MTB and ST × (1 – High_MTB) is –207 in the CIQ-based sample, versus –156 in the main sample. For LT1AT, the coefficient difference is –329 in the CIQ-based sample and –224 in the main sample. Finally, for STDEBT, the difference is –32 versus –19. These results are available in Table OA4 of the Online Appendix.

239.76∗ ∗ ∗ (0.01) Yes Yes Yes Yes Yes 3669 0.44

249.20∗ ∗ (0.05) Yes Yes Yes Yes Yes 3669 0.44

57.47∗ ∗ ∗ (0.01) −21.35 (0.44) 78.82∗ ∗ (0.03) Yes Yes Yes Yes Yes 3669 0.45

ratios of bank debt to total assets for the unrated sample, the speculative grade sample (i.e., firms with ratings lower than BBB − ), and the investment grade sample. The distributions are plotted in Fig. 2. 22 As expected, unrated firms have the highest ratio of bank debt to total assets. A similar pattern is observed in the speculative grade subsample, indicating some demand for bank financing by these firms. Investment grade firms reveal minimal ratios of bank debt to total assets. Consequently, we focus on speculative grade firms to conduct our analysis of rated firms. The bank dependence analysis, using the speculative grade subsample, produces the regression results in Columns 4–6 of Table 8. The coefficient of ST × Bank_Dep (i.e., bank-dependent firms) is positively significant; the coefficient of ST × (1 – Bank_Dep) is negatively significant (Column 4). In addition, the coefficient difference between the two interaction variables (Coef.) is highly significant at the 5% (or better) level. When we replace ST with LT1AT or STDEBT, we find similar results (Columns 5–6): Bank-dependent, speculative grade firms pay significantly higher interest on bank loans when they have more shorter debts in further support of Hypothesis 1. 4.3.4. All-in-undrawn spreads and credit lines Unlike all-in-drawn spreads, undrawn fees include both the commitment and annual fees that the borrower must pay to banks for funds committed for credit lines but not taken. That is, undrawn fees compensate banks for the liquidity risks they incur by guaranteeing borrowers access to funding, over the life of the credit lines. Therefore, the rollover risk hypothesis should hold when we consider undrawn fees. We rerun the Hypothesis 1 tests by replacing all-in-drawn spreads with undrawn fees (i.e., all-inundrawn spread in the Dealscan database). The results, presented in Panel A of Table 9, show that the coefficients of ST, LT1AT, and STDEBT are positive and highly significant. That is, the short-term

22 The data we use in this part of the analysis came from the Capital IQ database, so the sample period for these subsamples is 2002 to 2014. Of total loan level observations, 3949 are unrated firms, 4183 are speculative grade firms, and 2330 are investment grade firms.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF 14

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

0.6

Distribution 0.5

Percent

0.4

0.3

Unrated Speculative-grade

0.2

Investment-grade 0.1

0 0

5

10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100

Bank debt to total assets ratio (%) Fig. 2. Distribution of bank debt to total assets ratio This figure presents the distribution of the ratio of bank debt to total assets, based on the Capital IQ data set. The sample period is from 2002 to 2014. The sample contains 3949, 4183, and 2330 loan-level observations for unrated, speculative grade, and investment grade firms, respectively. The black line represents the distribution for the unrated subsample; the gray line represents the distribution for the speculative grade subsample; and the dotted line represents the distribution for the investment grade subsample.

debt ratio has an amplifying effect on not only the cost of credit to corporations but also the cost they pay to guarantee their access to liquidity. Furthermore, all-in-drawn spreads on credit lines compensate the bank for the credit risk it incurs when the borrower draws down its credit line in the future. The rollover risk hypothesis postulates that the conflict between shareholders and debtholders will increase the likelihood of default in the future, though not necessarily at the current time. Given a similar increase in short-term debt, if all-in-drawn spreads are significantly higher for credit lines than for other types of loans, Hypothesis 1 would be further supported. We examine this prediction by including two interaction terms in which the debt maturity proxy interacts with the dummy CREDITLINE and (1 − CREDITLINE) variables. The results, in Panel B of Table 9, confirm our prediction. 5. Endogeneity concerns and robustness tests 5.1. Endogeneity The empirical results indicate a strong, consistent association between short-term debt and loan pricing. However, similar to other empirical studies, our findings are vulnerable to endogeneity concerns. For example, firms with higher borrowing costs likely are restricted to using longer maturity debt, indicating a reverse causality problem. The loan spreads and short-term debt also may be determined simultaneously, by unobserved risk factors. Before conducting our checks, we note that, relative to other studies, the simultaneity issue is minimal in our research, because the loan spreads are set by the firms’ creditors in reaction to competitive market forces (i.e., they are observed outcomes rather than firm choices). Furthermore, we include firm and time fixed effects in our regressions to control for the time-invariant and time-varying factors that may affect both the debt maturity structure and loan spreads. However, completely eliminating endogeneity biases in empirical studies is unlikely. Therefore, we use a system simultaneous equations model (SEM) approach to reduce potential concerns about reverse causality and the simultaneous determination of loan spreads, debt maturity, and leverage.

To start, we apply the SEM model to the consolidated sample (i.e., firm-year observations), because the short-maturity debt variables and leverage are measured at the firm level. Consistent with Graham et al. (2008), we construct the consolidated sample by aggregating loan facilities to deals, using loan size–weighted averages of the relevant terms, comprising the 5946 firm-year observations. We expand the loan spread equation model by adding the debt maturity and leverage equations, as follows:

Spreadi,t = α10 + α11 × STi,t + Xi,t−1 + LIBORt (3) +Indust ry F ixed F f f ect s + Year F ixed E f f ects + εi,t STi,t = α20 + α21 × Spreadi,t + α22 × leveragei,t + α23 × ASSET _MATi,t + α24 × (stock vol atil ity)i,t + α25 × (l og sal es )i,t + α26 ×LSALES_squaredi,t + α27 × MT Bi,t + α28 × (excess stock return )i,t +LIBORt + Indust ry F ixed F f f ect s + Year F ixed E f f ects + εi,t (4) Leveragei,t = α30 + α31 × Spreadi,t + α32 × STi,t + α33 ×(P ro f it margin )i,t +α34 ×(stock vol atil ity)i,t +α35 × (l og sal es )i,t +α36 × MT Bi,t + α37 × (excess stock return )i,t + α38 ×F IX ED_ASSE Ti,t + Indust ry F ixed F f f ect s + Year F ixed E f f ect s +εi,t (5) where Spreadi , t is the weighted average of all-in-drawn spreads, based on the loan size in a given year and for a given firm. In addition, X represents a vector of firm-level control variables, as described for Eq. (1). Then ASSET_MAT is the measure of asset maturity, LSALES_squared is the square value of log sales, FIXED_ASSET represents a firm’s fixed assets, and the other variables are as defined in Appendix B. To capture industry fixed effects, we use single-digit SIC dummies, consistent with Acharya et al. (2013). We follow Johnson (2003) (see also Datta et al., 2005; Billett et al., 2007) to estimate leverage and debt maturity jointly. To estimate the SEM model, we use a generalized method of moments (GMM), with the exogenous variables as instruments in the moment conditions. The GMM ensures that the standard

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22 Table 9 All-in-undrawn spreads and credit lines. This table presents the benchmark regression results using all-in-undrawn spreads as the dependent variable (Panel A), instead of the all-in-drawn spreads in the main analysis. All-in-undrawn spreads refer to undrawn fees and include both commitment and annual fees that the borrower must pay the bank for funds committed for the credit line but not taken. This table also presents regression results that examine whether the effect of short-maturity debt variables on loan spreads is more pronounced for credit lines (Panel B). CREDITLINE is a dummy variable equals1 when the loan type is a credit line, and 0 otherwise. Results of the tests of the differences between coefficients on the interaction terms are in the row titled Coef. ∗, ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. Panel A: All-in-undrawn spreads (1) ST

(2)

(3)

13.74∗ ∗ ∗ (0.00) 15.22∗ ∗ (0.02)

LT1AT STDEBT CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects # of observations R-squared

Yes Yes Yes Yes Yes 5787 0.29

Yes Yes Yes Yes Yes 5787 0.3

2.60∗ ∗ ∗ (0.01) Yes Yes Yes Yes Yes 5787 0.3

Panel B: Credit lines (1) ST × CREDITLINE ST × (1–CREDITLINE)

(2)

153.76∗ ∗ ∗ (0.00) 99.85∗ ∗ (0.02) 180.62∗ ∗ ∗ (0.00) 54.2 (0.29)

LT1AT × CREDITLINE LT1AT × (1–CREDITLINE) STDEBT × CREDITLINE STDEBT × (1–CREDITLINE)

Coef. CONSTANT Firm variables Loan variables (except “Credit line”) Year fixed effects Firm fixed effects # of observations R-squared

(3)

53.91∗ ∗ (0.03) Yes Yes Yes Yes Yes 8882 0.46

126.42∗ ∗ ∗ (0.00) Yes Yes Yes Yes Yes 8882 0.46

21.34∗ ∗ ∗ (0.00) 13.9 (0.11) 7.44 (0.27) Yes Yes Yes Yes Yes 8882 0.46

errors of the estimates are heteroskedasticity and autocorrelation consistent.23 A two-equation SEM, which includes the loan spread and the debt maturity equations, produces the results in Table 10. 24 Shortterm debts and loan spreads exhibit a significant, bi-directional relation. The amplifying effect of short-term debt on the cost of bank loans remains robust after accounting for endogeneity. We perform 23 We do not report the R2 values for our estimated equations, because as Goldberger (1991) observes, R2 values reported in system estimation techniques might not fall between 0 and 1. Unfortunately, there is no widely accepted goodness-of-fit measure for nonlinear system estimates. Other instrumental variable techniques, such as two-stage least squares (2SLS), are special cases of GMM. For example, Greene (2002) and Kennedy (2003) report that, compared with 2SLS estimates, GMM estimates are more efficient when regression errors are heteroskedastic and/or autocorrelated and that GMM estimates coincide with 2SLS estimates otherwise. 24 We present results from the second-stage regression estimation. For the spread regression, the result is based on the regression using loan spreads as the dependent variable, with predicted short-term debt variables estimated from the first stage as one of the control variables. The first-stage regressions used to generate the estimated values of short-term debt variables are not reported.

[m5G;November 2, 2017;7:1] 15

a three-equation SEM by adding the leverage equation; again, we observe a positive and significant bi-directional relation between the short-term debt ratios and loan spreads. In the leverage equation, the coefficients for the short-term debt variables are negative and significantly different from 0, consistent with the liquidity risk effect suggested by Johnson (2003) and the single-equation results offered by Barclay and Smith (1995), who show that firms with longer maturity debt have higher leverage. In the short-maturity debt equation, the proportion of short-term debt also relates positively to the MTB ratio; this result is consistent with the predicted positive relation documented by Barclay et al. (2003). Finally, the coefficients for the other variables in the leverage and maturity equations are generally consistent with those reported by Johnson (2003) and Barclay et al. (2003).25 We use the SEM framework to examine whether short-term debt ratios are positively associated with the cost of bank loans. This framework controls for the possible effects of unobservable factors and a potential reverse causality bias. Our results strongly confirm that short-term debt ratios are positively associated with loan spreads, reinforcing our conclusion that a firm’s short-maturity debt structure is significant for determining the cost of bank loans. 5.2. Robustness checks 5.2.1. Alternative model specifications To estimate our baseline regressions, we use the panel data model with time and firm fixed effects, but for robustness, we consider other model specifications in this section. First, we estimate the baseline model with pooled ordinary least squares (PoolOLS) regressions, using standard errors adjusted for heteroskedasticity and firm clustering, while also adjusting for the industry fixed effect. Second, we adopt a random fixed effect model, in which we include industry dummies and clustered standard errors at the firm level. Table 11 presents the results; the estimated coefficients are systematically significant for all debt maturity proxies and models, implying that our results are robust to various model specifications. 5.2.2. Largest facility and consolidated sample In our main analysis, each observation represents a single loan facility. Yet a deal package could contain multiple loan facilities, which might reflect the deal structure, such that they are not entirely independent observations. Regarding these loans as independent facilities could inflate the statistical significance of our results, so we conduct two tests. First, we use the largest facility that the borrower receives in a specific year, as suggested by Hertzel and Officer (2012) and Houston et al. (2014). The sample thus decreases from 9941 to 5940. According to the results in Table 12, Panel A, increasing ST by one standard deviation leads to an increase in loan spreads by 4.98% (or 10.01 basis points). This coefficient is significant at the 1% level, and the economic impact is similar to the increase of 5.66% (or 11.44 basis points) in the baseline analysis in Model 4 of Table 4. Panels B and C confirm that the results based on LT1AT and STDEBT also are robust. Second, we follow Graham et al. (2008) and aggregate loan facilities to deals using loan size–weighted averages of the relevant terms. This consolidated sample contains 5946 firm-year observations. The short-term debt proxies are positive and highly

25 We also reexamine Hypothesis 2 with an SEM approach. The results continue to support our prediction that short-term debt is crucial for alleviating asset substitution problem and that banks perceive this effect and charge lower interest rates. The detailed results are available in Table OA5 of the Online Appendix.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF 16

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22 Table 10 SEM model: Consolidated sample using firm-year observations. This table presents the results of a SEM that includes the loan spread, debt maturity, and leverage equations (Eqs. 3, 4, and 5, respectively). The two-equation SEM includes the loan spread and shortmaturity debt equations. The three-equation SEM includes all three equations. The model is applied on the consolidated sample (firm-year sample). We estimate the SEM with a GMM, using the exogenous variables as instruments in the moment conditions. The GMM estimation method ensures that the standard errors of the estimates are heteroskedasticity and autocorrelation consistent. ∗, ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. Panel A: ST Two-equation system Spread Spread ST Leverage ASSET_MAT

0.0 0 07 (0.00) 180.7599 ∗ ∗ ∗ (0.00) 75.4510 ∗ ∗ ∗ (0.00) −0.0 0 03

Three-equation system

ST

Spread ∗∗∗

−0.0265 (0.11)

Log sales

−5.1682∗ ∗ ∗ (0.00) −22.3232∗ ∗ ∗ (0.00)

LSALES_squared

−0.0023 (0.77) 0.0015∗ ∗ (0.01)

170.6869∗ ∗ ∗ (0.00) 132.2358∗ ∗ ∗ (0.00) 0.0 0 01

Profit margin Interest coverage Net working capital Tangibility R&D Advertising Stock volatility Excess stock return Distance-to-default LIBOR CONSTANT Year fixed effects Industry fixed effects

−9.3990 ∗ ∗ ∗ (0.00) −15.5945∗ ∗ (0.04) −0.0482∗ ∗ ∗ (0.00) 0.0269∗ (0.07) −10.9122∗ (0.07) −28.6297 (0.19) 135.2348∗ ∗ ∗ (0.00) 84.6584∗ ∗ ∗ (0.00) −7.1017∗ ∗ ∗ (0.00) −2.6777∗ ∗ ∗ (0.00) −4.0443 (0.14) 261.6121∗ ∗ ∗ (0.00) Yes Yes

0.0041∗ ∗ (0.03)

−0.0578 ∗ ∗ ∗ (0.00) 0.0078∗ ∗ ∗ (0.01)

0.0062∗ (0.06) −0.0803∗ (0.08) Yes Yes

∗∗∗

0.0059∗ ∗ ∗ (0.00) −0.8583∗ ∗ ∗ (0.00)

−0.1485∗ ∗ ∗ (0.00) (0.64)

−2.2058∗ ∗ ∗ (0.00) −22.6560∗ ∗ ∗ (0.00)

FIXED_ASSET MTB

Leverage

0.0015 (0.00)

(0.16) Log age

ST

−10.7546∗ ∗ ∗ (0.00) −13.3022∗ ∗ ∗ (0.01) −0.0150∗ ∗ ∗ (0.00) 0.0293∗ ∗ ∗ (0.00) −8.9218∗ ∗ (0.01) −1.1955 (0.63) 29.9387∗ ∗ ∗ (0.00) 113.7636∗ ∗ ∗ (0.00) −13.0226∗ ∗ ∗ (0.00) −0.9871∗ ∗ ∗ (0.00) −1.3534∗ ∗ ∗ (0.00) 188.5836∗ ∗ ∗ (0.00) Yes Yes

0.0347∗ ∗ (0.02) −0.0 0 01 (0.95) 0.0881∗ ∗ ∗ 0.0138∗ ∗ ∗ (0.00)

0.1389∗ ∗ ∗ (0.00)

(0.00) 0.0634∗ ∗ ∗ (0.00) 0.0743∗ ∗ (0.02)

−0.1610∗ ∗ ∗ (0.00) 0.0214∗ ∗ ∗ (0.00)

−0.7471∗ ∗ ∗ (0.00) 0.0905∗ ∗ ∗ (0.00)

0.0121 (0.32) −0.3431 (0.10) Yes Yes

−0.9156∗ ∗ ∗ (0.00) Yes Yes

Panel B: LT1AT Two-equation system

Three-equation system

Spread

Spread

0.0 0 05∗ ∗ ∗ (0.00)

Spread LT1AT Leverage ASSET_MAT

LT1AT

274.3730∗ ∗ ∗ (0.00) 86.0844∗ ∗ ∗ (0.00) −0.0 0 02

−0.0219∗ (0.05)

256.7646∗ ∗ ∗ (0.00) 141.5287∗ ∗ ∗ (0.00) 0.0 0 02

(0.23) Log age Log sales

−5.3560∗ ∗ ∗ (0.00) −22.1614∗ ∗ ∗ (0.00)

LSALES_squared

0.0081 (0.13) 0.0 0 01 (0.87)

−9.8741∗ ∗ ∗ (0.00)

0.0043∗ ∗ ∗ (0.00)

Leverage

0.0 0 09∗ ∗ ∗ (0.00)

0.0058∗ ∗ ∗ (0.00) −1.2920∗ ∗ ∗ (0.00)

−0.1039∗ ∗ ∗ (0.00) (0.12)

−1.8894∗ ∗ ∗ (0.00) −22.3523∗ ∗ ∗ (0.00)

FIXED_ASSET MTB

LT1AT

−11.0333∗ ∗ ∗ (0.00)

0.0370∗ ∗ ∗ (0.00) −0.0015∗ (0.07) 0.0870∗ ∗ ∗ 0.0101∗ ∗ ∗ (0.00)

0.1341∗ ∗ ∗ (0.00)

(0.00) 0.0644 (0.00)

(continued on next page)

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

17

Table 10 (continued) Profit margin Interest coverage Net working capital Tangibility R&D Advertising Stock volatility Excess stock return Distance-to-default LIBOR CONSTANT Year fixed effects Industry fixed effects

−14.5273∗ ∗ (0.04) −0.0372∗ ∗ (0.01) 0.0294∗ ∗ (0.03) −12.0207∗ ∗ (0.04) −21.4583 (0.31) 59.4038 (0.13) 90.4599∗ ∗ ∗ (0.00) −8.9685∗ ∗ ∗ (0.00) −2.3924∗ ∗ ∗ (0.00) −3.1362 (0.25) 242.6693∗ ∗ ∗ (0.00) Yes Yes

−0.0497∗ ∗ ∗ (0.00) 0.0083∗ ∗ ∗ (0.00)

−0.0 0 03 (0.91) −0.0515 (0.10) Yes Yes

−15.8295∗ ∗ ∗ (0.00) −0.0107∗ ∗ ∗ (0.00) 0.0178∗ ∗ ∗ (0.00) −10.7209∗ ∗ ∗ (0.00) −3.8266∗ (0.08) 18.9746∗ ∗ ∗ (0.00) 116.0280∗ ∗ ∗ (0.00) −13.8846∗ ∗ ∗ (0.00) −0.8657∗ ∗ ∗ (0.00) −1.0346∗ ∗ ∗ (0.00) 175.9991∗ ∗ ∗ (0.00) Yes Yes

0.0887∗ ∗ ∗ (0.00)

−0.1063∗ ∗ ∗ (0.00) 0.0158∗ ∗ ∗ (0.00)

−0.7365∗ ∗ ∗ (0.00) 0.0918∗ ∗ ∗ (0.00)

0.0128 (0.10) −0.3565∗ ∗ ∗ (0.01) Yes Yes

−0.8581∗ ∗ ∗ (0.00) Yes Yes

Panel C: STDEBT Two-equation system Spread Spread STDEBT Leverage ASSET_MAT

0.0055 (0.00) 66.9721∗ ∗ ∗ (0.00) 142.8408∗ ∗ ∗ (0.00) 0.0 0 03

Three-equation system

STDEBT

Spread

∗∗∗

−1.1306∗ ∗ ∗ (0.00)

0.0068 (0.00) 65.0958∗ ∗ ∗ (0.00) 162.2282∗ ∗ ∗ (0.00) 0.0010∗ ∗

(0.52) Log age Log sales

−5.6242∗ ∗ ∗ (0.00) −22.7495∗ ∗ ∗ (0.00)

LSALES_squared

0.1024∗ ∗ ∗ (0.00) 0.0020 (0.15)

Profit margin Interest coverage Net working capital Tangibility R&D Advertising Stock volatility Excess stock return Distance-to-default LIBOR CONSTANT Year fixed effects Industry fixed effects

−12.0653∗ ∗ ∗ (0.00) −13.9385∗ ∗ ∗ (0.00) −0.0151 (0.16) 0.0410∗ ∗ ∗ (0.00) −13.1061∗ ∗ ∗ (0.00) −8.1587 (0.46) 74.9612∗ ∗ ∗ (0.00) 87.8219∗ ∗ ∗ (0.00) −7.5006∗ ∗ ∗ (0.00) −2.3211∗ ∗ ∗ (0.00) −3.7717 (0.17) 225.4160∗ ∗ ∗ (0.00) Yes Yes

0.0838∗ ∗ ∗ (0.00)

−0.5244∗ ∗ ∗ (0.00) 0.0544∗ ∗ ∗ (0.00)

0.0235 (0.18) −0.6594∗ ∗ ∗ (0.00) Yes Yes

∗∗∗

Leverage 0.0050∗ ∗ ∗ (0.00) −0.3804∗ ∗ ∗ (0.00)

−1.4049∗ ∗ ∗ (0.00) (0.04)

−3.7109∗ ∗ ∗ (0.00) −23.0057∗ ∗ ∗ (0.00)

FIXED_ASSET MTB

STDEBT

−12.3470∗ ∗ ∗ (0.00) −9.5717∗ ∗ ∗ (0.00) −0.0182∗ ∗ ∗ (0.00) 0.0369∗ ∗ ∗ (0.00) −11.9607∗ ∗ ∗ (0.00) −4.7814 (0.19) 38.6651∗ ∗ ∗ (0.00) 100.3851∗ ∗ ∗ (0.00) −10.1204∗ ∗ ∗ (0.00) −1.6041∗ ∗ ∗ (0.00) −1.5370∗ (0.06) 185.9958∗ ∗ ∗ (0.00) Yes Yes

0.1472∗ ∗ ∗ (0.00) 0.0 0 08 (0.51) 0.0713∗ ∗ ∗ 0.0993∗ ∗ ∗ (0.00)

0.1217∗ ∗ ∗ (0.00)

(0.00) 0.0663 (0.00) 0.0334∗ ∗ ∗ (0.01)

−0.7070∗ ∗ ∗ (0.00) 0.0783∗ ∗ ∗ (0.00)

−0.5826∗ ∗ ∗ (0.00) 0.0671∗ ∗ ∗ (0.00)

0.0131 (0.12) −0.8014∗ ∗ ∗ (0.00) Yes Yes

−0.6993∗ ∗ ∗ (0.00) Yes Yes

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF 18

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22 Table 11 Alternative model specification. This table presents regression results for the baseline model with alternative model specifications. Columns 1–3 reflect Pool-OLS regressions with standard errors adjusted for heteroskedasticity and within-firm clustering, including industry fixed effects. Columns 4–6 represent the random fixed effect model, in which we include industry dummies and clustered standard errors at the firm level. The industry fixed effects are captured using single-digit SIC industry dummies. ∗, ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. Model specification Pool-OLS (1) ST

Random effects (2)

(3)

∗∗

102.09∗ ∗ ∗ (0.00)

58.97 (0.09)

STDEBT Yes Yes Yes Yes Yes 8882 0.5

significant in Column 3 of Table 12, and as short-term debt increases, low-growth firms experience a greater increase in loan spreads than do high-growth firms (Column 4). Overall, these results indicate that our main findings and implications remain robust at the deal level.26 5.2.3. Logarithm loan spreads and newly listed firms Previous studies suggest that the logarithm of the loan spread can mitigate the effect of data skewness (e.g., Campello et al., 2011). For robustness, we rerun the baseline regressions by replacing the raw spreads with the natural logarithm of spreads. Custódio et al. (2013) also document that the use of short-term debts is higher among recently listed firms. Because such firms are less transparent, our results may be subject to a sample selection bias that supports the rollover risk hypothesis. For robustness, we exclude firms that are 4 years or younger and rerun our baseline regressions. Using the natural logarithm of spread and excluding newly listed firms leads to results that are qualitatively similar to our main findings (see Table OA8 of the Online Appendix). 5.2.4. Alternative controls Acharya et al. (2013) report that riskier firms rely more on cash than bank loans for liquidity and that growth firms hold more cash than other firms. An alternative explanation to our Hypothesis 2 is that growth firms hold more cash to prepare for acquisitions and therefore need less short-term debt and are less bank dependent. We explore this possibility by adding controls for cash-to-asset ratios and acquisitions in the past year as a percentage of assets. The results are qualitatively similar to our main findings; this alternative explanation does not affect our findings. We report the results in Table OA9 of the Online Appendix. Chen et al. (2012) suggest that firms with higher systematic risk (i.e., betas) use more long-term debt. Asset maturities may affect debt maturity decision. For robustness, we repeat our baseline regressions with additional controls for asset betas (estimated based on the Merton model) and asset maturities. In addition, we use LIBOR to account for the effect of intertemporal economic shocks; 26

(6)

86.62 (0.00) ∗

Yes Yes Yes Yes Yes 8882 0.5

(5) ∗∗∗

45.02 (0.05)

LT1AT

CONSTANT Firm variables Loan variables Year fixed effects Industry fixed effect Observations R-squared

(4)

We also rerun regressions with alternative model specifications (i.e., Pool-OLS and random effects) with both the largest loan facility and the consolidated samples. The results are qualitatively similar to our main findings. See Tables OA6 and OA7 of the Online Appendix.

11.57∗ ∗ (0.01) Yes Yes Yes Yes Yes 8882 0.5

Yes Yes Yes Yes Yes 8882 0.49

Yes Yes Yes Yes Yes 8882 0.49

15.38∗ ∗ ∗ (0.00) Yes Yes Yes Yes Yes 8882 0.49

however, another useful measure to capture macroeconomic conditions is the VXO index (Acharya et al., 2013). When we replace LIBOR with VXO in the baseline regression, the coefficients of the short-term debt variables remain significantly positive, which confirms our hypotheses.27 The results with these additional controls remain similar to our main findings (see Table OA10 of the Online Appendix). We consider the possibility that unobserved heterogeneity among lenders may have an impact on our results. For example, Paligorova and Santos (2017) suggest that banks that rely on wholesale funding (cf. insured deposits) tend to offer shorter maturity loans. We thus control for unobserved lenders’ characteristics by adding bank fixed effect to our model. In line with Goss and Roberts (2011), we classify the administration agent in each deal as the lead bank in the syndicate and identify the ultimate parents of each lead bank. Then, we retain banks with more than 10 loans in our sample and create dummies that reflect these identified banks in the baseline regressions.28 The results are consistent with our baseline results without controlling for bank fixed effect (see Table OA11 of the Online Appendix). 5.2.5. Financial constraints and macroeconomic shocks To provide additional evidence for the rollover risk hypothesis, we investigate the effect of short-term debts on loan spreads for subgroups of firms, according to their degree of financial constraints. Constrained firms suffer higher refinancing risk (Almeida et al., 2012), so we expect a more pronounced rateincreasing effect of short-term debts on loan spreads for them. The popular Kaplan-Zingales (KZ) index (Kaplan and Zingales, 1997) provides a measure of financial constraints: A higher KZ index indicates that the firm is more financially constrained. We replace ST in the baseline regression (Eq. 1) with two interaction terms: ST × KZ_dummy and ST × (1 − KZ_dummy), where the KZ_dummy variable equals 1 if the firm’s KZ value is greater than the median value of the variable for all firms in a given year, and 0 otherwise. In the Online Appendix, Table OA12 contains the

27 We use VXO (the implied volatility of S&P 100 options) instead of VIX (the implied volatility of S&P 500 options) because VXO has a longer time series than VIX. 28 We thus identify 46 different banks. The top 3 banks, granting the most loans during our sample period (1990–2014) are (1) Bank of American Merrill Lynch, (2) JP Morgan, and (3) Wells Fargo & Co.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

19

Table 12 Largest loan facility and consolidated samples. This table presents the regression results for the subsample of only the largest facility in a given loan deal, with a sample size of 6603 (Columns 1–2). It also presents the firm-level regression results for the consolidated sample, constructed by taking the weighted average of loan spreads for a given year in a given firm, which renders the sample as the firm-year observations in Columns 3–4. The consolidated sample comprises 5946 firm-year observations. We estimate models with firm- and loan-specific variables, firm and year fixed effects, and LIBOR. We only report the results of our main explanatory variables. The p-values in parenthesis were obtained by considering clustered standard errors at the firm level. ∗, ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. Panel A: ST

ST

Largest facility (1)

Largest facility (2)

Consolidated sample (3)

Consolidated sample (4)

117.33∗ ∗ ∗ (0.00)

140.12∗ ∗ ∗ (0.00)

Yes Yes Yes Yes Yes 5940 0.48

174.37∗ ∗ ∗ (0.00) −4.38 (0.28) −136.59∗ ∗ ∗ (0.00) Yes Yes Yes Yes Yes 5940 0.49

Yes Yes Yes Yes No 5946 0.39

193.82∗ ∗ ∗ (0.00) −2.79 (0.50) −131.39∗ ∗ ∗ (0.00) Yes Yes Yes Yes No 5946 0.39

Largest facility (1)

Largest facility (2)

Consolidated sample (3)

Consolidated sample (4)

119.26∗ ∗ ∗ (0.00)

148.74∗ ∗ ∗ (0.00)

Yes Yes Yes Yes Yes 5940 0.48

202.65∗ ∗ ∗ (0.00) −4.85 (0.22) −211.60∗ ∗ ∗ (0.00) Yes Yes Yes Yes Yes 5940 0.48

Yes Yes Yes Yes No 5946 0.39

236.53∗ ∗ ∗ (0.00) −2.51 (0.53) −231.12∗ ∗ ∗ (0.00) Yes Yes Yes Yes No 5946 0.40

Largest facility (1)

Largest facility (2)

Consolidated sample (3)

Consolidated sample (4)

14.19∗ ∗ ∗ (0.01)

23.84 (0.00) −6.33 (0.13) −18.01∗ (0.07) Yes Yes Yes Yes Yes 5940 0.49

20.53∗ ∗ ∗ (0.00)

29.31∗ ∗ ∗ (0.00) −4.93 (0.26) −16.50 (0.12) Yes Yes Yes Yes No 5946 0.40

High_MTB ST × High_MTB CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects #of observations R-squared Panel B: LT1AT

LT1AT High_MTB LT1AT × High_MTB CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects #of observations R-squared Panel C: STDEBT

STDEBT High_MTB STDEBT × High_MTB CONSTANT Firm variables Loan variables Year fixed effects Firm fixed effects #of observations R-squared

Yes Yes Yes Yes Yes 5940 0.48

estimation results. The estimated coefficient is statistically significant (at the 1% level) for the interaction of the short-term debt ratios and KZ_dummy, but not for the interaction of the short-term debt ratios and (1 – KZ_duumy), consistent with our expectation that the rate-increasing effect of short-term debts is significant mainly for financially constrained firms. We also examine whether macroeconomic shocks amplify the rate-increasing effect of short-term debts on loan spreads (i.e., the rollover risk hypothesis). Based on the sample period from 1990 to 2014, we identify macroeconomic conditions according to recessionary years, as classified by the National Bureau of Economic Re-

Yes Yes Yes Yes No 5946 0.40

search: 1990, 1991, 2001, 20 02, 20 08, and 20 09. The dummy variable Recession equals one if an observation occurs in one of these years, and zero otherwise. We also examine how the recent financial crisis affects the relationship between short-term bank loan and its cost. To do so, we create the dummy variable Crisis that equals to one if observations are during the period from the fourth quarter of 2007 to the fourth quarter of 2009; and zero otherwise. We replace ST in the baseline regression (Eq. 1) with two interaction terms: ST × Recession and ST × (1 − Recession). We also replace ST in the baseline regression (Eq. 1) with two interaction terms: ST × Crisis and ST × (1 − Crisis). Table OA13 in the Online Appendix

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF 20

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

contains the estimation results. Although ST is positively associated with loan spreads during both recessions and expansions, the magnitude of the effect is greater during recession and the financial crisis. This result highlights that macroeconomic shocks amplify the rate-increasing effect of the short debt maturity structure.29 5.2.6. Rollover risk measured with longer lags Finally, we explore our hypotheses using an alternative rollover risk measure with longer lags, as suggested by Gopalan et al. (2014). The measure LT2AT is the ratio of total long-term debt payable in year t, based on information in the firm’s year t – 2 balance sheet (Compustat item DD2), over the book value of total assets at the beginning of year t. The main difference between LT1AT and LT2AT is that whereas the former uses information available at the end of year t – 1, the latter uses information available at the end of year t – 2, which is less likely to correlate with changes in the firm’s credit quality around year t. With LT2AT, we can make a stronger argument regarding erogeneity. However, LT2AT also is a less precise proxy for the extent of rollover risk faced by the firm, compared with the short-term debt variables we use in our main analysis. When we re-estimate all the regressions in Tables 4 using LT2AT as the main independent variable, the results indicate a strong positive relation between LT2AT and loan spreads. This relation is significant only for firms with low growth opportunities (Table OA14 of the Online Appendix).30 We thus obtain robust evidence in support of our rollover risk and asset substitution hypotheses. 6. Conclusion Do banks charge higher loan rates to compensate for rollover risk? To address this question, we examine syndicated loans during the period from 1990 to 2014 in the U.S. market and provide strong empirical evidence that the short-term debt ratio is a critical determinant of loan spreads, after accounting for various firm- and loan-specific variables, as well as firm and year fixed effects. We also examine the asset substitution prediction that shortterm debts lead to safer investments, lower firm risk, and therefore lower bank loan spreads, especially for firms with strong incentives to pursue risky investments. Given the same increase in the short-maturity debt ratio, high-growth firms (with strong riskshifting incentives) experience a significantly smaller increase in the loan spread than low-growth firms. The result suggests that short-term debts alleviate the asset substitution problem, leading to a decrease in loan rates. This effect is especially strong for highgrowth firms. Upon further examination, we find that a longer debt maturity structure is particularly imperative for firms that are high-risk, bank-dependent, and committed to credit lines, to reduce their borrowing costs. Our results remain consistent when we use alternative debt maturity proxies, conduct various analyses to address 29 Following a reviewer’s suggestion, we consider the possibility that the impact of the macroeconomic shocks on our analyses. In particular, it is possible that in the low rate environment, companies with stronger credit may use relatively cheap long term debt, which is likely to lead to a two-way relationship between credit and rollover risks. Consequently, we conjecture that the rollover effect is likely to vary between very distinct macroeconomic conditions, e.g., crisis versus non-crisis periods. In particular, we examine whether the rollover effect differs between recessionary periods and non-recessionary periods as well as between the financial crisis period and non–crisis periods. We use a slightly broader definition of the crisis period to fully capture the impact of the crisis (e.g., Chen et al., 2012). 30 As Gopalan et al. (2014) point out, LT2AT (using long-term debts maturing in two years) is a noisier measure of rollover risk, so we should expect the magnitude of the rollover risk effect on loan spreads to be smaller. Our finding is consistent with this expectation: A one-standard-deviation increase in LT2AT (0.062) leads to an increase in loan spreads by 4.28 basis points (0.06×71.37), compared with the corresponding estimate of 6.32 basis points with LT1AT.

endogeneity concerns, and perform a number of robustness tests. Collectively, our results confirm that banks consider a firm’s debt maturity structure when determining the pricing of bank loans, in addition to conventional pricing factors. Further research can explore the effects of other attributes of debt structure (e.g., debt concentration) on the cost of bank loans. Appendix A. Literature on the determinants of loan duration on loan spreads

Literature Campello et al. (2011) Houston et al. (2014)

Expected sign + +

Dennis et al. (20 0 0)

( + /–)

Santos (2011)

( + /–)

Brockman et al. (2010)

Goss and Roberts (2011) Santos and Winton (2008)

+

– ( + /–)

Methodology

Data / Data period

Single equation Single equation

All loans / 1996–2002 All loans on S&P 500 companies / 20 03–20 08 Bank revolving credit / 1987–1995 All loans / 20 02–20 08 Corporate bonds / 1994–2005

Positive (not significant) Positive (significant)

All loans / 1991–2006 All loans / 1987–2002

Negative (not significant) Negative (significant)

1. System of equations 2. Single equation Single equation 1. Single equation 2. System of equations Single equation Single equation

Empirical result

Negative (significant)

Negative (significant) Positive (significant)

Appendix B. Variable descriptions

Variable Dependent variable Spread

Definition

Source

Loan spread over LIBOR plus fees at the issue date in basis points (Dealscan item all-in-drawn spread).

Debt maturity variables ST Proportion of short-term debt to total assets LT1AT Proportion of long-term debts maturing in 1 year to total assets STDEBT Proportion of short-term debt to total debts ST3 Proportion of total debt that matures within 3 years ST4 Proportion of total debt that matures within 4 years ST5 Proportion of total debt that matures within than 5 years MAT Book-value weighted numerical estimate of debt maturity, based on the assumption that the average maturities of the 6 Compustat maturity categories are 0.5 year, 1.5 years, 2.5 years, 3.5 years, 4.5 years, and 10 years. Firm variables Log age Logarithm of age

Dealscan

Compustat Compustat Compustat Compustat Compustat Compustat Compustat

Compustat Log sales

Logarithm of sales

Leverage

Ratio of total debts to total assets

MTB (market-to-book)

Ratio of market value to book value

Compustat Compustat Compustat (continued on next page)

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

Variable Variable

Definition

Profit margin

Ratio of net income to sales

Source

Other variables LIBOR

Compustat Interest coverage Tangibility

Net working capital

R&D Advertising

Logarithm of 1 plus EBITDA divided by interest expense truncated at 0 Ratio of inventories plus plant, property, and equipment to total assets Ratio of networking capital (current assets less current liabilities) to total debt Research and development expense divided by sales Advertising expense divided by sales

Compustat

High_MTB

Compustat

Compustat

STOCKVOL-A50

Compustat Compustat

Excess stock return

Stock volatility

Distance-to-default

ASSET_MAT

FIXED_ASSET

KZ index

Loan variables Log loan size Log loan duration Secure Senior Dividend rest

Corporate purposes

Debt repay Working capital

Term loan Bridge loan Credit line Log number of lenders

Excess stock return (relative to the market) over the past 12 months Standard deviation of a firm’s excess stock returns over the past 12 months KMV distance-to-default, based on Vassalou and Xing (2004) Weighted average of the maturity of long-term assets and current assets. The maturity of long-term assets is measured as gross property, plant, and equipment divided by depreciation; the maturity of current assets is current assets divided by the cost of goods. The weight for long-term assets is the share of gross property, plant, and equipment in total assets, and the weight for current assets is the share of current assets in total assets. Ratio of net property, plant, and equipment to the book value of total assets. The KZ (Kaplan and Zingales, 1997) index is defined as − 1.002 × Cash flow + 0.283 × Tobin’s q + 3.139 × Debt − 39.368 × Dividends − 1.315 × Cash holdings Loan facility amount in $ millions (Dealscan item Tranche Amount). Logarithm of duration of the loan in years. Dummy variable that takes the value of 1 if loan is secured by collateral. Dummy variable that takes the value of 1 if loan is senior. Dummy variable that takes the value of 1 if loan has restrictions on paying dividends. Dummy variable that takes the value of 1 if loan is for corporate purposes. Dummy variable that takes the value of 1 if loan is to repay existing debt. Dummy variable that takes the value of 1 if loan is for working capital purposes. Dummy variable that takes the value of 1 if loan is a term loan. Dummy variable that takes the value of 1 if loan is a bridge loan. Dummy variable that takes the value of 1 if loan is a credit line. Logarithm of number of lenders.

Compustat / CRSP

ZSCORE-B50

Compustat / CRSP Compustat / CRSP

DTD-B50

Compustat LINTEREST_COV-B50

Definition

Three-month U.S. London Interbank Offer Rate at the end of the month of deal signing. Dummy variable that takes the value of 1 if a firm has an MTB value above the median value of MTB among all firms in a given year and 0 otherwise. Dummy variable that takes the value of 1 if a firm has a stock volatility value above the median value of the variable among firms for a given year (higher-risk firms) and 0 otherwise. Dummy variable that takes the value of 1 if a firm has an Altman Z-score below the median value of the variable among firms for a given year (higher-risk firms) and 0 otherwise. Dummy variable that takes the value of 1 if a firm has a distance-to-default value below the median value of the variable among firms for a given year (higher-risk firms) and 0 otherwise. Dummy variable that takes the value of 1 if a firm has an interest coverage value below the median value of the variable among firms for a given year (higher-risk firms) and 0 otherwise.

21

Source

British Banker’s Association Compustat / CRSP

CRSP

Compustat

Compustat / CRSP

Compustat

Supplementary materials

Compustat

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.jbankfin.2017.10.008.

Compustat

References

Dealscan Dealscan Dealscan Dealscan Dealscan

Dealscan

Dealscan Dealscan

Dealscan Dealscan Dealscan Dealscan

(continued on next page)

Acharya, V., Almeida, H., Campello, M., 2013. Aggregate risk and the choice between cash and lines of credit. J. Finance 68, 2059–2116. Almeida, H., Campello, M., Laranjeira, B., Weisbenner, S., 2012. Corporate debt maturity and the real effects of the 2007 credit crisis. Crit. Finance Rev. 1, 3–58. Arena, M.P., 2011. The corporate choice between public debt, bank loans, traditional private debt placements, and 144a debt issues. Rev. Quant. Finance Account. 36, 391–416. Barclay, M.J., Marx, L.M., Smith, C.W., 2003. The joint determination of leverage and maturity. J. Corp. Finance 9, 149–167. Barclay, M.J., Smith, C.W., 1995. The maturity structure of corporate debt. J. Finance 50, 609–631. Bharath, S.T., Sunder, J., Sunder, S.V., 2008. Accounting quality and debt contracting. Account. Rev. 83, 1–28. Billett, M.T., King, T.H.D., Mauer, D.C, 2007. Growth opportunities and the choice of leverage, debt maturity, and covenants. J. Finance 62, 697–730. Brockman, P., Martin, X., Unlu, E., 2010. Executive compensation and the maturity structure of corporate debt. J. Finance 65, 1123–1161. Campello, M., Lin, C., Ma, Y., Zou, H., 2011. The real and financial implications of corporate hedging. J. Finance 66, 1615–1647. Campbell, T.S., Kracaw, W.A., 1990. Corporate risk management and the incentive effects of debt. J. Finance 45, 1673–1686. Chava, S., Livdan, D., Purnanandam, A., 2008. Do shareholder rights affect the cost of bank loans. Rev. Financ. Stud. 22, 2973–3004. Chava, S, Roberts, M.R., 2008. How does financing impact investment? The role of debt covenants. J. Finance 63, 2085–2121. Chen, H., Xu, Y., Yang, J., 2012. Systematic risk, debt maturity, and the term structure of credit spreads. No. w18367. National Bureau of Economic Research. Custódio, C., Ferreira, M.A., Laureano, L., 2013. Why are US firms using more short-term debt. J. Financ. Econ. 108, 182–212. Datta, S., Iskandar-Datta, M., Raman, K., 2005. Managerial stock ownership and the maturity structure of corporate debt. J. Finance 60, 2333–2350. Denis, D.J., Mihov, V.T., 2003. The choice among bank debt, non-bank private debt, and public debt: evidence from new corporate borrowings. J. Financ. Econ. 70, 3–28.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008

ARTICLE IN PRESS

JID: JBF 22

[m5G;November 2, 2017;7:1]

C.-W. Wang et al. / Journal of Banking and Finance 000 (2017) 1–22

Dennis, S., Nandy, D., Sharpe, L.G., 20 0 0. The determinants of contract terms in bank revolving credit agreements. J. Financ. Quant. Anal. 35, 87–110. Dhaliwal, S.D., Khurana, K.I., Pereira, R., 2004. Costly public disclosure and the choice between private and public debt. Working paper, University of Arizona. Eisdorfer, A., 2008. Empirical evidence of risk shifting in financially distressed firms. J. Finance 63, 609–637. Francis, B., Hasan, I., Koetter, M., Wu, Q., 2012. Corporate boards and bank loan contracting. J. Financ. Res. 35, 521–552. Goldberger, A.S., 1991. A Course in Econometrics. Harvard University Press, Cambridge, MA). Gopalan, R., Song, F., Yerramilli, V., 2014. Debt maturity structure and credit quality. J. Financ. Quant. Anal. 49, 817–842. Goss, A., Roberts, G.S., 2011. The impact of corporate social responsibility on the cost of bank loans. J. Bank. Finance 35, 1794–1810. Graham, J., Li, S., Qiu, J., 2008. Corporate misreporting and bank loan contracting. J. Financ. Econ. 89, 44–61. Greene, W.H., 2002. Econometric Analysis. Prentice Hall, Englewood Cliffs, NJ). Hale, G., Santos, J.AC, 2009. Do banks price their informational monopoly. J. Financ. Econ. 93, 185–206. Hasan, I., Hoi, C.K., Wu, Q., Zhang, H., 2014. Beauty is in the eye of the beholder: the effect of corporate tax avoidance on the cost of bank loans. J. Financ. Econ. 113, 109–130. Hasan, I., Hoi, C., Wu, Q., Zhang, H., 2017. Social capital and debt contracting: evidence from bank loans and public bonds. J. Financ. Quant. Anal. 52 (3), 1017–1765. He, Z., Xiong, W., 2012. Rollover risk and credit risk. J. Finance 67, 391–430. Hertzel, M.G., Officer, M.S., 2012. Industry contagion in loan spreads. J. Financ. Econ. 103, 493–506.

Houston, J.F., Jiang, L., Lin, C., Ma, Y., 2014. Political connections and the cost of bank loans. J. Account. Res. 52, 193–243. Jensen, M.C., Meckling, W.H., 1976. Theory of the firm: managerial behavior, agency costs, and capital structure. J. Financ. Econ. 3, 305–360. Johnson, S.A., 2003. Debt maturity and the effects of growth opportunities and liquidity risk on leverage. Rev. Financ. Stud. 16, 209–236. Kaplan, S.N., Zingales, L., 1997. Do investment-cash flow sensitivities provide useful measures of financing constraints. Q. J. Econ. 112, 169–215. Kennedy, P., 2003. A Guide to Econometrics. The MIT Press, Cambridge, MA. Khang, K., King, T.D., Nguyen, H., 2016. What determines outstanding corporate debt mix? Evidence from fractional multinomial logit estimation. Appl. Econ. 48, 276–291. Paligorova, T., Santos, J.A.C., 2017. Banks’ exposure to rollover risk and the maturity of corporate loans. Rev. Finance 21 (4), 1739–1765. Petersen, M.A., 2009. Estimating standard errors in finance panel data sets: comparing approaches. Rev. Financ. Stud. 22, 435–480. Santos, J.A., 2011. Bank corporate loan pricing following the subprime crisis. Rev. Financ. Stud. 24, 1916–1943. Santos, J.A., Winton, A., 2008. Bank loans, bonds, and information monopolies across the business cycle. J. Finance 63, 1315–1359. Valenzuela, P., 2016. Rollover risk and credit spreads: evidence from international corporate bonds. Rev. Finance 20, 631–661. Vassalou, M., Xing, Y., 2004. Default risk in equity returns. J. Finance 59, 831–868. Wang, C.-W., Chiu, W.-C., Peña, J.I., 2017. Effect of rollover risk on default risk: Evidence from bank financing. Int. Rev. Financ. Anal. 54, 130–143.

Please cite this article as: C.-W. Wang et al., Debt maturity and the cost of bank loans, Journal of Banking and Finance (2017), https://doi.org/10.1016/j.jbankfin.2017.10.008