Debt Specialization *

Transcription

1 Debt Specialization * Paolo Colla Università Bocconi Filippo Ippolito Università Bocconi Kai Li University of British Columbia This version: September, 2010 First version: October, 2009 * We thank Miguel Ferreira, Mark Flannery, Emilia Garcia, Mike Lemmon, Dave Mauer, Michael Meloche, Gordon Phillips, Zacharias Sautner, Philip Valta, seminar participants at the New University of Lisbon (Nova), University Carlos III de Madrid, Universitat Pompeu Fabra, Stockholm School of Economics and SIFR, and conference participants at the 6 th Portuguese Finance Network Conference (Azores), the 2010 China International Conference in Finance (Beijing), the ESSFM Conference (Gerzensee), and the European Finance Association Meetings (Frankfurt) for helpful comments. We thank Milka Dimitrova and Huasheng Gao for excellent research assistance. Li acknowledges the financial support from the Social Sciences and Humanities Research Council of Canada. All remaining errors are our own. Department of Finance-2-D2-08, Università Bocconi, Via G. Röntgen, Milano, Italy, (+39) , paolo.colla@unibocconi.it. Department of Finance-2-D2-02, Università Bocconi, Via G. Röntgen, Milano, Italy, (+39) , filippo.ippolito@unibocconi.it. Sauder School of Business, University of British Columbia, 2053 Main Mall, Vancouver, BC V6T 1Z2, , kai.li@sauder.ubc.ca.

2 Debt Specialization Abstract This paper provides the first large sample evidence on the patterns and determinants of debt structure using a new database of publicly listed U.S. firms. Within what is generally referred to as debt financing, we are able to distinguish between commercial paper, revolving credit facilities, term loans, senior and subordinated bonds and notes, and capital leases. We first show that most of the sample firms concentrate their borrowing in only one of these debt instruments, and only the low growth, low risk large firms with high profitability and the highest level of leverage borrow through multiple debt instruments. We then show that the extent of debt specialization is increasing in firm growth opportunities, cash flow risk, and asset maturity, while decreasing in asset tangibility. Finally, we find that firm characteristics that are known to be associated with their leverage decisions also affect their usage of different debt instruments. Our paper suggests that debt structure decisions, like capital structure decisions, are made based on a cost and benefit analysis to maximize firm value. Keywords: debt specialization, debt structure, commercial paper, revolving credit facilities, term loans, senior bonds and notes, subordinate bonds and notes, capital leases JEL classification: G32

3 I. Introduction Much attention has been devoted to the issues of why firms choose between equity and debt, and how optimal capital structure is designed to minimize the cost of financing (see the survey by Frank and Goyal (2008) of the voluminous literature on capital structure). In this paper, we focus on a much less studied topic in capital structure, namely debt structure. Our goals are to explore the types of debt commonly employed by public U.S. companies, and to relate their usage to the costs and benefits of different types of debt financing. To our knowledge, our paper is the first to provide large sample evidence on the subject. The existing literature suggests a number of important and as of yet unanswered questions concerning the patterns and determinants of debt structure: How are different types of debt used in practice to meet corporate funding needs? Do firms tend to specialize in one or two debt instruments, or do they borrow simultaneously from a variety of sources? How do these choices vary with firm characteristics, such as their asset maturity structure and their access to the public bond markets? To answer these questions, we take advantage of a new database available through Capital IQ, an affiliate of the Standard and Poor s, to examine debt structure of publicly listed U.S. firms. 1 Within what is generally referred to as debt financing, we are able to distinguish between commercial paper, revolving credit facilities, term loans, senior and subordinated bonds and notes, and capital leases. After merging the Capital IQ database with the Compustat database, we end up with a large panel data set that comprises 16,115 firm-year observations involving 3,296 unique firms for the period In addition to information on debt structure, the sample also contains leverage and other firm characteristics (e.g., firm size and profitability). 1 The SEC mandated electronic submission of SEC filings in Capital IQ has been collecting information about debt structure since then. The coverage has much improved since 2002 which is the starting point of our sample period.

4 Our first main finding is that firms specialize in their borrowing: Most sample firms concentrate their borrowing in only one of the above debt instruments. As primarily an exploratory exercise, we use cluster analysis to search for patterns of debt structure. We identify six distinctly different groups of firms: Five of them are predominant users of one single debt instrument, while only one group relies simultaneously on more than one debt instrument. This last group is mainly composed of the low growth, low risk large firms with high profitability and the highest level of leverage. The evidence is suggestive that the average public U.S. firm specializes in borrowing one type of debt to meet its funding needs. To further corroborate the above finding, we analyze conditional debt structure. We require firms to allocate a significant fraction of their debt to a given type of debt, and then examine the composition of their debt structure under this condition. We find that the majority of firms rely overwhelmingly on only one type of debt; specifically, the one which we have conditioned upon. For example, conditioning for firms to have more than 30% of their debt in term loans, we find that among this subset of firms term loans represent over 70% of their debt. We show that this result is robust to different specifications of the conditioning threshold. Second, we find that a key factor for understanding debt structure is credit quality. We show that debt structure varies substantially between not only rated and unrated firms but also across firms with different credit ratings: Large and high credit quality firms tend to have access to different sources of financing, while small and unrated firms rely exclusively on either capital leases or bank debt for financing. Further, there are some significant non-linear relations between actual credit ratings and types of debt beyond the usual categorization of firms being rated or not: For example, the amount of senior bonds and notes is increasing in credit quality, peaks at the rating of A, and then is decreasing in credit quality as the latter further improves. Faulkender and Petersen (2006) show that firms that do not have access to the public debt markets, as measured by not having a debt rating, tend to have lower debt ratios. Our finding highlights that the actual 2

5 credit rating, a comprehensive measure of firm credit worthiness, affects firm access to different sources of financing as well. Third, we analyse the determinants of debt specialization and find that the extent of debt specialization is decreasing in asset tangibility, while increasing in market-to-book (M/B) ratios, cash flow volatility, and asset maturity. Interestingly, we also observe that the most recent financial crisis had a strong impact on firms increasing adoption of debt specialization. Finally, we address the question of how choices of different debt instruments vary with firm characteristics. We rely on some recent papers in capital structure (see for example, Fama and French (2002) and Lemmon, Roberts and Zender (2008)) to identify firm characteristics that are known to be associated with cross-sectional variations in leverage, including profitability, asset tangibility, M/B ratios, firm size, cash flow volatility, the dividend payer dummy variable, asset maturity, and firm age. We find that the effects these firm characteristics have on leverage and on different debt instruments can vary substantially. For example, the previous literature has established that profitability is negatively and significantly associated with leverage as predicted by the pecking order theory. Our analysis further shows that this negative association is mainly driven by three types of debt: senior and subordinated bonds and notes, and capital leases. In contrast, profitability is positively and significantly associated with commercial paper and revolving lines of credit. We also show that asset maturity is negatively and significantly associated with revolving lines of credit and capital leases, while positively and significantly associated with senior and subordinated bonds and notes. These results indicate that using a gross measure of leverage such as total debt can be misleading, as it hides heterogeneity across different debt instruments. The evidence supports our conjecture that debt structure decisions, like capital structure decisions, are made based on a cost and benefit analysis to maximize firm value. Our paper is closely related to and motivated by Rauh and Sufi (2010) who examine types, sources, and priorities of debt using a sample of 305 randomly selected non-financial rated 3

6 public firms for the period (2,453 firm-year observations). They show that almost three quarters of their firm-year observations employ more than two different debt instruments, and that a quarter of the sample firms has no significant one-year change in their level of debt but significant changes in their debt composition. Further, high credit quality firms (BBB and higher) primarily use two tiers of capital: equity and senior unsecured debt. Low credit quality firms (BB and lower) tend to use several tiers of debt including secured, senior unsecured, and subordinated issues. Our work differs from Rauh and Sufi (2010) in the following aspects. First, our much larger sample allows us to examine the financing patterns of unrated firms as well these represent over 60% of our firm-year observations and 9% of the total assets of our sample firms, and thus account for an important part of the overall economy. By contrast, Rauh and Sufi s sample is limited to rated firms. Second, as a result of our broader sample, we are able to uncover the phenomenon of debt specialization in firm financing behavior, which is otherwise unobservable. Indeed, Rauh and Sufi conclude that financing through multiple sources of debt is the norm among large and rated firms in their sample, which we confirm only among a subsample of our firms. Moreover, we also show that smaller firms with no or poor credit ratings borrow through only one type of debt. Finally, we examine the determinants of this debt specialization phenomenon that we uncover. The outline of the paper is as follows. Section II reviews the related literature and develops our hypotheses. Section III describes our data and sample, and provides an overview of debt structure in our sample firms. Section IV presents evidence on the extent of debt specialization. Section V provides our explanations for the observed financing pattern. Section VI carries out various robustness checks on our main results. Finally, Section VII summarizes our findings and concludes. II. Literature Review and Hypothesis Development 4

7 How should firms choose their capital structure? In the ideal world of Modigliani-Miller (1958, 1963), capital structure choices are irrelevant for firm value and so any choice is optimal. Starting from Miller (1977), finance researchers have reassessed the irrelevance result by reintroducing various market frictions that make firm value dependent on capital structure choices. Among these frictions, bankruptcy costs, taxes, agency problems, and information asymmetries are the most well studied (see Myers (2001) for an excellent review). By trading off the costs and benefits of debt financing, firms with different characteristics employ different levels of debt at which firm value is maximized. Extending the optimal capital structure perspective to debt structure choices, we expect debt structure to vary across firms with different characteristics that capture different costs and benefits associated with a particular debt type. Indeed, the existing literature on firms usage of different types of debt shows that debt instrument choices depend on, by and large, the same factors that affect capital structure choices. In this section we review the literature that suggests that debt specialization happens from a cost and benefit analysis based on firm heterogeneity. A. Commercial Paper Although to our knowledge there is no theoretical work specific to commercial paper, there is ample empirical evidence identifying firm size and default risk as the main firm characteristics associated with its usage: Only large firms with strong credit ratings have the opportunity to access the commercial paper market. Kahl, Shivdasani, and Wang (2008) suggest that commercial paper represents a state-contingent funding source that partially substitutes the need for precautionary cash holdings. Gao and Yun (2009) examine the interplay between precautionary cash holdings and commercial paper during the recent financial crisis. They find that the aggregate commercial paper borrowing declined 15% after the collapse of Lehman Brothers, but the effect was concentrated among firms with high default risk. These high default 5

8 risk firms drew heavily from existing lines of credit to substitute lost borrowing from the commercial paper market. Using large firm size and old firm age to proxy for low default risk, our first hypothesis with respect to commercial paper is: H1: The use of commercial paper is increasing in firm size, age, and credit ratings. B. Revolving Credit Facilities The literature on revolving credit examines why firms are granted a credit line and why they draw down on it. Holmström and Tirole (1998), DeMarzo and Sannikov (2006), and DeMarzo and Fishman (2010) show that credit lines help reduce the trade-off between inefficient liquidation and moral hazard. A firm cannot wait to borrow funds after a liquidity shock is realized. The optimal liquidity policy can be implemented either in terms of precautionary cash holdings, or in terms of an irrevocable line of credit. Empirically, Jiménez, López, and Saurina (2008) and Sufi (2009) show that drawdowns on credit lines increase as firm financial conditions worsen. Ivashina and Scharfstein (2009) further show that drawdowns on existing credit lines increased drastically during the 2008 financial crisis. This literature suggests that firms with high agency costs and high cash flow uncertainty are more in need of a credit line. However, these very firms are also more likely to draw down on the credit line when their financial conditions worsen. Anticipating this, banks offer a credit line only to firms with low agency costs and low cash flow uncertainty. Using the M/B ratio to proxy for agency costs, and cash flow volatility to proxy for cash flow uncertainty, our second hypothesis with respect to revolving credit facilities is thus: H2: The use of revolving lines of credit is decreasing in M/B ratios and cash flow volatility. C. Bank versus Bond Financing 6

9 There is a large theoretical literature that examines the trade-off between bank versus bond financing. Diamond (1991) first introduces the idea that the choice between borrowing directly (through issuing corporate bonds, without monitoring) and borrowing through a bank where monitoring is primarily driven by the lender s need to reduce moral hazard. Chemmanur and Fulghieri (1994) further demonstrate that banks desire to acquire a reputation for making the right renegotiation versus liquidation decision provides them incentives to devote a larger amount of resources than bondholders toward such evaluations. In equilibrium, bank loans dominate bonds from the point of view of minimizing inefficient liquidation; however, firms with a lower probability of financial distress choose bonds over bank loans. Along similar lines, Bolton and Freixas (2000) show that firms turn to banks as a source of financing mainly because banks are good at helping them through times of financial distress. In equilibrium the riskier firms prefer bank loans, the safer ones tap the bond markets, and the ones in between issue both equity and bonds. The empirical evidence mostly provides support for the above key idea that banks specialize in lending to firms with higher agency and liquidation costs. Houston and James (1996) show that reliance on bank borrowing is decreasing in firm size, growth opportunities, leverage, and the firm s access to public debt markets. Johnson (1997) finds that firms use more public debt if they have lower information and monitoring costs, lower likelihood and costs of inefficient liquidation, and weaker incentives to harm the lenders. Hadlock and James (2002) show that firms are more likely to choose bank loans when asymmetric information problems are elevated. Cantillo and Wright (2000) find that large companies with abundant cash and collateral tap credit markets directly; and these markets cater to safe and profitable industries. In summary, the prior literature suggests that firms with lower monitoring needs and better fundamentals prefer public debt to bank financing. We expect monitoring needs to be lower for firms with larger size, higher asset tangibility, higher growth opportunities, and higher 7

10 credit ratings. Hence our third hypothesis with respect to the choice between bank debt and public debt is: H3: The use of bank debt is decreasing in asset tangibility, M/B ratios, firm size, and credit ratings; while the use of public debt is increasing in asset tangibility, M/B ratios, firm size, and credit ratings. D. Capital Leases The final strand of literature examines leases and shows that the use of leases induces a trade-off between lower liquidation costs and higher agency costs. Eisfeldt and Rampini (2010) show that there are two mechanisms at work in a lease: 1) leasing allows the lessor (owner) to repossess the asset when the lessee files for bankruptcy, which gives the lessor a stronger claim than that of secured lenders; 2) leasing entails agency costs, as the lessee retains control while the lessor keeps ownership (separation of ownership and control). They predict that firms that are financially constrained prefer to lease an asset, while firms that are less constrained prefer to own the asset. Krishnan and Moyer (1994) document that lessee firms have low retained earnings, high growth rates, low coverage ratios, and in general a high bankruptcy risk. Yan (2006) further shows that leases and debt are substitutes and that in those firms with more growth options or larger marginal tax rates, or in those firms paying no dividends, the substitutability is more pronounced. In summary, the existing theory and evidence on capital leases suggest that financially constrained firms with high growth and high risk are the best candidates for borrowing through leases, and hence our final hypothesis with respect to capital leases is: H4: The use of capital leases is decreasing in profitability, dividend payout, and credit ratings, while increasing in M/B ratios and cash flow volatility. 8

11 III. Data Overview A. Sample Description We start with U.S. firms traded on AMEX, NASDAQ, and NYSE covered by both Capital IQ (CIQ) and Compustat from 2002 to CIQ compiles detailed information on capital structure and debt structure by going through financial footnotes contained in firms 10K SEC filings. 2 We remove utilities (SIC codes ) and financials firms (SIC codes ) and end up with 27,802 firm-year observations. We further remove 1) firm-years with missing value of total assets (26,729 observations remaining); 2) firm-years with zero total debt (19,142 observations remaining); 3) firm-years with market or book leverage outside the unit interval (as in Lemmon, Roberts and Zender (2008), 17,572 observations remaining); and 4) firm-years for which the difference between total debt as reported in Compustat and the sum of debt types as reported in CIQ exceeds 10% of total debt. Our final sample has 16,115 firm-year observations involving 3,296 unique firms. In constructing our variables we use the same definitions as in Lemmon, Roberts and Zender (2008). Table A1 in the Appendix provides a detailed description of the variables used in our analysis. Total assets are expressed in millions of 2002 U.S. dollars. Table 1 presents descriptive statistics. Columns (1) and (2) report means and medians of the key firm characteristics aggregated across all years for our sample firms. Columns (3) and (4) present means and medians of the same key firm characteristics for the Compustat leveraged firms, i.e., firms with positive debt. This Compustat sample is formed by imposing similar filters as to our sample except filter 4) above. Our sample covers approximately 90% of the Compustat leveraged firms. Columns (5) and (6) test whether our sample is different from the Compustat leveraged firms. 2 Regulation S-X requires firms to detail their long-term debt instruments. Regulation S-K requires firms to discuss their liquidity, capital resources, and operating results. As a result of these regulations, firms detail their long-term debt issues and bank revolving credit facilities. Firms often also provide information on notes payable within a year (Rauh and Sufi (2010)). 9

12 We show that over the sample period , the mean (median) market leverage as measured by the ratio of total debt to the sum of total debt and market value of equity is (0.195). Using a sample of non-financial, non-utility firms from Compustat over the period , Faulkender and Petersen (2006) report that the mean (median) market leverage is (0.183) for leveraged firms. In comparison to the Compustat leveraged firm sample, our sample firms are not significantly different in most dimensions compared except dividend payout (although the economic significance of this difference is small). We conclude that our sample is representative of the Compustat leveraged firms. B. Debt Structure Overview CIQ decomposes total debt into seven mutually exclusive debt types: commercial paper (CP), drawn revolving credit facilities (RC), term loans (TL), senior bonds and notes (SBN), subordinated bonds and notes (SUB), capital leases (CL), 3 and other debt (Other). 4 Table A2 in the Appendix provides an example illustrating how CIQ collects and constructs the various debt types. Table 2 provides detailed summary statistics for debt types (normalized by total debt). Panel A of Table 2 shows that first, the majority of sample firms rely on senior bonds and notes for financing. The sample mean (median) ratio of senior bonds and notes to total debt (TD) is (0.208). Second, the median ratios of both revolving credit and term loans to total debt are close to zero or zero, while the 75 th percentiles are much greater than zero; suggesting that between a quarter and half of the sample firms rely on revolving credit facilities or term loans, both provided by banks. When adding up both debt instruments to obtain total bank debt, we find 3 Capital leases are different from operating leases. While in an operating lease, lease expenses are treated as an operating cost, a capital lease is recognized both as an asset and as a liability on the balance sheet, and is thus subject to depreciation. Typically, firms prefer to keep leases off the books and defer expenses, which gives them the incentive to report all leases as operating leases. As a result, the Financial Accounting Standards Board has ruled the conditions under which a lease should be reported as a capital lease. Though often disregarded in the existing literature, the distinction between capital and operating leases is important for our purposes. In our analysis of debt we will only consider capital leases, as operating leases are not reported as debt on the balance sheet. 4 Other debt mostly consists of short-term borrowings. Occasionally, it takes other forms such as deferred credits, fair value adjustments related to hedging contracts, or trust-preferred securities. 10

13 that more than half of the sample firms employ bank debt, with the sample mean (median) at (0.322) (untabulated). Third, more than a quarter of the sample firms employ capital leases, even though they are much less important on average than bank debt. The sample mean ratio of capital leases to total debt is 0.054, while that of revolving credit (term loans) is (0.212). Fourth, less than a quarter of the sample firms use subordinated bonds and notes. Lastly, the 90 th percentile of commercial paper is zero (untabulated), suggesting that less than 10% of the sample firms use commercial paper for financing. Total adjustment is the difference between the total debt variable obtained from Compustat and the sum of seven debt instruments from CIQ. When forming our sample, we have imposed the filter that the total adjustment for firms in the sample be less than 10% of total debt. After applying this filter, there is little discrepancy between the sum of debt instruments from CIQ and total debt from Compustat: Both the mean and median ratios of total adjustment to total debt are zero, and the 1 st and 99 th percentiles are and 0.038, respectively. These statistics, together with our sample s broad coverage of the Compustat leveraged firms, are reassuring about the CIQ s data quality. To capture the extent of firms concentrated use of different debt instruments, for firm i in year t we first compute the sum of the squared ratios of seven debt instruments to total debt outstanding: SS SBN TD CP TD 2 2 SUB TD RC TD 2 2 CL TD TL TD 2 2 Other TD 2 (1) We then normalize SS to get the normalized Herfindahl-Hirschman Index (henceforth simply referred to as HHI): SS 1 HHI 7 (2) i,t

14 Thus, a firm that employs exclusively one single debt instrument has HHI equals one, while HHI equals zero for a firm that simultaneously employs all seven debt instruments in equal proportions. Higher HHI values indicate firms tendency to specialize in fewer debt instruments. Complementary to HHI, we also compute another specialization measure, the Shannon Entropy (henceforth referred to as Entropy): Ent SBN TD CP TD ln ln SBN TD CP TD SUB TD RC TD ln ln SUB TD RC TD CL TD TL TD ln ln CL TD TL TD Other TD ln Other TD (3) This ranges from zero (for firms specialized in borrowing through one debt instrument for all their financing needs) to ln(7) (for firms diversified in borrowing across all seven debt instruments in equal proportions). Table 2 Panel A shows that the mean (median) HHI is (0.729) while for Entropy the mean (median) is (0.426). Moreover, about one fourth of our sample firms have HHI values close to unity (or equivalently Entropy values close to zero), suggesting that firms have a great tendency to specialize in one debt instrument. Table 2 Panel B presents the time series evidence on debt types as well as our two specialization measures. We find that firms appear to rely more on bank financing (both revolving credit and term loans) and less on commercial paper, subordinated bonds and notes, and capital leases over the sample period. Their use of senior bonds and notes and other debt has been stable over time. Furthermore, there is a moderate increase in the degree of debt specialization over time: the mean HHI (Entropy) is (0.476) in 2002 and increases (decreases) to (0.407) in The observed temporal variation motivates our introduction of year fixed effects in our multivariate analysis later. In summary, although there are many different debt instruments, for our sample of firms, senior bonds and notes are the most commonly employed debt instrument, followed by revolving 12

15 credit and term loans. Moreover, many sample firms tend to rely heavily on only one debt instrument for meeting their financing needs. In the rest of the paper, we provide a more detailed investigation of the patterns and determinants of debt structure. C. Credit Ratings and Debt Structure The literature has previously examined the relation between credit ratings and leverage. Diamond (1991), Chemmanur and Fulghieri (1994), and Bolton and Freixas (2000) have shown that credit quality is the primary source of variation driving a firm s optimal choices of different types of debt. Faulkender and Petersen (2006) examine the role of the source of capital in firms financing decisions, and show that firms with access to public bond markets (as measured by being rated) have substantially more debt. Kisgen (2006) finds that firm credit ratings affect capital structure decisions: Firms near a credit rating upgrade or downgrade issue less debt. Lemmon and Zender (2009) use the likelihood of being rated as a proxy for debt capacity and show that after accounting for debt capacity, the pecking order appears to be a good description of a firm s financing behavior. Rauh and Sufi (2010) document that high credit quality firms (BBB and higher) rely almost exclusively on two tiers of capital equity and senior unsecured debt while lower credit quality firms (BB and lower) use multiple tiers of debt including secured, senior unsecured, and subordinated issues. Motivated by this prior body of work, Table 3 presents an overview of the relation between credit ratings and debt structure. We consider a firm-year to be rated if the firm has at least one monthly Standard & Poor s long-term issuer rating, as recorded in Compustat (data item 280). About 40% of our firm-year observations are unrated. In untabulated analysis, we find that there is little temporal variation in the fraction of firms being rated over time. 5 5 Using Compustat firms over the period , Faulkender and Petersen (2006) show that only 19% (21%) of firms (with positive debt) have debt ratings. They conclude that public debt is uncommon. In unreported analysis, we find that rated firms in our sample are larger (both in terms of sales and book value of assets), and have a higher book leverage ratio than those in Faulkender and Petersen (2006). This is consistent with Lemmon and Zender s 13

16 Panel A of Table 3 presents differences in the use of various debt types between unrated and rated firms in our sample. We show that revolving credit facilities and term loans together, on average, account for more than half of unrated firms total debt, while senior bonds and notes account for slightly less than 25% of their total debt. Rated firms are much heavier users of senior and subordinated bonds and notes. Unrated firms are much heavier users of capital leases. Further, we show that unrated firms are more specialized in borrowing than rated firms, as captured by both the HHI and Entropy measures. At the bottom of Panel A, we also present sample mean (median) market leverage of unrated and rated firms. Consistent with Faulkender and Petersen (2006), we show that unrated firms with a mean (median) market leverage ratio of (0.148) tend to employ less debt than rated firms who have a mean (median) market leverage ratio of (0.275). Overall, unrated firms use significantly less commercial paper as well as senior and subordinated bonds and notes; and significantly more revolving credit, term loans, and capital leases than their rated counterparts suggesting that both banks and lessors have a comparative advantage in dealing with information asymmetry and moral hazard associated with unrated firms (Diamond (1991), Chemmanur and Fulghieri (1994), and Eisfeldt and Rampini (2010)). To further examine the relation between actual credit ratings and debt structure, we first assign to each monthly S&P letter rating class an integer number ranging from 1 (AAA) to 22 (D). Then, for each rated firm-year we round the average monthly numeric rating class to the nearest integer, and refer to it as the firm rating in a given year. We find that about 6% of our firm-years have a credit rating of A and higher, while close to 16% of our sample firms have investment grade ratings (equal to or higher than BBB-, untabulated). 6 (2009) finding that large firms with high leverage are more likely to be rated. Focusing on 305 randomly chosen Compustat firms with a long-term issuer rating in at least one year from , Rauh and Sufi (2010) show that three-quarters of their firm-year observations are rated. 6 Using Compustat firms from , Kisgen (2006) shows that 44.2% of his sample firms have a credit rating of A and higher, and 69.1% of his sample have investment grade ratings. Rauh and Sufi (2010) report 21.7% of their firms have a credit rating of A and higher, and close to half have investment grade ratings. The difference in rating 14

17 Panel B of Table 3 provides an overview of differences in the usage of various debt types (as a share of total debt) across a broad rating spectrum. We first show that commercial paper is used almost exclusively by investment grade firms, especially AAA- and AA-rated firms. Second, term loans and subordinated bonds and notes are most heavily used by speculative grade (equal to or lower than BB-) firms. Third, we document a non-linear relation between credit quality and the amount of senior bonds and notes used by our sample firms: The amount of senior bonds is increasing in credit quality, peaks at the rating of A, and then is decreasing in credit quality. Finally, we find that firms belonging to lower rating classes are characterized by higher values of HHI (and lower values of Entropy), suggesting that the extent of debt specialization is decreasing in credit quality. We conclude that credit quality affects both the composition of debt structure and the extent of debt specialization. Later in our multivariate analyses, we will include dummy variables for each different rating class and for firms being unrated to control for the complex relation between credit ratings and debt structure. 7 D. Industries and Debt Structure Table 4 presents an overview of the relation between two-digit SIC codes and debt structure and debt specialization measures. The top three most represented industries are Chemicals and Allied Products (10.45%), Business Services (9.80%), and Electronic Equipment (8.17%). These three industrial sectors exhibit similar levels of specialization, as captured by both the HHI and Entropy measures. The three industries with the highest level of debt specialization are: Legal Services, Furniture and Home Furnishings, and Apparel and Accessory Stores. The three industries with the lowest level of debt specialization are: Local Passenger distributions between the Kisgen s sample and our sample is probably due to the fact that his sample includes financials and utilities which tend to have better ratings than industrial firms. 7 We assign an integer equal to 23 to the variable Rating for an unrated firm-year observation. 15

18 Transit, Auto Repair and Services, and Electric, Gas, and Sanitary Services. Overall, Table 4 shows significant variation in debt structure and debt specialization across different industries, which motivates the inclusion of industry dummies in our multivariate analyses later. IV. Debt Specialization In this section, we present comprehensive evidence on the extent of debt specialization in our sample firms. A. Cluster Analysis Our first piece of evidence on debt specialization comes from cluster analysis, which is commonly used to discover unknown structures in data by maximizing variance (in terms of the Euclidean distance) between clusters and minimizing it within clusters. Once we separate data in clusters, we effectively remove much of the variance in any of the debt types. Table 5 presents the mean and median values for different debt types and key firm characteristics across the identified clusters using firm-year observations, sorted according to ascending group mean firm size. 8 Figure 1 presents the distribution of different debt types within each cluster (mean ratios are used). There are a total of six clusters for our sample firms. Across these six clusters, the extent of debt specialization as captured by the HHI and Entropy measures are fairly similar across the first five clusters, while it decreases drastically for cluster 6 which includes the largest firms in the sample. More specifically, cluster 1 includes the smallest firms in our sample with market leverage close to the sample average. These firms tend to use predominantly revolving lines of credit for financing. The group mean (median) revolving lines of credit to total debt ratio is Firm characteristics are measured contemporaneously. Using lagged measures gives qualitatively the same results except that sample size is slightly smaller. 16

19 (0.90). Cluster 2 includes the second smallest firms in our sample with the lowest level of market leverage. These firms tend to use predominantly capital leases for financing. The group mean (median) capital leases to total debt ratio is 0.88 (1.00). Cluster 3 firms tend to use predominantly term loans for financing. The group mean (median) term loans to total debt ratio is 0.82 (0.88). Cluster 4 firms tend to use predominantly subordinated bonds and notes for financing. The group mean (median) subordinated bonds and notes to total debt ratio is 0.79 (0.83). Firms in cluster 5 are considerably bigger and these firms tend to use predominantly senior bonds and notes for financing. The group mean (median) senior bonds and notes to total debt ratio is 0.91 (0.95). Finally, cluster 6 includes some of the largest firms in the sample. These firms tend to use a mix of senior bonds and notes, revolving credit, and term loans. The group mean (median) senior notes and bonds, revolving credit, and term loans to total debt ratio is 0.48 (0.52), 0.17 (0.10), and 0.14 (0.02), respectively. It is worth noting that this group of firms has the lowest M/B ratios and cash flow volatility, and the highest level of market leverage among the sample firms. In summary, the evidence from cluster analysis suggests that the vast number of sample firms specialize in borrowing from one type of debt, and only the largest and least risky firms simultaneously employ multiple types of debt. Moreover, we do not observe asset maturity to vary considerably across clusters, which suggests that debt specialization cannot be the mere byproduct of firms matching the maturity of their liabilities to that of their assets. Our evidence thus far is consistent with the prior theoretical literature suggesting that the use of different types of debt is determined by firm characteristics including their credit quality and ability to access the public debt market, and within a large sample of firms like ours, there would be some level of debt specialization based on firms costs and benefits analysis. What is striking in our findings is that debt specialization seems to be a dominant phenomenon. Our findings on debt specialization are in stark contrast to Rauh and Sufi (2010), who show that their average sample firm simultaneously employs multiple types of debt. We attribute 17

20 the difference in findings to the different samples examined in their paper and ours. They focus primarily on large and rated firms, while in our sample only about 40% of the firms are rated. B. Conditional Debt Structure Our second piece of evidence on debt specialization comes from examining conditional debt structures. Table 6 Panel A presents the shares of firm-year observations conditional on a particular debt type exceeding 30% of total debt (i.e., the significant user). Looking across the rows in Panel A, we find that significant users of one debt type are rarely significant users of any other debt types. This is true with the exception of the significant users of commercial paper which simultaneously employ a significant amount of senior bonds and notes. In all other cases, Panel A indicates that if a firm s use of a particular type of debt exceeds 30% of its total debt, that type is likely to be its only source of debt financing. These results provide further support for the phenomenon that firms specialize in borrowing from one type of debt. If firms were simultaneously employing multiple types of debt, we would have observed few firms exceeding 30% of their debt from a single source of financing. Panel B presents both the mean and median ratios of each debt type to total debt conditional on a particular debt type exceeding 30% of total debt. Specifically, we first impose the condition that a firm s use of a particular debt type exceeds 30% of its debt, thus identifying a subset of firms. Then, for this subset we compute mean and median ratios of all debt types to total debt and test the null hypothesis that the mean (median) ratio is no greater than 30%. We also report the number of firm-year observations whose particular debt type is strictly greater than 30% of total debt. For example, in the first row we require that the amount of commercial paper exceeds 30% of debt. This leaves us with 142 observations. For these observations the mean (median) ratio of commercial paper to total debt is (0.468), the mean (median) ratio of revolving credit to total debt is (0.000), and so on for all other types. 18

21 Examining the numbers in bold face along the diagonal line of Panel B, we show that the ratio of a given debt type to total debt is between 65% and 80%, conditional on the ratio of that particular type of debt to total debt exceeding the threshold of 30% (again with the notable exception of the significant users of commercial paper). Further, the t- and median tests strongly reject the null that the mean and median ratios of various debt types are no greater than 30%. The off-diagonal numbers reveal that any significant reliance on more than one debt type is rarely observed: The exception is that the significant users of commercial paper are also significant users of senior bonds. The results in Panel B highlight the general phenomenon that very few firms use other sources of debt over and beyond the one which we condition upon. This is strong evidence of firms relying primarily on a single debt instrument. Figure 2 presents the distribution of different debt types conditional on a particular debt type exceeding 30% of total debt. With the exception of the significant users of commercial paper, we observe the conditional mean ratio of various debt types easily exceeding 70% (as shown along the vertical axis). We conclude that specialization not diversity in types of debt is the dominant phenomenon for debt structure. The natural question to ask next is what drives debt specialization. V. Explaining Debt Specialization and Debt Structure The vast literature on capital structure has shown that the optimal capital structure decision is made based on the trade-off between costs and benefits of debt financing. Extending this line of thinking, we expect that the extent of debt specialization is also driven by firm characteristics that proxy for costs and benefits of debt specialization. A. Determinants of Debt Specialization 19

22 Since we are the first to examine the determinants of debt specialization, we rely on prior work in capital structure to decide on possible explanatory variables (see for example, Fama and French (2002), Lemmon, Roberts and Zender (2008), and Frank and Goyal (2008)). We estimate the following regression: Debt Specialization P Profitability 1 T Tangibility 1 MB M / B 1 S Firm Size 1 V CF Volatility 1 D Dividend Payer 1 (4) M Asset Maturity 1 A Firm Age Year FE Industry FE Rating FE where the dependent variables are the two measures of the extent of debt specialization: HHI and Entropy. Table 7 Columns (1) and (3) present the OLS regression results. For both specialization measures, we show that asset tangibility is negatively and significantly, while M/B ratios, cash flow volatility, and asset maturity are positively and significantly, associated with the extent of debt specialization. Our findings suggest that when there are low monitoring costs as captured by high asset tangibility, firms are able to borrow from a variety of different sources of financing; leading to reduced reliance on a few debt instruments. On the other hand, when there are severe information asymmetry and high default risk as captured by high growth and high cash flow risk, firms are forced to concentrate their borrowing to fewer different sources of financing such as public debt or capital leases (see our hypotheses H3 and H4). The positive and significant association between asset maturity and debt specialization is mostly driven by senior bonds and notes, which is shown to be the predominant source of financing (Table 2 Panel A) and positively associated with asset maturity (Table 5, see cluster 5). Interestingly, we also show that in the midst of the 2008 financial crisis, firms scrambled to borrow from as many sources of financing as possible; leading to a significant decline in the degree of debt specialization, consistent with the evidence reported in Ivashina and Scharfstein (2009) and Gao and Yun (2009). Afterwards, the degree of debt specialization reverts back. 20

23 To get a sense of the extent to which firm characteristics affect the degree of debt specialization, we consider four alternative specifications of Equation (4) by separately including firm characteristics, year, industry, or rating fixed effects. Table 7 Columns (2) and (4) present the OLS regression results when including firm characteristics only, while the lower part in Table 7 reports the adjusted R2 for the different fixed effects specifications. The effects of firm characteristics on debt specialization are fairly similar whether the different fixed effects are included or not. When considering firm characteristics only, large firms appear to be less specialized, but this effect is not robust to the inclusion of various fixed effects. The full model specification in Equation (4) achieves an adjusted R2 of and for HHI and Entropy, respectively, while the goodness-of-fit of a model with firm characteristics only is about 50% lower. Industry and rating fixed effects contribute similar levels of explanatory power achieving an adjusted R2 in the range of 0.32 to 0.45, while year fixed effects alone are not overall effective in explaining debt specialization. All fixed effects together account for about three-quarters of the total explanatory power achieved by the full model specification in Equation (4). In summary, consistent with the theoretical predictions on the relation between the use of different debt types and firm characteristics reviewed in Section II, we show that the extent of debt specialization is also driven by concerns of information asymmetry and agency problems. B. Determinants of Debt Structure model: To examine the determinants of debt structure, we estimate the following regression Debt Type P Profitability 1 T Tangibility 1 MB M / B 1 S Firm Size 1 V CF Volatility 1 D Dividend Payer 1 (5) M Asset Maturity 1 A Firm Age Year FE Industry FE Rating FE 21

24 where the dependent variables debt types are calculated as a fraction of total debt. As we have shown in Table 2 Panel A, the distribution of debt types has a mass point of observations at zero for over half of the firm-year observations when a particular debt type is not employed, which may raise some concerns about the conditional normality of the dependent variable in Equation (5). As a result, we estimate Equation (5) with a Tobit specification that accounts for two-sided censoring of the dependent variables at zero and one, and present results in Table 8. 9 We show that firm size and age is positively and significantly associated with the use of commercial paper, consistent with our first hypothesis (H1). Moreover, we show that profitability, tangibility, M/B ratios, and the dividend payer dummy are positively and significantly associated with the usage of commercial paper (Column (1)). Given that profitability and tangibility are important determinants of credit ratings and that dividend payments may signal profitability, the evidence further corroborates our first hypothesis. We show that M/B ratios and cash flow volatility are negatively and significantly associated with revolving credit facilities, consistent with our second hypothesis (H2). Outside the predictions of our second hypothesis, we also show that profitability, asset tangibility, and the dividend payer dummy are positively and significantly, while firm size and asset maturity are negatively and significantly, associated with revolving credit facilities (Column (2)). The former is consistent with banks close monitoring of line of credit drawdowns (Jiménez, López, and Saurina (2008) and Sufi (2009)) they only provide credit lines to firms that are sufficiently profitable, i.e., paying dividends. The latter is also intuitive as small firms tend to have more volatile cash flows leading to lower usage of revolving credit; and as revolving credit facilities 9 In untabulated analysis, we first establish a benchmark by regressing market leverage on the same set of firm characteristics. We find that consistent with prior research such as Frank and Goyal (2008) and Lemmon, Roberts, and Zender (2008), profitability, M/B, cash flow volatility, and the dividend payer dummy are negatively and significantly associated with market leverage; while asset tangibility and firm size are positively and significantly associated with market leverage. 22