ASSET REDEPLOYABILITY AND THE CHOICE BETWEEN BANK DEBT AND PUBLIC DEBT*

Transcription

1 ASSET REDEPLOYABILITY AND THE CHOICE BETWEEN BANK DEBT AND PUBLIC DEBT* Chelsea Chen a, David A. Maslar b, and Matthew Serfling c August 2018 ABSTRACT A firm with less redeployable assets, which are assets that have fewer alternative uses outside the firm, is more likely to borrow from banks than issue public debt, especially when credit market conditions are tight. These findings are consistent with firms with less redeployable assets valuing the ability to renegotiate bank debt contracts instead of liquidating assets in the event of default. Consistent with this mechanism, firms with lower asset redeployability are more likely to modify covenants during a loan renegotiation event and sell fewer assets following covenant violations. Our results contribute to work on the determinants of which debt markets a firm chooses to borrow from and the role that banks play as intermediaries. Keywords: Asset redeployability, Debt structure, Bank debt, Public debt, Debt issuance JEL Classifications: D92, G32, G33 * We are grateful for the helpful comments and suggestions from Douglas Fairhurst, He Li (discussant), Sarah Shaikh, seminar participants at the University of Tennessee, and conference participants at the 2017 Southern Finance Association annual meeting. a Department of Finance, Ohio University, hchen@ohio.edu. b Department of Finance, University of Tennessee, dmaslar@utk.edu, tel: c Corresponding Author: Department of Finance, University of Tennessee, Stokely Management Center, 916 Volunteer Boulevard, Knoxville, TN 37996, mserflin@utk.edu, tel:

2 1. Introduction An important decision a firm s financial manager has to make is how to finance the firm s investments. While there is extensive work examining the choice between issuing equity and debt, most research treats all debt as uniform. However, a large body of theoretical work (e.g., Diamond (1991, 1993), Bolton and Scharfstein (1996), Bolton and Freixas (2000), Park (2000), and DeMarzo and Fishman (2007)) and recent empirical work (e.g., Rauh and Sufi (2010) and Colla, Ippolito, and Li (2013)) recognize heterogeneity in the sources of debt financing and seek to understand why firms choose to raise funds from different sources. Bolton and Scharfstein (1996) go so far as to suggest that, given the economic magnitude of the debt market and the observation that firms tend to utilize debt more often, (pp. 2): it may be more important to understand the structure of debt financing than the choice between debt and equity. 1 An important determinant of a firm s borrowing choice is the firm s underlying asset structure, typically measured by the fraction of its assets comprised of property, plant, and equipment. However, both theoretical and empirical work offer mixed findings as to how asset structure affects which debt market a firm borrows from. We propose that a potential reason for the mixed findings is that prior studies do not recognize and account for the substantial heterogeneity among different types of physical assets. Like human capital, some physical assets are firm and project specific, while other assets can be used by multiple firms for various projects. If a project fails and lenders take control of an asset, or a firm goes through the process of costly reorganization and must sell some of its assets, the ability to liquidate the firm s assets will be an important determinant of whether creditors recover the loan principal or whether the firm can successfully restructure. Thus, due to market frictions, even if a firm holds a large quantity of physical assets, the potential benefits of holding more physical assets may diminish if these assets are 1 In addition, U.S. corporations have increased their use of debt over the past century, with debt-to-capital ratios averaging about 12% in the 1920s and about 28% in the early 2000s (Graham, Leary, and Roberts (2015)). Consistent with this finding, the amount of new debt issues relative to new equity issues has been increasing over time. In 1987, U.S. corporations issued $326 billion of bonds and $66 billion of equity. In 2015, these values grew to $1,611 billion in new bonds and $174 billion in new equity (see Board of Governors of the Federal Reserve System (1989, 2017)). 1

3 highly specialized and cannot be used by other firms. In this study, we provide new insights into debt financing decisions by capturing the heterogeneity in a firm s underlying asset structure using a measure of the redeployability of its physical assets. We use this measure to examine whether the redeployability of a firm s assets affects its choice between raising bank debt and issuing public bonds. We use the asset redeployability measure from Kim and Kung (2017), which captures the extent to which a firm s underlying assets have alternative uses both within and across industries and is therefore related to the liquidity and resale value of those assets. 2 Kim and Kung (2017) construct the measure using data from the capital flow table from the Bureau of Economic Analysis (BEA), which breaks down expenditures on new equipment, software, and structures by asset class and industry. The measure incorporates the dimensions of both asset specificity and asset market thickness. Supporting the validity of the measure, firms with more redeployable assets are more active in asset sales and exhibit higher recovery rates in default (Kim and Kung (2017)). We focus on the choice between bank and public debt because these are two important sources of debt financing (e.g., Johnson (1997) and Rauh and Sufi (2010)). How might having physical assets and the redeployability of these assets affect which debt market a firm raises capital from? While prior work leads to different findings and conclusions, one commonality in these studies is that they argue that having more physical assets increases liquidation and collateral values, which in turn affects the debt issuance decision. On the one hand, as collateral values increase due to having more physical assets, firms should prefer bank debt over public debt. The underlying argument is that a bank s ability to efficiently liquidate a defaulted firm is more valuable in cases in which the firm s underlying assets are worth more (Berlin and Loeys (1988) and Park (2000)). Consistent with this view, Lin (2016) finds that an increase in the collateral value of a firm s real estate is associated with a greater use of bank debt but not public debt. Thus, to the extent 2 Kim and Kung (2017) argue that asset redeployability represents a cost that creates a wedge between the purchase price of an asset and the asset s liquidation value (investment reversibility). This friction is also closely related to asset specificity as described in Williamson (1983, 1988). 2

4 that more redeployable assets have higher collateral values, we should find that firms with lower asset redeployability prefer public debt over bank debt. On the other hand, as physical assets are easier to value and require less external monitoring, firms with fewer physical assets have higher bankruptcy costs, which makes it harder for these firms to borrow from arms-length lenders in public debt markets (Hoshi, Kashyap, and Scharfstein (1993) and Cantillo and Wright (2000)). Further, firms with fewer physical assets that are harder to sell have higher liquidation costs and should value the option to renegotiate loan terms instead of selling assets in the event of default that bank debt provides. Compared with public debt with dispersed ownership, bank debt has concentrated holdings and is easier to renegotiate in the event of default. 3 Consistent with these predictions, Denis and Mihov (2003), Cantillo and Wright (2000), and Rauh and Sufi (2010) find that firms with fewer physical assets use more bank debt. Cvijanović (2014) also finds that a decrease in the market value of a firm s real estate is associated with a decrease in both bank and public debt. Overall, to the extent that less redeployable assets are harder to liquidate in the event of default, we should find that firms with lower asset redeployability prefer bank debt over public debt. To explore the relation between the redeployability of a firm s assets and its choice to borrow from a bank or issue public debt, we examine the incremental borrowing choice using a sample of 20,178 new bank loans and bond issues. This approach is similar to the one used by Denis and Mihov (2003) and Morellec, Valta, and Zhdanov (2015) and allows us to link a firm s characteristics just prior to the borrowing decision to the choice between borrowing from a bank and issuing public debt. After controlling for several characteristics known to affect the choice of debt financing type and whether the firm has access to both private and public debt markets, we find that firms with lower asset redeployability are more likely to borrow from banks than issue public debt. Importantly, in all of our 3 With respect to renegotiation, in the case of public debt, the Trust Indenture Act of 1939 requires that firms must receive the unanimous consent of public bondholders to alter any of the material terms of the bond indenture. Given that public debt tends to be owned by many investors, this dispersed ownership makes renegotiation of public bond terms costly and often infeasible. 3

5 tests, we control for the proportion of a firm s assets that are fixed, and therefore the effect of asset redeployability on the debt issuance decision is over and above the effect of having more fixed assets. Further, the economic magnitude of the effect of asset redeployability on the debt issuance decision is similar to the effect of other determinants, such as profitability and default probability. In our analyses, endogeneity could arise in two ways. The first potential endogeneity concern is related to selection bias, as firms must first decide to raise external capital before they choose whether to borrow from banks or issue public bonds. The second potential concern is that our results could suffer from an omitted variables bias. By construction, the asset redeployability measure is largely industry-specific and time-invariant, which precludes the inclusion of industry or firm fixed effects in our regression models. It is therefore possible that our results are driven by other industryor firm-level heterogeneity that is correlated with the asset redeployability measure. Further, our models could insufficiently account for differences between firms with low and high asset redeployability, and it is these differences that drive the relation between asset redeployability and the debt issuance decision. We address these concerns in three ways. First, we show that our main results are robust to using a Heckman selection model. In the first stage, we estimate the likelihood that a firm raises capital from the external debt market. We then reestimate our main results after controlling for the inverse Mill s ratio we obtain from the estimates in the first-stage selection model. In the first stage model, we include a measure of whether a firm s investment needs exceed its internally generated cash flows, which captures a firm s need for external financing. We exclude this variable from the second stage. This variable is correlated with the decision to raise external capital but unlikely correlated with whether the firm borrows from a bank or issues public debt after controlling for measures of the firm s financial condition. Second, we attempt to identify situations in which the marginal effect of asset redeployability on which debt market the firm borrows from is greater. One possible such situation is based on credit market conditions. For example, when credit market conditions are tighter, the potential buyers of a 4

6 firm s assets have less access to capital, and the firm will find it harder to sell its assets. Thus, we expect that if firms with lower asset redeployability prefer bank debt over public debt due to the high liquidation costs associated with selling their assets, then the positive relation between low asset redeployability and the choice to raise bank debt over public debt should be stronger when credit markets are tighter. Importantly, we are able to exploit the cross-sectional variation of these tests and control for firm fixed effects in the regressions. We use four credit spread measures to proxy for economy-wide borrowing conditions: (1) the TED spread, (2) the excess bond premium spread, (3) the Gilchrist and Zakrajšek (2012) credit spread index, and (4) the Baa-Aaa corporate bond spread. We also capture state-wide borrowing conditions using the Black and Strahan (2001) index of state-level bank branching deregulation. Consistent with the prediction, we find that firms with lower asset redeployability are even more likely to borrow from banks instead of issuing public debt when credit market conditions are tighter. Third, we perform a propensity score matched sample analysis. We match firms in the lowest tercile of asset redeployability to firms in the highest tercile based on firm and industry characteristics. The results from diagnostic tests suggest that the matching procedure is successful, and we continue to find that firms with lower asset redeployability are more likely to choose bank debt over public debt. Our results are consistent with the underlying mechanism that firms with less redeployable assets choose bank debt over public debt due to the ability to renegotiate debt contracts instead of selling assets when they default. We test this conjecture and find that our results are consistent with this underlying channel in two ways. First, we use a sample of loan renegotiations from Roberts (2015) and show that firms with less redeployable assets are more likely to modify covenants during renegotiation events. Second, we show that firms with lower asset redeployability are also less likely to sell assets following covenant violations. The assumption underlying this second test is that if a firm defaults and it does not liquidate assets, the firm is renegotiating its debt terms. While our primary research question focuses on the relation between asset redeployability and 5

7 the debt issuance decision, we also examine the relation between asset redeployability and the cost of borrowing from banks and public debt markets. To account for asset redeployability affecting which debt market a firm chooses to borrow from, we estimate an endogenous switching regression model. Our results suggest that firms with less redeployable assets are charged higher spreads by banks and in public debt markets. We also address a number of residual concerns. First, as a falsification test, we show that the relation between asset redeployability and debt financing decisions is more pronounced and largely holds only for firms that have potential access to both public debt markets and bank loans. Second, our results are robust to using alternative measures of asset redeployability. Last, we find that lower asset redeployability has a similar effect on a firm s choice to issue non-bank private debt (e.g., Rule 144A bonds) over public debt as it does on the firm s decision to raise bank debt over public debt. Our study contributes to the literature in three ways. First, our paper relates to a growing body of empirical work examining the determinants of which debt markets a firm s financial managers choose to borrow from. While bank loans and public bonds collectively represent a substantial portion of a firm s capital structure, an open question in the literature is: why do some firms utilize credit from private sources while other firms borrow from arms-length lenders in the public markets? Factors shown to affect this decision include, but are not limited to, credit quality, growth opportunities, competition, governance, information asymmetry, collateral values, corporate ownership structure, and earnings quality (e.g., Denis and Mihov (2003), Bharath, Sunder, and Sunder (2008), Lin, Ma, Malatesta, and Xuan (2013), Dey, Nikolaev, and Wang (2015), Morellec, Valta, and Zhdanov (2015), and Lin (2016)). We build on this work by showing that the redeployability of a firm s assets is also important and incremental in these decisions. Second, because understanding a firm s debt structure is also important for understanding the firm s capital structure (e.g., Bolton and Scharfstein (1996), Rauh and Sufi (2010), and Colla, Ippolito, and Li (2013)), our paper more broadly contributes to work on how firms make capital structure 6

8 decisions. In particular, our paper relates to how asset structure and the liquidity of these assets affect capital structure decisions (e.g., Morellec (2001), Sibilkov (2009), Liu and Wong (2011), Campello and Giambona (2013), and Kim and Kung (2017)). Third, our results contribute to a better understanding of the role that financial institutions play in credit markets and the benefits they provide to borrowers (e.g., Jayaratne and Strahan (1996) and Butler, Cornaggia, and Gurun (2017)). Our findings imply that the ability to renegotiate bank debt contracts provided by the concentrated holdings of this debt by financial institutions is a valuable option for firms with assets that are costlier to liquidate. 2. Sample Selection and Empirical Methodology 2.1. Sample Selection To construct our sample of new bank debt and public debt issues, we follow Bharath, Sunder, and Sunder (2008) with a few modifications. We use Loan Pricing Corporation s Dealscan database to obtain data on new bank loans. This database consists of detailed information on commercial loans made to corporations. Dealscan data are compiled from SEC filings and public documents (10Ks, 10Qs, 8Ks and registration statements), loan syndicators, and other internal sources. The data are organized by package, which is a contract signed between a borrower and a lender at a particular date. These packages can contain multiple loans (i.e., facilities ). Each of the loans can have different maturities, different contract terms and characteristics (e.g., revolvers or term loans), and serve different purposes (e.g., used for takeovers or working capital needs). In this paper, we treat each package as a unit of observation. To enter the sample, the package must have at least one facility with non-missing data on loan amount, interest rate spread (all-in-drawn spread over LIBOR), and maturity. We obtain bond issuance data from Thomson Reuters Securities Data Corporation (SDC) database. To enter the sample, the bond must have non-missing data on amount, yield, and maturity. We create a sample of public bonds by following prior work and excluding convertible bonds, callable bonds, private placements, and Rule 144A eligible bonds from the sample (e.g., Bharath, Sunder, and 7

9 Sunder (2008) and Lin, Wang, and Wu (2011)). In Section 4.3, we add Rule 144A bonds and other private placements back to our sample to examine how asset redeployability affects the probability of issuing non-bank private debt over public debt as well as bank debt over public debt in a multinomial logistic regression. We obtain firm financial statement data from the CRSP-Compustat merged sample for the years 1987 to We exclude utilities (SIC ), financial firms (SIC ), and any firm-year observation with missing data for our main variables of interest. After merging this data with SDC and Dealscan, our sample consist of 20,178 new debt issues of which 17,378 are for bank loans and 2,800 are for public bonds Construction of the Asset Redeployability Measure In this section, we describe the construction of the asset redeployability measure from Kim and Kung (2017). 4 A key advantage of their redeployability measure is that it accounts for asset use across as well as within industries. While industry peers are more likely to place a higher value on a firm s assets, they are also more likely to face the same financial and operational difficulties as the firm liquidating its assets. Thus, capturing the cross-industry dimension of asset use is important for measuring a firm s asset redeployability. As we use the data provided by Kim and Kung (2017), we only briefly describe the process that the authors use to construct the measure and highlight the components that directly affect our analyses. The construction of firm-level asset redeployability involves three steps. First, the authors use the 1997 BEA capital flow table to calculate a redeployability score for each asset category in the BEA table. The table breaks down expenditures on new equipment, software, and structures by 180 assets for 123 industries. Each asset s redeployability score is the sum of weights of industries that use 4 We refer the reader to Kim and Kung (2017) for a more in-depth discussion of the construction of the asset redeployability measure. We thank Hyunseob Kim and Howard Kung for making their asset redeployability measure available through The Review of Financial Studies on the publisher s website. 8

10 the asset among the 123 industries in the BEA table, where weights are the sum of the market capitalizations of all public firms in an industry over the sum of market capitalizations across all public firms. As Kim and Kung (2017) note, the asset-level redeployability scores appear reasonable. For example, the score is 0.66 for industrial trucks, trailers, and stackers, which are used in a wide range of industries, while it is 0.02 for drilling oil and gas wells, which are predominantly used in the oil and gas industries. Second, Kim and Kung (2017) aggregate the asset redeployability scores across the 180 different asset categories in order to construct an industry-level redeployability index. This index equals the value-weighted average of each asset redeployability score used by the industry, in which weights equal the industry s expenditures on a given asset category divided by the industry s total capital expenditures from the BEA table. Last, Kim and Kung (2017) calculate firm-level asset redeployability as the value-weighted average of the industry-level redeployability indexes across a firm s business segments. The authors test the external validity of their asset redeployability measure. Consistent with more redeployable assets having higher liquidation values, they show that this measure is positively correlated with creditors recovery rates after a firm defaults as well as associated with more active transactions in used asset markets General Empirical Methodology To test whether the redeployability of a firm s assets is related to its choice of borrowing from a bank or issuing public debt, we estimate the following probit regression model: Pr( Bank Debt ) Φ( α Low Redeployability Xβ υ ε ), (1) i, t 1 1 i, t t t i, t 1 where Bank Debt i,t+1 is an indicator variable that is set to one if firm i raises bank debt during fiscal year t+1 and equals zero if firm i issues a public bond during fiscal year t+1. Low Redeployability is our variable of interest and captures the extent to which a firm s assets are less usable by other firms. 9

11 Low Redeployability equals one minus the redeployability score of a firm s assets from Kim and Kung (2017). Thus, higher values of Low Redeployability indicate that the firm s assets are less redeployable. If firms with less redeployable assets are more likely to choose to raise bank debt over public debt, then the estimated coefficient α 1 will be positive. In all reported results, we tabulate marginal effects. In Eq. (1), X is a set of firm-level control variables that account for a number of firm-specific factors that have been shown to influence a firm s choice between raising bank debt and issuing public bonds (e.g., Denis and Mihov (2003) and Morellec, Valta, and Zhdanov (2015)). Specifically, we control for book value of assets, existing financial leverage, profitability, the market-to-book ratio, the fraction of assets that are fixed, cash flow volatility, expected default probability (following the method in Bharath and Shumway (2008)), and whether the firm has a long-term debt rating (a control for whether a firm has access to both bank and public debt markets). We also control for the number of business and operating segments the firm reports. In constructing this list of control variables, we largely follow prior work, but note that these controls are often motivated from theoretical models of bank specialization (Diamond (1991) and Rajan (1992)). Because the asset redeployability measure varies largely at the industry level, as a first pass at controlling for industry characteristics that could be correlated with asset redeployability and a firm s debt issuance decision, we also control for average industry-level financial leverage, profitability, and market-to-book ratio based on 4-digit SIC industries. Importantly, we examine the effect of asset redeployability on the choice of which type of debt a firm issues after controlling for the fixed asset ratio. 5 Thus, the coefficient estimate α 1 is interpreted as the effect of asset redeployability on the debt issuance decision, holding the portion of assets that are fixed as well as other firm characteristics constant. Following Denis and Mihov (2003), we consider the incremental choice of whether a firm selects bank or public debt. We thus are able to link the bank-public debt issuance decision to the 5 In untabulated analyses, we verify that our main result from Table 2 is robust to controlling for not only the proportion of assets that are fixed based on book values, but also the value of a firm s real estate assets based on market values (e.g., Chaney, Sraer, and Thesmar (2012), Cvijanović (2014), and Lin (2016)). 10

12 characteristics of a firm at the time the decision is made. As Rauh and Sufi (2010) show, after classifying a firm s debt into bank debt, straight bond debt, convertible bond debt, program debt, mortgage debt, and all other debt, (pp. 4243) for almost 70% of firm-year observations in our sample, balance-sheet debt comprises significant amounts of at least two of these types. Furthermore, the authors discuss that even in cases in which a firm s total debt levels do not change, firms often adjust their debt composition by altering their debt outstanding in the various categories. In taking an incremental approach, we are able to examine the characteristics of the firm at each point in time when the choice is made to borrow. In all of our tests, we also include year indicator variables to account for nation-wide factors and trends affecting the choice to raise bank debt versus public debt over time. We also correct for heteroskedasticity and serial correlation by clustering standard errors at the firm level. Finally, we winsorize continuous variables at their 1st and 99th percentiles, and all dollar values are expressed in 2015 dollars. Panel A of Table 1 presents detailed definitions and summary statistics for our main variables for the full sample. In our sample, 86.1% of debt issues are bank debt. Panel B compares the means of the variables for firms with low asset redeployability (bottom tercile) to firms with high asset redeployability (top tercile). Several variables are significantly different across the two samples. As a first pass at controlling for these differences, we include each variable as a control variable. In Section 3.2.3, we more directly address this issue by performing a propensity score matched sample analysis. 3. Empirical Results 3.1. Asset Redeployability and the Choice between Bank and Public Debt We begin by testing the relation between asset redeployability and whether a firm prefers bank debt or public debt. Table 2 presents the results of this analysis. Consistent with firms with lower asset redeployability valuing the option to renegotiate debt contracts instead of liquidating assets in the event of default, column 1 shows that firms with lower asset redeployability are more likely to raise bank debt over issuing public debt. The estimated coefficient on Low Redeployability of implies that 11

13 a one standard deviation lower value of asset redeployability is associated with a 1.7 (= ) percentage point higher likelihood of choosing bank debt. This value is economically significant when compared to the effect of the other determinants of the debt issuance decision. For example, a one standard deviation increase in the natural logarithm of book assets, return on assets, and fixed assets is associated with a decrease in the likelihood of raising bank debt of 10.9, 2.2, and 1.6 percentage points, respectively. A one standard deviation increase in default probability is associated with an increase in the probability of raising bank debt of 2.0 percentage points, and firms with a debt rating are 9.1 percentage points less likely to borrow from a bank. In our sample, firms may choose to re-enter debt markets multiple times per year. To alleviate the concern that a small number of repeat issuers are driving our results, we repeat the analysis in column 1 after collapsing the sample to one observation per firm per year and report the results in column 2. We redefine the dependent variable to equal the dollar amount of all bank debt raised during the year divided by the total dollar amount of all debt issued during the year. Because the dependent variable is bounded between zero and one, we estimate the regression in column 2 with a Tobit model. Consistent with the results in column 1, column 2 also shows a positive relation between low asset redeployability and the amount of bank debt raised Addressing Endogeneity The first potential endogeneity concern is related to selection bias, as firms must first decide to raise external capital before they choose whether to borrow from banks or issue public bonds. The second potential concern is that our results could suffer from an omitted variable bias. By construction, a limitation of the asset redeployability measure is that it is largely industry-level and time-invariant. As such, we find that 4-digit SIC industry and firm fixed effects explain 90% and 95% of the variation in the measure, respectively. This relative stickiness precludes the inclusion of industry or firm fixed effects in our regression models. Because of this feature of the measure, we focused on cross-sectional variation in the relation between asset redeployability and the debt issuance choice instead of time- 12

14 series variation. Consequently, it is possible that our results are driven by other industry- or firm-level heterogeneity that is correlated with the asset redeployability measure. In the following sections, we address these potential concerns in three ways. First, we reestimate our main analysis using a Heckman selection model. Second, we estimate models in which the redeployability measure is interacted with measures of economy- or state-wide borrowing conditions to capture situations in which for any given level of asset redeployability, a firm is more likely to borrow from a bank than issue public debt. By exploiting the cross-sectional nature of these tests, we are able to include firm fixed effects (which also control for industry fixed effects) in the regressions and examine how changes in credit market conditions affect firms based on their degree of asset redeployability. Third, we implement a propensity score matched sample analysis that corrects for endogenous selection on observed variables (Rosenbaum and Rubin (1983) and Dehejia and Wahba (2002)). In the absence of an instrument for asset redeployability, a partial remedy is to match firms with high asset redeployability to firms with low asset redeployability along observable dimensions so that differences in asset redeployability is the only source of observable variation across the two groups. The hope is that, once the two groups are made similar along observable dimensions, other systematic variation due to unobservable differences will also be minimized Heckman Two-Step Selection Models We first examine whether our main results in columns 1 and 2 of Table 2 are robust to correcting for selection bias in which the redeployability of a firm s assets first affects whether the firm chooses to raise capital from external debt markets. After correcting for this potential bias, we then reexamine whether asset redeployability affects which debt market the firm chooses to borrow from. In the first stage of the Heckman selection model, we estimate the likelihood that a firm raises capital from the external debt market in a given fiscal year. We report the results in column 3 of Table 2. This sample includes all of the firms in the CRSP-Compustat merged database for the years 1987 to 2015 (excluding financials and utilities). 13

15 To implement the Heckman selection model, we need a variable that is correlated with the decision to raise external debt capital but is not correlated with whether the firm borrows from a bank or issues public debt after controlling for the other variables in the model (exclusion restriction). One possible variable is the extent to which a firm s investment needs exceed its internally generated cash flows. When investment needs exceed internally generated cash flows, a firm will need to raise capital in external markets (e.g., Rajan and Zingales (1998) and Albring, Banyi, Dhaliwal, and Pereira (2016)). However, it is not obvious why this measure would be correlated with which debt market the firm borrows from after controlling for measures of the firm s financial condition, such as profitability, default probability, and market-to-book ratio. Following Rajan and Zingales (1998), we measure a firm s dependence on external finance as the firm s capital expenditures less its cash flows from operations. We calculate operating cash flows following Byoun (2008). We set the indicator variable External Finance Dependence to one if a firm s capital expenditures exceed its operating cash flows and zero otherwise. The results in column 3 show a positive correlation between a firm needing external capital and the likelihood that it raises capital from debt markets. The coefficient estimate of on External Finance Dependence implies that firms in which capital expenditures exceed operating cash flows are 1.7 percentage points more likely to raise debt capital, which translates into an increase of 6.8% relative to the 24.9% of firms that issue debt in any given year (=0.017/0.249). Firms with less redeployable assets are also less likely to raise debt capital. Columns 4 and 5 repeat the analyses in columns 1 and 2 after controlling for the inverse Mill s ratio obtained from the estimates in column 3. Overall, the results in columns 4 and 5 imply that our main result that firms with less redeployable assets are more likely to raise bank debt than issue public bonds is robust to controlling for the potential selection bias arising from asset redeployability also affecting the decision to raise external debt capital. 6 6 The sample sizes of new bank loans and public bond issues in columns 4 and 5 are smaller than those in columns 1 and 2 due to the requirement that data on capital expenditures and operating cash flows must not be missing. 14

16 Effect of Credit Market Conditions We next examine how credit market conditions affect the relation between asset redeployability and the choice between borrowing from a bank and issuing public debt. When credit market conditions are tighter, the potential buyers of a firm s assets have less access to capital. As a result, if liquidation concerns drive our results, then for any given level of asset redeployability, the marginal impact of asset redeployability will be larger when credit markets are tighter. To test this prediction, we interact the measure of asset redeployability with four measures of economy-wide borrowing conditions and one-measure of state-wide financing constraints. For these tests, we estimate linear probability models to allow for firm and year fixed effects and to better interpret the marginal effects on the interaction terms. Columns 1-4 of Table 3 present the results of the analysis examining the effect of economywide borrowing conditions on the relation between asset redeployability and the choice to raise bank debt over public debt. We capture credit market conditions using four credit spread measures. Our first measure is the TED Spread (column 1). The TED Spread is the difference between the 3-month LIBOR and the 3-month U.S. Treasury Bill rate. It is used as a conventional gauge of credit risk. It measures a bank s frictions in obtaining funding, and there is a large literature documenting that frictions in the supply of capital to banks spill over to the funding of firms (e.g., Ivashina and Scharfstein (2010)). Our second and third measures of credit market conditions are from Gilchrist and Zakrajšek (2012). Gilchrist and Zakrajšek (2012) create a spread between corporate bonds and synthetic risk-free securities that mimic the cash flows of the corresponding corporate debt (GZ credit spread). This measure offers a significant improvement in predicting future economic activity compared to existing measures. From this credit spread, they also create the excess bond premium (EBP) spread. They provide evidence that an increase in the EBP spread reflects a reduction in the risk-bearing capacity of the financial sector, and as a result, a contraction in the credit supply. Column 2 captures economywide borrowing conditions using the EBP spread, and as robustness, column 3 uses the GZ credit 15

17 spread. Our final measure is the Moody s Baa-Aaa corporate bond spread the difference between the yields on indexes of seasoned long-term Baa- and Aaa-rated corporate bonds (column 4). Columns 1-4 present results from regressions that include year fixed effects, firm fixed effects, and control variables. Overall, the results are consistent with our prediction that the positive relation between low asset redeployability and the choice of borrowing from banks over issuing public debt is stronger when credit market conditions are tighter. Column 5 presents the results of the analysis examining the effect of state-wide borrowing conditions on the relation between asset redeployability and the choice to raise bank debt over public debt. To capture variation in state-wide external financial market frictions, we exploit the extent of bank branch deregulation across U.S. states over the 1987 to 1998 period. 7 As bank branching becomes less regulated, competition increases and frictions in financial markets decrease, making it easier for firms to obtain external capital. Thus, the positive relation between low asset redeployability and a firm s choice to raise bank debt instead of public debt should be weaker when bank branching is less restricted. To measure the extent to which states allow bank branching (and hence banking deregulation) across states over time, we follow Black and Strahan (2001) and Hombert and Matray (2016) and construct a deregulation index as the sum of the ways in which banks can expand their operations. This index ranges from zero to three and captures whether states allow: (1) multibank holding companies, (2) intrastate branching via mergers and acquisitions, and (3) unrestricted (de nova) intrastate branching. Due to the staggered timing of deregulation across states and the inclusion of firm fixed effects in the regression, column 5 examines the effect of banking deregulation on the relation between asset redeployability and the debt issuance choice in a difference-in-differences setting. The results show that, while firms with less redeployable assets are more likely to raise bank debt than public debt, this relation is attenuated when access to credit is greater. 7 State-wide bank branching deregulation began in the 1970s and ended in the 1990s. Thus, for this analysis, we follow Hombert and Matray (2016) and end our sample period in 1998 one year after the end of the deregulation period. 16

18 Propensity Score Matched Sample Analysis To implement the propensity score matched sample analysis, we match firms in the bottom tercile of asset redeployability to firms in the top tercile of asset redeployability (firms with asset redeployability in the middle tercile are excluded from the sample). To do so, we use a probit regression model and first regress an indicator variable for whether the firm has asset redeployability in the bottom tercile on our previously used control variables and estimate the probability (i.e., propensity score) that a firm has low asset redeployability. Column 1 in Panel A of Table 4 reports the marginal effects from this regression. Next, we match each observation when a firm has low asset redeployability to an observation when a firm has high asset redeployability. We match without replacement and require the propensity scores for each matched pair to be within ±1.0% of each other. The resulting sample consists of 4,903 debt issuances in which a firm has low asset redeployability matched to 4,903 debt issuances in which a firm has high asset redeployability. Following Fang, Tian, and Tice (2014) and Dhaliwal, Judd, Serfling, and Shaikh (2016), we perform several diagnostic tests to evaluate the successfulness of the matching procedure. If the matching procedure is successful, then we should find that: (1) the control variables in the matched sample do not explain any variation in whether a firm has low asset redeployability, (2) the difference in the propensity scores of firms with low and high asset redeployability is negligible, and (3) the means of the matched variables are not statistically different for firms with low and high asset redeployability. We test these predictions in three ways. First, we rerun the same model specification as in column 1 of Panel A for the matched sample and report the results in column 2. The results show that all of the control variables are statistically insignificant and the pseudo-r 2 drops to less than 0.1%, indicating that the control variables do not explain any variation in whether a firm has low asset redeployability. Second, we examine the difference between the propensity scores of firms with low and high asset redeployability and tabulate the results in Panel B. The mean difference is less than 17

19 0.001 and therefore trivial. Third, Panel C reports the univariate comparisons of the means of each matched variable between firms with low and high asset redeployability. The results show that the means of the matched control variables are statistically the same across the two samples, except for the number of business segments, but this difference is negligible. Collectively, these diagnostic tests suggest that the matching procedure is successful. The univariate results in Panel C also show that, for the matched sample, firms with low asset redeployability are 2.2 percentage points more likely to raise bank debt over public debt when compared to firms with high asset redeployability. Columns 1 and 2 of Panel D present the multivariate results from our base regression in column 1 of Table 2 for the matched sample in which we measure asset redeployability with the indicator variable for whether the firm has asset redeployability in the bottom tercile as well as with the continuous variable Low Redeployability. Consistent with our earlier findings, the results continue to show that firms with lower asset redeployability are more likely to borrow from a bank than issue public debt Asset Redeployability and the Ability to Modify Loan Covenants Our finding that firms with less redeployable assets are more likely to borrow from banks than issue public debt is consistent with these firms valuing the option to renegotiate debt contracts instead of liquidating assets in the event of default. If this interpretation is correct, then we would expect that firms with less redeployable assets are more likely to renegotiate covenants. We examine this channel in two ways in the following sections Asset Redeployability and Covenant Modifications In our first approach, we use the debt renegotiation and covenant modification sample from Roberts (2015) to directly examine the relation between asset redeployability and the likelihood that a covenant is modified during a loan renegotiation event. The data, available on Roberts website, are a random sample of 114 Compustat firms with loan data in Dealscan. Roberts (2015) uses Securities and 18

20 Exchange Commission filings to track the path of each loan over the loan s life, identifying all instances of originations, renegotiations, and terminations. Renegotiation events include amendments, amended and restated agreements, and rollovers. Each renegotiation leads to changes in terms, such as changes to contract terms (e.g., yield spread, maturity, and amount of loan) or modifications to convent structures (e.g., accounting measures, dividend distributions, and investments). We merge this data with our data on asset redeployability and exclude rollover renegotiation events (following Roberts (2015)). Our final sample consists of 106 unique firms, 237 unique loans, and 1,069 renegotiation events of which 51.6% have at least one covenant modification. Table 5 presents results from probit models regressing an indicator variable that is set to one if a covenant is modified during a renegotiation event and zero otherwise on our redeployability measure, the same set of control variables used in Roberts (2015), and year fixed effects (column 2). In these regressions, we also control for firm size and the firm s expected default probability. Overall, the results show that, compared to firms with more redeployable assets, firms with less redeployable assets are more likely to modify covenants during a loan renegotiation. In terms of economic significance, the coefficient estimates in column 2 imply that firms with a one standard deviation lower asset redeployability score (0.114) are 5.4 percentage points more likely to modify a covenant during a renegotiation event (= ) Asset Redeployability, Covenant Violations, and Asset Sales In our second approach, we examine whether firms with less redeployable assets are less likely to sell their assets after violating a loan covenant. This test serves as an indirect test as to whether the redeployability of a firm s assets affects the likelihood that the firm is able to renegotiate debt contracts instead of liquidating its assets. Underlying this test is the assumption that, if a firm defaults and does not liquidate assets, the firm renegotiated. The advantage of this test is that it allows us to test the channel using a large sample of firms. Table 6 presents the results of this analysis. To construct a sample of firms that do and do not violate covenants, we follow the methodology 19

21 in Chava and Roberts (2008). First, we create a quarterly sample of firms from the merged CRSP- Compustat database, excluding financials and utilities. Second, we merge this sample with loan information from Dealscan. We limit the sample to firms in Dealscan because Dealscan contains covenant thresholds that we can use to determine if a firm violates a covenant. 8 In addition, we focus on Dealscan data as technical defaults occur almost exclusively in bank debt issues, which contain more and tighter covenants compared to public debt issues (Kahan and Tuckman (1995)). Third, we limit the sample to firm-quarter observations in which the firm is bound by either a current ratio or net worth covenant over the life of the loan, which starts with the earliest start date in the loan package and ends with the latest maturity date. We focus on these two financial covenants because they are the most frequently used and the accounting measures used for these two covenants are standardized and unambiguous. Finally, we record that a firm is in violation of a covenant during a quarter if the firm s current ratio or net worth drops below the covenant threshold. The sample period we use for this analysis is from 1994 to 2015 because information on covenants is fairly limited prior to After deleting observations with missing data, we have 28,463 to 42,626 firm-quarter observations depending on the model specification. Firms are in violation of a covenant in about 16% of these quarters, which is similar to the rate found in Chava and Roberts (2008). We capture asset sales with three measures. In column 1 of Table 6, we measure asset sales as the percentage decrease in gross PP&E from quarter t-1 to t. Specifically, we calculate the one-quarter percentage change in gross PP&E, set positive changes to zero, and then multiply the remaining value by negative one so that increases in this variable can be interpreted as a decrease or sale of PP&E. The construction of the asset sale measure in column 2 is the same as that in column 1 except that we use 8 We do not use SDC for this analysis because it has limited coverage of covenant usage and does not have detailed data on covenant thresholds. 9 Like Chava and Roberts (2008), we make three corrections to the Dealscan loan data. First, when two loans overlap, we use the strictest covenant threshold in determining whether a firm violates a covenant during a quarter. Second, when the loan contains dynamic covenants that change over the life of the loan, we linearly interpolate the covenant thresholds over the life of the loan. Third, we correct loan maturity dates in cases when there is post-origination amendments to the loan maturity. 20

22 net PP&E instead of gross PP&E. Last, in column 3, we measure asset sales by scaling the amount of funds received from the sale of PP&E during quarter t by beginning of quarter PP&E. In all of the models, we interact the variable Low Redeployability in year t-1 with the indicator variable Violate that is set equal to one if the firm violates a current ratio or net worth covenant during quarter t and zero otherwise. In these tests, we standardize Low Redeployability to have a mean of zero and a standard deviation of one to ease the interpretation of the coefficient estimates. In all of the models, we also include year-quarter fixed effects to control for time-specific trends, fiscal-quarter fixed effects to control for effects related to the firm s fiscal period, firm fixed effects, and the same set of control variables as in Table 2 measured in t-1. The results in Table 6 show that when firms with highly redeployable assets violate a covenant, they are more likely to sell assets. However, firms with less redeployable assets are significantly less likely to sell assets when they violate a covenant. The coefficient estimates in column 1 imply that firms with an average value of asset redeployability sell 0.71% of PP&E in quarters when they violate a covenant based on the percentage decrease in gross PP&E measure. However, firms with a one standard deviation lower value of asset redeployability sell 0.28 percentage points fewer assets when they violate a covenant. Similarly, firms with an average value of asset redeployability sell 1.10% (0.05%) of PP&E in quarters when they violate a covenant based on the percentage decrease in net PP&E measure in column 2 (the funds received from the sale of PP&E measure in column 3). Yet, the coefficient estimates on the interaction terms in columns 2 and 3 imply that firms with a one standard deviation lower value of asset redeployability sell 0.60 and 0.11 percentage points fewer assets when they violate a covenant, respectively. Relative to the amount of assets sold by firms with an average value of asset redeployability when they violate a covenant, these magnitudes in columns 1-3 are economically significant. Overall, the results in Tables 5 and 6 support the interpretation of our findings that firms with less redeployable assets choose to borrow from banks instead of issuing public debt because they are 21