The Financial Review. The Debt Trap: Wealth Transfers and Debt-Equity Choices of Junk-Grade Firms

Transcription

1 The Financial Review The Debt Trap: Wealth Transfers and Debt-Equity Choices of Junk-Grade Firms Journal: The Financial Review Manuscript ID: FIRE--0-0.R Manuscript Type: Paper Submitted for Review Keywords: capital structure, wealth transfer effects, junk bond issues

2 Page of The Financial Review The Debt Trap: Wealth Transfers and Debt-Equity Choices of Junk-Grade Firms Abstract If outstanding debt is risky, issuing equity transfers wealth from equity holders to debt holders. If existing leverage is high and bankruptcy costs are small, this wealth transfer effect outweighs the gains to stockholders from optimizing firm value. Empirically, we find that for investmentgrade firms, higher leverage implies a greater likelihood of issuing equity, as expected in a standard trade-off model. However, consistent with the impact of wealth transfer effects, for junk-grade firms higher leverage implies a greater likelihood of issuing debt. The analysis implies an additional route through which historical shocks determine firms financing choices.

3 The Financial Review Page of Introduction The traditional trade-off theory of capital structure suggests that firms optimize their value by choosing a leverage which weighs the tax benefits of debt against the deadweight costs of financial distress. More recent papers analyze the dynamics of capital structure decisions taking into account the firm s existing leverage, tax status, and issuance costs (see, for instance, Hennessy and Whited, 0, or DeAngelo, DeAngelo, and Whited, ). In this paper, we add to the existing literature by showing how the choice between debt and equity issuance is affected by wealth transfers from existing equity holders to debt holders. Specifically, stockholders of highly levered firms have an incentive to avoid secondary equity issuances as these imply a transfer of wealth to bondholders. While additional equity could increase total firm value, this increase in wealth is not always sufficient to offset the loss faced by existing equity holders. Myers () shows that this wealth transfer implies that firms do not undertake some positive NPV projects, and here we show that it also creates an asymmetry in the way firms adjust their capital structures. While most firms rebalance to an optimal capital structure, junk-grade firms with high debt will more frequently choose additional debt rather than rebalancing by issuing equity. We begin by providing a theoretical model of the firm s decision to use debt or equity financing. We show that the wealth transfer effect implies that riskier firms are more likely to use additional debt financing. We test this theoretical proposition and find confirming evidence. For investment-grade firms, the probability of using debt decreases with existing leverage, suggesting adjustment toward target capital structure ratios. In contrast, junk-grade firms with a higher leverage ratio are more likely to issue debt than those with a lower leverage ratio. This Consistent with this theory, investment-grade firms issue debt more often than equity while junk-grade firms issue equity more often.

4 Page of The Financial Review finding is consistent with Lemmon, Roberts, and Zender (0) who note that while leverage ratios exhibit some mean-reversion, there is a clear tendency for highly levered firms to remain highly levered. Our model additionally implies that the wealth transfer effect would be larger if the financing need is greater (a larger issue size relative to existing firm value) and if more of the outstanding debt is long-term. We find support for both the relative size effect and the effect of debt maturity in our analysis of financing choice. These results add to the literature on the importance of historical shocks in determining capital structure. In particular, they suggest the existence of a debt trap wherein firms with high leverage ratios have insufficient incentives to reduce leverage back to levels which maximize firm value. Thus a negative shock can have longer-term consequences, as some firms choose to retain high leverage ratios after a sufficiently negative shock, and this can reduce firm value (see van Binsbergen, Graham, and Yang, 0, for estimates of expected bankruptcy costs and other costs associated with debt). Hennessy and Whited (0) also provide a model which generates hysteresis effects; however, their effects are driven by tax effects related to existing debt and cash flows, rather than by wealth transfers associated with long-term risky debt. Becker and Stromberg () find that, consistent with a reduction in wealth transfer concerns, a change in Delaware law which increased fiduciary duty towards bondholders led to higher leverage and a reduced reliance on covenants. However, Becker and Stromberg do not directly address whether marginal financing decisions by risky firms reflect wealth transfer effects. Prior studies find evidence for wealth transfers from stockholders to bondholders by examining stock price reactions to announcements of seasoned equity offerings (Eberhart and

5 The Financial Review Page of Siddique, 0; Elliott, Prevost, and Rao, 0). A separate literature tests trade-off versus pecking order behavior in firms. For instance, Leary and Roberts (0) and Kayhan and Titman (0) find evidence consistent with an optimal capital structure and significant adjustment costs. Flannery and Rangan (0) find that firms partially adjust to a target capital structure, and Oztekin and Flannery () find that adjustment speed varies with tax benefits and bankruptcy costs in a manner consistent with a dynamic trade-off model. Our paper is instead the first to demonstrate in a dynamic model that junk-grade firms capital structure choices are influenced by wealth transfer concerns. Consistent with this notion, in their survey of CFOs, Graham and Harvey (0) find that investment-grade firms are more likely to adhere to a target capital structure than junk-grade firms. The rest of the paper is organized as follows. Section is a brief summary of the related capital structure literature. Section provides the theoretical analysis. Section describes the data used in this paper. Section presents the main empirical results, and section considers two alternative hypotheses. Section contains the concluding remarks.. Prior literature on capital structure. Studies of incremental financing decisions Most capital structure studies examine leverage, which measures the cumulative impact of the firm s financing decisions as well as adjustment costs. However, a smaller group of studies addresses the debt-equity choice when a firm raises external capital. This approach yields different insights into the capital structure decision and resolves some puzzling results documented by cross-sectional studies of leverage. For instance, Mackie-Mason (0) shows A related literature also provides evidence of wealth transfers from bondholders to stockholders with repurchases (see Maxwell and Stephens, 0).

6 Page of The Financial Review that firms facing higher tax rates are more likely to issue debt. Kisgen (0, 0) reports evidence that firms marginal capital structure decisions are driven by rating considerations rather than target leverage.. Market timing, equity undervaluation, and capital structure The strong relationship between stock performance and equity issuance raises the question of whether managers make security choices based on market timing considerations. As Hovakimian, Opler, and Titman (0) point out, superior stock performance prior to the issue could indicate increased growth opportunities. Firms would issue equity rather than debt to minimize agency costs of debt in the face of these increased investment opportunities. A better test of market timing considerations is provided by stock performance after the security issue. If managers private information plays a significant role in their security choice decisions, then they would avoid issuing equity when they believe it to be undervalued. Assuming prices return to equilibrium in the long run, the equity undervaluation explanation for debt issuance predicts that, on average, stock returns following a debt issuance should be positive. In contrast with this prediction, Spiess and Affleck-Graves () show considerable long-run post-issue underperformance by firms making straight and convertible debt offerings. This underperformance is more severe for firms that issue junk-grade debt.. The Pecking Order hypothesis DeAngelo and Roll () argue that time-series variation in leverage is as important as cross-sectional variation in leverage. Some other studies that examine the marginal decision to issue debt or equity include Marsh (), Hovakimian (0), and Hovakimian, Hovakimian, and Tehranian (0). Baker and Wurgler (0) argue that market timing by managers explains capital structure. Butler, Grullon, and Weston (0, 0) dispute the evidence on managers market timing ability.

7 The Financial Review Page of Myers and Majluf () argue that in the presence of informational asymmetries between the firm and outside investors, the firm selects financing mechanisms using the following order of preference: internal funds, debt, and external equity. A number of studies have examined the validity of the pecking order hypothesis, and there has been considerable debate about the correct specifications for testing. Leary and Roberts (0) provide a brief review of the conclusions in these papers. They conclude that there is limited evidence for the pecking order hypothesis but more support for modified models which include features such as bankruptcy costs. Under the pecking order hypothesis, firms typically issue equity only if they have exhausted debt capacity. Hence, we would expect pre-issuance leverage for equity issuers to be higher than for debt issuers. In our analysis, we examine whether junk-grade firms that issue debt have less leverage and higher interest-coverage ratios compared to junk-grade equity issuers.. Dynamic capital structure models Hennessy and Whited (0) introduce a dynamic capital structure model where there is no optimal capital structure, but firm s histories help determine the marginal benefits to using internal cash, debt, or equity to finance new projects. DeAngelo, DeAngelo, and Whited () propose a dynamic model in conjunction with an optimal capital structure. In this model, temporary debt financing is used to finance new projects but firms return, albeit slowly, to their target leverage ratios. DeAngelo, DeAngelo, and Whited () state that their model produces speeds of adjustment which are consistent with prior empirical studies (such as Flannery and Firms may choose to issue equity if managers believe it to be overvalued. We examine this possibility explicitly in our tests.

8 Page of The Financial Review Rangan, 0), which found that speeds of adjustment toward target leverage averaged between one-third and one-twelfth per year. However, Hennessy and Whited use only single period riskless debt, whereas DeAngelo, DeAngelo, and Whited () use perpetual riskless debt, and thus neither of these models captures the wealth transfer effects which we consider. Our model is more limited in scope than these dynamic models, but it illustrates how existing risky debt can affect leverage choice, particularly for junk-grade firms. Prior studies, such as Hovakimian, Opler, and Titman (0) and Denis and McKeon () find that debt issuances push firms away from their target leverage ratios. However, these studies do not examine the difference in issuance decisions between investment-grade and junk-grade issuers.. Wealth transfer to current bondholders and security choice The focus of our model and empirical tests is the potential wealth transfer to existing bondholders while raising new capital. Eberhart and Siddique (0) provide evidence for the hypothesis that seasoned equity offerings (SEOs) transfer wealth from shareholders to bondholders because SEOs reduce default risk. They examine the relationship between stock and bond returns for the issuing firms and conclude that there is a partial wealth transfer from stockholders to bondholders. More closely related to our study, Hovakimian, Opler, and Titman (0) report that firms with operating losses are more likely to issue equity if a greater proportion of their debt is short-term. This finding is consistent with wealth transfers to shortterm debt being minimal. Becker and Stromberg () study the impact of a Delaware court ruling in which stated that corporate directors of financially distressed firms had fiduciary duties not only to stockholders but also to bondholders. They report that distressed firms

9 The Financial Review Page of incorporated in Delaware were more likely to issue equity following this ruling, consistent with a focus on firm value rather than equity value.. Issuance costs and security choice One alternative hypothesis which could explain the choice of debt by junk-grade firms is that debt issuance costs could be relatively lower than equity issuance costs for these firm compared to investment-grade firms. However, for issuance costs to explain our findings, the relative cost of issuing equity rather than debt would have to increase as ratings decreased, and this is counter to the issuance cost pattern found in the literature. Specifically, Lee, Lochhead, Ritter, and Zhao () show that debt issuance costs increase as ratings decrease, and the literature has found no such corresponding pattern for equity issuance costs (see also the discussion in Leary and Roberts, 0). Thus, existing estimates of relative issuance costs would suggest a strict decrease in debt issuance as ratings decrease, whereas the wealth transfer hypothesis provides a rationale for why junk-grade firms with more debt are more likely to issue debt.. Theoretical analysis In this section we develop a theoretical model for the debt or equity issuance decision of an existing firm. Specifically, consider a firm with a pre-existing capital structure that wishes to finance a new project. We assume for the sake of convenience that the project is profitable even given the potential disincentive to invest due to pre-existing debt described by Myers (). We consider a simple one-period model with risk-neutral investors for simplicity, and the firm The bulk of our sample period (-0) occurs after the Delaware ruling. Thus, this change in the legal environment of Delaware firms is not expected to significantly affect our empirical analysis (which also includes year-specific fixed effects).

10 Page of The Financial Review chooses either to finance the new project with debt or equity. We assume that all parties concerned have the same information, and thus the model s results are driven by the moral hazard from existing debt. There are three possible outcomes in our model, as illustrated by the following table: Probability Total Payoff Value of Old Debt if Equity Financing p=-p-p H D D Value of Old Debt if Debt Financing p M D (M-C)*D/(D+N) p L L-C (L-C)*D/(D+N) Here D is the face value of previously issued debt, C is bankruptcy costs (which must be less than or equal to L as firm value cannot go negative), H, M, and L are high, middle, low payoffs; and the probabilities associated with these states of the world are p, p, and p. N is the face value of additional debt raised, and we assume that the new debt has equal seniority with the existing debt. Note that if the low payoff state occurs (with payoff L and probability p), the firm goes into bankruptcy regardless of whether debt or equity is used to finance the additional project. If the medium payoff state occurs (with payoff M and probability p), the firm goes bankrupt if it used additional debt but not if it used equity financing. If the high payoff state occurs (with payoff H and probability -p-p), then the firm does not go bankrupt. If the firm goes bankrupt, equity holders receive nothing and debt holders receive the residual value of the firm less bankruptcy costs. In the context of a more general multi-period model, if the new debt has lower seniority than the old debt, this would decrease, but not eliminate, the wealth transfer from existing bond holders. This follows because new bondholders will receive coupon payments (and potentially principal payment) before the claims of the old bondholders are met. If the new debt instead has greater seniority than existing debt, this would increase the wealth transfer from existing debt holders.

11 The Financial Review Page 0 of To further simplify this problem, we assume there are no taxes or tax-benefits to debt (although this would make debt more attractive), and that no other agency issues are associated with either debt or equity. Instead, we assume the total future value of cash flows (including cash going to pay bankruptcy costs) is the same under both debt and equity financing. Let FV d equal the future value of expected cash flows if additional debt financing is used and FV e equal the future value of expected cash flows if additional equity financing is used, then our assumption states that FV d = FV e. These future values can be broken down into the following: FV e = D e + E e + E new + C e () and FV d = D d + E d + D new + C d () where an e subscript denotes the future value of cash flows under equity financing and a d subscript denotes the future value of cash flows under debt financing. Since the firm finances the same amount (call this investment amount I 0 ) regardless of whether it issues debt or equity, E new = D new. That is, we assume that the firm has a project of a particular size to undertake, and the amount of capital raised to finance this project is assumed to be the same regardless of whether debt or equity financing is used. As the project is undertaken regardless of whether debt or equity financing is used, the total future values of the expected cash flows (including those paid in bankruptcy costs) are equal. Thus: FV d = FV e () and this can be written as: D d + E d + C d = D e + E e + C e ()

12 Page of The Financial Review To determine whether the firm prefers debt or equity financing, we consider X = E e E d = D d D e + C d C e () X is the additional value of existing equity if the firm chooses to issue equity rather than debt. Thus, if X is positive the firm will choose to issue equity, whereas if X is negative the firm will choose to issue debt. Given the payoffs presented above, we can write X as: X = [p(l-c)*d/(d+n) + p(m-c)*d/(d+n) + p*d] - [p(l-c) + p*d + p*d] + [p*c + p*c] p*c () To properly sign X, we also need to consider the breakeven equation for additional debt. In addition to risk-neutral investors, we assume a zero interest rate and that debt markets are perfectly competitive. Thus the present value of the amount raised equals the future value of cash flows to debt. These assumptions imply the following break-even equation for the amount of new debt raised: I o = p(l-c)*n/(d+n) + p(m-c)*n/(d+n) + p*n () Note that firms issuing more debt will face a higher nominal interest rate, although the expected return to debt holders is constant by construction. We proceed by considering the comparative statics of how X changes with p, p, and C. Our results are summarized in the propositions below, and Appendix A provides the proofs. Proposition I: All else equal, higher bankruptcy costs imply a preference for equity over debt financing (i.e., dx/dc > 0). 0

13 The Financial Review Page of This finding is expected, since new debt holders pass on the additional costs of debt to existing shareholders, larger bankruptcy costs imply a greater preference for equity financing. This result also holds in a trade-off model. Proposition II: All else equal, the higher the initial probability that the firm goes bankrupt, the more likely the firm is to use additional debt financing (i.e., dx/dp < 0). This finding reflects the transfer of wealth from existing bondholders to stockholders that a new debt issue entails (see Asquith and Wizman, 0). That is, if the firm issues additional debt, there is a wealth transfer from old bondholders since part of the cash flow flowing to new bondholders comes at their expense. Thus, all else equal, firms with more risky initial debt are more likely to use debt financing. Proposition III: If bankruptcy costs are small, then a greater increase in the probability of bankruptcy from issuing additional debt implies a greater preference for debt financing (i.e., dx/dp < 0). The variable p represents the probability of bankruptcy that is avoided if the firm issues equity rather than debt. A larger p implies a greater wealth transfer to existing bondholders from an equity issue (relative to a bond issue) as the probability of bankruptcy is lowered. Thus, a larger p implies a preference by the existing equity holders to issue debt rather than equity. In practice, we interpret this proposition as implying that large issues, which are associated with

14 Page of The Financial Review larger wealth transfers, are more likely to be financed with debt than equity. However, if bankruptcy costs are sufficiently large, this preference can change as bankruptcy costs are passed on to equity holders through a higher cost of debt. We test these three propositions by examining which firm characteristics determine the choice of a firm using debt or equity when financing an additional project. Consistent with Proposition II, we find that for junk-grade firms, more debt is associated with an increase in the probability of using debt rather than equity financing. Consistent with Proposition III, we find that for junk-grade firms, a larger financing is more likely to be done with debt. Testing our Proposition I is trickier, as there are no direct measures of ex ante bankruptcy cost magnitudes. However, we find that firm size is positively related to whether the firm chooses a debt issue, and larger firm size could proxy for proportionately lower bankruptcy costs (see Bris, Welch, and Zhu, 0). Additionally, whereas both the theoretical and empirical findings related to our Propositions II and III are novel, Proposition I is consistent with prior findings. Additionally, note that the existing debt in our theoretical model is long-term as it matures after the new financing decision is undertaken. If the firm instead held short-term debt, this wealth transfer effect would not apply. We therefore test whether more short-term debt decreases the probability of the firm issuing additional debt, and we find some confirmatory evidence of this effect for junk-grade issuers. We next turn to describing the data set we use to test these hypotheses as well as the most likely alternative hypotheses.. Data and sample construction Note that C measures the costs in bankruptcy, rather than the expected bankruptcy costs which additionally capture the probability of bankruptcy.

15 The Financial Review Page of We examine the determinants of the debt versus equity choice given that firms have decided to access external capital markets for their financing needs. Several of the studies examining incremental financing decisions rely on financial statement data from COMPUSTAT to identify the financing obtained during a year, but we instead use the Securities Data Company (SDC) database to identify all equity and non-convertible debt issues during the -0 period. Leary and Roberts (0) state that the prior studies find similar results whether they use COMPUSTAT or the SDC database. A drawback of the SDC data is that it does not include data on bank loans. Thus, the sample captures all instances of firms raising external capital except bank loans. The exclusion of data on bank loans from our analysis can be justified as our main emphasis is on understanding why firms with junk-grade ratings raise capital in the form of debt rather than equity. If firms face restrictions in accessing capital markets and thus cannot issue equity, they could be forced to take bank loans because they have no access to public bond markets. However, junk-grade firms that issue debt in our sample have demonstrated that they can access capital markets. Hence, it is harder to argue that their decision to not issue equity is due to lack of access to capital markets. After excluding financial, utility, regulated, and foreign firms, we merge the SDC data with the COMPUSTAT database to get the key financial characteristics that explain the firm s capital structure choice. Frank and Goyal (0) evaluate the relative importance of the variables that have been used in the extant literature and identify a few key factors that reliably explain variation in leverage. The impact of these variables on leverage is noted within parentheses as reported by Frank and Goyal: median industry debt ratio (+), market-to-book ratio (-), tangibility (+), profitability (-), log of assets (+), and expected inflation/level of interest rates (-). We We screen the data to exclude debt issues which are used to refinance existing debt. This screen relies on COMPUSTAT data and makes our data more similar to other studies that use COMPUSTAT data.

16 Page of The Financial Review therefore require that firms retained in the sample have data in the COMPUSTAT database on the following variables: total assets, collateral, profitability, market-to-book, and leverage. We also add the most recent marginal tax rates from WRDS as compiled by Blouin, Core, and Guay (0). Next, we identify debt ratings from the SDC and COMPUSTAT databases. Issues that are rated BBB or better are classified as investment-grade, while those with lower debt ratings are classified as junk-grade. 0 We classify equity issuers by applying the same criterion to the COMPUSTAT ratings of a firm s overall creditworthiness. Firm-events for which no rating is available in either database are dropped from the sample. Next, we drop events where the firm issued debt and equity on the same date (0 cases). In a few instances, we observe a firm having multiple debt issues within a few days. Typically, the debt issues differ in their maturity. Rather than treating these cases as independent decisions to issue debt, we consider them to be the consequence of a single decision to seek debt financing. Accordingly, we consolidate debt issues occurring within trading days (roughly equivalent to calendar month) after the previous debt issue in the sample. Price data are acquired from the Center for Research in Security Prices (CRSP). We require a firm to have at least one year (0 trading days) of price data available on CRSP before and after the issue date. This requirement removes firms selling equity through initial public offerings, since this decision could be driven by the need to ensure liquidity for current shareholders rather than capital structure related reasons. A key aim of the study is to examine whether firms issue debt because equity is undervalued. To obtain evidence on this issue, we evaluate post-issue stock performance during the year after issue. Requiring returns for 0 days 0 We primarily use the S&P rating, but use the equivalent Moody s rating if the S&P rating is not available.

17 The Financial Review Page of after each issue allows for evaluation of post-offer performance over a uniform period for all sample events. Our final sample consists of investment-grade equity issues,, investment-grade debt issues,,0 junk-grade equity issues, and 0 junk-grade debt issues. Investment-grade firms more frequently issue debt than equity, while junk-grade firms more frequently issue equity. This relation is consistent with the trade-off hypothesis which suggests that the net benefit of debt is smaller for firms facing a higher probability of financial distress. Table contains the distribution of the sample firms across the Fama-French industry groups. With a few exceptions, most industries are represented in both the investment-grade and junk-grade categories. One potential concern with the SDC data is that it also captures debt issues that are conducted to refinance existing debt. These issues constitute a continuation of the existing debt policy rather than a conscious decision about the form of additional external capital used to finance new investments. The data suggest that junk-grade firms debt issuances are more frequently used to refinance existing debt. Since the use of proceeds information is not available for all issues and general corporate purpose could include debt refinancing, we classify a debt issue as a refinancing if the change in total debt during the fiscal year corresponding to the issue date is less than or equal to % of the total debt in the prior fiscal year. We can attribute a little more than % of debt issues to be merely refinancing events rather than leverage increasing Eleven firms were delisted within 0 days after the security ( equity and debt issues) issuance. These firms appear to be in the midst of a merger/takeover. There were only () firm-events in the junk-grade (investmentgrade) group that were delisted between and 0 days after the issue. Retaining these firms for everything but the stock performance analysis does not alter any of our conclusions. Change in total debt is calculated as total debt in the issue year minus total debt in the fiscal year prior to the issue date. Total debt is defined as debt in current liabilities + long-term debt.

18 Page of The Financial Review issues. Our emphasis is on the subsample that excludes refinancing debt issues; however, we also replicate the analysis using all debt issues. Table shows the frequency distribution of annual offerings of debt and equity offerings by investment and junk-grade firms during the years to 0. Examining the annual ratio of debt issues to equity issues, Table shows that junk-grade firms issued equity more frequently during the 0- recession period. We include year dummies in our subsequent analysis of debt-equity choice to correct for trends across time.. Empirical analysis. Univariate analysis Table provides summary statistics for the determinants of capital structure suggested by Frank and Goyal (0) as well as several other variables relevant to our study. Appendix B contains the definitions for these variables. Variables constructed from COMPUSTAT data are as of the last fiscal year prior to the issue date. The number of observations applies to all the variables except marginal tax rates and debt due in one through three years, as these variables are missing for some firms. As expected, junk-grade firms are more levered than investment-grade firms. Additionally, the investment-grade equity issuers have slightly more leverage than the investment-grade debt issuers, and this is consistent with the trade-off model. However, this pattern does not hold for junk-grade firms. Junk-grade debt issuers have roughly one-third more leverage than junk-grade equity issuers. This use of additional debt by more highly levered junk-grade firms is consistent with a wealth transfer effect (and our Proposition II).

19 The Financial Review Page of Figure presents the percent of net external financing in the form of debt for investmentgrade and junk-grade issuers in our sample. We segment the investment-grade and junk-grade categories into low-leverage and high-leverage groups based on the median leverage for each group at the end of the fiscal year prior to security issuance;, we examine the composition of the net external financing raised over the following,, and years. Consistent with the trade-off theory, junk-grade firms use less debt than investment-grade firms, and high-leverage investment-grade firms issue less debt than low-leverage investment-grade firms. However, consistent with wealth transfer considerations, high-leverage junk-grade firms issue more debt than low-leverage junk-grade firms. Investment-grade firms overall are larger, more profitable, have higher marginal tax rates, and smaller relative issue size. The univariate analysis also shows that both investment-grade and junk-grade issuers of debt are larger and more profitable than issuers of equity. Junk-grade equity issuers have higher market-to-book ratios and lower collateral that junk-grade debt issuers. Consistent with a more important wealth-transfer effect for larger issues (and our Proposition III), relative issue size is larger for junk-grade debt issuers than for junk-grade equity issuers. In contrast, equity issues are somewhat larger than debt issues for investment-grade firms. Additionally, junk-grade equity issuers have a higher ratio of debt due in one, two, or three years. More short-term debt decreases the potential wealth transfer effect, and again this pattern of more short-term debt for equity issuers is not present in the investment-grade sample. We next consider a multivariate analysis of the debt/equity choice to detect the influence of wealth transfer effects. We calculate net external financing as in Frank and Goyal (0). We truncate negative values for net debt and net equity financing at zero to constrain the proportion of debt financing to be between zero and one. Negative values for these variables can yield meaningless ratios.

20 Page of The Financial Review Multivariate analysis of debt-equity choice In Table we present our logit model estimates of debt-equity choice. The dependent variable is set to (0) for debt (equity) issues. We eliminate refinancing debt issues in the models presented, although including refinancing debt issues has a negligible effect on our results. In Model, we are interested in examining how existing firm leverage affects the choice between debt and equity for new financing. We measure firm leverage at the end of the fiscal year prior to the issue date. We include an interaction between leverage and junk-grade to assess the differential impact of leverage between investment-grade and junk-grade firms, We demean all interacted variables as suggested by Balli and Sørensen (). To control for the impact of target capital structure, we include the median industry leverage (see Frank and Goyal, 0; and Lemmon, Roberts, and Zender, 0). To control for debt market conditions, we include interest rate levels and spreads; specifically, the T-bill rate at the time of the issue as well as the spread between Baa and Aaa rated bonds. To control for equity market conditions, we include the prior -year (0 trading days) firm-specific and market-wide returns based on the equally-weighted CRSP index. We also include year dummies in all the specifications. In Model of Table, we add the firm s marginal tax rate using the Blouin, Core, and Guay (0) calculation. In Model, we add the relative size of the issue, the fraction of debt due in one year, and interactions between these variables and whether the firm is a junk-grade issuer. Models and are similar to Model, but they include, respectively, debt due in two years and three years rather than debt due in one year. Note that we do not expect endogeneity, at least in terms of a simultaneity bias, to be a concern in this analysis as the independent variables (with the exception of relative size in Model through ) are all pre-determined, and The results are similar if we use the marginal tax rate provided by Graham at and supplemented with the procedures described in Graham and Mills (0) for filling in missing data.

21 The Financial Review Page of our dependent variable is the marginal decision to issue debt or equity. That is, this research design largely precludes the possibility of reverse causality as a concern because we do not expect next year s financing decision to cause a change in the existing leverage (for instance) this year. We include a variety of control variables to reduce omitted variable bias. The estimated coefficient on leverage in all specifications supports the basic trade-off theory. That is, for the investment-grade firms in the sample, leverage has a significant negative effect on the probability of issuing more debt. This evidence is also consistent with Lemmon, Roberts, and Zender s (0) findings about firms tendency to revert to a mean capital structure. On the other hand, we find that the coefficient on the interaction between leverage and junk-grade is positive and significant. The sum of the coefficients on leverage and leverage interacted with junk-grade is also positive and significant at the % level. Thus junk-grade firms with higher leverage are more likely to issue debt. This increased likelihood of issuing debt for junk-grade firms with higher existing leverage is consistent with the wealth transfer effect as stated in Proposition II, but counter to the trade-off theory, which would predict a movement toward an optimal capital structure. In terms of economic impact, the estimates from Model imply that for the investmentgrade firms a move from one standard deviation below the mean in leverage to one standard deviation above the mean is associated with a decrease in the probability of issuing debt from.% to.%. However, a similar move for the junk-grade firms implies an increase in the probability of issuing debt from.% to.%. Model of Table adds the marginal tax rates compiled by Blouin, Core, and Guay (0). In this specification, a higher marginal tax rate implies a higher probability of issuing debt; however this effect is not significant. In Model of Table we add the relative size of the

22 Page of The Financial Review issue and the fraction of existing long-term debt due in one year, as well as interactions between these terms and whether the firm is a junk-grade issuer. Firms that have larger financing needs are more likely to issue debt, and this effect is larger for junk-grade firms than for investmentgrade firms. This result is consistent with our Proposition III as junk-grade firms avoid the larger wealth transfer effects from a greater equity issue. Additionally, the results in Model of Table show that firms with more debt due in one year are more likely to issue debt. However, the interaction between junk-grade and existing debt due in year is negative and significant at the % level. This supports the notion that there is a smaller wealth-transfer effect if existing debt is primarily short-term. Model of Table includes debt due in two years, and this decreases the magnitude of the interaction between short-term debt and junk-grade status. Similarly, Model considers debt due in three years, and this further decreases the magnitude of the short-term debt and junk-grade interaction. Thus, junk-grade firms with more debt are more likely to issue additional debt, and this effect diminishes if the firm has more short-term debt, and diminishes more for one-year debt than for two-year or three-year debt. The estimated coefficients on our control variables are largely as expected. Median industry leverage has a strong, positive association with the probability of debt issue. Both higher interest rates and spreads are associated with a lower probability of a debt issue, but these effects are mostly insignificant. This lack of significance could be due to our year dummies capturing many of these macroeconomic effects. Consistent with the survey results in Graham and Harvey (0), companies experiencing high firm-specific stock returns during the year prior to the issue date tend to issue equity. Further, the evidence suggests that an improved We consider only the portion of long-term debt due in year, because the structure of short-term debt could be affected by working capital management considerations. If we instead consider all debt due in year, our results are similar and stronger.

23 The Financial Review Page of outlook for the economy, as indicated by high market index returns, also encourages firms to issue equity. These results could be due to higher stock returns being driven not only by higher expected profitability of existing operations but also by prospects for better growth opportunities (Hovakimian, Opler, and Titman, 0). In this case, the negative relationship between stock returns and debt usage would be consistent with the trade-off hypothesis as firms use less debt in the presence of growth options (see Barclay, Morellec, and Smith, 0). Table presents similar specifications, but instead splits the sample into separate investment-grade and junk-grade subsamples, and this allows for different coefficients on the control variables between the subsamples. Consistent with the results in Table, the coefficient on leverage is negative and significant for the investment-grade subsample regressions, but positive and significant for the junk-grade subsample regressions. Similarly, and again consistent with Proposition III, the coefficient on relative size is positive for junk-grade firms. Additionally, a greater fraction of debt due in year implies a greater probability of the firm issuing debt for investment-grade firms, but a lower probability of debt issues for junk-grade firms. The magnitudes of the coefficients debt due in years and years are smaller than for year debt, and these results are consistent with larger wealth transfer effects for junk-grade firms if the debt is longer. Overall, our findings on how past leverage affects debt-equity choice support our theory and the importance of a wealth transfer effect for junk-grade issuers. robustness tests of our results. We next consider some We obtain similar results if we replace market returns and firm-specific returns with the firm s raw holding-period returns. An implicit assumption here is that, on average, the projects faced by different firms have similar risk. The alternative explanation, that firms finance low-risk issues with debt and higher-risk issues with equity, would require that high-debt firms have lower risk projects than low-debt firms, and that this phenomenon does not hold for investment-grade issuers.

24 Page of The Financial Review Robustness tests In unreported regressions, instead of the median industry leverage we use dummy variables based on Fama-French industry classifications and find similar results. As our primary analysis is a logit, the use of too many dummies causes a drop in sample size because some dummy variables become perfect predictors. Nevertheless, our results are maintained under this alternative specification. We also consider a sample which includes debt refinancing, and these results are similar to those reported. We further test whether our results could be due to a nonlinear relation between leverage and the probability of using debt financing. We therefore include the square of leverage in our specification, but this variable is not significant and our previous results do not change. In an additional specifications, we add firm-specific variables based on Frank and Goyal (0). Specifically, we include the firm s market-to-book, collateral, profitability, and log(assets) from the year before the new issue in Table. Taken together, these additional firm-specific variables explain a relatively large portion of existing leverage, and therefore we expect our results to be weaker under these specifications. Model of Table includes the marginal tax rate, and Models through add relative size, the fraction of short-term debt, and the interactions of these variables with junk-grade. Our primary result,that junk-grade firms with more leverage are more likely to issue debt, is unchanged in these specifications. The relative size and short-term debt effects are also consistent with our prior results. To reduce the impact of outliers, we winsorize leverage, market-to-book ratios, collateral, profitability, relative size, and the prior year firm-specific return in our analyses. When corresponding regressions are estimated separately for investment-grade and firm-grade subsamples, the coefficient for leverage is negative and significant for investment-grade firms as predicted by the trade-off theory. However, the coefficient for leverage is close to zero and insignificant for junk-grade firms; this result is consistent with the trade-off theory effects being offset by wealth-transfer effects for junk-grade firms.

25 The Financial Review Page of The evidence on the additional firm-specific control variables is largely consistent with the trade-off theory of leverage and prior empirical findings. One surprising result is that a higher marginal tax rate implies a lower probability of issuing debt. The negative coefficient on marginal tax rate in these models appears to be due to the high correlation between the tax rate and firm profitability. Eliminating profitability from these regressions makes the tax coefficient insignificant, and further removing the junk dummy variable so as to make the analysis similar to prior studies, makes the estimated coefficient on the marginal tax rate positive and significant. If we replace the Blouin, Core, and Guay (0) tax rate with that calculated by Graham, the coefficients on marginal tax rates are also not significant in any of the models.. Alternative explanations for junk-grade firms issuing debt Consistent with the trade-off hypothesis of capital structure, most junk-grade firms issue equity. In this section, we consider two alternative hypotheses for why some junk-grade firms choose to issue debt rather than equity: the pecking order hypothesis and the equity undervaluation hypothesis. However, note that neither of these hypotheses can explain the impact of leverage on debt-equity choices of junk-grade firms presented above.. The Pecking Order hypothesis A prediction of the pecking order hypothesis is that firms issue equity only as a last resort. This implies that equity issuers, especially junk-grade issuers, should have high leverage ratios. However, Table reveals that among junk-grade firms, equity issuers have much lower leverage compared to debt issuers. Possibly, the leverage measure does not adequately capture cash flow constraints faced by firms. The high M/B ratios enjoyed by equity issuers relative to debt issuers

26 Page of The Financial Review could translate into lower leverage ratios, even though they are more severely constrained in terms of cash flows. To address this possibility, we examine cash-flow based measures of leverage. Table displays the evolution of the interest coverage ratio (ICR) for our sample, from three years pre-issue to three years post-issue. In Panel A, ICR is defined as the ratio of operating income to interest expenses. In Panel B we consider the ratio of operating income to interest expense plus short-term debt, and Panel C we further add rent expenses to the denominator. The results are consistent across all definitions. While investment-grade firms show no particular pattern for ICR before or after the issue, junk-grade firms that issue debt have lower ICR before and after issue relative to junk-grade firms that issue equity. This trend is not consistent with the pecking order hypothesis which predicts that more financially constrained firms are forced to issue equity. However, this evidence is consistent with wealth transfer effects influencing security choice (and our Proposition II). That is, some junk-grade firms issue debt despite having a low coverage ratio as these firms are cautious about the potential wealth transfer from existing equity holders to bondholder caused by issuing equity.. Alternative explanations undervalued equity Another potential explanation for junk-grade firms issuing debt is that their equity is undervalued at the time of the security issue. The positive link between prior stock performance and the likelihood of equity issuance observed in Tables through reinforces the notion of market timing by managers who exploit their private information in making security issuance decisions. A crucial prediction of the market timing hypothesis is that stocks of debt issuers At time, the differences are statistically significant at the % level for junk-grade firms in all panels; at time 0, the differences are significant in Panels A and B. For investment-grade firms, none of the differences are significant at the % level.

27 The Financial Review Page of should outperform stocks of equity issuers, especially for junk-grade firms, since debt issuers in this category go against the expectations that they will issue equity. Spiess and Affleck-Graves () report negative equity returns over a -year holding period after debt issues. However, it is possible that managers have a shorter forecast horizon when they make issuance decisions. We therefore examine descriptive statistics of post-issue stock performance during the year (0 trading days) after the issue date. Table presents the analysis of various measures of post-issue stock performance for equity and debt issuers in both debt rating categories. Post-issue raw returns are positive and significant for all four groups. For the junk-grade firms, debt issuers have higher returns (.%) than the equity issuers (.%), and this difference is even larger when comparing median values (.% versus.%). While this difference is consistent with managers of junkgrade firms choosing to issue debt due to stock undervaluation, these results are purely driven by market movements, not by firm-specific returns. The argument that managers use their firmspecific private information while selecting financing instruments suggests that we should see superior firm-specific performance. In contrast, differences in market returns suggest a pattern driven by equity financing waves (see Pastor and Veronesi, 0, for a theory of IPO waves). For junk-grade issuers, we find that market returns of debt issuers are higher than market returns of equity issuers. However, firm-specific returns of junk-grade debt issuers are significantly lower than firm-specific returns of equity issuers. The evidence discussed above is subject to the critique of Schultz (0) and Butler, Grullon, and Weston (0) that returns on event-time portfolios can generate spurious evidence on market-timing abilities of managers. Hence, following Schultz (0) and Butler, Grullon, and Weston (0), we conduct the analysis using calendar-time portfolios. We isolate firm-