Journal of Financial Economics 68 (2003) 75–109

Capital structure choice: macroeconomic conditions and financial constraints$ Robert A. Korajczyka, Amnon Levyb,c,* b

a Kellogg School of Management, Northwestern University, Evanston, IL 60208-2001, USA Haas School of Business, University of California Berkeley, Berkeley, CA 94720-1900, USA c Moody’s KMV, San Francisco, CA 94111-1016, USA

Received 29 August 2001; accepted 21 March 2002

Abstract This paper provides new evidence of how macroeconomic conditions affect capital structure choice. We model firms’ target capital structures as a function of macroeconomic conditions and firm-specific variables. We split our sample based on a measure of financial constraints. Target leverage is counter-cyclical for the relatively unconstrained sample, but pro-cyclical for the relatively constrained sample. Macroeconomic conditions are significant for issue choice for unconstrained firms but less so for constrained firms. Our results support the hypothesis that unconstrained firms time their issue choice to coincide with periods of favorable macroeconomic conditions, while constrained firms do not. r 2002 Elsevier Science B.V. All rights reserved. JEL classification: G32; G1 Keywords: Capital structure; Business cycles; Financial constraints

$ We thank Susan Chaplinsky, Matthew Clayton, Kent Daniel, Michael Fishman, Chris Hennessy, Laurie Hodrick, Armen Hovakimian, Ravi Jagannathan, Deborah Lucas, Hamid Mehran, Mitchell Petersen, Todd Pulvino, Anna Scherbina, Bill Schwert (the editor), Jeremy Stein, and an anonymous referee for helpful comments. We would also like to thank seminar participants at the AFA Annual Meetings, Columbia/New York University joint seminar, the Federal Reserve Bank of San Francisco, Moody’s KMV, the New York University Macro Lunch, and the University of California Berkeley. *Corresponding author. Present address: Moody’s KMV, 1620 Montgomery Street, Suite 140, San Francisco, CA 94111-1016, USA. E-mail address: [email protected] (A. Levy).

0304-405X/02/$ - see front matter r 2002 Elsevier Science B.V. All rights reserved. PII: S 0 3 0 4 - 4 0 5 X ( 0 2 ) 0 0 2 4 9 - 0

76

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

1. Introduction Capital structure choice varies over time and across firms. For example, aggregate equity issues vary pro-cyclically and aggregate debt issues vary counter-cyclically for firms that access public financial markets. Meanwhile, firms that exhibit higher degrees of financial constraints do not exhibit these pronounced counter-cyclical debt issue patterns.1 In addition, firms are more likely to issue equity following an abnormal increase in their own price of equity (e.g., Korajczyk, et al., 1990). Such observations suggest that both macroeconomic conditions and firm-specific factors drive variations in financing choices and that these variations differ with the degree of financial market access. In this paper we quantify the relative importance of these factors by performing a variance decomposition for the time variation in financing choices on a sample of firms that is split on a measure of financial constraints. We investigate the role of macroeconomic conditions and financial constraints in determining capital structure choice since these can induce time-series and crosssectional heterogeneity in firm behavior. Firms facing financial constraints do not choose capital structure in the same manner as unconstrained firms.2 Similarly, time variation in macroeconomic conditions, such as changes in the relative pricing of asset classes, can lead a given firm to choose different capital structures at different points in time, other things being equal. This allows us to investigate alternative capital structure theories in light of the effects of financial constraints and macroeconomic conditions. Two important theories of capital structure are the tradeoff theory and the pecking order theory. In the tradeoff theory, the benefits of increased leverage (for example, tax benefits or reductions in agency costs) are weighed against the costs of increased leverage (for example, deadweight bankruptcy costs) in order to determine the optimal amount of leverage. In the pecking order theory, external financing is more expensive for riskier securities (possibly due to informational asymmetries between managers and security holders). Thus, firms prefer to finance first with internal funds, then with debt, and lastly with equity. Our approach is similar to Hovakimian et al. (2001), who look at the relation among firm-specific variables, target leverage, and issue choice. However, we split our sample into two subsets, financially constrained and financially unconstrained. Theoretically, we define financially constrained firms as the set of firms that do not have sufficient cash to undertake investment opportunities and that face severe agency costs when accessing financial markets (we use retention rates and that investment opportunities as proxies). Moreover, we estimate the relation between 1 Specifically, Choe et al. (1993) show that aggregate seasoned primary equity issues are pro-cyclical and debt issues are counter-cyclical. Korajczyk et al. (1990) show that aggregate equity issues are positively related with equity market performance. Gertler and Gilchrist (1993) show that aggregate net debt issues (public and private) increase for large firms but remain flat for small firms following recessions associated with a monetary contraction. Gertler and Gilchrist (1994) show that aggregate net short-term debt is more stable over the business cycle for small firms. 2 None of the firms in our sample are completely shut out of financial markets since they are firms that choose to access public capital markets over the sample period. We use the terms constrained and unconstrained to denote a relative relation.

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

77

firms’ debt ratio and (1) firm-specific variables and (2) macroeconomic conditions. We use the fitted values of this relation to estimate firms’ target capital structures. We then investigate the relation between security issuances/repurchases, the deviation from target leverage, and both firm-specific and macroeconomic variables. Empirically, the relation between firm-specific variables and target leverage is consistent with some elements of both the pecking order theory and the tradeoff theory of capital structure. However, the relation is also inconsistent with some elements of each theory. Larger firms and those with more tangible assets tend to have higher leverage. Firms with unique assets tend to have lower leverage. Consistent with the tradeoff theory, firms with large depreciation tax-shields have lower target leverage. Also consistent with the tradeoff theory, deviations from our estimated target leverage explain firms’ choice of security issuance. However, the negative relation between operating income and leverage and the negative relation between the macroeconomic variables and leverage seem consistent with a pecking order theory, particularly for unconstrained firms. Fig. 1 illustrates the systematic peaks in corporate leverage ratios that occur during economic downturns over the last 50 years. A tradeoff model would imply pro-cyclical leverage since during expansions (when the equity market is performing well, expected bankruptcy costs are lower, firms are more likely to have taxable income to shield, and firms have more free cash) debt should be more attractive for unconstrained firms (see Jensen and Meckling, 1976; Gertler and Hubbard, 1993; or Zwiebel, 1996). Pulvino (1998) provides evidence that firm health and market performance influence bankruptcy costs for the airline industry. Titman and Wessels (1988) relate expected bankruptcy costs (using asset specificity as a metric) to capital structure. Mackie-Mason (1990)

Debt to Asset Ratio

1

0.75

0.5

0.25

0 1952

1957

1962

1967

1972

1977

1982

1987

1992

1997

Date Fig. 1. Aggregate nonfinancial corporate debt to asset ratio across NBER expansions (shaded) and contractions (light). Debt to asset ratio is measured as the total credit instruments of nonfinancial corporations measured at book value, divided by the sum of credit market instruments and the market value of equity, as reported in Board of Governors of the Federal Reserve System, ‘‘Flow of Funds Accounts.’’

78

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

and Graham (1996) both provide evidence that tax shields impact capital structure choice. Moreover, since these firms seem far from bankruptcy, asset substitution (Jensen, 1986) or debt overhang (Myers, 1977) are unlikely to be driving the observed patterns. A pecking order model would be consistent with these patterns since firms prefer using internally generated funds to finance investment and have more internal funds during expansions. The fact that our results with firm-specific variables are consistent with elements of both tradeoff and pecking order theories is consistent with previous empirical work that studies cross-sectional target leverage ratios such as Titman and Wessels (1988), Hovakimian et al. (2001), and Fama and French (2002). Our results indicate that, after correcting for firm-specific variables, the macroeconomic variables help explain some of the counter-cyclical leverage patterns for the unconstrained sample. These patterns are consistent with some recent theoretical work (e.g., Levy, 2001) that relates capital structure to macroeconomic conditions. Levy (2001) develops an agency model in which debt aligns managers’ interests, which include private benefit extraction, with those of the outside shareholders. In recessions, levered managers’ wealth is reduced relative to outside shareholders. This shift in relative wealth exacerbates the agency problem and increases the optimal amount of leverage in order to realign managers’ incentives with those of the shareholders. This leads to counter-cyclical leverage for those firms that are not severely constrained. The model provides one motivation for our use of businesscycle variables that proxy for relative aggregate managerial wealth when estimating target leverage. The results add to the credit channel literature that analyzes the relation between debt issues, financial constraints, monetary policy, credit conditions, and the business cycle. The literature, which often uses size or the degree of bank dependance as proxies for the level of financial constraints, generally agrees with the proposition that firms that face greater financial constraints find it difficult to borrow to smooth cash flows following negative shocks to the economy. Gertler and Gilchrist (1993) find that aggregate net debt issues, following recessions associated with a monetary contraction, increase for large firms but remain stable for small firms that rely on private debt. Similarly, Gertler and Gilchrist (1994) show that aggregate net shortterm debt issues are less sensitive to the business cycle for small firms. The literature debates whether these patterns are due to the effect of monetary policy on firms’ debt issue patterns through the bank lending channel or through the balance sheet channel. Bernanke and Gertler (1995) provide a description of the debate as well as a review of the literature. The bank lending channel theory focuses on the possible effects of monetary policy actions on the supply of loans by depository institutions (e.g., Kashyap et al., 1993). The balance sheet channel theory stresses the potential impact of an economic slowdown on borrowers’ balance sheets. For example, Kiyotaki and Moore (1997) and Suarez and Sussman (1999) develop general equilibrium models where constrained firms (farmers in the Kiyotaki and Moore model) are always up against their borrowing constraints. The pro-cyclical value of collateral, against which they borrow, results in pro-cyclical leverage. Similar to Levy (2001), the Kiyotaki and Moore model predicts that relative

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

79

aggregate managerial wealth determines optimal leverage. We find that our constrained sample exhibits pro-cyclical leverage, which supports the Kiyotaki and Moore model. As demonstrated in Fischer et al. (1989) or Leland (1994, 1998), frictions will result in a firm deviating from its target leverage. In a second stage we, therefore, estimate how firms’ security issue choices (debt versus equity) vary with (1) the distance between actual and target leverage, (2) macroeconomic conditions, and (3) firm-specific variables. Macroeconomic and firm-specific variables are included in this second stage. These variables account for variations in the costs of issuing or repurchasing debt or equity due to adverse selection, for example. These variables also account for variations in the marginal costs and benefits of debt and equity financing that have been shown to perform well in describing issue choice. For example, Mackie-Mason (1990) provides evidence that marginal tax shields can help describe issue choice. Although we find that security issue and repurchase choice is sensitive to deviations from the target for both samples, it is particularly sensitive in the constrained sample. The unconstrained sample’s issue choice is substantially more sensitive to variations in macroeconomic conditions than the constrained sample, consistent with arguments that unconstrained firms can deviate from their target capital structure in order to time their issues to periods when market conditions are most favorable (i.e., periods when the relative pricing of the security issued is favorable). We find that macroeconomic variables are a significant determinant of repurchase choice for our unconstrained sample. As with the target regressions, our firm-specific results are consistent with previous studies. Our issue choice results are consistent with several empirical and theoretical papers that study how aggregate security issues vary across time. These studies, in general, focus on aggregated debt or equity issues, or on equity issues alone, and have not explicitly tested for the role of macroeconomic conditions in financing choices. (Exceptions are Marsh (1982), who includes a forecast of aggregate debt and equity issues as a measure of ‘‘market conditions’’ in estimating issue choice, and Bayless and Chaplinsky (1991), who include a measure of equity market performance and the change in the T-bill in estimating issue choices.) This literature focuses on settings similar to the Myers and Majluf (1984) pecking order theory, where insiders know more about the prospects of their firms than outsiders. Managers who have current equity owners’ interests in mind avoid issuing equity when they believe their shares are underpriced. Therefore, equity issues convey unfavorable news about the firm’s prospects. Consistent with this theory, there is a substantial negative price reaction to equity issue announcements (Masulis and Korwar, 1986) that are less negative following credible releases of information (Korajczyk et al., 1991). Jensen (1986) argues that a free cash flow story where managers can indulge themselves when they are not forced to make interest payments can explain these empirical observations. However, the fact that the average price reaction is the same for purely secondary issues (Korajczyk et al., 1990) is not consistent with the free cash flow story being the sole cause of the price decline. Moreover, debt financing has an issue announcement effect that is less negative than

80

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

equity issues; this is consistent with its value having a lower sensitivity to information asymmetries (Chaplinsky and Hansen, 1993). Lucas and McDonald (1990) extend Myers and Majluf’s (1984) theory to a dynamic setting where managers have private information about their company’s value. They optimally choose to delay equity issues until they have an investment opportunity and their stock price rises to (or above) its true value. Correlated market prices across firms result in equity issues clustering around market peaks. Similar to Korajczyk et al., (1990), who provide evidence of equity issue clusters following a run-up in the equity market, we find that firms prefer equity over debt financing following such a run-up. Choe et al. (1993) argue that adverse selection costs vary counter-cyclically to explain the general increase in equity issues during expansions. A model with endogenous counter-cyclical adverse selection is derived in Eisfeldt (2001). Bayless and Chaplinsky (1996) argue that ‘‘windows of opportunity’’ in which capital can be raised at favorable terms result in observed periods of extreme equity issue volume (or ‘‘hot’’ equity markets) as firms seek to exploit these opportunities. Two interpretations of a window of opportunity include (i) a behavioral argument where the market is particularly exuberant over equity issues and (ii) the relative pricing of asset classes (i.e., debt and equity), due to the severity of adverse selection for example, is such that a large number of firms prefer to issue equity. Empirically, price reactions to equity issue announcements are less negative during these periods. Motivated by the theoretical arguments and existing evidence, we estimate the average recent price reaction to equity issue announcements and find that it helps explain issue choice for the unconstrained sample. Although these descriptions of macroeconomic episodes in which equity issues cluster are useful starting points, they are not completely consistent with the data. For example, during the second expansion of the 1970s, the equity market performed poorly and the average price drop upon an equity issue announcement was relatively large, yet equity issues as a fraction of total outside funding was relatively high. In fact, Bayless and Chaplinsky (1996) classify the period between 1976 and 1979 as a cold market. The average price reaction to an equity issue announcement was
3.6% during this period, compared with
2.0% during their classified ‘‘normal’’ markets. Meanwhile, Choe et al. (1993) show that the dollar amount of equity issues, as a fraction of the sum of equity and straight debt issues for companies listed in on the NYSE, AMEX, and NASDAQ, was 40% between April 1975 and January 1980. It was 18% for the entire sample period between January 1971 and December 1991. They also show that between January 1971 and November 1973 the ratio was 35%, despite poor performance in the equity market. Unfortunately, we do not have readily available data regarding price reactions prior to 1974. Our results suggest that this pattern may be due to firms being far from their target leverage ratios because of the unusually high leverage observed for that period (see Fig. 1). The rest of this paper is organized as follows. Section 2 discusses the data set and empirical specification. Section 3 describes our estimation results. Section 4 provides checks for robustness. Section 5 concludes with directions for future research.

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

81

2. Data and empirical specification All series are converted to real values in 1980 dollars using the consumer price index (CPI) inflation series (Ibbotson Associates, 2001). Similar to the procedures used in Hovakimian et al. (2001), we examine the determinants of financial choices when firms make significant changes to their capital structure. In order to be included in our sample, a firm must have either the net value of equity (common and preferred) issued, repurchased, or paid out as a dividend; or the change in the book value of debt (straight and convertible) of at least 5% of the book value of assets in the previous quarter. Since the value of preferred stock is rarely reported and convertible debt is never reported in COMPUSTAT at the quarterly frequency, we can disentangle neither net common equity from preferred equity nor convertible debt from straight debt issues and repurchases. Hovakimian et al. (2001) disentangle such issues by using the annual COMPUSTAT files. Quarterly COMPUSTAT data (Primary/Secondary/Tertiary, Full Coverage, and Research) and monthly Center for Research in Security Prices (CRSP) data from 1984:1 to 1999:3 are used. Our methodology requires a firm to have reported COMPUSTAT data items for eight quarters before, and eight quarters after, changing its capital structure. This methodology filters out firms with unstable financial or operating status whose financing decisions are influenced by factors other than those analyzed in our paper.3 For example, Asquith et al. (1994) show that financial instability is related to restructuring of assets and liabilities. In addition, Welch (1989, 1996) finds evidence that firms conduct seasoned ‘‘follow-on’’ equity issues in conjunction with their initial public offerings. Since the motivation for such follow-on offerings is likely to be different than the typical seasoned offering, we use only firms for which we have at least 48 months of CRSP data prior to the event of changing their capital structure. A 48 month cutoff was chosen by looking at Figure 4 of Welch (1989) and Table 3, Panel C of Welch (1996), which show that the probability of a firm conducting a seasoned issue drops substantially four years after its initial public offering. Consistent with findings in Welch (1989, 1996), the young firms’ unconditional issue choices are economically and statistically different from those of the older firms. Moreover, both firm-specific and macroeconomic variables are much less informative in describing capital structure choices for the young sample. Although including young firms does not

3

As in Hovakimian et al. (2001), we exclude firms-quarters with outlying firm-specific values (those with the highest 0.5% and, for some variables, lowest 0.5% of values) to minimize the influence of extreme values on the analysis. Moreover, since selling, general and administrative expenses, and book assets are used as normalizing variables, they are required to be positive. Outliers are excluded when the market-tobook ratio is greater than 19.51 or less than 0.31; when the average operating income/assets over the previous four quarters are greater than 0.35 and less than
0.12; when the average selling expense/sales over the previous four quarters is greater than 1.18 and less than 0.019; when accounts receivables/assets are greater than 0.74; when the average depreciation/assets over the previous four quarters are greater than 0.042; when the average income taxes/assets over the previous four quarters are greater than 0.046 and less than
0.016; and when one year excess stock returns are greater than 338% and less than
112%.

82

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

change the qualitative nature of our results, the overall results are noisier since they comprise approximately 20% of the sample. Because of the importance of regulatory factors in capital structure choice for utilities and financial firms, they are excluded from our sample. In total, our sample contains 5,623 event quarters with significant capital structure changes.

2.1. Financial constraints Given the evidence from the credit channel literature that firms with differential access to financial markets have different debt issue patterns, we split our sample into two categories that we will refer to as financially constrained and financially unconstrained. Theoretically, we define a firm as financially constrained if it does not have sufficient cash to undertake investment opportunities and if it faces severe agency costs when accessing financial markets. Empirically, we use a high retention rate combined with the existence of investment opportunities to identify financially constrained firms. Since dividends and security repurchases compete with investment for funds, firms that have investment opportunities and face relatively high costs of external finance should choose to retain net income for investment. Therefore, the specific criteria for a firm-event window to be labeled financially constrained are: (1) the firm does not have a net repurchase of debt or equity and does not pay dividends within the event window, and (2) the firm’s Tobin’s Q, defined as the sum of the market value of equity and the book value of debt, divided by the book value of assets, at the end of the event quarter should be greater than one. A firm-event window is labeled as financially unconstrained if it does not meet these two criteria. Our methodology is similar to that used by Fazzari et al. (1988), who use the level of dividend distributions to categorize firms into groups facing differing levels of financial constraints. However, we also condition on Tobin’s Q to ensure that firms in our constrained sample have investment opportunities and are not financially distressed. A number of alternative definitions of financial constraints have been proposed. For our sample, firms in the most constrained decile, according to the augmented Kaplan and Zingales (1997) measure used in Lamont et al. (2001), are approximately 15% more likely to make a major security repurchase than the firms classified as less constrained. Kashyap et al. (1994) use the existence of a bond rating to categorize unconstrained versus constrained firms. This classification algorithm would lead to a much larger fraction of our sample being categorized as constrained since only approximately 20% of the sample has a bond rating. In total, we classify 565 firm events as financially constrained and 5,059 as financially unconstrained. Only eight of the financially constrained firm-event windows have investment grade bonds rated by S&P; four have a rating of BBB, four have a rating of AA, and the rest have speculative grade debt or no rating at all. It is worth noting that although this classification is persistent, firms do switch classification. Approximately one-third of firms that re-enter the sample switch classification.

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

83

Several summary statistics for nine quarter event-windows in each class, presented in Table 1, suggest that we are in fact proxying for financial constraints. As in other studies measuring financial constraints (e.g., Fazzari et al., 1988), our constrained firms are smaller on average. Since one of the requirements is to have a market-tobook ratio greater than one after an issue, it is no surprise that our constrained sample has a substantially higher overall average. It is interesting that the constrained sample has substantially higher capital expenditures (as a fraction of assets), despite having lower average income and approximately the same median income. This suggests that we are capturing firms with investment opportunities. The constrained sample has a lower debt ratio when measured using market values, but approximately the same average leverage ratio when measured using book values. This is not surprising since book values proxy for tangible assets (that can be used more easily as collateral) and is expected since our sample is stratified on Q. The constrained sample has a higher beta, suggesting that the relative movement in equity values is not driving the difference in leverage patterns across the two samples.4 Finally, motivated by papers such as Fazzari et al. (1988), Table 1’s second row from the bottom relates the sensitivity of investment to cash flow. We measure cash flow sensitivity by regressing investment (capital expenditures divided by property plant and equipment) on cash flow (net income plus depreciation), lagged Tobin’s Q, and lagged cash as a fraction of assets. Only the cash flow coefficients and standard errors (in parentheses) are presented, along with a test of coefficient equality. In the absence of frictions, cash flows should not influence investment decisions, although there is some debate about the exact nature of the relation between financial constraints and investment-cash flow sensitivity (e.g., Fazzari et al., 2000; Kaplan and Zingales, 2000; Abel and Eberly, 2002). Consistent with the findings of Fazzari et al. (1988), investment for our constrained sample is sensitive to variations in cash flows, while investment for our unconstrained sample is not sensitive to variations in cash flows. 2.2. Empirical specification We use both firm-specific data and aggregated/macroeconomic data to isolate macroeconomic conditions in firms’ financing choices. The procedure is similar to Gertler and Hubbard (1993), who estimate macroeconomic determinants of dividend payments. Variables are lagged one quarter since their public release dates are not immediately known. Similar to Marsh (1982), Bayless and Chaplinsky (1991), and Hovakimian et al. (2001), we assume that firms’ capital structure choices can be described by the following two equations, Levi;t ¼ MacroTi
3 a þ Xi;t
3 b þ fi þ qt þ fqt þ d86t 4

ð1Þ

To estimate beta, up to 60 months of CRSP data prior to the change in capital structure are used. When 60 months are not available, we use the data that are available (at least 48 months).

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

84

Table 1 Summary statistics of firm-specific variables for unconstrained and constrained firms The table presents means, medians (in brackets), and p-value of mean (in parentheses) and distribution (in brackets) equality across unconstrained and constrained firms. The sample includes all firm event windows that had CRSP return data 48 months prior to, and COMPUSTAT items eight quarters prior to and eight quarters after, a net debt or equity (including dividends) issue or repurchase of at least 5% of book assets between 1985:1 and 1998:3. Constrained firms do not pay dividends, do not have a net equity or debt purchase over the event window, and have a Tobin’s Q of greater than one at the end of the event quarter. Unconstrained firms are firms that are not labeled as constrained. Beta is measured using at least four (and up to five) years of data prior to the event quarter. The row labeled cash flow sensitivity presents the coefficient and standard error (in parentheses) on cash flow (net income plus depreciation) from an OLS regression of investment (capital expenditures less depreciation divided by property plant and equipment) on cash flow, lagged Tobin’s Q, and lagged cash as a fraction of assets. The test of distribution equality is a Kolmogorov–Smirnov nonparametric test. Variable Book assets (deflated to 1980 dollars) ST+LT Debt/market assets (percent) ST+LT Debt/book assets (percent) Operating income/book assets (percent) Market/book Capital expend./book assets (percent) PPE/book assets (percent) Beta Cash flow sensitivity Sample size

Unconstrained firms

p-values of mean and distribution equality

Constrained firms

426.89 [70.75] 27.98 [24.00] 26.98 [25.55] 3.65 [3.64] 1.37 [1.04] 4.27 [2.67] 31.83 [27.02] 1.03 [1.04] 0.02 (0.02) 45,443

(0.000) [0.000] (0.000) [0.000] (0.965) [0.000] (0.000) [0.000] (0.000) [0.000] (0.000) [0.000] (0.000) [0.000] (0.000) [0.000] (0.003)

93.00 [20.82] 18.69 [14.71] 27.62 [24.68] 2.07 [3.29] 2.17 [1.57] 5.83 [3.18] 28.11 [21.47] 1.13 [1.17] 0.15 [0.04] 5,055

and Prðyit ¼ 1Þ ¼ F ½dðLevit
Levit
3 Þ þ MacroIit
3 Z þZit
3 g þ qt þ fqt þ d86t :

ð2Þ

The first equation describes a firm’s expected optimal leverage ratio in month t (Lev*it) as a function of known macroeconomic target variables (MacroTit
3); known firm-specific target variables (Xit
3); a fixed effect (fi); calendar quarter dummy variables (q1,t, q2,t, and q3,t, for the first through third quarters of the calendar year); a financial quarter dummy (fq1,t, fq2,t, and fq3,t, for the first through third quarters of the fiscal year); and a pre-1986 Tax Act dummy that equals one for years prior to 1987 (d86t). Lev*it is a firm’s expected optimal leverage ratio in the absence of the information asymmetries or transaction costs discussed in the introduction. The

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

85

second equation describes a firm’s expected issue (repurchase) choice (yit=1 for debt and 0 for equity) as a function of the distance the existing capital structure is from the expected target (Lev*it
Levit
3); known macroeconomic issue variables (MacroIit
3); known firm-specific issue variables (Zit
3); and a calendar quarter, a financial quarter, and a pre-1986 Tax Act dummy. Firm-specific and macroeconomic conditioning variables are discussed at length below. 2.3. Target leverage variables We explain firms’ target leverage using firm-specific and macroeconomic variables. 2.3.1. Firm-specific target leverage variables The choice of firm-specific target variables, Xit
3, is meant to capture the nature (tangible vs. intangible) of the firms’ assets; profitability; the existence of alternative tax shields; and size. Since firm-specific variables are not the primary focus of the paper, we do not include summary statistics for these variables. Unless specified, these variables are measured as a fraction of book assets. Net property plant and equipment is included as a measure of collateral (Titman and Wessels, 1988). The mean tax payment and depreciation over the previous four quarters are included to proxy for potential tax shields and tax losses carried forward. COMPUSTAT does not report tax losses carried forward at the quarterly frequency. However, Hovakimian et al. (2001) find that annual net operating loss carry-forwards are significantly positively correlated with operating profits. Following Titman and Wessels, the mean selling expense, as a fraction of sales over the previous four quarters, are included as a proxy for uniqueness. Measures of uniqueness proxy for costs associated with liquidating specialized assets. As a firm becomes more unique, it is more difficult and costly to liquidate its assets in the event of bankruptcy. The mean operating income over the previous four quarters and the firm’s market-tobook ratio are included as measures of long-term performance and growth opportunities (Titman and Wessels, 1988). A fifth order polynomial of market-tobook is included since its relation with leverage is highly nonlinear. The order is chosen based on the observation that higher orders have no substantive effect on the adjusted R2 or the other coefficients. Attempting to minimize Akaike’s information criterion, or the Schwartz criterion, suggests an order of greater than 15. The Akaike criterion leads to over fitting-models, even asymptotically (Greene, 1993, p. 517). In addition, we correct for firm size, as do Hovakimian et al. (2001) and Fama and French (2002). They argue that cash flows of larger, more diversified firms are less volatile, increasing the probability that interest tax shields will be realized and reducing expected bankruptcy losses. Market capitalization is non-stationary (e.g., the average market capitalization of NYSE/AMEX/NASDAQ firms in January 1985 was $289 million while the average market capitalization in December 1999 was $1,884 million). To correct for size in a stationary fashion, we include the ratio of the firm’s assets to the average market equity capitalization of CRSP stocks traded on NYSE, AMEX, and NASDAQ. Firm-level fixed-effects are included in most regressions to correct for idiosyncratic factors that influence capital structure.

86

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

2.3.2. Macroeconomic target variables In Kiyotaki and Moore (1997) and Levy (2001), the aggregate distribution of wealth between managers and outside shareholders determines the degree of agency problems and optimal leverage. Using the fact that manager compensation is tied to corporate profits (due to bonuses) and equity performance (due to equity and option compensation), we construct three series that are included in MacroTit
3 to proxy for this distribution effect. The two-year aggregate domestic nonfinancial corporate profit growth is computed using quarterly data from the Flow of Funds and is matched with the firm quarter with the most overlap. The two-year equity market return is computed from the CRSP value-weighted index of stocks traded on NYSE, AMEX, and NASDAQ. The annualized rate on three-month commercial paper over the rate on the three-month Treasury bill is used as a forward-looking measure of corporate profits and therefore a forward-looking measure of managerial compensation.5 The commercial paper spread is frequently argued to proxy for the market’s expectation of commercial paper issuances as firms smooth shocks to cash flows, and is shown to widen dramatically prior to an economic slowdown. Bernanke and Blinder (1992) observe that the commercial paper spread over Treasury bills sends to rise most sharply during credit-crunches induced by the Federal Reserve. Friedman and Kuttner (1993) provide a discussion of why the commercial paper spread performs well in predicting economic activity. Since issuances are endogenous, we use instrumental variables estimation when the commercial paper spread is included in the regressions.6 Unconditional sample correlations of leverage measures and the target macroeconomic variables, lagged one quarter, are presented in Table 2. Consistent with findings in the credit channel literature, unconditional debt ratios are generally more counter-cyclical for our unconstrained sample. In fact, for our constrained sample unconditional debt ratios have a pro-cyclical relationship with the two-year equity market return when measuring assets with book values.

2.4. Issue and repurchase variables In a world with no frictions one would expect firms to immediately adjust to their target leverage. This would imply that we could predict security issuance by knowing the current deviation from the target. With frictions, however, firms might choose to 5

The commercial paper rate and the Treasury Bill rate were obtained from the Federal Reserve Board’s web page at http://www.federalreserve.gov/releases. Prior to August 1997, the commercial paper rate is a rate on short-term negotiable promissory notes issued by financial and nonfinancial companies with AA bond ratings. After September 1997 the rate is on commercial paper issued by nonfinancial companies only. 6 The instruments used are the macroeconomic variables included in the regression, the term spread (defined and discussed in Section 2.4.2), and the spread between the annualized six-month Treasury-bill rate and the Federal Funds rate (which proxies for future expected monetary policy). Both the six-month Treasury-bill rate and the Federal Funds rate are obtained from the Federal Reserve Board’s web page at http://www.federalreserve.gov/releases.

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

87

Table 2 Correlation of macroeconomic ‘‘target’’ variables with leverage The table presents the correlation of leverage measures with lagged macroeconomic variables used in estimating ‘‘target’’ leverage, as well as p-values (in parentheses) from a test of correlation equality across the constrained and unconstrained sample. The sample includes all firm event windows that had CRSP return data 48 months prior to, and COMPUSTAT items eight quarters prior to and eight quarters after, a net debt or equity (including dividends) issue or repurchase of at least 5% of book assets between 1985:1 and 1998:3. Constrained firms do not pay dividends, do not have a net equity or debt purchase over the event window, and have a Tobin’s Q of greater than one at the end of the event quarter. Unconstrained firms are firms that are not labeled as constrained. Debt is measured as the sum of short- and long-term debt. Market assets are measured as the sum of the market value of equity and the book value of debt. The two-year corp. profit growth is real aggregate domestic nonfinancial corporate profit growth using quarterly data from the Flow of Funds and is matched with the firm quarter with the most overlap. The two-year equity market return is the real return on CRSP value-weighted index of stocks traded on NYSE, AMEX, and NASDAQ. The commercial paper spread is the annualized rate on three-month commercial paper over the rate on the three-month Treasury bill. The test for correlation equality is based on 1þr# Unconst 1þr# Const 1 1 1 1 2 ln 1
r# Unconst
2 ln 1
r# Const ; which is distributed normally with mean 0 and variance NUnconstr
3 þ NConstr
3 under the null of correlation equality, where r# is the estimated correlation (see Freund, 1992, p. 524). Unconstrained firm-windows

Constrained firm-windows

(N=45,976)

(N=5,055)

Variable

Debt Book assets

Debt Market assets

Debt Book assets

Debt Market assets

2-year corp. profit growth


0.052 (0.002)
0.010 (0.000) 0.071 (0.247)


0.078 (0.322)
0.053 (0.001) 0.084 (0.000)


0.009


0.071

0.067


0.007

0.060

0.162

2-year equity market return Commercial paper spread

deviate from target leverage. In modeling issue choice, we include firm-specific and macroeconomic variables that proxy for such frictions. 2.4.1. Firm-specific issue and repurchase variables As with our firm-specific target variables, the choice and results of firm-specific issue variables, Zit
3, are similar to extant research. Hence, we do not include summary statistics. Zit
3 represents a vector of firm-specific variables that, unless specified, are measured as a fraction of book assets. Ambarish et al. (1987) generalize Myers and Majluf (1984) to show that the stock price response to new equity financing depends on the relative asymmetric information of the value of assets in place versus growth opportunities. When the predominant source of information is growth opportunities (assets in place), the announcement effect is positive (negative). Empirical support is found by Pilotte (1992) who uses measures of growth opportunities as proxies for the relative information. Motivated by Pilotte (1992), we include proxies for growth opportunities in Zit
3 such as the mean of capital expenditures over the four quarters prior to the event; the mean of selling expenses as a fraction of sales over the four quarters prior to the event; and Tobin’s Q. Pilotte

88

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

(1992) shows that the ratio of research and development expenditures to sales, R&D/ Sales (a proxy for growth opportunities), is positively related to a firm’s two-day equity announcement period abnormal return. When R&D/Sales is high, the negative announcement effect is smaller. We use selling expense rather than R&D since selling expense includes R&D and is more consistently reported at a quarterly frequency. We get similar results when we use R&D/Sales on the subsample of reporting firms. We also include proxies for net operating loss carry-forwards (used in Hovakimian et al., 2001, but not reported in the COMPUSTAT quarterly data set) in Zit
3 since issue decisions will be influenced by tax considerations. The proxies include mean operating income (also argued to proxy for growth opportunities), mean tax payments, and mean depreciation over the four quarters prior to the event. As above, we correct for size in a stationary manner by including the ratio of the firm’s assets to the average equity market capitalization of CRSP stocks traded on NYSE, AMEX, and NASDAQ. Firm fixed effects are included in most regressions to correct for idiosyncratic factors that influence capital structure. The one-year abnormal return, using a market model with 48 to 60 months of return data prior to the issue, is included to correct for the observed equity price runup prior to equity issue announcement (see Korajczyk et al., 1990; and Graham and Harvey, 2001). This run-up in price prior to an equity issue is argued to reflect either investment opportunities or management timing equity issues to periods when they believe it is overpriced. Both Graham and Harvey (2001) and Baker and Wurgler (2002) provide discussions of equity issue timing. The dollar value of the issue (or repurchase) is included to correct for potential interactions between the issue size and price reaction in addition to the possibility that slightly under-levered firms might issue equity when they issue a large amount of only one security. Following Hovakimian et al. (2001), who argue that managers’ issue choice is influenced by reported accounting numbers, we include a dilution dummy variable that equals one if an equity issue dilutes the firm’s earnings per share more than a debt issue (also see Graham and Harvey, 2001). The firm’s earnings-price ratio is defined as E/P=(EBIT
Debt rd)(1
Tc)/market value of equity. Issuing equity dilutes the earnings-price ratio more than issuing debt if the first derivative of E/P with respect to issuing equity is more negative than the first derivative with respect to debt, which reduces to the condition that E/P>rd(1
Tc) (see Hovakimian et al., 2001, footnote 19). We use the yield on Moody’s Baa rated debt as rd, and Tc is set equal to 50% prior to 1987 and 34% afterwards. We also use a dummy variable that equals one if an equity issue will dilute the firm’s book equity value per share (the dummy equals one when the firm’s market value of equity is less than its book value). 2.4.2. Macroeconomic issue and repurchase variables Our macroeconomic issue variables are meant as proxies for frictions that influence issue choice. The choice of variables is motivated by theories that relate financing decisions to variations in adverse selection and bankruptcy costs. Korajczyk et al. (1992), Choe et al. (1993), Bayless and Chaplinsky (1996), and Eisfeldt (2001) argue that firms issue equity when adverse selection costs are low and,

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

89

therefore, the price reaction to an equity issue announcement is less negative. We measure the severity of the average adverse selection cost by a three-month moving average of two-day cumulative abnormal prediction error (CAPE) to seasoned equity issue announcements using data from Securities Data Corp. (SDC) and CRSP. Following Bayless and Chaplinsky (1996), CAPE are estimated for day 0 and day +1 using the market model where registration dates are used as announcement dates. To determine the effects of using registration dates, they collected announcement dates from NEXIS for 1989 and found a comparable CAPE based on announcement dates and registration dates. Returns are obtained from CRSP. Utilities and financial firms are excluded from the estimation. Since the construction of the price reaction variable resulted in a noisy variable, due in part to the low number of issues during certain periods, we use instrumental variables estimation when the price reaction variable is included in regressions. The instruments we use are the three-month lag of the included macroeconomic variables, the three-month lag of corporate profit growth over the previous three months, and the lagged growth rate of leading economic variables over the previous three months. This is the first variable included in MacroIit
3. Motivated by arguments in Gertler and Hubbard (1993), where macroeconomic conditions are credible indicators of investment opportunities, we use the term spread and the run-up in the equity market as credible signals of economic performance and expected growth opportunities. Stock and Watson (1998) provide a discussion of variables that predict economic activity. The term spread is measured as the difference between the Ibbotson Associates’ (2001) long-term (approximately 20 years) government bond yield series and the short-term (approximately one month) Treasury-bill rate series. The use of the equity market run-up is also due to the theoretical arguments in Lucas and McDonald (1990) and the empirical findings in Korajczyk et al. (1990) that equity issue clusters follow increases in the equity market. We included the three-month CRSP value-weighted equity market return in MacroIit
3. Finally, we are motivated by Pulvino (1998), who provides evidence that bankruptcy costs are counter-cyclical for the airline industry. We use the default spread, that is, an average yield on Baa less Aaa Moody’s rated corporate bonds with maturity of approximately 20–25 years, as a proxy for time variation in expected bankruptcy costs. A similar measure is used by Fama and French (1989).7 Table 3 presents sample mean and median values for issue choice macroeconomic variables, lagged one quarter, across the unconstrained and constrained firmquarters that have either a net equity or debt issue of larger than 5% of book assets in the previous quarter, but not both. Tests of mean equality or distribution equality across debt and equity issuers suggest that unconstrained firms issue equity when a larger than average run-up in the equity market has occurred, when the average price reaction to equity issue announcements is less negative than average, and when the term spread is higher than average (i.e., when economic prospects are good). 7

Both corporate bond series were obtained from the Federal Reserve Board’s web page at http:// www.federalreserve.gov/releases.

(0.002) [0.009] (0.005) [0.001] (0.002) [0.002] (0.645) [0.591]


1.78 [
1.88] 2.61 [2.83] 4.18 [4.08] 0.89 [0.85]

Price reaction to equity issue announcements

Default spread

Three-month equity market return

Term spread

p-values of mean and distribution equality

Equity issuers (N=392)

Variable


1.94 [
1.92] 2.40 [2.48] 2.78 [3.42] 0.89 [0.83]

Debt issuers (N=2,285)

Unconstrained firm quarters


1.80 [
1.89] 2.49 [2.29] 4.57 [4.13] 0.84 [0.73]

Equity issuers (N=94)

(0.148) [0.358] (0.952) [0.195] (0.011) [0.023] (0.724) [0.369]

p-values of mean and distribution equality


1.98 [
1.98] 2.49 [2.64] 2.96 [3.46] 0.83 [0.73]

Debt issuers (N=349)

Constrained firm quarters

(2.51)

(6.61)

(1.36)

(0.99)

Overall standard deviation

Table 3 Summary statistics of macroeconomic ‘‘issue’’ variables for unconstrained and constrained equity and debt issuers The table presents mean and median (in brackets) values of lagged macroeconomic variables for unconstrained and constrained firms that issue equity or debt. p-values of mean (in parentheses) and distribution (in brackets) equality across equity and debt issues are also presented. The last column presents the standard deviation of the macro variable. The sample includes all firms that had CRSP return data 48 months prior to, and COMPUSTAT items eight quarters prior to and eight quarters after, a net debt or equity issue of at least 5% of book assets between 1985:1 and 1998:3. Constrained firms do not pay dividends, do not have a net equity or debt purchase over the event window, and have a Tobin’s Q of greater than one at the end of the event quarter. Unconstrained firms are firms that are not labeled as constrained. The price reaction to equity issue announcements is the average CAPE estimated for day 0 and day +1 using a market model across all nonfinancial and nonutility equity issuers on SDC. The term spread is the difference between the long term yield on government bonds and the Treasury-bill rate. The three-month equity return is the real return on CRSP value-weighted index of stocks traded on NYSE, AMEX, and NASDAQ. The commercial paper spread is over the Treasury-bill rate. The default spread is the difference between an average yield on Baa less Aaa Moody’s rated bonds. The test of distribution equality is a Kolmogorov–Smirnov nonparametric test.

90 R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

91

Although the bivariate analysis suggests that the default spread (the cost of financial distress) is indistinguishable across unconstrained equity and debt issuers, the multivariate analysis, reported later, suggests that it is relevant. The results for the constrained sample are much noisier, with the mean equity market return the only variable that is statistically different for constrained equity and debt issuers. Moreover, the difference in mean and median point estimates are frequently opposite those of the unconstrained sample. Table 4 presents sample mean and median values for repurchase choice macroeconomic variables, lagged one quarter, across the unconstrained firmquarters that have either a net equity or debt repurchase of larger than 5% of book assets in the previous quarter, but not both. Since open market repurchases take several months (and sometimes several years) to complete, we are probably picking up tender offers in our definition of share repurchases. Fried (2001) discusses open market repurchases. Although the ratio of equity to debt repurchases will be biased downward, it is not clear that the coefficients on the explanatory macroeconomic- or firm-specific variables will be biased. Although these bivariate results suggest that

Table 4 Summary statistics of macroeconomic ‘‘repurchase’’ variables for unconstrained equity and debt repurchasers The table presents mean and median (in brackets) values of lagged macroeconomic variables for unconstrained firms that repurchase equity or debt. p-values of mean (in parentheses) and distribution (in brackets) equality across equity and debt repurchasers are also presented. The sample includes all firms that had CRSP return data 48 months prior to, and COMPUSTAT items eight quarters prior to and eight quarters after, a net debt or equity issue of at least 5% of book assets, and had at least 5% leverage previous to the repurchase between 1985:1 and 1998:3. Unconstrained firms are firms that are not labeled as constrained, where constrained firms do not pay dividends, do not have a net equity or debt purchase over the event window, and have a Tobin’s Q of greater than one at the end of the event quarter. The price reaction to equity issue announcements is the average CAPE estimated for day 0 and day +1 using a market model across all nonfinancial and nonutility equity issuers on SDC. The term spread is the difference between the long-term yield on government bonds and the Treasury-bill rate. The three-month equity return is the real return on CRSP value-weighted index of stocks traded on NYSE, AMEX, and NASDAQ. The default spread is the difference between an average yield on Baa less Aaa Moody’s rated bonds. The test of distribution equality is a Kolmogorov–Smirnov nonparametric test. Unconstrained firm quarters Variable

Price reaction to equity issue announcements Term spread 3-month equity market return Default spread

Equity repurchase (N=164)
1.90 [
1.90] 2.37 [2.19] 2.49 [3.43] 0.89 [0.90]

p-values of mean and distribution equality

Debt repurchase (N=1,769)

(0.506) [0.427] (0.912) [0.640] (0.461) [0.101] (0.734) [0.373]


1.96 [
1.91] 2.38 [2.31] 2.88 [3.41] 0.91 [0.90]

92

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

repurchase choices are unaffected by macroeconomic conditions, the multivariate analysis, reported below, suggests that they are relevant.

3. Estimation 3.1. Estimating target leverage We assume a firm’s actual leverage ratio (Levit) is equal to its optimal target leverage ratio (Lev*it), plus measurement error that is orthogonal to the explanatory variables. This implies that we can estimate coefficients and standard errors from Eq. (1) using the following specification: Levit ¼ MacroTit
3 a þ Xit
3 b þ fi þ qt þ fqt þ d86t þ eit :

ð3Þ

When estimating Eq. (3), four different measures of leverage are used. In all cases, debt is measured using book values. The first measure of leverage is the ratio of short-term plus long-term debt to market value of assets, measured as the sum of the market value of common equity and book value of debt. The second measure is the ratio of long-term debt to market value of assets. The third is the ratio of short-term plus long-term debt to book value of assets. The fourth measure is the ratio of shortterm plus long-term debt, less cash and marketable securities, to market value of assets, less cash and marketable securities. Our results regarding firm-specific variables are similar to those found in previous studies, such as Hovakimian et al. (2001) and Fama and French (2002). Target leverage is negatively related to firm profitability, the extent of non-interest tax shields, the level of taxes paid, the level of intangible versus tangible assets, and the market-to-book ratio. Target leverage is positively related to firm size. Some of our results are consistent with tradeoff theories (e.g., the negative relation between target leverage and non-interest tax shields). Some of our results are consistent with pecking order theories (e.g., the negative relation between target leverage and profitability). Finally, some of our results seem to indicate that some variables are proxies for deviations from target, such as the negative relation we find between target leverage and taxes paid. Hovakimian et al. (2001) interpret the relation between target leverage and net operating loss carry-forwards similarly. The secondstage issue choice coefficients on loss carry-forwards and taxes paid are consistent with this interpretation, since carry-forwards decrease, and taxes paid increase, the probability of issuing debt rather than equity. Because of the similarity of our results for firm-specific variables to those in the literature, we report the coefficients only for the macroeconomic variables. Regression coefficients, standard errors, and the time-series variance decomposition results can be found in Table 5 for a variety of specifications. Results for the unconstrained sample are presented in Panel A, and results for the constrained sample are presented in Panel B. Standard errors are robust to heteroskedasticity and are clustered at the overlapping event-level, which assumes error independence only across non-overlapping events (Rogers, 1993).

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

93

Table 5 Macroeconomic determinants of target leverage The table presents coefficients on lagged macroeconomic variables, standard errors (in parentheses), and a variance decomposition from estimating the determinants of target leverage for unconstrained (Panel A) and constrained (Panel B) event windows. Panel A also presents p-values of coefficient equality across unconstrained and constrained firms (in brackets). The sample includes all firm-event windows that had CRSP return data 48 months prior to, and COMPUSTAT items eight quarters prior to and eight quarters after, a net debt or equity (including dividends) issue or repurchase of at least 5% of book assets between 1985:1 and 1998:3. Constrained firms do not pay dividends, do not have a net equity or debt purchase over the event window, and have a Tobin’s Q of greater than one at the end of the event quarter. Unconstrained firms are firms that are not labeled as constrained. The two-year corp. profit growth is real aggregate domestic nonfinancial corporate profit growth using quarterly data from the Flow of Funds and is matched with the firm quarter with the most overlap. The two-year equity market return is the real return on CRSP value-weighted index of stocks traded on NYSE, AMEX, and NASDAQ. The commercial paper spread is the annualized rate on three-month commercial paper over the rate on the three-month Treasury bill, and is instrumented using the included macroeconomic variables, the term spread and the difference between the six-month Treasury bill rate and the Federal Funds rate. All regressions include, but do not report coefficients on a pre-1986 Tax Act dummy, financial and calendar quarterly dummies, the lag of property plant and equipment as a fraction of book assets, the mean operating income as a fraction of assets over the previous four quarters, the mean selling expense as a fraction of sales over the previous four quarters, the mean operating income over the previous four quarters, a fifth-order polynomial of the firm’s market-to-book ratio, the ratio of the firm’s assets-to-the market value of CRSP stocks, and either firm or four-digit SIC fixed effects. Standard errors are robust to heteroskedasticity, making no overlapping eventlevel error structure assumptions. Following Campbell (1991), we calculate asymptotic standard errors for the variance decomposition using the delta method. We treat regression coefficients and the covariance matrix of the macroeconomic and firm components as parameters jointly estimated using the Generalized Methods of Moments of Hansen (1982). When cash is netted out, the denominator in our measure of leverage is occasionally negative or very small, resulting in extreme observations. Therefore, we require a firm to have an absolute leverage ratio of less than three and require assets, net of cash holdings, be positive throughout the event window. Coefficients that are statistically significantly different from zero at the 1%, 5% and 10% level are marked with ***, **, and * respectively. Lev Panel A: unconstrained firms 2-year corp. profit growth 2-year equity market return Commercial paper spread Fixed effect Number of obs. Adjusted R2 Macroeconomic variance Firm variance 2 covariance Panel B: constrained firms 2-year corp. profit growth

STþLT Debt Market assets

STþLT Debt Market assets

LT Debt Market assets

STþLT Debt Book assets

STþLT Debt
cash Market assets
cash


0.083*** (0.013)
0.053*** (0.007) 3.585*** (0.552)


0.106*** (0.019)
0.050*** (0.009) 5.345*** (0.934)


0.044*** (0.012)
0.014** (0.006) 1.547*** (0.530)


0.062*** (0.015)
0.012** (0.008) 5.546*** (0.562)


0.035* (0.020)
0.057*** (0.010) 3.349*** (0.892)

Firm 45,443 0.825 0.272*** (0.010) 0.558*** (0.012) 0.171*** (0.006)

4 digit SIC 45,443 0.543 0.301*** (0.013) 0.488*** (0.015) 0.211*** (0.009)

Firm 45,443 0.775 0.120*** (0.012) 0.707*** (0.019) 0.172*** (0.017)

Firm 45,443 0.755 0.510*** (0.017) 0.579*** (0.017)
0.089*** (0.003)

Firm 44,882 0.803 0.144*** (0.012) 0.740*** (0.016) 0.115*** (0.010)

0.099*** (0.027) [0.000]

0.218*** (0.047) [0.000]

0.192*** (0.054) [0.000]

0.103*** (0.030) [0.000]

0.024 (0.036) [0.002]

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

94

Table 5 (continued) Lev

STþLT Debt Market assets

STþLT Debt Market assets

LT Debt Market assets

STþLT Debt Book assets

STþLT Debt
cash Market assets
cash

2-year equity market return

0.047*** (0.016) [0.000]
1.217 (1.209) [0.000] Firm 5,055 0.831 0.234*** (0.031) [0.253] 0.846*** (0.031) [0.000]
0.080*** (0.011) [0.000]

0.036** (0.019) [0.000]
0.046 (1.964) [0.013] 4 digit SIC 5,055 0.662 0.042** (0.019) [0.000] 0.946*** (0.028) [0.000] 0.011** (0.005) [0.000]

0.032** (0.016) [0.005]
1.116 (1.152) [0.036] Firm 5,055 0.780 0.286*** (0.038) [0.000] 0.766*** (0.037) [0.160]
0.052*** (0.007) [0.000]

0.178*** (0.026) [0.000] 1.759 (1.906) [0.057] Firm 5,055 0.782 0.411*** (0.022) [0.000] 0.363*** (0.021) [0.000] 0.225*** (0.011) [0.000]

0.061*** (0.026) [0.000]
3.141 (2.054) [0.003] Firm 5,040 0.786 0.286*** (0.035) [0.000] 0.859*** (0.031) [0.000]
0.145*** (0.018) [0.000]

Commercial paper spread

Fixed effect Number of obs. Adjusted R2 Macroeconomic variance

Firm variance

2 covariance

The last three rows of Table 5’s Panels A and B present results from decomposing the time-series variation in target leverage into macroeconomic and firm-specific factors. This decomposition provides a summary measure of the relevance of each factor. In computingP the decomposition, we construct Macrot ¼ PI # where I denotes the number of firms in # and Firmt ¼ Ii¼1 Xit
3 b; MacroT it
3 a i¼1 the sample at time t. We then define the variance decomposition as: Macro Variance ¼

Firm Variance ¼

Covariance ¼

Variance½Macrot  ; Variance½Macrot þ Firmt 

Variance½Firmt  ; Variance½Macrot þ Firmt 

Covariance½Firmt ; Macrot  : Variance½Macrot þ Firmt 

ð4Þ

ð5Þ

ð6Þ

Firm- or industry-fixed effects are included in the regressions, but are not included in the variance decomposition. Therefore, the variance decomposition does not incorporate variations due to firms with different average leverage entering and exiting the sample as they access financial markets. Firmt includes only deviations from a firm’s average (industry) leverage. Industry composition is determined by the firm’s four-digit Standard Industrial Classification (SIC) code. As in Campbell (1991), we calculate the asymptotic standard errors for the variance decomposition using the delta method. We treat the regression coefficients

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

95

and the covariance matrix of Firmt and Macrot as parameters to be jointly estimated using the Generalized Method of Moments of Hansen (1982). Looking at the coefficients for the macroeconomic variables, the unconstrained sample has both statistically and economically significant counter-cyclical coefficients (Table 5, Panel A), but the constrained sample generally has pro-cyclical coefficients that are statistically different from the unconstrained sample (see Table 5, Panel B). This corresponds with predictions in Kiyotaki and Moore (1997), Suarez and Sussman (1999), and Levy (2001). Kiyotaki and Moore (1997) predict that procyclical collateral values result in pro-cyclical leverage patterns for constrained firms. This suggests a positive relation between leverage and the product of collateral measures, such as property plant and equipment, and our measures of macroeconomic conditions. When we include such product terms, we indeed find a positive but statistically insignificant relationship. We suspect the lack of significance stems from the fact that as collateral increases, firms tend to be less constrained, resulting in a noisy experiment. The variance decomposition results are given in the last three rows of Panels A and B of Table 5. They show that macroeconomic conditions account for a substantial proportion of the time variation in target leverage, particularly for the unconstrained sample. The firm-specific variables also account for a substantial proportion of target leverage variation. The proportion of variance attributable to the firm-specific variables is significantly larger for the constrained sample when we use market-value leverage ratios. As expected, target leverage variations due to macroeconomic and firm-specific variables co-vary positively for the unconstrained sample. During expansions, firms have positive earnings and tend to have lower leverage as in Titman and Wessels (1988), for example. An exception occurs when assets are measured with book values where the negative covariation is driven by the unusual increase in leverage during the 1980s (also see Section 4.3 and Bernanke and Campbell, 1988). The signs of the target leverage coefficients generally are the same regardless of whether assets are measured in market values or book values. The major exception is for the market-to-book ratio. Fama and French (2002) provide an extensive analysis of book versus market debt ratios. The covariance component is generally small for the constrained firms. 3.2. Estimating the debt-equity issue choice In this section we describe the results from estimating the specification described in Eq. (2) for firms that have a net equity or debt issue of at least 5% of their book assets. Hovakimian et al. (2001) use a similar method to classify issues, and obtain results similar to ours using issue data from SDC. However, they did find instances of misclassification of equity issues, due to debt conversion or a transfer of an equity account from a subsidiary to a parent, and similar misclassification of debt issues. Throughout the analysis we consider two separate measures of debt: long-term debt and the sum of short-term and long-term debt. Table 6 presents the coefficients and standard errors (robust to heteroskedasticity using the Huber/White/sandwich estimator) from estimating a probit model of the probability of issuing debt versus

OLS

1.21*** (0.08) Price reaction to equity issue announcements
4.38*** (1.59)

(Lev*t
Levt
3)

Variable 1.07*** (0.08)
3.60** (1.53)

Probit

ST or LT debt vs. equity

1.51*** (0.12)
8.22*** (2.74)

OLS

1.73*** (0.17)
9.29*** (3.25)

Probit

LT debt vs. equity

Unconstrained firms

2.77*** (0.34)
6.56 (5.33)

OLS

2.85*** (0.39)
3.87 (4.90)

Probit

ST or LT debt vs. equity

Probit 3.02*** 4.00*** (0.56) (0.82)
5.06
9.29 (7.99) (8.05)

OLS

LT debt vs. equity

Constrained firms

Table 6 Determinants of issue choice The table presents coefficients ð@Pr½debt issue=@VariableÞ; standard errors (in parentheses), and a variance decomposition from estimating the determinants of issue choice for unconstrained and constrained firms. The sample includes all firms that had CRSP return data 48 months prior to, and COMPUSTAT items eight quarters prior to and eight quarters after, a net debt or equity issue of at least 5% of book assets between 1985:1 and 1998:3. Constrained firms do not pay dividends, do not have a net equity or debt purchase over the event window, and have a Tobin’s Q of greater than one at the end of the event quarter. Unconstrained firms are firms that are not labeled as constrained. Lev*t
Lev t
3 is the difference between the fitted values from the first-stage target regression and actual leverage in the quarter previous to the event, where leverage is measured using the ratio of short- and long-term debt to the sum of short-term debt, long-term debt, and the market value of equity. The three-month equity return is the 3-month CRSP value-weighted equity market return. The price reaction to equity issue announcements is the average CAPE estimated for day 0 and day +1 using a market model across all nonfinancial and nonutility equity issuers on SDC. The three-month lag of the included macroeconomic variables, the three-month lag of corporate profit growth over the previous three months, and the lagged growth rate of leading economic variables over the previous three months were used as instruments for the price reaction variable, which is probably measured with error. Although the fitted values have the same unconditional mean, the series is smoother with a standard deviation of 0.006. The term spread is the difference between the long-term yield on government bonds and the Treasury-bill rate. The three-month equity return is the real return on CRSP valueweighted index of stocks traded on NYSE, AMEX, and NASDAQ. The commercial paper spread is over the Treasury-bill rate. The default spread is the difference between an average yield on Baa less Aaa Moody’s rated bonds. The one-year abnormal return uses a market model with 60 months of data prior to the issue. All regressions include, but do not report, coefficients on a pre-1986 Tax Act dummy, financial and calendar quarterly dummies, the mean of capital expenditures as a fraction of assets over the four quarters prior to the event, the mean of selling expenses as a fraction of sales over the four quarters prior to the event, Tobin’s Q, mean operating income as a fraction of assets over the four quarters prior to the event, mean tax payments as a fraction of assets over the four quarters prior to the event, mean depreciation as a fraction of assets over the four quarters prior to the event, the ratio of the firm’s assets to the market value of CRSP stocks, the dollar value of the issue as a fraction of assets, a dilution dummy that equals one if an equity issue dilutes the firm’s earnings per share more than a debt issue, and a market-to-book ratio dummy that equals one if an equity issue will dilute the firm’s book value per share. Probit regressions also include residuals from first stage regression to correct for the bias introduced by including an estimate. Standard errors are robust to heteroskedasticity using the Huber/White/sandwich estimator. Coefficients that are statistically significantly different from zero at the 1%, 5%, and 10% level are marked with ***, **, and *, respectively.

96 R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

2 cov(macro,firm)

2 cov(target,firm)

2 cov(target,macro)

Firm variation

Macroeconomic variation

Number of observations Adjusted R2/pseudo R2 Target deficit variation

One year abnormal return

Default spread

Three month equity market return

Term spread


1.89*** (0.48)
0.16* (0.10)
0.06** (0.03)
0.11*** (0.01) 2,677 0.23 0.21*** (0.03) 0.38*** (0.09) 0.41*** (0.05)
0.10*** (0.01) 0.02** (0.01) 0.08*** (0.02) —

—

—

—

—


1.92*** (0.45)
0.17** (0.09)
0.02 (0.03)
0.07*** (0.01) 2,677 0.28 —


3.09*** (0.82)
0.30** (0.16)
0.13*** (0.05)
0.14*** (0.02) 1,542 0.27 0.15*** (0.03) 0.48*** (0.08) 0.36*** (0.05)
0.08*** (0.02) 0.03*** (0.01) 0.06*** (0.02) —

—

—

—

—


3.79*** (0.97)
0.35* (0.20)
0.15*** (0.06)
0.13*** (0.02) 1,542 0.26 —


2.24 (1.69)
0.53* (0.33)
0.08 (0.09)
0.00 (0.02) 443 0.34 0.47*** (0.10) 0.22* (0.12) 0.40*** (0.08)
0.04 (0.03)
0.10** (0.05) 0.05* (0.03) —

—

—

—

—


2.32 (1.68)
0.62** (0.34)
0.09 (0.07)
0.01 (0.02) 443 0.39 —


2.91 (2.55)
0.97** (0.49)
0.15 (0.14)
0.01 (0.03) 294 0.31 0.43*** (0.11) 0.24* (0.13) 0.35*** (0.09)
0.07** (0.06)
0.04 (0.04) 0.08** (0.03) —

—

—

— —


1.69 (3.90)
1.23** (0.63)
0.36 (0.24)
0.02 (0.04) 294 0.28 — R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109 97

98

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

equity as a function of the independent variables.8 Additionally, OLS estimates are presented and used to construct the variance decomposition, which requires a linear relationship between the fitted values and the independent variables. Higher firm profitability and the level of taxes paid increase the probability of issuing debt even though the coefficients of these variables in the target leverage regression are negative. The change in sign on profitability between the target leverage and issue choice regressions is consistent with a short-run pecking order model (internally generated profits are used rather than debt). It is also consistent with a long-run tradeoff model (when a profitable firm eventually accesses the public capital markets, it is more likely to issue debt). Similar results are found by Hovakimian et al. (2001). As discussed above, the change in sign between the target leverage and issue choice regressions of the coefficient on taxes paid is similar to the change in sign on loss carry-forwards found by Hovakimian et al. (2001) and is consistent with these variables proxying for deviations from target leverage. As with the target regressions, since our results regarding firm-specific variables are similar to those found in previous studies, we report only coefficients on macroeconomic variables. Issue patterns for unconstrained and constrained firms are substantially different. The leverage deficit is the difference between the expected target leverage ratio at the end of the issue quarter (Lev*it) and the actual leverage ratio at the beginning of the issue quarter. The target is estimated from the fitted values of Eq. (3). As expected, the coefficients on the leverage deficit are positive and statistically significant for both classes of firms; a leverage deficit increases the probability of issuing debt. We STþLT Debt use Market Assets as the measure of leverage when analyzing short-term or long-term LT Debt debt issues, and Market Assets when analyzing only long-term debt issues. This yields results consistent with those found in Hovakimian et al. (2001), who find that firms move more actively toward their target than suggested by Shyam-Sunder and Myers (1999). The latter argue that the pecking order theory provides a better empirical description of capital structures than theories arguing that firms target their leverage ratio. Notice that constrained firms are much more sensitive to deviations from their target. The difference is generally statistically significant, but not reported in Table 6 for brevity. At the margin, a one-standard-deviation increase in the average target deficit results in an increase in the probability of issuing (debt rather than equity) of 10% to 15% for unconstrained firms, and 16% to 23% for the constrained firms. The standard deviation of the leverage deficit, measured using the sum of short-term and long-term debt, is 0.09 for unconstrained and 0.06 for constrained firms. Looking at the unconstrained sample, all of the macroeconomic variables have the expected signs and are generally statistically significant. Meanwhile, the only macroeconomic variable that is significant at predicting issue choice for the constrained sample is the return on the equity market. The difference across the two samples is generally not statistically significant and is not reported in the table for 8

Standard errors that were estimated using a bootstrap technique to account for the estimation of the leverage deficit and instrumented price reaction (see Table 6) were virtually identical to the ones reported.

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

99

brevity. It is interesting that the significant variables for both samples become larger in magnitude when equity issues are compared to long-term debt issues, rather than total debt issues, for both the constrained and unconstrained firms. This may be a result of short-term debt issues representing short-term financing needs as opposed to long-term capital structure decisions. The variance decomposition is estimated in the same manner as in Section 3.1. The difference between the unconstrained and constrained sample is striking. As suggested by the coefficients, the unconstrained sample is much more sensitive to macroeconomic conditions while the constrained sample is much more sensitive to deviations from the ‘‘target.’’ Moreover, macroeconomic conditions are only marginally statistically significant for the constrained sample. Meanwhile, the variance attributed to time variation in firm-specific variables is substantial for both samples. Similar to the macroeconomic variables, the one-year abnormal return on the firm’s equity, the only firm-specific variable we report, is insignificant for the constrained sample. In general, the results support the hypothesis that constrained firms do not time issues to periods when conditions are favorable while the unconstrained firms are able to time issues. Moreover, since constrained firms are much more sensitive to deviations from their target, and deviate from their target by less (the target deviation has a smaller standard deviation), this timing ‘‘indifference’’ suggests that constrained firms take what they can get. 3.3. Estimating the debt-equity repurchase choice Similar to Section 3.2, this section describes results from the second stage choice regressions for unconstrained firms that have a net equity (preferred and common) or debt repurchase of at least 5% of their book assets. Throughout the analysis we consider two separate measures of debt: long-term debt and the sum of short-term and long-term debt.9 For a firm to be included in the sample, its relevant debt ratio must be at least 5% in the quarter previous to the repurchase. Table 7 presents coefficients and standard errors (robust to heteroskedasticity using the Huber/ White/sandwich estimator) from estimating the probability of repurchasing debt versus equity as a function of the independent variables using a probit specification. Standard errors that were estimated using a bootstrap technique to account for the estimation of the leverage deficit and instrumented price reaction (see Table 6) were virtually identical to the ones reported. OLS estimates are also presented and used to construct the variance decomposition. The leverage deficit coefficient has the predicted sign and is statistically significant in that a leverage deficit decreases the probability that a firm will repurchase debt. Although the return on the equity market and the default spread are not statistically significant, the term spread, a measure of future macroeconomic performance, is significant. The positive and significant coefficient on the price reaction variable 9

Including large dividends along with equity repurchases made virtually no difference. Results are excluded for brevity.

100

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

Table 7 Determinants of repurchase choice The table presents coefficients ð@Pr½debt issue=@ Variable), standard errors (in parentheses), and a variance decomposition from estimating the determinants of repurchase choice for unconstrained firms. The sample includes all firms that had CRSP return data 48 months prior to, and COMPUSTAT items eight quarters prior to and eight quarters after, a net debt or equity repurchase of at least 5% of book assets, and had at least 5% leverage previous to the repurchase between 1985:1 and 1998:3. Unconstrained firms are firms that are not labeled as constrained, where constrained firms do not pay dividends, do not have a net equity or debt purchase over the event window, and have a Tobin’s Q of greater than one at the end of the event quarter. Lev*t
Levt
3 is the difference between the fitted values from the first-stage target regression and actual leverage in the quarter previous to the event, where leverage is measured using the ratio of short- and long-term debt to the sum of short-term debt, long-term debt, and the market value of equity. The three-month equity return is the three-month CRSP value-weighted equity market return. The price reaction to equity issue announcements is the average CAPE estimated for day 0 and day +1 using a market model across all nonfinancial and nonutility equity issuers on SDC. The three-month lag of the included macroeconomic variables, the three-month lag of corporate profit growth over the previous three months, and the lagged growth rate of leading economic variables over the previous three months were used as instruments for the price reaction variable, which is probably measured with error. Although the fitted values have the same unconditional mean, the series is smoother with a standard deviation of 0.006. The term spread is the difference between the long-term yield on government bonds and the Treasury-bill rate. The three-month equity return is the real return on CRSP value-weighted index of stocks traded on NYSE, AMEX, and NASDAQ. The default spread is the difference between an average yield on Baa less Aaa Moody’s rated bonds. The commercial paper spread is over the Treasury-bill rate. The default spread is the difference between an average yield on Baa less Aaa Moody’s rated bonds. The one-year abnormal return, uses a market model with 60 months of data prior to the issue. All regressions include, but do not report coefficients on a pre-1986 Tax Act dummy, financial and calendar quarterly dummies, the mean of capital expenditures as a fraction of assets over the four quarters prior to the event, the mean of selling expenses as a fraction of sales over the four quarters prior to the event, Tobin’s Q, mean operating income as a fraction of assets over the four quarters prior to the event, mean tax payments as a fraction of assets over the four quarters prior to the event, mean depreciation as a fraction of assets over the four quarters prior to the event, the ratio of the firm’s assets to the market value of CRSP stocks, the dollar value of the repurchase as a fraction of assets, a dilution dummy that equals one if an equity repurchase dilutes the firm’s earnings per share more than a debt issue, and a market-to-book ratio dummy that equals one if an equity repurchase will dilute the firm’s book value per share. Probit regressions also include residuals from first-stage regression to correct for the bias introduced by including an estimate. Standard errors are robust to heteroskedasticity using the Huber/White/sandwich estimator. Coefficients that are statistically significantly different from zero at the 1%, 5%, and 10% level are marked with ***, **, and * respectively. ST or LT debt vs. equity Variable (Lev*t
Levt
3) Price reaction to equity issue announcements Term spread Three-month equity market return Default spread One year abnormal return

OLS

Probit


0.74*** (0.09) 5.10*** (1.74) 0.85* (0.52)
0.01 (0.09)
0.01 (0.03) 0.04*** (0.01)


0.61*** (0.07) 2.99*** (1.06) 0.73** (0.33) 0.00 (0.06)
0.00 (0.02) 0.02** (0.01)

LT debt vs. equity OLS

Probit


1.12*** (0.14) 8.61** (3.41) 2.37** (1.01) 0.06 (0.19)
0.06 0.06 0.07*** (0.02)


1.23*** (0.17) 6.92** (2.84) 2.49*** (0.84) 0.07 (0.14)
0.02 (0.05) 0.04** (0.02)

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

101

Table 7 (continued) ST or LT debt vs. equity Variable Number of observations Adjusted R2/pseudo R2 Target deficit variation Macroeconomic variation Firm variation 2 Cov(target,macro) 2 Cov(target,firm) 2 Cov(macro,firm)

OLS 1,933 0.13 0.47*** (0.11) 0.71*** (0.13) 0.29*** (0.08)
0.33*** (0.04) 0.07*** (0.02)
0.19*** (0.03)

Probit 1,933 0.24 — — — — — —

LT debt vs. equity OLS 1,933 0.17 0.28*** (0.08) 0.51*** (0.12) 0.23*** (0.07)
0.16*** (0.04) 0.13*** (0.03)
0.00 (0.02)

Probit 861 0.23 — — — — — —

suggests that the price reaction variable is a proxy for more than the simple window of opportunity in which equity capital can be raised at favorable terms (Bayless and Chaplinsky, 1996). To ensure that the significance of the coefficient on the price reaction variable is not due to firms timing equity for debt recapitalizations, we estimate the repurchase choice model excluding observations that have an equity issue and a debt repurchase in the same quarter (less than 10% of the repurchase sample had an equity issue and debt repurchase). We find that the coefficients on the price reaction variable are not economically or statistically different from the full sample. The results are excluded for brevity. In general, the effect of variation in short-term macroeconomic variables on repurchases complement the results on issues in exhibiting a move toward debt financing (debt issues or equity repurchases) during economic downturns (i.e., when the price reaction variable is more negative, when the return on the equity market has been low, and when the term spread is high). Moreover, as above, the target leverage deviation and the significant macroeconomic variables become more economically relevant when comparing equity issues to long-term debt issues rather than total debt issues. Finally, the variance decomposition results are similar to those found above.

4. Robustness checks 4.1. Product market competition Several empirical and theoretical studies have suggested that capital structure affects competition in the product market. For example, Campello (2000) finds that a

102

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

firm’s leverage, relative to that of its industry rivals, will affect its incentives in the product market following aggregate demand shocks. These findings suggest that variables included on the right-hand side of the target leverage regression (Eq. (3)) that proxy for profitability are endogenous. To ensure that this potentially spurious correlation is not driving the results in Table 5, we use the fact that capital structure can interact with product market competition only when firms have market power. Specifically, target leverage is estimated for a subsample of unconstrained firms in competitive industries. We were not able to conduct the same exercise for the constrained sample due to its small sample size. Using the classification procedure of Rotemberg and Saloner (1986), we split our sample of manufacturing firms on the basis of the two-digit SIC four-firm concentration ratio. Industries with a concentration ratio of less than the median of 35.4 are labeled as competitive, and those with a ratio of 35.4 or more are labeled as noncompetitive.10 As seen in the first column of Table 8, the coefficients for competitive manufacturing firms are very similar to the coefficients for the noncompetitive manufacturing subsample reported in the second column. Moreover, statistical tests, not reported in the table for brevity, indicate that the coefficients are not statistically different across the two subsamples. 4.2. Debt overhang and endogenous growth Several empirical and theoretical studies have suggested that debt overhang (Myers, 1977; or Hennessy, 2001) can have detrimental effects on investment. Lang et al. (1996) use Tobin’s Q as a proxy for growth opportunities and find a negative relation between investment and leverage for manufacturing firms whose growth opportunities are either not recognized or not sufficiently valuable to overcome the effects of their debt overhang. They find no such relationship for growth firms. Their results suggest that proxies for growth included on the right-hand side of the target leverage regression may be endogenous for firms having low Tobin’s Q ratios, potentially biasing our results in Panel A of Table 5. The third and fourth columns of Table 8 present results from estimating target leverage for the subsamples of unconstrained manufacturing firm quarters with Tobin’s Q of greater than one and with Tobin’s Q of less than or equal to one. The basic results are unchanged. 4.3. The unusual leverage of the 1980s As noted in Bernanke and Campbell (1988), leverage was unusually high throughout the 1980s. This is surprising given that corporate profits and the return 10 The sample includes manufacturing firms in food and kindred products (SIC 2000-2099), textile products (SIC 2200-2299), apparel and related products (SIC 2300-2399), lumber and wood products (SIC 2400-2499), furniture and fixtures (SIC 2500-2599), paper and allied products (SIC 2600-2699), petroleum and coal products (SIC 2900-2999), leather and leather products (SIC 3100-3199), and fabricated metal industries (SIC 3400-3499). Printing and publishing (SIC 2700-2799) is excluded, despite a concentration measure of 18.90, since newspapers and magazines are ‘‘highly concentrated, once location in space or type are taken into account’’ (Rotemberg and Saloner, 1986, p. 401).

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

103

Table 8 Macroeconomic determinants of target leverage for subsamples of unconstrained manufacturing firms The table presents coefficients on lagged macroeconomic variables, standard errors (in parentheses), and a variance decomposition from estimating the determinants of target leverage, measured as the ratio of book value of short- and long-term debt to the sum of short-term debt, long-term debt, and the market value of equity, for unconstrained event windows. Unconstrained firms are firms that are not labeled as constrained, where constrained firms do not pay dividends, do not have a net equity or debt purchase over the event window, and have a Tobin’s Q of greater than one at the end of the event quarter. The sample includes all firm event windows that had CRSP return data 48 months prior to, and COMPUSTAT items eight quarters prior to and eight quarters after, a net debt or equity (including dividends) issue or repurchase of at least 5% of book assets between 1985:1 and 1998:3. Competitive industries includes firms with SICs in the range of 2000-2099, 2200-2699, 2900-2999, 3100-3199, and 3400-3499. Noncompetitive industries include firms with SICs in the range of 2000-3999 that are not competitive. Growth firms are defined as manufacturing firms (SICs in the range of 2000-3999) that have a Tobin’s Q of greater than one, and non-growth firms are defined as manufacturing firms (SICs in the range of 2000-3999) that have a Tobin’s Q of less than or equal to one. The two-year corp. profit growth is real aggregate domestic nonfinancial corporate profit growth using quarterly data from the Flow of Funds and is matched with the firm quarter with the most overlap. The two-year equity market return is the real return on CRSP valueweighted index of stocks traded on NYSE, AMEX, and NASDAQ. The commercial paper spread is the annualized rate on three month commercial paper over the rate on the three month Treasury bill, and is instrumented using the included macroeconomic variables, the term spread and the difference between the six-month Treasury-bill rate and the Federal Funds rate. All regressions include, but do not report coefficients on a pre-1986 Tax Act dummy, financial and calendar quarterly dummies, the lag of property plant and equipment as a fraction of book assets, the mean operating income as a fraction of assets over the previous four quarters, the mean selling expense as a fraction of sales over the previous four quarters, the mean operating income over the previous four quarters, a fifth order polynomial of the firm’s marketto-book ratio, the ratio of the firm’s assets to the market value of CRSP stocks, and either firm or fourdigit SIC fixed effects. Standard errors are robust to heteroskedasticity, making no overlapping event-level error structure assumptions. Following Campbell (1991), we calculate asymptotic standard errors for the variance decomposition using the delta method. We treat regression coefficients and the covariance matrix of the macroeconomic and firm components as parameters jointly estimated using the Generalized Methods of Moments of Hansen (1982). When cash is netted out, the denominator in our measure of leverage is occasionally negative or very small, resulting in extreme observations. Therefore, we require a firm to have an absolute leverage ratio of less than three and require assets, net of cash holdings be positive throughout the event window. Coefficients that are statistically significantly different from zero at the 1%, 5%, and 10% level are marked with ***, **, and * respectively.

Two-year corp. profit growth Two-year equity market return Commercial paper spread Fixed effect Number of obs. Adjusted R2 Macroeconomic variance Firm variance 2 covariance

Competitive industries

Noncompetitive industries


0.072*** (0.029)
0.066*** (0.015) 2.552** (1.245) Firm 8,528 0.849 0.295*** (0.026) 0.666*** (0.028) 0.039*** (0.003)


0.116*** (0.018)
0.056*** (0.009) 4.862*** (0.788) Firm 20,268 0.812 0.260*** (0.011) 0.428*** (0.013) 0.311*** (0.012)

Growth firms

Nongrowth firms


0.037** (0.016)
0.022** (0.009) 2.81*** (0.700) Firm 15,664 0.810 0.182*** (0.017) 0.619*** (0.024) 0.198*** (0.019)


0.156*** (0.022)
0.094*** (0.012) 1.873** (0.973) Firm 13,132 0.811 0.664*** (0.019) 0.393*** (0.019)
0.058*** (0.002)

104

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

on equity was high, and the average price reaction to equity issues was relatively small in absolute value. Between the first quarter of 1985 and the fourth quarter of 1988 the average price reaction was
1.7%. Meanwhile, the average price reaction during the expansion in the 1990s (between the first quarter of 1992 and the third quarter of 1998) was
2.1%. The first column of Table 9 presents target leverage estimates for the unconstrained sample using data up to the first quarter of 1988. We were not able to conduct the same exercise for the constrained sample due its small sample size. All three macroeconomic coefficients are larger in absolute value than in the full sample. Both the two-year equity market return and two-year corporate profits are statistically more negative than the full sample (test statistics are not reported for brevity). The substantive difference, between this sample and the full sample, in the variance decomposition is the strong negative covariance of macroeconomic and firm variables, which is expected given that leverage was increasing during this period. The results suggest that the increase in leverage was due to firm-specific variables. 4.4. Time-varying volatility It has long been argued that optimal leverage should be inversely related to the return volatility of the underlying asset. Since volatility moves counter-cyclically (Campbell et al., 2001), variation in volatility is not driving the counter-cyclical leverage observed in the unconstrained sample. To ensure that counter-cyclical volatility is not driving our results for the constrained sample, we run the target regressions for the subperiod between 1991:2 and 1998:3 since Campbell et al. (2001, Figs. 2–4) show this sub-period has relatively stable volatility. The second column of Table 9 shows that the results for the macroeconomic variables are more pronounced for this subperiod.

5. Conclusion In this study we examine the determinants of time variation in firms’ leverage ratios and security issue choices between 1984 and 1998. The sample is divided on the basis of a measure of financial constraints faced by the firms. We find that the leverage of firms in our financially unconstrained sample varies counter-cyclically with macroeconomic conditions. This supports the premise of Levy’s (2001) model of capital structure choice in which managers prefer debt financing when their compensation is relatively low, that is, following low returns in the equity market or low corporate profits. Moreover, macroeconomic conditions account for 12% to 51% of the time-series variation in these firms’ leverage. Meanwhile, we find that firms in the financially constrained sample have pro-cyclical leverage with macroeconomic conditions accounting for 4% to 41% of the time series variation. This is consistent with the balance sheet credit channel models of Kiyotaki and Moore (1997) and Suarez and Sussman (1999) in which constrained

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

105

Table 9 Macroeconomic determinants of target leverage for various subperiods The table presents coefficients on lagged macroeconomic variables, standard errors (in parentheses), and a variance decomposition from estimating the determinants of target leverage, measured as the ratio of book value of short- and long-term debt to the sum of short-term debt, long-term debt, and the market value of equity, for unconstrained event windows. Unconstrained firms are firms that are not labeled as constrained, where constrained firms do not pay dividends, do not have a net equity or debt purchase over the event window, and have a Tobin’s Q of greater than one at the end of the event quarter. The sample includes all firm event windows that had CRSP return data 48 months prior to, and COMPUSTAT items eight quarters prior to and eight quarters after, a net debt or equity (including dividends) issue or repurchase of at least 5% of book assets between 1985:1 and 1998:3. The two-year corp. profit growth is real aggregate domestic nonfinancial corporate profit growth using quarterly data from the Flow of Funds and is matched with the firm quarter with the most overlap. The two-year equity market return is the real return on CRSP value-weighted index of stocks traded on NYSE, AMEX, and NASDAQ. The commercial paper spread is the annualized rate on three-month commercial paper over the rate on the three-month Treasury bill, and is instrumented using the included macroeconomic variables, the term spread and the difference between the six-month Treasury-bill rate and the Federal Funds rate. All regressions include, but do not report coefficients on a pre-1986 Tax Act dummy, financial and calendar quarterly dummies, the lag of property plant and equipment as a fraction of book assets, the mean operating income as a fraction of assets over the previous four quarters, the mean selling expense as a fraction of sales over the previous four quarters, the mean operating income over the previous four quarters, a fifth-order polynomial of the firm’s market-to-book ratio, the ratio of the firm’s assets-to-the market value of CRSP stocks, and either firm or four-digit SIC fixed effects. Standard errors are robust to heteroskedasticity, making no overlapping event-level error structure assumptions. Following Campbell (1991), we calculate asymptotic standard errors for the variance decomposition using the delta method. We treat regression coefficients and the covariance matrix of the macroeconomic and firm components as parameters jointly estimated using the Generalized Methods of Moments of Hansen (1982). When cash is netted out, the denominator in our measure of leverage is occasionally negative or very small, resulting in extreme observations. Therefore, we require a firm to have an absolute leverage ratio of less than three and require assets, net of cash holdings, be positive throughout the event window. Coefficients that are statistically significantly different from zero at the 1%, 5%, and 10% level are marked with ***, **, and * respectively.

Subperiod Two-year corp. profit growth Two-year equity market return Commercial paper spread Fixed effect Number of obs. Adjusted R2 Macroeconomic variance Firm variance 2 covariance

Unconstrained firms

Constrained firms

Prior to 1988:1
0.211*** (0.030)
0.109*** (0.008) 7.839*** (1.429) Firm 5,531 0.876 0.848*** (0.018) 0.214*** (0.014)
0.062*** (0.002)

1992:2 to 1998:4 0.249*** (0.035) 0.122*** (0.020)
4.365*** (1.180) Firm 3,651 0.815 0.568*** (0.020) 0.204*** (0.014) 0.228*** (0.008)

106

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

firms borrow more when collateral values are highest, that is, following high returns in the equity market or high corporate profits. Our results support arguments that, at the issue-choice stage, firms consider how far they are from their target leverage as well as the marginal costs associated with issuing one security over another. In particular, it seems that unconstrained firms are able to time their issues to periods when the relative pricing of the asset is favorable. Deviations from the target account for 15% to 21% and macroeconomic conditions account for approximately 38% to 48% of their time-series variation in issue choice. Meanwhile, we find much less evidence of macroeconomic timing on the part of constrained firms. The results indicate that constrained firms deviate from their target by less and their issue choices are much more sensitive to deviations from their target, which accounts for 43% to 47% of the time-series variation in issue choice. Meanwhile, macroeconomic conditions are only marginally statistically significant and account for 22% to 24% of their time-series variation. In short, it seems constrained firms take what they can get. Our results on repurchases for our unconstrained sample complement those found on issues. Deviations from the target account for 28% to 47% of the time variation in repurchases and macroeconomic conditions account for 51% to 71% of the variation. We find that the average price reaction to equity issue announcements significantly influences repurchase decisions. This suggests that time-series variation in the price reaction is capturing more than ‘‘windows of opportunity,’’ described in Bayless and Chaplinsky (1996). Our results are consistent with elements of both tradeoff and pecking order theories. For example, the fact that deviations from target leverage explain issue choice is consistent with the tradeoff theory while the negative relation between profitability and target leverage is consistent with the pecking order theory. Several of our results indicate that constrained firms fit the pecking order theory less well than do unconstrained firms: (1) deviations from target leverage explain a larger fraction of the issue choice for constrained firms than unconstrained firms; (2) constrained firms have pro-cyclical leverage, but unconstrained firms have countercyclical leverage; and (3) macroeconomic conditions play a smaller role in both target leverage and issue choice for constrained firms than for unconstrained firms. Our results raise a number of interesting issues that can be addressed in future work. First, the results suggest that financing decisions reflect the state of the economy. Since information regarding security issues is contemporaneous, it may be more useful in describing economic conditions than, for example, infrequently reported information in earnings releases. Such results are of interest to policy makers as well as to the asset pricing literature. Baker and Wurgler (2000) and Kaplin and Levy (2001) find that the ratio of aggregate equity issues to total external financing predicts asset returns. Our results suggest that this is due to the covariance between the equity financing ratio and business conditions combined with the documented covariance of risk premia and business conditions (e.g., Fama and French, 1989). It would be interesting to address how macroeconomic conditions influence not only the level of debt in the capital structure, but also the maturity structure of debt chosen by firms. This maturity structure, at the aggregate level,

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

107

varies over time (Gertler and Gilchrist, 1993, 1994) and predicts asset returns (Kaplin and Levy, 2001).

References Abel, A.B., Eberly, J.C., 2002. Q theory without adjustment costs & cash flow effects without financing constraints. Unpublished working paper, Northwestern University, Evanston, IL. Ambarish, R., John, K., Williams, J., 1987. Efficient signaling with dividends and investment. Journal of Finance 42, 321–344. Asquith, P., Gertner, R., Scharfstein, D., 1994. Anatomy of financial distress: an examination of junkbond issuers. Quarterly Journal of Economics 109, 625–658. Baker, M., Wurgler, J., 2000. The equity share in new issues and aggregate stock returns. Journal of Finance 55, 2219–2257. Baker, M., Wurgler, J., 2002. Market timing and capital structure. Journal of Finance 57, 1–32. Bayless, M., Chaplinsky, S., 1991. Expectations of security type and the information content of debt and equity offers. Journal of Financial Intermediation 1, 195–214. Bayless, M., Chaplinsky, S., 1996. Is there a window of opportunity for seasoned equity issuance? Journal of Finance 51, 253–278. Bernanke, B.S., Blinder, A.S., 1992. The federal funds rate and the channels of monetary policy. American Economic Review 82, 901–921. Bernanke, B.S., Campbell, J.Y., 1988. Is there a corporate debt crisis? Brookings Papers on Economic Activity 1, 83–139. Bernanke, B.S., Gertler, M., 1995. Inside the black box: the credit channel of monetary policy transmission. Journal of Economic Perspectives 9, 27–48. Campbell, J.Y., 1991. A variance decomposition for stock returns. Economic Journal 101, 157–179. Campbell, J.Y., Lettau, M., Malkiel, B., Xu, Y., 2001. Have individual stocks become more volatile? An empirical exploration of idiosyncratic risks. Journal of Finance 56, 1–43. Campello, M., 2000. Capital structure and product markets interactions: Evidence from business cycles. Unpublished working paper, University of Illinois at Urbana-Champaign. Chaplinsky, S., Hansen, R.S., 1993. Partial anticipation, the flow of information and the economic impact of corporate debt sales. Review of Financial Studies 6, 209–732. Choe, H., Masulis, R.W., Nanda, V., 1993. Common stock offerings across the business cycle. Journal of Empirical Finance 1, 3–31. Eisfeldt, A.L., 2001. Endogenous liquidity in asset markets. Unpublished working paper #281. Northwestern University, Evanston, IL. Fama, E.F., French, K.R., 1989. Business conditions and expected returns on stocks and bonds. Journal of Financial Economics 25, 23–49. Fama, E.F., French, K.R., 2002. Testing tradeoff and pecking order predictions about dividends and debt. Review of Financial Studies 15, 1–33. Fazzari, S.M., Hubbard, R.G., Petersen, B.C., 1988. Financial constraints and corporate investment. Brookings Papers on Economic Activity 2, 141–195. Fazzari, S.M., Hubbard, R.G., Petersen, B.C., 2000. Investment-cash flow sensitivities are useful: A comment on Kaplan and Zingales. Quarterly Journal of Economics 115, 695–706. Fischer, E.O., Heinkel, R., Zechner, J., 1989. Dynamic capital structure choice: theory and tests. Journal of Finance 44, 19–40. Freund, J., 1992. Mathematical Statistics, Fifth Edition. Prentice-Hall, New Jersey. Fried, J., 2001. Open market repurchases: signaling or managerial opportunism? Unpublished working paper. Boalt Hall School of Law, U.C. Berkeley. Friedman, M., Kuttner, K., 1993. Why does the paper-bill spread predict real economic activity? In: Stock, J., Watson, M., (Eds.), Business Cycles, Indicators and Forecasting, Studies in Business Cycles, 28. University of Chicago Press for the NBER, Chicago.

108

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

Gertler, M., Gilchrist, S., 1993. The role of credit market imperfections in the monetary transmission mechanism: arguments and evidence. Scandinavian Journal of Economics 95, 43–63. Gertler, M., Gilchrist, S., 1994. Monetary policy, business cycles, and the behavior of small manufacturing firms. Quarterly Journal of Economics 109, 309–340. Gertler, M., Hubbard, R.G., 1993. Corporate financial policy, taxation and macroeconomic risk. RAND Journal of Economics 24, 286–303. Graham, J., 1996. Debt and the marginal tax rate. Journal of Financial Economics 41, 41–73. Graham, J.R., Harvey, C.R., 2001. The theory and practice of corporate finance: evidence from the field. Journal of Financial Economics 60, 187–243. Greene, W.H., 1993. Econometric Analysis, Second Edition. MacMillan Publishing Company, New York. Hansen, L.P., 1982. Large sample properties of generalized method of moments estimators. Econometrica 50, 1029–1054. Hennessy, C., 2001. Tobin’s Q, debt overhang, and investment. Unpublished working paper, Haas School of Business, U.C. Berkeley. Hovakimian, A., Opler, T., Titman, S., 2001. The debt equity choice. Journal of Financial and Quantitative Analysis 36, 1–24. Ibbotson Associates. 2001. Stocks, bonds, bills, and inflation: 2001 Yearbook. Ibbotson Associates, Chicago. Jensen, M.C., 1986. Agency costs of free cash flow, corporate finance, and takeovers. American Economic Review 76, 323–329. Jensen, M.C., Meckling, W.H., 1976. Theory of the firm: managerial behavior, agency costs, and capital structure. Journal of Financial Economics 3, 305–360. Kaplin, A., Levy, A., 2001. Corporate security issues and asset returns. Unpublished working paper. Haas School of Business, U.C. Berkeley. Kaplan, S.N., Zingales, L., 1997. Do investment-cash flow sensitivities provide useful measures of financing constraints? Quarterly Journal of Economics 112, 169–216. Kaplan, S.N., Zingales, L., 2000. Investment-cash flow sensitivities are not valid measures of financing constraints. Quarterly Journal of Economics 115, 707–712. Kashyap, A.K., Lamont, O.A., Stein, J., 1994. Credit conditions and the cyclical behavior of inventories. Quarterly Journal of Economics 109, 565–592. Kashyap, A.K., Stein, J., Wilcox, D.W., 1993. Monetary policy and bank lending. American Economic Review 83, 78–98. Kiyotaki, N., Moore, J., 1997. Credit cycles. Journal of Political Economy 105, 211–248. Korajczyk, R.A., Lucas, D., McDonald, R.L., 1990. Understanding stock price behavior around the time of equity issues. In: Hubbard, R.G. (Ed.), Asymmetric Information, Corporate Finance and Investment. NBER and University of Chicago Press, Chicago. Korajczyk, R.A., Lucas, D.J., McDonald, R.L., 1991. The effect of information releases on the price and timing of equity issues. Review of Financial Studies 4, 685–708. Korajczyk, R.A., Lucas, D.J., McDonald, R.L., 1992. Equity issues with time-varying asymmetric information. Journal of Financial and Quantitative Analysis 27, 397–417. Lamont, O., Polk, C., Sa!a-Requejo, J., 2001. Financial constraints and stock returns. Review of Financial Studies 14, 529–554. Lang, L., Ofek, E., Stultz, R., 1996. Leverage, investment, and firm growth. Journal of Financial Economics 40, 3–29. Leland, H.E., 1994. Corporate debt value, bond covenants, and optimal capital structure. Journal of Finance 49, 1213–1252. Leland, H.E., 1998. Agency costs, risk management, and capital structure. Journal of Finance 53, 1213–1243. Levy, A., 2001. Why does capital structure choice vary with macroeconomic conditions? Unpublished working paper. Haas School of Business, U.C. Berkeley. Lucas, D., McDonald, R.L., 1990. Equity issues and stock price dynamics. Journal of Finance 45, 1019–1043.

R.A. Korajczyk, A. Levy / Journal of Financial Economics 68 (2003) 75–109

109

Mackie-Mason, J., 1990. Do taxes affect corporate financing decisions? Journal of Finance 45, 1471–1495. Marsh, P., 1982. The choice between equity and debt: an empirical study. Journal of Finance 37, 121–144. Masulis, R.W., Korwar, A.N., 1986. Seasoned equity offerings. Journal of Financial Economics 15, 91–118. Myers, S.C., 1977. Determinants of corporate borrowing. Journal of Financial Economics 5, 147–175. Myers, S., Majluf, N., 1984. Corporate financing decisions when firms have information investors do not have. Journal of Financial Economics 13, 187–221. Pilotte, E., 1992. Growth opportunities and the stock price response to new financing. Journal of Business 65, 371–394. Pulvino, T.C., 1998. Do asset fire sales exist? an empirical investigation of commercial aircraft transactions. Journal of Finance 53, 939–978. Rogers, W., 1993. Regression standard errors in clustered samples. Stata Technical Bulletin 13, 88–94. Rotemberg, J.J., Saloner, G., 1986. A supergame theoretic model of price wars during booms. American Economic Review 76, 390–407. Shyam-Sunder, L., Myers, S.C., 1999. Testing static tradeoff against pecking order models of capital structure. Journal of Financial Economics 51, 219–244. Stock, J., Watson, M., 1998. Business cycle fluctuations in US macroeconomic time series, NBER working paper 6528. Suarez, J., Sussman, O., 1999. A stylized model of financially driven business cycles. Unpublished working paper 9722, Centro de Estudios Monetarios y Financieros, Madrid. Titman, S., Wessels, R., 1988. The determinants of capital structure choice. Journal of Finance 43, 1–19. Welch, I., 1989. Seasoned offerings, imitation costs, and the underpricing of initial public offerings. Journal of Finance 44, 421–449. Welch, I., 1996. Equity offerings following the IPO theory and evidence. Journal of Corporate Finance 2, 227–259. Zwiebel, J., 1996. Dynamic capital structure under managerial entrenchment. American Economic Review 86, 1197–1215.