This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and

Transcription

1 This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier s archiving and manuscript policies are encouraged to visit:

2 Journal of Financial Economics 110 (2013) Contents lists available at ScienceDirect Journal of Financial Economics journal homepage: Human capital, capital structure, and employee pay: An empirical analysis $ Thomas J. Chemmanur a,n, Yingmei Cheng b, Tianming Zhang c a Boston College, Carroll School of Management, 440 Fulton Hall, Chestnut Hill, MA 02467, USA b The Florida State University, Department of Finance, USA c The Florida State University, Department of Accounting, USA article info Article history: Received 26 June 2010 Received in revised form 28 November 2012 Accepted 4 December 2012 Available online 1 August 2013 JEL classification: G32 Keywords: Capital structure Human capital Labor costs abstract We test the predictions of Titman (1984) and Berk, Stanton, and Zechner (2010) by examining the effect of leverage on labor costs. Leverage has a significantly positive impact on cash, equity-based, and total of chief executive officers (CEOs). Compensation of new CEOs hired from outside the firm is positively related to prior-year firm leverage. In addition, leverage has a positive and significant impact on average employee pay. The incremental total labor expenses associated with an increase in leverage are large enough to offset the incremental tax benefits of debt. The empirical evidence supports the theoretical prediction that labor costs limit the use of debt. & 2013 Elsevier B.V. All rights reserved. 1. Introduction The trade-off theory of capital structure points to bankruptcy costs as the main reason that firms in many industries do not assume higher levels of leverage to take advantage of the corporate tax saving benefits of debt. For helpful comments and discussions, we thank Jonathan Berk, John Graham, Bing He, Michael Roberts, Zacharias Sautner, and participants in conference presentations at the American Accounting Association Northeast Region meeting (Best Paper Award), the Center for Research in Security Prices Forum at the University of Chicago, the Financial Management Association meetings, the Financial Management Association European meetings, the Western Finance Association meetings, and the American Finance Association meetings. Special thanks to an anonymous referee and the editor, Bill Schwert, for helpful comments and suggestions that greatly improved the paper. We also appreciate comments and suggestions from seminar participants at the University of Massachusetts at Amherst, the University of Texas at Arlington, Boston College, Florida State University, University of Florida, Qinghua University, and Zhejiang Business University. n Corresponding author. Tel.: ; fax: address: chemmanu@bc.edu (T.J. Chemmanur). However, considerable empirical evidence indicates that the magnitude of direct bankruptcy costs is too low to be a sufficient disincentive preventing firms from taking on higher levels of debt. Some authors have, therefore, suggested indirect bankruptcy costs as a solution to the puzzle of the observed underleveraging of firms in many industries. In an important paper, Titman (1984) develops a model in which a firm's liquidation decision is causally linked to its bankruptcy status. He argues that customers, workers, and suppliers of firms that produce unique or specialized products are likely to suffer high costs in the event of liquidation. In particular, in a setting where employees have firm-specific human capital, the fact that bankruptcy can impose significant costs on employees (by reducing the value of their human capital) can significantly affect firms' capital structures. 1 Formalizing the Titman (1984) arguments, Berk, Stanton, and Zechner (2010; BSZ (2010) hereafter) develop a model incorporating the idea 1 For an excellent review of empirical research on capital structure, see Parsons and Titman (2008) X/$ - see front matter & 2013 Elsevier B.V. All rights reserved.

3 T.J. Chemmanur et al. / Journal of Financial Economics 110 (2013) that human capital costs associated with financial distress and bankruptcy could be large enough to be a disincentive for firms to issue debt. The objective of this paper is to empirically analyze, for the first time in the literature, whether human capital costs are an important determinant of the capital structure of firms as postulated by the theoretical literature. We do this by examining the relation between the observed capital structures of firms and the of their chief executive officers (CEOs), as well as the relation between observed capital structures and the average wages of their work forces. While we use CEO to measure the pay of a critical employee, we use the average employee wage to measure the of a collective employee. In the model of BSZ (2010), each firm faces a risk-averse employee and risk-neutral investors. In the optimal labor contract between firms and employees, a firm with higher leverage pays a higher wage to its employee to compensate him for the expected bankruptcy costs that will be borne by the employee, because the employee is unable to fully insure his human capital risk. Firms, therefore, choose not to increase leverage beyond the point where the marginal tax benefits of debt are offset by the incremental labor costs associated with higher levels of debt. The empirical implication here is that, in the cross section, firms with higher leverage are associated with higher employee pay. 2 We test this prediction ( the Titman-BSZ prediction ) in our empirical analysis. We also study whether the magnitude of the additional associated with an increase in leverage is large enough to at least partially explain the underleveraging of firms. In contrast to the theories that focus on the ex ante relation between leverage and employee pay, Perotti and Spier (1993) focus on the ex post effect of leverage on employee pay. 3 In particular, they argue that firms are able to use leverage strategically when current profits are low and future investment is necessary to guarantee full payment of the union's claim (wages). By retiring equity through a junior debt issue, shareholders can credibly threaten not to undertake valuable new investments unless the union agrees to wage reductions. The implication of the argument is that, under suitable conditions, firms with high leverage are associated with lower employee pay. The ex post relation between leverage and employee pay implied by the model of Perotti and Spier (1993), however, is not inconsistent with the ex ante relation between the same variables in the Titman-BSZ prediction. As Perotti and Spier (1993) point out, if workers anticipate that equity holders could attempt to use higher leverage to negotiate their wages downward ex post, they will demand higher expected wages ex ante to compensate them for bearing this risk. Perotti and Spier (1993) also 2 The models of Jaggia and Thakor (1994) and Berkovitch, Israel, and Spiegel (2000) also have somewhat similar predictions. 3 Several other papers make similar arguments. See, e.g., Baldwin (1983), Bronars and Deere (1991), Perotti and Spier (1993), Dasgupta and Sengupta (1993), Hennessy and Livdan (2009), and Brown, Fee, and Thomas (2009). point out that a firm will not be able to use leverage as a bargaining tool to reduce employee wages if their profits from existing assets are large (i.e., the firm does not face a significant probability of financial distress). We make use of these results to empirically disentangle the ex ante effects suggested by the Titman-BSZ prediction from the ex post effects suggested by Perotti and Spier (1993). We accomplish this by splitting our sample between firms approaching financial distress (distressed firms) and those that do not face a significant probability of distress (safe firms). We find that the debt ratio of a firm positively affects the magnitude of its CEO. Firms with higher leverage pay their CEOs more, in terms of total, cash pay, and equity-based pay. In our ordinary least squares (OLS) regressions, an increase in market leverage by one standard deviation is associated with an increase of more than 8% in CEO total, a magnitude that is economically significant. We recognize that unobserved CEO characteristics could influence firm leverage as well as CEO pay, so that the direction of causality can be ambiguous. For example, CEOs who have had more interaction with the board (and, therefore, have more influence) could have greater ability to affect their own pay and at the same time choose the firm's leverage level. To address this issue, we study the relation between the first-year of newly appointed CEOs who are hired from outside and firm leverage in the year prior to their appointment. Clearly, newly appointed CEOs who are hired from outside should have no influence on the firm's leverage in the year prior to their appointment. We show that, even in the case of new CEOs hired from the outside, is positively related to leverage. We also find that leverage has a positive and significant impact on average employee pay. Further, the incremental labor expenses associated with an increase in leverage are large enough to offset all of the incremental tax benefits arising from such an increase. For a firm with median values of leverage, average employee pay, total labor expenses, and total debt, if the market leverage ratio increases by one standard deviation, total labor expenses increase by $14.01 million, holding the number of employees constant. Assuming 6% as the average return on debt in our sample from 1992 to 2006 and assuming a tax rate of 35%, the tax benefits of debt increase by $5.09 million, smaller than the increase in total labor expenses of $14.01 million. This supports the hypothesis that the incremental labor costs associated with an increase in leverage are economically significant and large enough in magnitude to limit the use of debt. One potential concern with our baseline analysis is the endogeneity of leverage. In particular, the assets of a given firm could be such that they can support a high level of leverage (for example, the proportion of tangible assets could be high) and could also require highly paid employees to operate these assets, thus generating a positive correlation between leverage and employee pay. To deal with this potential endogeneity problem, we employ an instrumental variable, namely, the marginal corporate tax rate, to generate an exogenous variation in leverage. The theoretical literature in corporate finance suggests

4 480 T.J. Chemmanur et al. / Journal of Financial Economics 110 (2013) that the tax benefit of debt is positively related to a firm's marginal tax rate, thus resulting in a positive correlation between a firm's marginal tax rate and its leverage ratio. The empirical literature also supports the positive relation between marginal tax rate and leverage (see, e.g., Leary and Roberts, 2010). At the same time, no theoretical or empirical literature indicates that the marginal corporate tax rate directly affects employee pay. Using the marginal corporate tax rate as the instrument, we study the relation between leverage and average employee pay in a twostage least squares (2SLS) regression framework. The results confirm that, even after accounting for the potential endogeneity of leverage, firms with a higher level of leverage are associated with a higher level of average employee pay. 4 Using the sample of manufacturing firms in the US over , Titman and Wessels (1988) find that firms with more specialized labor have lower debt ratios. Because more specialized workers are paid more, this suggests a negative relation between leverage and wages. If labor specialization is related to both leverage and employee pay, the omission of labor specialization from the regression of employee pay could cause a bias in the estimated coefficient of leverage. We address this issue by examining the quits rate, the percentage of the industry's total work force that voluntarily left their jobs in the sample years. Following Titman and Wessels (1988), we use quits rate as our proxy for labor specialization. A lower quits rate corresponds to greater labor specialization. We find that the quits rate is negatively correlated with average employee pay, consistent with the notion that more specialized workers are paid more. However, we find that the correlation between leverage and the quits rate is not statistically significant. Furthermore, in our multivariate regression of average employee pay in which the quits rate is included as an explanatory variable, the quits rate is insignificant, and the coefficient of leverage remains positive and significant. We also empirically disentangle the ex ante relation between leverage and employee pay from the ex post effects suggested by Perotti and Spier (1993). To accomplish this, we split our sample based on each firm's Altman Z-score and study safe and distressed firms separately. Consistent with the Titman-BSZ prediction, the relation between leverage and average employee pay is positive and significant in the sample of safe firms. Meanwhile, the coefficient of leverage is negative in the distressed sample, but not statistically significant. This suggests that, while the ex ante relation between leverage and employee pay suggested by Titman-BSZ prediction dominates in our entire sample and in the subsample of safe firms, in distressed firms the ex post relation postulated by Perotti and Spier (1993) could partially or fully offset the effect of firms compensating employees for their human capital risk due to higher leverage. This is not surprising, because it is precisely in distressed firms that we expect the ability 4 However, the instrumental variable that we use for leverage, the marginal tax rate, has some limitations. We discuss these limitations in Section 8. of firms to use leverage as a bargaining tool with employees to be the strongest [as pointed out by Perotti and Spier (1993)]. Labor expenses, which we use to compute average employee pay, are missing for a number of firms in the Compustat database. This creates a potential sample-selection bias if firms selectively decide whether or not to report this information. To adjust for this potential selection bias, we adopt a Heckman (1979) two-step analysis. Our results are robust to the Heckman procedure. The second stage of our Heckman two-step analysis indicates that, even after controlling for potential sample selection, leverage has a positive effectonaverageemployeepay. Employee entrenchment is an important element in the model of BSZ (2010). Entrenchment in their model means that employees are unable to fully insure their human capital risk. BSZ (2010) argue that employee entrenchment is the reason that an employee demands a higher wage from a firm with higher leverage. This allows us to conduct yet another test of the Titman-BSZ prediction. We expect to observe a stronger effect of leverage on labor costs when the employee is more entrenched. To empirically test the effect of employee entrenchment on the leveragewage relation, we examine technology versus nontechnology firms. Existing evidence (e.g., Anderson, Banker, and Ravindran, 2000) suggests that employees in nontechnology firms are more entrenched than in technology firms (in the sense that the potential reduction in employees' human capital if their firm goes bankrupt is greater). Given this, the impact of leverage on employee in nontechnology firms can be expected to be greater than in technology firms. We, therefore, split our sample between technology and nontechnology firms and conduct our analysis separately on these two subsamples. We find that the influence of leverage on the cash, equitybased, and total of CEOs is positive and significant in nontechnology firms. In technology firms, leverage affects the cash pay of CEOs, but it does not have significant effects on their total or equity-based. The leverage ratio also has a positive and significant effect on average employee pay in nontechnology firms, but not in technology firms. Thus, the effect of leverage on CEO as well as on average employee pay is greater for nontechnology firms than for technology firms, consistent with the Titman-BSZ prediction. Our paper is related to the empirical literature examining thenotionthatleveragecouldserveasabargainingtoolfor firms against labor and could thereby have a disciplining effect on labor. See, e.g., Benmelech, Bergman, and Enriquez (2009), who show that airlines in financial distress obtain wage concessions from employees whose pension plans are underfunded; Matsa (2010), who finds that firms characterized by greater union bargaining power use greater leverage; and Hanka (1998), who shows that firms using higher levels of debt reduce employment more often and use more parttime or seasonal employees. Our empirical results do not necessarily contradict those of the above cited literature. As pointed out by Perotti and Spier (1993), the disciplining effect of debt on labor is greater in firms with a significant chance of financial distress and can coexist with employees demanding greater wages ex ante (to induce them to join

5 T.J. Chemmanur et al. / Journal of Financial Economics 110 (2013) firms with greater leverage ratios). These greater wages could be required not only to compensate employees for the potential loss of their human capital in the event that the firm goes bankrupt [as suggested by Titman (1984) and BSZ (2010)], but also for the potential reduction in wages or other benefits arising from their lower bargaining power ex post if the firm enters financial distress subsequent to their joining it. The fact that the positive relation we find between leverage ratios and employee pay arises mostly from the subsample of safe firms (in which the disciplining effects of debt on the employment relation is likely to be the least), suggests that both of the effects could be operating in employee firm relations in practice. 5 Our paper contributes to the literature by showing, for the first time, that leverage has a positive impact on employee (as measured by either CEO or average employee pay) and that, at the existing median debt level, the incremental labor costs associated with an increase in leverage are sufficient to offset the incremental tax benefits of debt. Our study helps to establish the importance of labor costs in capital structure decisions and, thus, advance our understanding of the determinants of corporate leverage. Finally, ours is the first paper that explicitly analyzes the relation between executive and capital structure. While a large prior literature exists on executive as reviewed by, e.g., Frydman and Jenter (2010), to our best knowledge, no prior research has empirically analyzed the relation between executive and capital structure. The rest of this paper is organized as follows. Section 2 reviews the relevant theory in more detail and develops testable hypotheses. Section 3 describes our data and sample selection procedures. Section 4 presents our empirical analysis of the relation between capital structure and CEO. Section 5 presents our empirical analysis of the relation between capital structure and average employee pay. Section 6 compares our empirical results for technology versus nontechnology firms. Section 7 presents some additional robustness tests. Section 8 summarizes our results, discusses the limitations of our instrumental variable analysis, and concludes. 2. Development of hypotheses Titman (1984) develops a model in which a firm's liquidation decision is causally linked to its bankruptcy status. He argues that customers, workers, and suppliers of firms that produce unique or specialized products are likely to suffer high costs in the event of liquidation. In particular, in a setting where employees have firm-specific human capital, the fact that bankruptcy can impose significant costs on employees (through reducing the value 5 Our paper is also broadly related to the large literature studying the factors that could contribute to the apparent underleveraging of firms. See, e.g., Graham and Tucker (2006), who find that tax sheltering activities help to explain the low debt ratio of the firms in their sample. The literature on the role of human capital in asset pricing is also indirectly related. See, e.g., Fama and Schwert (1977). of their human capital) can significantly affect firms' capital structures. The model of BSZ (2010) formalizes the arguments of Titman (1984). In their model, each firm has only one employee, who is risk averse; investors in the firm are risk neutral. The employee is averse to bearing his own human capital risk. It is also assumed that the firm operates in competitive capital and labor markets. If the firm is in financial distress, the employee has to take a pay cut to ensure full repayment of debt. Further, if the firm is forced into bankruptcy, the employee could be terminated. Therefore, the employee faces substantial costs in the event of financial distress and bankruptcy. Because a higher debt level implies a higher probability of bankruptcy and the employee is unable to insure fully his human capital risk, firms with higher leverage have to pay, in equilibrium, a higher wage to the employee to compensate him for the expected bankruptcy costs borne by him. We make use of two measures of labor costs to test the above theories: CEO and average employee pay. CEO measures the pay of the most important employee. In the model of BSZ (2010), there is only one employee per firm. A company's CEO plays a critical role in affecting corporate performance, and his productivity is more difficult to evaluate than that of lower level employees. Therefore, the single employee in the model of BSZ (2010) can be best interpreted as the CEO in empirical tests. Average employee pay measures the of a collective employee. Because average employee pay is calculated as total labor expenses divided by the number of employees, we are able to use this measure to directly derive the marginal impact of leverage on total labor expenses and, therefore, to compare the marginal effect of debt on labor costs with the incremental tax benefits of debt. Based on the implications of the theoretical models discussed above and using the above test variables, we have the following testable hypotheses. Hypothesis 1. Firms with higher leverage will incur larger CEO. Hypothesis 2. Firms with higher leverage will incur larger average employee pay. Hypothesis 3. At the existing debt level, the additional labor costs associated with an increase in leverage are large enough to offset the incremental tax benefits of debt. Perotti and Spier (1993) argue that labor unions will bargain less aggressively and could be more willing to take pay cuts if highly levered firms run a greater risk of bankruptcy. Although their model implies that workers, ex ante, will demand a higher expected wage in for bearing the risk (Proposition IV of their paper), another empirical implication of their theory is that, ex post, a negative correlation will exist between leverage and wage when a firm faces substantial financial distress. Thus we have the following testable hypothesis. Hypothesis 4. Firms with higher leverage will incur lower average employee pay when they are in financial distress.

6 482 T.J. Chemmanur et al. / Journal of Financial Economics 110 (2013) One important element of the model of BSZ (2010) is the degree of job entrenchment. Different from the same term used in the literature on corporate governance, entrenchment in this context means the degree to which employees are able to insure their human capital risk (lower their ability to insure, greater the extent of entrenchment). Job entrenchment in this sense is the reason that the employee demands a higher pay from a firm with higher leverage in BSZ (2010). To empirically study the impact of employee job entrenchment on the leverage-wage relation, we examine technology versus nontechnology firms. Evidence suggests that employees in nontechnology firms are more entrenched compared with those in technology firms. 6 Given this, we expect leverage to have a stronger impact on labor costs in nontechnology firms than in technology firms. This yields our fifth testable hypothesis. Hypothesis 5. The effect of leverage on CEO as well as on average employee pay will be greater in nontechnology firms than in technology firms. 3. Data and summary statistics In this section, we provide details of the sample selection and the preliminary summary of the variables in the sample Sample of CEO We gather information on CEO pay from the Execu- Comp database. It provides detailed information on the of the top five executives of Standard & Poor's (S&P) 1,500 firms since We focus on the CEOs. We merge ExecuComp with the Compustat Industrial Annual database from 1992 to We delete firms with nonpositive book value of equity and exclude financial and utilities companies. A total of 17,173 firm-year observations satisfy these criteria, and 14,891 observations have all the necessary information to be included in our OLS regressions of CEO. During our sample period ( ), there are 1,952 new CEOs. To determine whether a new CEO is an outside hire, we use the following two-step procedure. First, we search for his previous employer in the ExecuComp database. If his prior employer is not the same as the current firm, then he is an outside hire. Second, if we cannot identify his previous employer in the ExecuComp database (ExecuComp reports information only on the top five executives in S&P 1,500 firms), we search the Lexis Nexis Academic Universe by the name of the executive and of the company to determine whether he is hired from 6 Anderson, Banker, and Ravindran (2000) show that the demand for executives and other critical employees in technology firms is intense, leading to higher employee turnovers than in nontechnology firms. Ittner, Lambert, and Larcker (2003), using proprietary survey data, find that technology firms rank employee retention objectives as the most important goal of their equity grant program. Overall, this evidence indicates that employees in technology firms suffer a lower loss of human capital if their firms enter financial distress compared with those in nontechnology firms. outside or promoted from inside. 7 We identify 373 outside hires using this methodology Sample of average employee pay We use information from the Compustat Industrial Annual database between 1992 and 2006 to study the impact of leverage on average employee pay. 8 We exclude financial and utilities companies, and we exclude firms with fewer than one hundred employees. We also drop firms with nonpositive book values of equity. We calculate average employee pay as total labor expenses divided by the number of employees. Compustat provides labor and related expenses (data item 42) and the number of employees (data item 29). According to the Compustat data manual, data item 42 includes salaries and wages, pension costs, payroll taxes, incentive, profit sharing, and other benefit plans. Data item 42 thus represents a firm's total labor expenses. This suits our purpose, as we need to estimate the impact of leverage on total labor costs. About 10% of firms recorded in the Compustat have valid information on data item 42. This could introduce a sample-selection bias (see Section 5). There are 5,269 firm-year observations that have the necessary information to be included in our OLS regression of average employee pay Other data sources We obtain quits rates from the database of Job Openings and Labor Turnover Survey (JOLTS) provided by the U.S. Bureau of Labor Statistics. The quits rate is the number of quits (voluntary separations) during the entire year as a percent of annual average employment. The data are available at the industry level from The industry classification is based on the North American Industry Classification System (NAICS). Appendix Table A1 reports annual quits rates by industry and year. Corporate governance could play a role in CEO, and it could also matter in determining average employee pay. 9 Therefore, we examine whether corporate governance is a factor in determining average employee pay and CEO. We use the G-Index constructed by Gompers, Ishii, and Metrick (2003) as a measure of corporate governance. They compute the G-Index using a total of 24 possible antitakeover provisions. The data source is the Investor Responsibility Research Center (IRRC) database, which provides annual information on corporate antitakeover provisions for the years 1990, 1993, 1995, 1998, 2000, 2002, 2004, 2006, and We fill in observations in the missing years using information from the most recent year. 7 Lexis-Nexis Academic Universe provides comprehensive information contained in major US and world publications (including Wall Street Journal, New York Times, Washington Post, USA Today, among many others), Securities and Exchange Commission filings, news wire services, web publications, TV and radio broadcast transcripts, major company profiles and reports, court cases, law reviews, and even blogs. 8 This ensures that our samples of CEO and employee pay cover the same time period. 9 Cronqvist, Heyman, Nilsson, Svaleryd, and Vlachos (2009) find that CEOs with more control pay their workers more.

7 T.J. Chemmanur et al. / Journal of Financial Economics 110 (2013) For example, we use information from 2004 for year A greater value of the G-Index corresponds to weaker shareholder rights and stronger managerial power. Throughout our empirical analysis of both CEO and average employee pay, all dollar amounts are adjusted to 1992 dollars using the consumer price index (CPI). 10 We use the Fama and French 48-industry classification to categorize firms into their respective industries (the classification is obtained from Kenneth French's website) Empirical tests and results on capital structure and CEO In this section, we describe our empirical tests of the impact of leverage on the magnitude of CEO. We start with OLS regressions of CEO in the whole sample. We then perform additional tests to identify causality. To accomplish this, we examine the impact of leverage in the prior year on the of newly appointed CEOs who are hired from outside Summary statistics In Table 1, we present summary statistics for the variables used in our analysis of CEO. ExecuComp provides two measures of total : one includes the value of the options granted, and the other includes the value of options exercised. We use the total including the value of options exercised in our analysis. The results remain qualitatively the same when the value of options granted is considered. Cash is the sum of salary and bonus, as provided by ExecuComp. We compute equity-based as the total minus salary, bonus, other annual pay, and LTIP (long-term incentive plan). The most common forms of equity-based are stock options and restricted stocks. Market capitalization is computed as the stock price multiplied by the number of shares outstanding at the end of a fiscal year. Market-tobook ratio is the market capitalization divided by the book value of equity. All continuous variables except leverage are winsorized at the 1st and 99th percentiles. 12 Leverage is the variable of interest. We measure leverage in four ways. The market leverage, as used widely in the literature (e.g., Leary and Roberts, 2010), is computed as the total debt divided by the sum of total debt and market value of equity. The book leverage, also used commonly in the literature, is computed as the total debt divided by the sum of total debt and book value of equity. Total debt is the sum of long-term debt and debt in current liabilities (data item 9 plus data item 34). Debt in current liabilities (data item 34) includes notes payable (data item 10 CPI data are taken from the website of the Bureau of Labor Statistics: 11 French's website is ken.french/data_library.html. 12 Another way to identify outliers is by employing the Hadi (1992, 1994) procedure. The exclusion of outliers does not affect the results of our multivariate analysis. 206) and debt due in 1 year (data item 44). Welch (2011) argues that the liabilities that are nonfinancial debt should not be included in the computation of leverage ratio. We follow Welch (2011) and introduce two additional measures of leverage, which we refer to as alternative market leverage and alternative book leverage. We calculate alternative market leverage as (total long-term debt+debt due in one year)/(total long-term debt+debt due in one year+market value of equity) and calculate alternative book leverage as (total long-term debt+debt due in one year)/(total long-term debt+debt due in one year+book value of equity). 13 Due to space limitations, we report results only from our analysis using market leverage, alternative market leverage, and alternative book leverage. Results from our analysis using book leverage are available upon request. CEOs' cash (salary plus bonus) has a mean of $972,330 and a median of $736,490, with a 1% cutoff of $109,090 and 99% cutoff of $4.531 million. The equity-based has a larger mean but a smaller median than the cash. The reason is that equity-based pay has a wider range across firms than cash pay, and some CEOs have extremely large equity-based pay. For example, the 99% cutoff of equitybased is about $27 million, while the 1% cutoff is only $12,500. We use the natural log of the variables in our multivariate regression of CEO to reduce the potential impact of outliers. The one-year return to shareholders (including dividends), a measure of firm performance, has a median of 10.33%. Turning to CEO characteristics, the median CEO age is 65, and the median length of CEO tenure is four years. Only 2% of the CEOs in our sample are female, and 64% of the CEOs also serve as chairman of the board. The G-Index has a mean of 9.26 and a median of OLS regressions In our reduced form analysis, we model CEO as: CEOPay i;t ¼ γ 0 þ γ 1 Size i;t þ γ 2 Leverage i;t þ γ 3 MTB i;t þγ 4 RET i;t þ γ 5 Age i;t þ γ 6 Tenure i;t þ γ 7 Chair i;t þγ 8 MALE i;t þ ε i;t ; CEOPay i,t is the CEO of firm i in year t, and it is measured in three ways: cash, equity-based, and total. Size i,t is the natural log of market capitalization of firm i as of year t. We expect Size i,t to be a positive and significant determinant of CEO. As Murphy (1999) points out, the best-documented stylized fact regarding CEO pay is that CEO pay is higher in larger firms. Leverage i,t is the leverage ratio of firm i as of year t. If firms with higher leverage pay a higher wage to their CEOs, γ 2 is positive. MTB i,t is the market-to-book ratio of firm i as of year t, which is used as a proxy for firms' 13 We thank an anonymous referee for bringing Welch (2011) to our attention and for suggesting that we also report our analysis using these two alternative measures of market and book leverage. ð1þ

8 484 T.J. Chemmanur et al. / Journal of Financial Economics 110 (2013) Table 1 Summary statistics of variables used in the analysis of CEO. This table summarizes the variables used in the analysis of CEO. Cash (salary plus bonus) and total s are provided by Execucomp. We compute equity-based as the total minus salary, bonus, other annual pay, and long-term incentive plan. Market leverage is computed as (debts in current liabilities+long-term debt)/(debts in current liabilities+long-term debt+market value of equity). Alternative market leverage is computed as (debts due in one year+long-term debt)/(debts due in one year+long-term debt+market value of equity). Alternative book leverage is computed as (debts due in one year+long-term debt)/(debts due in one year+long-term debt+book value of equity). Market capitalization is computed as the stock price multiplied by the number of shares outstanding as of the end of a fiscal year. Market-to-book ratio is the market capitalization divided by the book value of equity. All continuous variables except leverage are winsorized at the 1st and 99th percentiles. All dollar amounts are adjusted to 1992 dollars using the consumer price index. The G-Index was constructed by Gompers, Ishii, and Metrick (2003) as a measure of corporate governance. N Mean Median Standard deviation 1% Cutoff 99% Cutoff Cash (salary+bonus) (thousands) 14, ,531 Equity-based (thousands) 14,891 1, , ,656 Total including options exercised (thousands) 14,891 2, , , ,099 Market leverage 14, Alternative market leverage 14, Alternative book leverage 14, Market capitalization (millions) 14,891 4, , ,204 Market-to-book ratio 14, One-year return to shareholders (%) 14, CEO age 14, CEO tenure (years as CEO in the firm) 14, CEO is male 14, CEO is also the chairman 14, G-Index 11, growth opportunities. RET i,t is the return to shareholders of firm i in year t, a popular measure of the performance of firm i in year t. The existing literature shows a positive relation between CEO pay and firm performance. 14 Hence, we expect γ 4 to be positive. In addition, we control for individual CEO characteristics that could affect CEO. Age i,t is the age of the CEO of firm i as of year t; Tenure i,t is the number of years the executive has acted as the CEO in firm i prior to year t; Chair i,t is one if the CEO is also the chairman and zero otherwise; and MALE i,t is one if the CEO is male and zero otherwise. We include year dummies to control for time-specific variation in CEO pay. As shown by the literature, CEO has increased tremendously during the past few decades. We include industry dummies due to the significant variation in CEO pay across industries. In Table 2, we report the estimated coefficients and standard errors obtained from the OLS regression of Eq. (1). The standard errors are clustered by firm. Estimation results from using market leverage, alternative market leverage, and alternative book leverage are reported in Panel A, Panel B, and Panel C, respectively. Columns 1 3 in each panel exclude the G-Index, and columns 4 6 in each panel include the G-Index. Including the G-Index in the regression reduces the sample size. Firm size has a positive impact on all three measures of CEO. A larger firm pays its CEO, on average, more than a smaller firm does, which is consistent with the literature. A higher one-year return to shareholders is associated with greater CEO pay. This is consistent with the positive relation between CEO pay and firm performance as shown by the literature. 14 Murphy (1999) provides a comprehensive review of the relation between firm performance and CEO. On average, an older CEO earns a larger pay. Being the chairman has a positive and significant effect on CEO. Gender does not have a significant effect on CEO pay. The coefficient on CEO tenure is not significant in the regression of total and cash, but it is negative and significant (at the 5% level) in that of equitybased. Market-to-book ratio is not significant in the regressions of total, but it is negative in the regression of cash pay and is positive in that of equitybased. This suggests that growth firms pay less cash but more stock-based to their CEOs than value firms. The leverage ratio has a positive and significant effect on cash, equity-based, and total s. According to Column 1 of Panel A, if market leverage goes up by o1 standard deviation (0.19, as reported in Table 1), the natural log of CEO total increases by ¼0.080, which translates to more than 8.3% increase in total pay. Therefore, starting at the median total CEO of $1.20 million, the total CEO pay increases by about $100,000, an economically significant amount. If market leverage increases by 1 standard deviation (0.19), the CEO's cash pay goes up by more than 12% and the CEO's equity-based pay goes up by more than 8%. In summary, the effect of leverage on CEO is economically as well as statistically significant. The G-Index is a positive and significant factor in determining CEOs' cash, equity-based, and total pay, suggesting that stronger managerial power is associated with greater CEO. Leverage continues to have a positive and significant effect on CEO in the presence of the G-Index. We also estimate Eq. (1) by year, in the spirit of Fama and Macbeth (1973). Table 3 reports the coefficient of leverage in the regression of CEO for every year between 1992 and The coefficients of all three

9 T.J. Chemmanur et al. / Journal of Financial Economics 110 (2013) Table 2 Ordinary least square regressions of CEO. This table reports the coefficients and standard errors obtained from OLS estimation of the following model: CEOPay i;t ¼ γ 0 þ γ 1 Size i;t þ γ 2 Leverage i;t þ γ 3 MTB i;t þ γ 4 RET i;t þ γ 5 Age i;t þγ 6 Tenure i;t þ γ 7 Chair i;t þ γ 8 MALE i;t þ ε i;t ; CEOPay i,t is measured in three ways: the log of CEO total, the log of CEO cash (salary plus bonus), and the log of equity-based. Size i,t is the log of market capitalization of firm i as of year t. Leverage i;t is the leverage ratio of firm i as of year t. Market leverage, alternative market leverage, and alternative book leverage are defined the same as in Table 1. MTB i;t is the market-to-book ratio of firm i as of year t. RET i,t is the return to the shareholders of firm i in year t. Age i,t is the age of the CEO of firm i as of year t; Tenure i,t is the number of years the executive has been acting as CEO in firm i prior to year t; Chair i,t is one if the CEO is also the chairman and zero otherwise; and MALE i,t is one if the CEO of firm i as of year t is male and zero otherwise. Numbers in the parentheses are the standard errors. The standard errors are clustered by firm and are also robust to heteroskedasticity. Regressions in Panel A use market leverage, regressions in Panel B use alternative market leverage, and regressions in Panel C use alternative book leverage. nnn, nn, and n indicate significance at the 1%, 5%, and 10% level, respectively. The G-Index was constructed by Gompers, Ishii, and Metrick (2003) as a measure of corporate governance. Total (1) Cash (2) Equity-based (3) Total (4) Cash (5) Equity-based (6) Panel A: Market leverage Market leverage 0.42 nnn 0.59 nnn 0.42 nn 0.43 nnn 0.55 nnn 0.45 nn (0.07) (0.05) (0.18) (0.07) (0.06) (0.20) Firm size 0.41 nnn 0.29 nnn 0.66 nnn 0.41 nnn 0.28 nnn 0.66 nnn (0.01) (0.01) (0.03) (0.01) (0.01) (0.03) Market-to-book ratio nnn nn nn 0.01 nn 0.02 nn (0.005) (0.003) (0.010) (0.005) (0.004) (0.01) One-year return to nnn nnn nnn nnn nnn nnn shareholders (0.0002) (0.0001) (0.0004) (0.0002) (0.0001) (0.001) CEO age nnn nnn nnn nnn nn nnn (0.002) (0.001) (0.005) (0.002) (0.002) (0.005) CEO tenure nn (0.003) (0.002) (0.006) (0.003) (0.002) (0.01) CEO is also the 0.18 nnn 0.14 nnn 0.24 nnn 0.18 nnn 0.14 nnn 0.24 nnn chairman (0.03) (0.02) (0.07) (0.03) (0.02) (0.07) CEO is male (0.13) (0.08) (0.31) (0.13) (0.09) (0.31) G-Index 0.02 nnn 0.02 nnn 0.05 nnn (0.005) (0.004) (0.01) Year effects Yes Yes Yes Yes Yes Yes Industry effects Yes Yes Yes Yes Yes Yes Intercept 3.15 nnn 3.73 nnn 1.57 nn 2.91 nnn 3.68 nnn 1.86 nn (0.23) (0.21) (0.62) (0.31) (0.30) (0.74) Number of observations 14,891 14,891 14,891 11,527 11,527 11,527 R-squared Panel B: Alternative market leverage Alternative market leverage 0.41 nnn 0.58 nnn 0.40 nn 0.43 nnn 0.54 nnn 0.45 nn (0.07) (0.05) (0.18) (0.08) (0.06) (0.20) Firm size 0.41 nnn 0.29 nnn 0.66 nnn 0.41 nnn 0.28 nnn 0.66 nnn (0.01) (0.01) (0.03) (0.01) (0.01) (0.03) Market-to-book ratio nnn nn nn 0.01 nnn 0.02 nn (0.005) (0.003) (0.010) (0.005) (0.003) (0.01) One-year return to shareholders (%) nnn nnn nnn nnn nnn nnn (0.0002) (0.0001) (0.0004) (0.0002) (0.0001) (0.001) CEO age nnn nnn nnn nnn nn nnn (0.002) (0.001) (0.005) (0.002) (0.002) (0.005) CEO tenure nn (0.003) (0.002) (0.006) (0.003) (0.002) (0.01) CEO is also the chairman 0.18 nnn 0.15 nnn 0.24 nnn 0.18 nnn 0.14 nnn 0.25 nnn (0.03) (0.02) (0.07) (0.03) (0.02) (0.07) CEO is male (0.13) (0.08) (0.31) (0.13) (0.09) (0.31) G-Index 0.02 nnn 0.02 nnn 0.05 nnn (0.005) (0.004) (0.01) Year effects Yes Yes Yes Yes Yes Yes Industry effects Yes Yes Yes Yes Yes Yes Intercept 3.16 nnn 3.74 nnn 1.56 nn 2.92 nnn 3.69 nnn 1.85 nn

10 486 T.J. Chemmanur et al. / Journal of Financial Economics 110 (2013) Table 2 (continued ) Panel B: Alternative market leverage (0.23) (0.21) (0.62) (0.31) (0.30) (0.74) Number of observations 14,891 14,891 14,891 11,527 11,527 11,527 R-squared Panel C: Alternative book leverage Alternative book leverage 0.25 nnn 0.49 nnn 0.29 nn 0.26 nnn 0.44 nnn 0.34 nn (0.06) (0.04) (0.15) (0.07) (0.05) (0.17) Firm size 0.40 nnn 0.28 nnn 0.65 nnn 0.40 nnn 0.27 nnn 0.65 nnn (0.01) (0.01) (0.02) (0.01) (0.01) (0.03) Market-to-book ratio nnn nnn 0.01 (0.005) (0.003) (0.01) (0.006) (0.004) (0.01) One-year return to shareholders (%) nnn nnn nnn nnn nnn nnn (0.0002) (0.0001) (0.0004) (0.0002) (0.0001) (0.0005) CEO age nnn nnn nnn nnn nn 0.02 nnn (0.002) (0.001) (0.005) (0.002) (0.002) (0.005) CEO tenure nn (0.002) (0.002) (0.006) (0.003) (0.002) (0.01) CEO is also the chairman 0.18 nnn 0.14 nnn 0.24 nnn 0.18 nnn 0.14 nnn 0.24 nnn (0.03) (0.02) (0.07) (0.03) (0.02) (0.07) CEO is male (0.13) (0.08) (0.31) (0.13) (0.09) (0.31) G-Index 0.02 nnn 0.02 nnn 0.05 nnn (0.005) (0.004) (0.01) Year effects Yes Yes Yes Yes Yes Yes Industry effects Yes Yes Yes Yes Yes Yes Intercept 3.23 nnn 3.83 nnn 1.49 nn 3.02 nnn 3.80 nnn 1.76 nn (0.22) (0.20) (0.62) (0.29) (0.29) (0.74) Number of observations 14,891 14,891 14,891 11,527 11,527 11,527 R-squared measures of leverage in the regression of CEOs' total pay and cash pay are positive in all of the 15 years. The coefficient of alternative book leverage is positive in the regression of CEOs' equity-based pay in 13 out of 15 years, and the other two measures of leverage have a positive coefficient in the regression of CEOs' equity-based pay in 14 out of 15 years New CEOs hired from outside Some unobservable and thus uncontrolled CEO characteristics could affect both leverage and in the same direction, thus resulting in the positive coefficient of leverage in the OLS regression of CEO. For example, CEOs who have had more interaction with the board (and, therefore, have more influence) could have greater ability to affect their own pay and at the same time choose the firm's leverage level. To address potential concerns regarding causality, we study the subset of newly appointed CEOs who are hired from outside. We examine how the first-year of these new CEOs is affected by firm leverage in the year prior to their appointment. CEOs hired from outside should have no influence on their firms' capital structure in the year prior to their appointment, so that this is a clean test of the relation between leverage and CEO, allowing us to deal with the potential causality problem. We model the relation between the first-year of newly appointed CEOs hired from outside and the leverage ratio in the year prior to their appointment as: CEOPay i;t ¼ γ 0 þ γ 1 Size i;t 1 þ γ 2 Leverage i;t 1 þ γ 3 MTB i;t 1 þγ 4 RET i;t þ γ 5 Age i;t þ γ 6 Chair i;t þ γ 7 MALE i;t þ ε i;t : ð2þ In Eq. (2), firm size, leverage, and market-to-book ratio are computed as of the fiscal year prior to the appointment of the new CEO. CEO tenure is omitted from Eq. (2), because we estimate Eq. (2) on the sample of newly appointed CEOs hired from outside (all of them have zero tenure, by definition). Titman (1984) and BSZ (2010) predict that a firm with higher leverage will pay its employees more. In the case of a newly hired CEO, he will demand and obtain a higher pay from a firm with higher leverage. Therefore, we expect γ 2 to be positive. In Table 4, we present the coefficients and standard errors obtained from estimating Eq. (2) on the subset of newly appointed CEOs who are hired from outside. Firm size is a strong factor in determining the pay of newly appointed CEOs (cash, equity-based, and total ). The coefficient on the market-to-book ratio is negative and significant in all three types of, suggesting that growth firms pay their new CEOs less than value firms do. The coefficient of stock return during the first year of a new CEO is positive and significant in the regression of equity-based. CEO age has a negative effect on equity-based, different from what is in Table 2. This is due to the differences between the underlying samples. In Table 2, the same CEO in the same firm appears in multiple years, and the pay