Human Capital, Capital Structure, and Employee Pay: An Empirical Analysis

Transcription

1 Human Capial, Capial Srucure, and Employee Pay: An Empirical Analysis Thomas J. Chemmanur a, * Yingmei Cheng b Tianming Zhang c a Boson College, Carroll School of Managemen, USA b The Florida Sae Universiy, Deparmen of Finance, USA c The Florida Sae Universiy, Deparmen of Accouning, USA ABSTRACT We es he predicions of Timan (1984) and Berk, Sanon, and Zechner (2010) by examining he effec of leverage on labor coss. Leverage has a significanly posiive impac on CEOs cash, equiy-based, and oal compensaion. Compensaion of new CEOs hired from ouside he firm is posiively relaed o prior-year firm leverage. In addiion, leverage has a posiive and significan impac on average employee pay. The incremenal oal labor expenses associaed wih an increase in leverage are large enough o offse he incremenal ax benefis of deb. The empirical evidence suppors he heoreical predicion ha labor coss limi use of deb. Keywords: Capial srucure; Human capial; Labor coss JEL classificaion: G32 * Corresponding auhor. Professor of Finance, Carroll School of Managemen, Boson College, 440 Fulon Hall, Chesnu Hill, MA chemmanu@bc.edu. Phone: (617) Fax: (617) For helpful commens and discussions, we hank Jonanhan Berk, John Graham, Bing He, Michael Robers, Zacharias Sauner and paricipans in conference presenaions a he AAA Norheas Region Meeing (Bes Paper Award), he CRSP Forum a he Universiy of Chicago, he Financial Managemen Associaion Meeings, he Financial Managemen Associaion European Meeings, he Wesern Finance Associaion Meeings, and he American Finance Associaion Meeings. Special hanks o an anonymous referee and he edior, Bill Schwer, for helpful commens and suggesions ha grealy improved he paper. We also appreciae commens and suggesions from seminar paricipans a he Universiy of Massachuses a Amhers, he Universiy of Texas a Arlingon, Boson College, The Florida Sae Universiy, Universiy of Florida, Qinghua Universiy, and Zhejiang Business Universiy.

2 1. Inroducion The radeoff heory of capial srucure poins o bankrupcy coss as he main reason why firms in many indusries do no ake on higher levels of leverage o ake advanage of he corporae ax saving benefis of deb. However, here is considerable empirical evidence indicaing ha he magniude of direc bankrupcy coss is oo low o be a sufficien disincenive prevening firms from aking on higher levels of deb. Some auhors have herefore suggesed indirec bankrupcy coss as a soluion o he puzzle of he observed underleveraging of firms in many indusries. In an imporan paper, Timan (1984) develops a model in which a firm s liquidaion decision is causally linked o is bankrupcy saus. He argues ha cusomers, workers, and suppliers of firms ha produce unique or specialized producs are likely o suffer high coss in he even of liquidaion. In paricular, in a seing where employees have firm-specific human capial, he fac ha bankrupcy can impose significan coss on employees (by reducing he value of heir human capial) can significanly affec firms capial srucures. 1 Formalizing Timan s (1984) argumens, Berk, Sanon, and Zechner (2010) develop a model incorporaing he idea ha human capial coss associaed wih financial disress and bankrupcy may be large enough o be a disincenive for firms o issue deb. The objecive of his paper is o empirically analyze, for he firs ime in he lieraure, wheher human capial coss are indeed an imporan deerminan of he capial srucure of firms as posulaed by he above heoreical lieraure. We do his by examining he relaion beween he observed capial srucures of firms and he compensaion of heir CEOs, as well as he relaion beween observed capial srucures and he average wages of heir work forces: while we use CEO compensaion o measure he pay of a criical employee, we use he average employee wage o measure he compensaion of a collecive employee. In he model of Berk, Sanon, and Zechner (2010), referred o as BSZ (2010) from now on, each firm faces a risk-averse employee and riskneural invesors. In he opimal labor conrac beween firms and employees, a firm wih higher leverage pays a higher wage o is employee o compensae him for he expeced bankrupcy coss 1 For an excellen review of empirical research on capial srucure, see Parsons and Timan (2008).

3 ha will be borne by he employee, since he employee is unable o fully insure his human capial risk. Firms will herefore choose no o increase leverage beyond he poin where he marginal ax benefis of deb are offse by he incremenal labor coss associaed wih higher levels of deb. The empirical implicaion here is ha, in he cross secion, firms wih higher leverage will be associaed wih higher employee pay. 2 We will es his predicion ( he Timan-BSZ predicion from now on) in our empirical analysis. We will also sudy wheher he magniude of he addiional compensaion associaed wih an increase in leverage is large enough o a leas parially explain he underleveraging of firms. In conras o he above heories ha focus on he ex ane relaionship beween leverage and employee pay, Peroi and Spier (1993) focus on he ex pos effec of leverage on employee pay. 3 In paricular, hey argue ha firms are able o use leverage sraegically when curren profis are low and fuure invesmen is necessary o guaranee full paymen of he union s claim (wages). By reiring equiy hrough a junior deb issue, shareholders can credibly hreaen no o underake valuable new invesmens unless he union agrees o wage reducions. The implicaion of he above argumen is ha, under suiable condiions, firms wih high leverage will be associaed wih lower employee pay. The above ex pos relaionship beween leverage and employee pay implied by he model of Peroi and Spier (1993), however, is no inconsisen wih he ex ane relaionship beween he same variables in he Timan-BSZ predicion. As Peroi and Spier (1993) poin ou in heir paper, if workers anicipae ha equiy holders may aemp o use higher leverage o negoiae heir wages downward ex pos, hey will demand higher expeced wages ex ane o compensae hem for bearing his risk. Peroi and Spier (1993) also poin ou ha a firm will no be able o use leverage as a bargaining ool o reduce employee wages if heir profis from exising asses are large (i.e., he firm 2 The models of Jaggia and Thakor (1994) and Berkovich, Israel, and Spiegel (2000) also have somewha similar predicions. 3 Several oher papers make similar argumens: see, e.g., Baldwin (1983), Bronars and Deere (1991), Peroi and Spier (1993), Dasgupa and Sengupa (1993), Hennessy and Livdan (2009), and Brown, Fee, and Thomas (2009). 2

4 does no face a significan probabiliy of financial disress). We make use of he laer resuls from Peroi and Spier (1993) o empirically disenangle he ex ane effecs suggesed by he Timan-BSZ predicion from he ex pos effecs suggesed by Peroi and Spier (1993). We accomplish his by spliing our sample beween firms approaching financial-disress ( disressed firms) and hose ha do no face a significan probabiliy of disress ( safe firms). We find ha he deb raio of a firm posiively affecs he magniude of is CEO compensaion. Firms wih higher leverage pay heir CEOs more, in erms of oal compensaion, cash pay, and equiy-based pay. In our OLS regressions, an increase in marke leverage by one sandard deviaion is associaed wih an increase of more han 8% in CEO oal compensaion, a magniude ha is economically significan. We recognize ha unobserved CEO characerisics may influence firm leverage as well as CEO pay, so ha he direcion of causaliy may be ambiguous. For example, CEOs who have had more ineracion wih he board (and herefore have more influence) may have greaer abiliy o affec heir own pay and a he same ime choose he firm s leverage level as well. In order o address his issue, we sudy he relaionship beween he firs-year compensaion of newly appoined CEOs who are hired from ouside and firm leverage in he year prior o heir appoinmen. Clearly, newly appoined CEOs who are hired from ouside should have no influence on he firm s leverage in he year prior o heir appoinmen. We show ha, even in he case of new CEOs hired from he ouside, compensaion is posiively relaed o leverage. We also find ha leverage has a posiive and significan impac on average employee pay. Furher, he incremenal labor expenses associaed wih an increase in leverage are large enough o offse all of he incremenal ax benefis arising from such an increase. For a firm wih median values of leverage, average employee pay, oal labor expenses, and oal deb, if he marke leverage raio increases by one sandard deviaion, oal labor expenses increase by $14.01 million, holding he number of employees consan. Assuming 6% as he average reurn on deb in our sample from and assuming a ax rae of 35%, he ax benefis of deb increase by $5.09 million, smaller han he increase in oal labor expenses of $14.01 million. This suppors he 3

5 hypohesis ha he incremenal labor coss associaed wih an increase in leverage are economically significan and large enough in magniude o limi he use of deb. One poenial concern wih our baseline analysis is he endogeneiy of leverage. In paricular, he asses of a given firm may be such ha hey can suppor a high level of leverage (for example, he proporion of angible asses may be high) and may also require highly paid employees o operae hese asses, hus generaing a posiive correlaion beween leverage and employee pay. To deal wih his poenial endogeneiy problem, we employ an insrumenal variable, namely, he marginal corporae ax rae, o generae an exogenous variaion in leverage: he heoreical lieraure in corporae finance suggess ha he ax benefi of deb will be posiively relaed o a firm s marginal ax rae, hus resuling in a posiive correlaion beween a firm s marginal ax rae and is leverage raio. The empirical lieraure also suppors he above view (see, e.g., Leary and Robers, 2010). A he same ime, here is no heoreical or empirical lieraure indicaing ha he marginal corporae ax rae will direcly affec employee pay. Using he marginal corporae ax rae as he insrumen, we sudy he relaionship beween leverage and average employee pay in a wo-sage leas square (2SLS) regression framework. The resuls of he above analysis confirm ha, even afer accouning for he poenial endogeneiy of leverage, firms wih a higher level of leverage are associaed wih a higher level of average employee pay. 4 Using he sample of manufacuring firms in he US over , Timan and Wessels (1988) find ha firms wih more specialized labor have lower deb raios. Since more specialized workers are paid more, his suggess a negaive relaion beween leverage and wages. If labor specializaion is relaed o boh leverage and employee pay, he omission of labor specializaion from he regression of employee pay may cause a bias in he esimaed coefficien of leverage. We address his issue by examining he quis rae, he percenage of he indusry s oal work force ha volunarily lef heir jobs in he sample years. Following Timan and Wessels (1988), we use quis 4 However, he insrumenal variable ha we use for leverage, he marginal ax rae, has some limiaions. We discuss hese limiaions in he concluding secion of his paper (Secion 8). 4

6 rae as our proxy for labor specializaion: a lower quis rae corresponds o greaer labor specializaion. We find ha he quis rae is negaively correlaed wih average employee pay, consisen wih he noion ha more specialized workers are paid more. However, we find ha he correlaion beween leverage and he quis rae is no saisically significan. Furhermore, in our mulivariae regression of average employee pay where he quis rae is included as an explanaory variable, he quis rae is insignifican, and he coefficien of leverage remains posiive and significan. As discussed earlier, we also empirically disenangle he ex ane relaionship beween leverage and employee pay from he ex pos effecs suggesed by Peroi and Spier (1993). To accomplish his, we spli our sample based on each firm s Alman s Z-score and sudy safe and disressed firms separaely. Consisen wih he Timan-BSZ predicion, he relaionship beween leverage and average employee pay is posiive and significan in he sample of safe firms. On he oher hand, he coefficien of leverage is negaive in he disressed sample, bu no saisically significan. This suggess ha, while he ex ane relaionship beween leverage and employee pay suggesed by Timan-BSZ predicion dominaes in our enire sample and in he subsample of safe firms, in disressed firms he ex pos relaionship posulaed by Peroi and Spier (1993) may parially or fully offse he above effec of firms compensaing employees for heir human capial risk due o higher leverage. This is no surprising, since i is precisely in disressed firms ha we expec he abiliy of firms o use leverage as a bargaining ool wih employees o be he sronges (as poined ou by Peroi and Spier, 1993). Labor expenses, which we use o compue average employee pay, are missing for a number of firms in he COMPUSTAT daabase. This creaes a poenial sample-selecion bias if firms selecively decide wheher or no o repor his informaion. To adjus for his poenial selecion bias, we adop a Heckman (1979) wo-sep analysis. Our resuls are robus o he Heckman procedure: he second sage of our Heckman wo-sep analysis indicaes ha, even afer conrolling for poenial sample selecion, leverage has a posiive effec on average employee pay. 5

7 Employee enrenchmen is an imporan elemen in he model of BSZ (2010). Enrenchmen in heir model means ha employees are unable o fully insure heir human capial risk. BSZ (2010) argue ha employee enrenchmen is he reason why an employee demands a higher wage from a firm wih higher leverage. This allows us o conduc ye anoher es of he Timan-BSZ predicion: we expec o observe a sronger effec of leverage on labor coss when he employee is more enrenched. To empirically es he effec of employee enrenchmen on he leverage-wage relaion, we examine echnology versus non-echnology firms. Exising evidence (e.g., Anderson, Banker, and Ravindran, 2000) suggess ha employees in non-echnology firms are more enrenched han in echnology firms (in he sense ha he poenial reducion in employees human capial if heir firm goes bankrup is greaer). Given his, he impac of leverage on employee compensaion in nonechnology firms may be expeced o be greaer han in echnology firms. We herefore spli our sample beween echnology and non-echnology firms and conduc our analysis separaely on hese wo subsamples. We find ha he influence of leverage on he cash, equiy-based, and oal compensaion of CEOs is posiive and significan in non-echnology firms. In echnology firms, leverage affecs he cash pay of CEOs, bu i does no have significan effecs on heir oal or equiy-based compensaion. The leverage raio also has a posiive and significan effec on average employee pay in non-echnology firms, bu no in echnology firms. Thus, he effec of leverage on CEO compensaion as well as on average employee pay is greaer for non-echnology firms han for echnology firms, consisen wih he Timan-BSZ predicion. Our paper is relaed o he empirical lieraure examining he noion ha leverage may serve as a bargaining ool for firms agains labor and may hereby have a disciplining effec on labor: see, e.g., Benmelech, Bergman, and Enriquez (2009), who show ha airlines in financial disress obain wage concessions from employees whose pension plans are underfunded; Masa (2010), who documens ha firms characerized by greaer union bargaining power use greaer leverage; and Hanka (1998), who documens ha firms using higher levels of deb reduce employmen more ofen 6

8 and use more par ime or seasonal employees. Our empirical resuls do no necessarily conradic hose of he above lieraure: as poined ou by Peroi and Spier (1993), he disciplining effec of deb on labor will be greaer in firms wih a significan chance of financial disress, and can co-exis wih employees demanding greaer wages ex ane (o induce hem o join firms wih greaer leverage raios). These greaer wages may be required no only o compensae employees for he poenial loss of heir human capial in he even ha he firm goes bankrup (as suggesed by Timan (1984) and BSZ (2010)), bu also for he poenial reducion in wages or oher benefis arising from heir lower bargaining power ex pos if he firm eners financial disress subsequen o heir joining i. The fac ha he posiive relaionship we documen beween leverage raios and employee pay arises mosly from he subsample of safe firms (where he disciplining effecs of deb on he employmen relaionship is likely o be he leas), suggess ha boh he above effecs may be operaing in employee-firm relaionships in pracice. 5 Our paper conribues o he lieraure by showing, for he firs ime, ha leverage has a posiive impac on employee compensaion (as measured by eiher CEO compensaion or average employee pay), and ha, a he exising median deb level, he incremenal labor coss associaed wih an increase in leverage are sufficien o offse he incremenal ax benefis of deb. Our sudy helps o esablish he imporance of labor coss in capial srucure decisions, and hus advance our undersanding of he deerminans of corporae leverage. Finally, ours is he firs paper ha explicily analyzes he relaion beween execuive compensaion and capial srucure. While here is a large prior lieraure on execuive compensaion as reviewed by, e.g., Frydman and Jener (2010), o our bes knowledge, no prior research has empirically analyzed he relaion beween execuive compensaion and capial srucure. 5 Our paper is also broadly relaed o he large lieraure sudying he facors ha may conribue o he apparen underleveraging of firms: see, e.g., Graham and Tucker (2006), who find ha ax shelering aciviies help o explain he low deb raio of he firms in heir sample. The lieraure on he role of human capial in asse pricing is also indirecly relaed: see, e.g., Fama and Schwer (1977). 7

9 The res of his paper is organized as follows. Secion 2 reviews he relevan heory in more deail and develops esable hypoheses. Secion 3 describes our daa and sample selecion procedures. Secion 4 presens our empirical analysis of he relaion beween capial srucure and CEO compensaion. Secion 5 presens our empirical analysis of he relaion beween capial srucure and average employee pay. Secion 6 compares our empirical resuls for echnology versus non-echnology firms. Secion 7 presens some addiional robusness ess. Secion 8 summarizes our resuls, discusses he limiaions of our insrumenal variable analysis, and concludes. 2. Developmen of Hypoheses Timan (1984) develops a model in which a firm s liquidaion decision is causally linked o is bankrupcy saus. He argues ha, cusomers, workers, and suppliers of firms ha produce unique or specialized producs are likely o suffer high coss in he even of liquidaion. In paricular, in a seing where employees have firm-specific human capial, he fac ha bankrupcy can impose significan coss on employees (hrough reducing he value of heir human capial) can significanly affec firms capial srucures. The model of BSZ (2010) formalizes he above argumens of Timan (1984). In heir model, each firm has only one employee, who is risk averse; invesors in he firm are risk neural. The employee is averse o bearing his own human capial risk. I is also assumed ha he firm operaes in compeiive capial and labor markes. If he firm is in financial disress, he employee has o ake a pay cu o ensure full repaymen of deb. Furher, if he firm is forced ino bankrupcy, he employee may be erminaed. Therefore, he employee faces subsanial coss in he even of financial disress and bankrupcy. Because a higher deb level implies a higher probabiliy of bankrupcy and he employee is unable o insure fully his human capial risk, firms wih higher leverage have o pay, in equilibrium, a higher wage o he employee o compensae him for he expeced bankrupcy coss borne by him. 8

10 We make use of wo measures of labor coss o es he above heories: CEO compensaion and average employee pay. CEO compensaion measures he pay of he mos imporan employee. In he model of BSZ (2010), here is only one employee per firm. A company s CEO plays a criical role in affecing corporae performance, and his produciviy is more difficul o evaluae han ha of lower level employees. Therefore, he single employee in he model of BSZ (2010) migh be bes inerpreed as he CEO in empirical ess. Average employee pay measures he compensaion of a collecive employee. Since average employee pay is calculaed as oal labor expenses divided by he number of employees, we are able o use his measure o direcly derive he marginal impac of leverage on oal labor expenses and herefore o compare he marginal effec of deb on labor coss wih he incremenal ax benefis of deb. Based on he implicaions of he above heoreical models and using he es variables discussed above, we have he following esable hypoheses. Hypohesis 1: Firms wih higher leverage will incur larger CEO compensaion. Hypohesis 2: Firms wih higher leverage will incur larger average employee pay. Hypohesis 3: A he exising deb level, he addiional labor coss associaed wih an increase in leverage are large enough o offse he incremenal ax benefis of deb. Peroi and Spier (1993) argue ha labor unions will bargain less aggressively and may be more willing o ake pay cus if highly levered firms run a greaer risk of bankrupcy. Alhough heir model implies ha workers, ex ane, will demand a higher expeced wage in compensaion for bearing he above risk (Proposiion IV of heir paper), anoher empirical implicaion of heir heory is ha, ex pos, here will be a negaive correlaion beween leverage and wage when a firm faces subsanial financial disress. Thus we have he following esable hypohesis: Hypohesis 4: Firms wih higher leverage will incur lower average employee pay when hey are in financial disress. One imporan elemen of he model of BSZ (2010) is he degree of job enrenchmen. Differen from he same erm used in he lieraure on corporae governance, enrenchmen in his 9

11 conex means he degree o which employees are able o insure heir human capial risk (lower heir abiliy o insure, greaer he exen of enrenchmen). Job enrenchmen in his sense is he reason why he employee demands a higher pay from a firm wih higher leverage in BSZ (2010). To empirically sudy he impac of employee job enrenchmen on he leverage-wage relaion, we examine echnology versus non-echnology firms. There is evidence suggesing ha employees in non-echnology firms are more enrenched compared o hose in echnology firms. 6 Given his, we expec leverage o have a sronger impac on labor coss in non-echnology firms han in echnology firms. This yields our fifh esable hypohesis: Hypohesis 5: The effec of leverage on CEO compensaion as well as on average employee pay will be greaer in non-echnology firms han in echnology firms. 3. Daa and Summary Saisics 3.1. Sample of CEO compensaion We gaher informaion on CEO pay from he Execucomp daabase. I provides deailed informaion on he compensaion of he op five execuives of S&P 1,500 firms since We focus on he CEOs. We merge Execucomp wih he COMPUSTAT Indusrial Annual Daabase from 1992 o We delee firms wih non-posiive book value of equiy and exclude financial and uiliies companies. 17,173 firm-year observaions saisfy hese crieria. 14,891 observaions have all he necessary informaion o be included in our OLS regressions of CEO compensaion. During our sample period ( ), here are 1,952 new CEOs. To deermine wheher a new CEO is an ouside hire, we use he following wo-sep procedure. Firs, we search for his previous employer in he Execucomp daabase. If his prior employer is no he same as he curren firm, hen he is an ouside hire. Second, if we canno idenify his previous employer in he 6 Anderson, Banker, and Ravindran (2000) documen ha he demand for execuives and oher criical employees in echnology firms is inense, leading o higher employee urnovers han in non-echnology firms. Iner, Lamber, and Larcker (2003), using proprieary compensaion survey daa, find ha echnology firms rank employee reenion objecives as he mos imporan goal of heir equiy gran program. Overall, his evidence indicaes ha employees in echnology firms will suffer a lower loss of human capial if heir firms ener financial disress compared o hose in non-echnology firms. 10

12 Execucomp daabase (Execucomp only repors informaion on he op five execuives in S&P 1,500 firms), we search he Lexis-Nexis Academic Universe by he name of he execuive and of he company o deermine wheher he is hired from ouside or promoed from inside. 7 We idenify 373 ouside hires using his mehodology Sample of average employee pay We use informaion from he COMPUSTAT Indusrial Annual daabase beween 1992 and 2006 o sudy he impac of leverage on average employee pay. 8 We exclude financial and uiliies companies, and we exclude firms wih less han 100 employees. We also drop firms wih nonposiive book values of equiy. We calculae average employee pay as oal labor expenses divided by he number of employees. COMPUSTAT provides labor and relaed expenses (daa iem 42) and he number of employees (daa iem 29). According o he COMPUSTAT daa manual, daa iem 42 includes salaries and wages, pension coss, payroll axes, incenive compensaion, profi sharing, and oher benefi plans. Daa iem 42 hus represens a firm s oal labor expenses. This suis our purpose, since we need o esimae he impac of leverage on oal labor coss. Abou 10% of firms recorded in he COMPUSTAT have valid informaion on daa iem 42. This may inroduce a sample-selecion bias and we discuss how we conrol for his in Secion 5. There are 5,269 firmyear observaions ha have he necessary informaion o be included in our OLS regression of average employee pay Oher daa sources We obain quis raes from he daabase of Job Openings and Labor Turnover Survey (JOLTS) on he websie of he U.S. Bureau of Labor Saisics. The quis rae is he number of quis (volunary separaions) during he enire year as a percen of annual average employmen. The daa is available a he indusry level from The indusry classificaion is based on he Norh 7 Lexis-Nexis Academic Universe provides comprehensive informaion conained in major U.S and world publicaions (including Wall Sree Journal, New York Times, The Washingon Pos, USA Today, among many ohers), SEC filings, news wire services, web publicaions, TV and radio broadcas ranscrips, major company profiles and repors, cour cases, law reviews, and even blogs. 8 This ensures ha our samples of CEO compensaion and employee pay cover he same ime period. 11

13 American Indusry Classificaion Sysem (NAICS). Appendix A repors annual quis raes by indusry and year. Corporae governance may play a role in CEO compensaion, and i may also maer in deermining average employee pay. 9 Therefore, we examine wheher corporae governance is a facor in deermining average employee pay and CEO compensaion. We use he G-Index consruced by Gompers, Ishi and Merick (2003) as a measure of corporae governance. They compue he G-Index using a oal of 24 possible aniakeover provisions. The daa source is he Invesor Responsibiliy Research Cener (IRRC) daabase, which provides annual informaion on corporae aniakeover provisions for he years 1990, 1993, 1995, 1998, 2000, 2002, 2004, 2006, and We fill in observaions in he missing years using informaion from he mos recen year: for example, we use informaion from 2004 for year A greaer value of he G-Index corresponds o weaker shareholder righs and sronger managerial power. Throughou our empirical analysis of boh CEO compensaion and average employee pay, all dollar amouns are adjused o 1992 dollars using he Consumer Price Index (CPI). 10 We use he Fama-French 48-indusry classificaion o caegorize firms ino heir respecive indusries (he classificaion is obained from Kenneh French s websie) Empirical Tess and Resuls on Capial Srucure and CEO compensaion In his secion, we describe our empirical ess of he impac of leverage on he magniude of CEO compensaion. We sar wih OLS regressions of CEO compensaion in he whole sample. We hen perform addiional ess o idenify causaliy: o accomplish his, we examine he impac of leverage in he prior year on he compensaion of newly appoined CEOs who are hired from ouside Summary saisics 9 Cronqvis e al. (2009) find ha CEOs wih more conrol pay heir workers more. 10 CPI daa is aken from he websie of he Bureau of Labor Saisics: hp:// 11 hp://mba.uck.darmouh.edu/pages/faculy/ken.french/daa_library.hml 12

14 In Table 1, we presen summary saisics for he variables used in our analysis of CEO compensaion. Execucomp provides wo measures of oal compensaion: one includes he value of he opions graned while he oher includes he value of opions exercised. We use he oal compensaion including he value of opions exercised in our analysis. The resuls remain qualiaively he same when he value of opions graned is considered. Cash compensaion is he sum of salary and bonus, as provided by Execucomp. We compue equiy-based compensaion as he oal compensaion minus salary, bonus, oher annual pay, and LTIP (Long-erm Incenive Plan). The mos common forms of equiy-based compensaion are sock opions and resriced socks. Marke capializaion is compued as he sock price muliplied by he number of shares ousanding a he end of a fiscal year. Marke-o-book raio is he marke capializaion divided by he book value of equiy. All coninuous variables excep leverage are winsorized a he 1 s and 99 h perceniles. 12 Leverage is he variable of ineres. We measure leverage in four ways: he marke leverage, as used widely in he lieraure (e.g., Leary and Robers, 2010), is compued as he oal deb divided by he sum of oal deb and marke value of equiy; he book leverage, also used commonly in he lieraure, is compued as he oal deb divided by he sum of oal deb and book value of equiy. Toal deb is he sum of long-erm deb and deb in curren liabiliies (daa 9 plus daa 34). Deb in curren liabiliies (daa 34) includes noes payable (daa 206) and deb due in 1 year (daa 44). Welch (2011) argues ha he liabiliies ha are non-financial deb should no be included in he compuaion of leverage raio. We follow Welch (2011) and inroduce wo addiional measures of leverage, which we refer o as alernaive marke leverage and alernaive book leverage respecively. We calculae alernaive marke leverage as (oal long-erm deb + deb due in 1 year)/(oal long-erm deb + deb due in 1 year + marke value of equiy) and calculae alernaive book leverage as (oal long-erm deb + deb due in 1 year)/(oal long-erm deb + deb due in 1 12 Anoher way o idenify ouliers is by employing Hadi s (1992, 1994) procedure. The exclusion of ouliers does no affec he resuls of our mulivariae analysis. 13

15 year + book value of equiy). 13 Due o space limiaions, we repor resuls only from our analysis using marke leverage, alernaive marke leverage, and alernaive book leverage. However, resuls from our analysis using book leverage are available upon reques. CEOs cash compensaion (salary plus bonus) has a mean of $972,330 and a median of $736,490, wih a 1% cuoff of $109,090 and 99% cuoff of $4.531 million. The equiy-based compensaion has a larger mean bu a smaller median han he cash compensaion. The reason is ha equiy-based pay has a wider range across firms han cash pay, and some CEOs have exremely large equiy-based pay. For example, he 99% cuoff of equiy-based compensaion is abou $27 million, while he 1% cuoff is only $12,500. We use he naural log of he compensaion variables in our mulivariae regression of CEO compensaion o reduce he poenial impac of ouliers. The one-year reurn o shareholders (including dividends), a measure of firm performance, has a median of 10.33%. Inser Table 1 here Turning o CEO characerisics, he median CEO age is 65, and he median lengh of CEO enure is 4 years. Only 2% of he CEOs in our sample are female. 64% of he CEOs also serve as Chairman of he board. The G-index has a mean of 9.26 and a median of OLS regressions In our reduced form analysis, we model CEO compensaion as he following: CEOPay = γ 0 + γ 1 Size + γ 2 Leverage + γ 3 MTB + γ 4 RET + γ 5 Age + γ 6 Tenure + γ 7 Chair + γ 8 MALE + ε. (1) CEOPay is he CEO compensaion of firm i in year and i is measured in hree ways: cash, equiy-based, and oal compensaion. Size is he naural log of marke capializaion of firm i as of year. We expec Size o be a posiive and significan deerminan of CEO compensaion. As Murphy (1999) has poined ou, he bes-documened sylized fac regarding CEO pay is ha CEO 13 We hank an anonymous referee for bringing Welch (2011) o our aenion and for suggesing ha we also repor our analysis using hese wo alernaive measures of marke and book leverage, respecively. 14

16 pay is higher in larger firms. Leverage i, is he leverage raio of firm i as of year. If firms wih higher leverage pay a higher wage o heir CEOs, γ 2 would be posiive. MTB i, is he marke-o- book raio of firm i as of year, which is used as a proxy for firms growh opporuniies. RET i, is he reurn o shareholders of firm i in year, a popular measure of he performance of firm i in year. The exising lieraure documens a posiive relaion beween CEO pay and firm performance. 14 Hence, we expec γ 4 o be posiive. In addiion, we conrol for individual CEO characerisics ha may affec CEO compensaion. Age i, is he age of he CEO of firm i as of year ; Tenure i, is he number of years he execuive has aced as he CEO in firm i prior o year ; CEO is also he Chairman, and zero oherwise; Chair i, is one if he MALE i, is one if he CEO is male, and zero oherwise. We include year dummies o conrol for ime-specific variaion in CEO pay. As documened by he lieraure, CEO compensaion has increased remendously during he pas few decades. We include indusry dummies due o he significan variaion in CEO pay across indusries. In Table 2, we repor he esimaed coefficiens and sandard errors obained from he OLS regression of Eq. (1). The sandard errors are clusered by firm. Esimaion resuls from using marke leverage, alernaive marke leverage, and alernaive book leverage are repored in Panel A, Panel B, and Panel C, respecively. Columns 1-3 in each panel exclude he G-Index, while columns 4-6 in each panel include he G-Index. Including he G-index in he regression reduces he sample size. Firm size has a posiive impac on all hree measures of CEO compensaion. A larger firm pays is CEO, on average, more han a smaller firm does, which is consisen wih he lieraure. A higher one-year reurn o shareholders is associaed wih greaer CEO pay. This is consisen wih he posiive relaion beween CEO pay and firm performance as documened by he lieraure. 14 Murphy (1999) provides a comprehensive review of he relaion beween firm performance and CEO compensaion. 15

17 On average, an older CEO earns a larger pay. Being he Chairman has a posiive and significan effec on CEO compensaion. Gender does no have a significan effec on CEO pay. The coefficien on CEO enure is no significan in he regression of oal and cash compensaion, bu is negaive and significan (a he 5% level) in ha of equiy-based compensaion. Marke-obook raio is no significan in he regressions of oal compensaion, bu i is negaive in he regression of cash pay and is posiive in ha of equiy-based compensaion. This suggess ha growh firms pay less cash bu more sock-based compensaion o heir CEOs han value firms. Inser Table 2 here The leverage raio has a posiive and significan effec on cash, equiy-based, and oal compensaions. According o Column (1) of Panel A, if marke leverage goes up by one sandard deviaion (0.19, as repored in Table 1), he naural log of CEO oal compensaion increases by 0.19*0.42 = 0.080, which ranslaes o more han 8.3% increase in oal pay. Therefore, if we sar a he median oal CEO compensaion of $1.20 million, he oal CEO pay increases by abou $100,000, an economically significan amoun. If marke leverage increases by one sandard deviaion, 0.19, he CEO s cash pay goes up by more han 12% and he CEO s equiy-based pay goes up by more han 8%. In summary, he effec of leverage on CEO compensaion is economically as well as saisically significan. The G-Index is a posiive and significan facor in deermining CEOs cash, equiy-based, and oal pay, suggesing ha sronger managerial power is associaed wih greaer CEO compensaion. Leverage coninues o have a posiive and significan effec on CEO compensaion in he presence of he G-Index. We also esimae Eq. (1) by year, in he spiri of Fama and Macbeh (1973). Table 3 repors he coefficien of leverage in he regression of CEO compensaion for every year beween 1992 and The coefficiens of all hree measures of leverage in he regression of CEOs oal pay and cash pay are posiive in all of he 15 years. The coefficien of alernaive book leverage is posiive in he regression of CEOs equiy-based pay in 13 ou of 15 years, while he oher wo measures of 16

18 leverage have a posiive coefficien in he regression of CEOs equiy-based pay in 14 ou of 15 years. Inser Table 3 here 4.3. New CEOs hired from ouside I is possible ha some unobservable and hus unconrolled CEO characerisics affec boh leverage and compensaion in he same direcion, hus resuling in he posiive coefficien of leverage in he OLS regression of CEO compensaion. For example, CEOs who have had more ineracion wih he board (and herefore have more influence) may have greaer abiliy o affec heir own pay and a he same ime choose he firm s leverage level as well. To address poenial concerns regarding causaliy, we sudy he subse of newly appoined CEOs who are hired from ouside. We examine how he firs-year compensaion of hese new CEOs is affeced by firm leverage in he year prior o heir appoinmen. CEOs hired from ouside should have no influence on heir firms capial srucure in he year prior o heir appoinmen, so ha his is a clean es of he relaionship beween leverage and CEO compensaion, allowing us o deal wih he poenial causaliy problem discussed above. We model he relaionship beween he firs-year compensaion of newly appoined CEOs hired from ouside and he leverage raio in he year prior o heir appoinmen as follows: CEOPay = γ 0 + γ 1 Size -1 + γ 2 Leverage -1 + γ 3 MTB -1 + γ 4 RET + γ 5 Age + γ 6 Chair + γ 7 MALE + ε. (2) In Eq. (2), firm size, leverage, and marke-o-book raio are compued as of he fiscal year prior o he appoinmen of he new CEO. CEO enure is omied from Eq. (2), since we esimae Eq. (2) on he sample of newly appoined CEOs hired from ouside (all of hem have zero enure, by definiion). Timan (1984) and BSZ (2010) predic ha a firm wih higher leverage will pay is employees more. In he case of a newly hired CEO, he will demand and obain a higher pay from a firm wih higher leverage. Therefore, we expec γ 2 o be posiive. 17

19 In Table 4, we presen he coefficiens and sandard errors obained from esimaing Eq. (2) on he subse of newly appoined CEOs who are hired from ouside. Firm size is a srong facor in deermining he pay of newly appoined CEOs (cash, equiy-based, and oal compensaion). The coefficien on he marke-o-book raio is negaive and significan in all hree ypes of compensaion, suggesing ha growh firms pay heir new CEOs less han value firms do. The coefficien of sock reurn during he firs year of a new CEO is posiive and significan in he regression of equiy-based compensaion. CEO age has a negaive effec on equiy-based compensaion, differen from wha we see in Table 2. This is due o he differences beween he underlying samples. In Table 2, he same CEO in he same firm appears in muliple years, and he pay ofen increases wih he CEO s age. In Table 4, all he CEOs are newly hired from ouside and he compensaion informaion is based on heir firs-year pay only. The negaive coefficien on CEO age in he regression of equiy-based compensaion suggess ha a younger newly hired CEO, in his firs year, earns more equiy-based pay han an older newly hired CEO, afer conrolling for oher facors. Inser Table 4 here The variable of ineres is he leverage raio. The coefficien on leverage is posiive and significan in he regressions of all hree forms of compensaion for he newly hired CEOs, for all hree measures of he leverage raio. The impac of leverage on CEO compensaion is also economically significan: a one sandard deviaion increase in marke leverage corresponds o an 19% increase in cash pay, a 27% increase in equiy-based pay, and an 18% increase in he oal compensaion of a new CEO. The resuls in Table (2)-(4) demonsrae ha leverage has a srong and posiive effec on he level of CEO compensaion, supporing Hypohesis 1. Firms wih higher leverage incur a greaer amoun of CEO compensaion, which is consisen wih he Timan-BSZ predicion discussed above. 18

20 5. Empirical Tess and Resuls on Capial Srucure and Average Employee Pay In his secion, we presen resuls on he effec of leverage on average employee pay. In he mulivariae analysis, we sar wih OLS regressions. We hen uilize insrumenal variable regressions of average employee pay o address poenial concerns abou he endogeneiy of leverage. Finally, we deal wih a poenial sample selecion problem using a Heckman (1979) wosep analysis Summary saisics Table 5 provides he summary saisics of he variables used in our analysis of average employee pay. Average employee pay is compued as labor expenses (daa iem 42) divided by he number of employees (daa iem 29). Leverage is he oal deb divided by he sum of oal deb and he book value of equiy. Marke capializaion is he sock price muliplied by he number of shares ousanding as of he fiscal year end. We compue average sales per employee by dividing he amoun of oal sales (daa iem 12) by he number of employees (daa iem 29). Marke-o-book raio is he marke capializaion divided by he book value of equiy. Physical capial inensiy is compued as gross propery, plan, and equipmen scaled by oal asses (daa iem 6). All coninuous variables excep leverage are winsorized a he 1 s and 99 h perceniles. The mean (median) of average employee pay is $32,760 ($32,000). The 1% cuoff is $1,490, and he 99% cuoff is $95,580. Marke capializaion has a wide range, from $2.08 million (he 1% cuoff) o $82,827 million (he 99% cuoff). To reduce he poenial influence of ouliers, we use he log of average employee pay and he log of marke capializaion in our analysis. The mean of sales per employee is abou $166,260. The marke-o-book raio has a mean of 2.94 and a median of On average, he gross amoun of propery, plan, and equipmen is abou 69% of oal asses. Inser Table 5 here 5.2. OLS regressions 19

21 Our objecive here is o esimae he effec of leverage on average employee pay. In our reduced form analysis (he base case), we use he following specificaion: AEP = β 0 + β 1 Size + β 2 Leverage + β 3 AvgSale + β 4 MTB + β 5 PCI + ε. (3) AEP is he naural log of average employee pay of firm i in fiscal year. Size i, is he log of marke capializaion of firm i a he end of year. Prior empirical sudies have documened ha larger firms end o pay higher wages o heir employees han smaller firms, so we expec β 1 o be posiive. AvgSale i, is he sales per employee. We use i AvgSale, o direcly measure he produciviy of he average employee of firm i in year, and we expec β 3 o be posiive. MTB is he marke-o-book raio of firm i as of year. We conrol for he marke-o-book raio, as i is a common proxy for a firm s growh opporuniy. PCI is he physical capial inensiy of firm i as of year. We include he measure of physical capial inensiy for wo reasons: firs, capial inensive firms end o be more producive (Cronqvis e al., 2009); second, BSZ (2010) predic a posiive correlaion beween physical capial inensiy and employee wage. We include he year dummies o conrol for he aggregae variaion in employee pay. We also include he indusry dummies because here is a grea deal of heerogeneiy in pay pracices across indusries. The effec of leverage on average employee pay is of paricular ineres. If firms of higher leverage pay heir employees more, β 2 will be posiive. Panel A of Table 6 presens he esimaed coefficiens and sandard errors obained from he OLS regression of Eq. (3) for all firms. The sandard errors are clusered by firm and are also robus o heeroskedasiciy. Larger firms pay heir employees more, consisen wih he lieraure (e.g., Brown and Medoff, 1989). Average sales per employee affecs average employee pay posiively, consisen wih our expecaion, since sales per employee is a measure of employee produciviy. Neiher physical capial inensiy nor he marke-o-book raio has a significan impac on average employee pay. Mos imporanly, afer conrolling for oher facors, he leverage raio has a posiive effec on average employee pay. The coefficiens on all hree leverage raios are posiive and 20

22 significan a he 1% or 5% level. This suppors Hypohesis 2. We find ha he G-Index is no a saisically significan facor in deermining average employee pay. Furher, even in he presence of he G-Index, leverage is a posiive and significan deerminan of average employee pay. Inser Table 6 here We now examine he subse of financially disressed versus safe firms. Since is inroducion by Alman (1968), he Z-score has been used for he predicion of bankrupcy. Following he original formula, we compue he Z-score as follows: Z = 1.2T T T 3 +.6T 4 + T 5. (4) Here T 1 = working capial / oal asses, where working capial is compued as curren asses minus curren liabiliies; T 2 = reained earnings / oal asses; T 3 = earnings before ineres and axes / oal asses; T 4 = marke value of equiy / book value of oal liabiliies; T 5 = sales/ oal asses. A lower Z-score corresponds o a greaer probabiliy of bankrupcy: firms wih a Z-score above 2.99 are considered o be safe, while hose a Z-score of 1.8 or lower considered o be disressed, and hose wih Z scores in beween he above wo hreshold values considered o be in he grey zone. Panel B of Table 6 repors he resuls from esimaing Eq. (3) on he wo subses: disressed firms and safe firms. When firms are financially disressed, average employee pay is no significanly relaed o leverage; when firms are safe, average employee pay increases wih leverage. In summary, he evidence supporing Hypohesis 4 is weak or nonexisen. This indicaes ha while he ex ane relaionship beween leverage and employee pay suggesed by Timan-BSZ predicion dominaes in our enire sample and in he subsample of safe firms, in disressed firms he ex-pos relaionship posulaed by Peroi and Spier (1993) may parially or fully offse he above effec of firms compensaing employees for he reducion in he value of heir human capial due o higher leverage. In Panel C, we include he quis rae in he OLS esimaion. Column 1 includes he quis rae only. The coefficien is significan and negaive, suggesing ha more specialized labor ges paid more. Column 2 adds indusry and year fixed effecs o he regression, and he coefficien of quis 21

23 rae becomes insignifican. This is no surprising, given ha he annual quis rae is measured a he indusry level. In columns 3-5, he coefficien of quis rae remains insignifican, bu ha of all hree leverage raios is sill posiive and saisically significan. As a robusness es, we also esimae Eq. (3) by year, in he spiri of Fama and Macbeh (1973). Table 7 repors he esimaed coefficien on leverage in every year during The coefficien on marke leverage ranges from o 0.66, and is posiive in 13 ou of 15 years. Is mean is 0.22, saisically larger han zero. The impac of leverage is somewha weaker prior o year 2000 han afer year To undersand why, we examine he percenage of non-echnology firms by year. We find ha he percenage of non-echnology firms is below sample mean in 5 ou of 9 years during while he percenage of non-echnology firms is below sample mean in only 2 ou of 6 years during As we documen laer, he effec of leverage on average employee pay is sronger in non-echnology firms, so ha he smaller coefficiens on leverage prior o he year 2000 may be due o he lower fracion of non-echnology firms prior o year Inser Table 7 here 5.3. Insrumenal variable regressions As we discussed in he inroducion, he asses of a given firm may be such ha hey can suppor a high level of leverage (for example, he proporion of angible asses may be high) and may also require highly paid employees o operae hese asses, hus generaing a posiive correlaion beween leverage and employee pay. To deal wih his poenial endogeneiy problem, we employ an insrumen variable, namely, he marginal corporae ax rae, o generae an exogenous variaion in leverage. A valid insrumenal variable (IV) for leverage needs o saisfy wo condiions: i is correlaed wih he leverage raio (he validiy requiremen), bu is uncorrelaed wih he residual in he regression of employee pay (he exclusion resricion). The insrumen we use, namely, he marginal corporae ax rae, saisfies boh requiremens. The heoreical lieraure in corporae finance indicaes ha he ax benefi of deb will be posiively relaed o a firm s marginal ax rae, hus resuling in a posiive correlaion beween a firm s marginal ax rae and is leverage 22

24 raio. The empirical lieraure suppors he above view (for example, Leary and Robers, 2010). A he same ime, here is no heoreical or empirical lieraure indicaing ha he marginal corporae ax rae will direcly affec average employee pay. Following Graham, Lemmon, and Schallheim (1998), we use he marginal ax raes based on income before ineres is deduced (MTRB), from he daabase of marginal ax raes provided by John Graham (for more deails, see Graham, 1996a and Graham, 1996b). When examining he effec of firms leverage on bond raings, Molina (2005) also uses marginal ax rae as an insrumen for leverage. We implemen he insrumenal variable regressions by using he 2SLS (wo-sage leas squares) procedure in STATA (Wooldridge, 2002). In he firs sage, leverage is regressed ono he insrumenal variable and conrol variables; in he second sage, average employee pay is regressed ono he insrumened leverage and conrol variables. The firs sage regression specificaion is given by: Leverage = α 0 + α 1 MRTB + α 2 Size + α 3 AvgSale + α 4 MTB The second sage regression specificaion is given by: + α 5 PCI + α 6 (EBIT/TA) + α 7 STD(EBIT/TA) + δ. (5) AEP = β 0 + β 1 Size + β 2 Leverage + β 3 AvgSale + β 4 MTB + β 5 PCI + β 6 (EBIT/TA) + β 7 STD(EBIT/TA) + ε. (6) MTRB is he marginal ax rae based on income before ineres is deduced. EBIT/TA is earnings before depreciaion, ineres, and axes divided by oal asses, and STD(EBIT/TA) is he sandard deviaion of EBIT/TA in he pas five years. The resuls from he above insrumenal variable regression are presened in Table 8. In he firs sage analysis (leverage is he dependen variable), marginal ax rae is an imporan deerminan of deb raio, significan a he 1% level. In heir survey of he weak-insrumen lieraure, Sock e al. (2002) develop benchmarks for he necessary magniude of he F-saisic. When he number of insrumens is 1, 2, 3, 5, and 10, he suggesed criical F-values are 8.96, 11.59, 12.83, 15.09, and 20.88, respecively. If he firs-sage parial F-saisic falls below hese criical values, he insrumens are considered o be weak and 23

25 inference problems are poenially serious. The parial F-saisics of our insrumen in he regressions of all hree leverage raios are above he criical value of The resuls in he firs sage and he parial F-es confirm ha he marginal ax rae is a srong insrumen (i.e., i saisfies he validiy requiremen). Inser Table 8 here In he second sage analysis, firm size and average sales per employee are posiive and significan deerminans of he average employee pay, consisen wih he resuls from our OLS regressions presened in Table 6. More imporanly, we find from our second sage regression ha, even afer accouning for he poenial endogeneiy of leverage, leverage coninues o be a posiive and significan deerminan of average employee pay. Noe ha in he firs-sage regression of alernaive book leverage, he marke-o-book raio has a posiive coefficien, which seems o conradic he negaive relaion beween leverage and he marke-o-book documened in he exising lieraure. 15 However, Chen and Zhao (2006) show ha he negaive relaion ha has been documened beween book leverage and marke-o-book raio is driven by a few small firms wih very large marke-o-book raios. In paricular, hey noe ha a posiive relaion beween marke-o-book and leverage holds for 88% of all firms, accouning for more han 95% of he oal marke capializaion Missing daa on labor expenses: a Heckman (1979) wo-sep analysis As we have menioned earlier, labor expenses are missing for a number of firms in COMPUSTAT. This creaes a poenial sample selecion bias, if firms selecively decide wheher or 15 I is worh poining ou ha even papers in he exising lieraure show ha he negaive coefficien on marke-o-book has a significanly smaller magniude in he regression of book leverage han in he regression of marke leverage. For example, in Table 5 of Hovakimian, Kayhan, and Timan (2012), he coefficien of marke-o-book is (z-sa is -3.8) in he regression of book leverage while i is (z-sa is -15.2) in he regression of marke leverage. The difference in he sign of he marke-o-book coefficien when using book leverage beween our resul and he previous work could be due o he fac ha our sample size is smaller: he limied availabiliy of CEO compensaion and average employee pay daa significanly reduces our sample size, compared o oher sudies. 16 The review aricle by Parsons and Timan (2008) has a deailed discussion of papers sudying he relaion beween leverage and he marke-o-book raio, including he paper by Chen and Zhao (2006). They sugges cauion when using and inerpreing marke-o-book raios in leverage specificaions. 24

26 no o repor labor expense informaion. To conrol for his poenial sample selecion bias, we adop a Heckman (1979) wo-sep analysis in his secion. In he firs sep, we esimae a probi model of wheher or no a firm repors labor expenses. The dependen variable is one if he daa on labor expenses is non-missing, and zero oherwise; he independen variables include he dummies of he firm s lising exchange, in addiion o he original conrol variables in he regression of average employee pay. The lising exchange is he idenifying variable: we assume ha firms on differen exchanges have differen reporing behavior (he resuls in he firs-sep probi analysis confirm his assumpion), while exchange lising does no affec he repored average employee pay (o verify his condiion, we add he dummies of exchange lising o he OLS regression of average employee pay, and find ha hey are joinly insignifican, wih an F- saisic of 1.29 and p-value of 0.28). In he second sep, we examine he effec of leverage on average employee pay. The inverse Mills raio (Lambda) derived from he selecion model is included in he second sep as a regressor, and all oher independen variables are as specified in Eq. (3). Inser Table 9 The esimaed coefficiens and sandard errors are repored in Table 9. From he esimaion of he selecion model in he firs sep, we observe ha larger firms wih higher leverage, lower sales per employee, lower marke-o-book raio, and higher physical capial inensiy are more likely o repor labor expenses. The exchange dummies are joinly significan. In he second sep, he coefficiens on firm size and average sales per employee are posiive and significan. More imporanly, he impac of leverage on average employee pay remains posiive and significan afer we conrol for he poenial sample-selecion bias. The Heckman wo-sep procedure produces consisen esimaion of parameers. The coefficien on he inverse Mills raio is saisically disinguishable from zero and negaively signed, suggesing ha he unobserved facors ha make reporing of labor expenses more likely end o be associaed wih lower average employee pay. 25

27 5.5. A comparison of he incremenal coss and benefis of leverage Boh our OLS and insrumenal variable regressions provide evidence supporing Hypohesis 2. Based on he resuls presened in Panel A of Table 6, we now compue he incremenal ax benefis and labor coss associaed wih an increase in leverage. For a firm wih he median values of he leverage raio, average employee pay, oal labor expenses, and oal deb, if he marke leverage raio increases by 0.23 (one sandard deviaion of leverage in he sample in Panel A of Table 6), he naural log of average employee pay will increase by 0.23*0.23= Saring a he median level of average employee pay of $32.00 (in housands), average employee pay hen becomes $33.79 (in housands), an increase of 5.60%. The median oal labor expenses is abou $250 million, so he increase in oal labor expenses is abou 250*5.60%= $14.01 million, assuming ha he number of employees does no change. The reurn on corporae bonds depends on various facors such as ineres rae, credi raing, and ime o mauriy, so ha we can only use an average ineres rae for our calculaion of he ax benefis of deb. We use 6% as he average rae of reurn on corporae bonds in our sample from The median level of deb is abou $120 million in our sample. Saring from he median leverage raio of 0.20, he level of deb goes up by 202% when we increase marke leverage raio by 0.23 (one sandard deviaion), holding everyhing else consan. If we assume a marginal ax rae of 35%, which is he median corporae marginal ax rae as compued by John Graham (Graham, 1996a and Graham, 1996b), ineres expenses will increase by $120*202%*0.06 = $14.54 million, and he ax benefis of deb would increase by 14.54*0.35 = $5.09 million, which is smaller han he increase in oal labor expenses ($14.01 million). The above calculaion shows ha he addiional labor coss associaed wih an increase of one sandard deviaion in he leverage raio offses all of he incremenal ax benefis associaed wih he leverage increase. 17 We believe ha 6% is a reasonable esimae. The compounded annual reurn for long-erm U.S governmen bonds has averaged less han 3% during he pas four decades, and we assume ha corporae bonds, on average, have a 3% premium over long-erm U.S governmen bonds. 26

28 We repea he above calculaion by increasing he marke leverage raio from he median level of 0.20 all he way o 0.68, an increase of more han wo sandard deviaions. Fig. 1 plos he changes in oal labor expenses and he ax benefis of deb as he leverage raio goes up. The graph shows ha he addiional labor expenses offse all of he incremenal ax benefis of deb even when he leverage raio is increased by as much as wo sandard deviaions. Inser Fig. 1 here The above analysis demonsraes ha he incremenal labor coss associaed wih an increase in leverage are economically significan. Furher, hese incremenal labor coss are greaer han he addiional ax benefis of deb associaed wih a wide range of changes in he leverage raio. Therefore, he above resuls suppor our hypoheses 2 and 3. Overall, his evidence is consisen wih he Timan-BSZ predicion, i.e., risk-averse employees demand greaer compensaion from firms wih higher leverage, and such indirec coss of bankrupcy are economically large enough o limi he use of deb by hese firms. 6. Technology firms versus non-echnology firms We now sudy CEO compensaion and average employee pay in wo subses of our sample: echnology firms versus non-echnology firms. The definiion of echnology firms and nonechnology firms follows ha in Anderson, Banker, and Ravindran (2000), Iner, Lamber, and Larcker (2003), and Murphy (2003). Technology firms are defined as companies in he compuer, sofware, inerne, elecommunicaions, or neworking fields. Non-echnology firms are firms wih SIC codes less han 4000 no oherwise caegorized as echnology firms. 18 We examine wheher he effec of leverage on employee pay is differen beween echnology and non-echnology firms since 18 Technology firms are defined as companies wih primary SIC designaions of 3570 (Compuer and Office Equipmen), 3571 (Elecronic Compuers), 3572 (Compuer Sorage Devices), 3576 (Compuer Communicaion Equipmen), 3577 (Compuer Peripheral Equipmen), 3661 (Telephone & Telegraph Apparaus), 3674 (Semiconducor and Relaed Devices), 4812 (Wireless Telecommunicaion), 4813 (Telecommunicaion), 5045 (Compuers and Sofware Wholesalers), 5961 (Elecronic Mail-Order Houses), 7370 (Compuer Programming, Daa Processing), 7371 (Compuer Programming Service), 7372 (Prepackaged Sofware), and 7373 (Compuer Inegraed Sysems Design). Non-echnology firms are firms wih SIC codes less han 4000 no oherwise caegorized as echnology firms. 27

29 employees in non-echnology firms are more enrenched han in echnology firms (in he sense discussed in Secion 2), so ha we expec he effec of leverage on employee pay in non-echnology firms o be greaer han ha in echnology firms (consisen wih Hypohesis 5) CEO compensaion in echnology versus non-echnology firms We firs examine CEO compensaion in echnology versus non- echnology firms. In Table 10, we compare CEO compensaion and various explanaory variables across echnology and nonechnology firms. The mean of oal compensaion for CEOs in echnology firms is greaer han ha in non-echnology firms, bu he median is smaller for CEOs in echnology firms. Alhough CEOs in echnology firms receive less cash compensaion han CEOs in non-echnology firms, he former have a greaer mean of equiy compensaion han he laer, consisen wih Anderson, Banker, and Ravindran (2000), Iner, Lamber, and Larcker (2003), and Murphy (2003). Technology firms have lower leverage han non-echnology firms. Furher, CEOs in echnology firms are younger han CEOs in non-echnology firms. Finally, CEOs in echnology firms are less likely o serve as he Chairmen of he board and are more likely o be female han hose in non-echnology firms. Inser Table 10 here Panel A of Table 11 repors he resuls from OLS regressions of CEO compensaion for nonechnology firms. Firm size is posiively relaed o all hree ypes of compensaion: cash, equiybased, and oal compensaion. Marke-o-book raio is negaively relaed o cash compensaion, bu posiively relaed o equiy-based pay. One-year reurn is posiive and significan in he regressions of all hree measures of CEO compensaion. CEO age is posiive and significan in he regression of oal and equiy-based compensaion. Serving as he Chairman of he board increases he CEO s cash, equiy-based, and oal compensaion. The leverage raio has a posiive and significan effec on CEOs cash, equiy-based, and oal compensaion, for all hree measures of leverage. Panel B of Table 11 presens he regression resuls for echnology firms. Size and CEO age are posiive and significan deerminans of all hree ypes of compensaion. One-year reurn has posiive and significan influences on oal and cash compensaion. A higher marke-o-book raio is 28

30 associaed wih a lower cash pay. Leverage has a posiive and significan effec on cash compensaion, bu no on equiy-based or oal compensaion. Inser Table 11 here We use a Wald es o examine wheher he coefficien of leverage is saisically differen across he wo groups. The value of chi-square is (10.89) wih a p-value of (0.0010) for he regression of oal compensaion (equiy-based compensaion) on marke leverage, and he value of chi-square is 1.24 wih a p-value of 0.26 for he regression of cash compensaion on marke leverage. The Wald ess sugges ha he effec of leverage on oal and equiy-based CEO compensaion is differen beween echnology and non-echnology firms, alhough he effec of leverage on CEO cash compensaion is no saisically differen beween he wo groups. Overall, he effec of leverage on CEO compensaion is greaer for non-echnology firms han for echnology firms, consisen wih our hypohesis 5, hus providing furher suppor for he Timan-BSZ predicion Average employee pay in echnology versus non-echnology firms In Table 12, we analyze he effec of leverage on average employee pay in echnology versus non-echnology firms. Consisen wih he exising lieraure, echnology firms have lower physical capial inensiy and lower leverage raio han non-echnology firms. Technology firms are also smaller han non-echnology firms, and hey have smaller sales per employee han non-echnology firms. The mean of average employee pay is no significanly differen beween he wo groups, bu he median of average employee pay is greaer for non-echnology firms. I is worh noing ha, alhough he mean leverage raio in echnology firms is low, he cross-secional variaion of leverage raio is sill large, e.g., alernaive book leverage ranges from 0 o 0.90 wih a sandard deviaion of 0.21 (no abulaed). Similar o he sample in Table 10, he echnology firms in Table 12 also have lower leverage han non-echnology firms. Differen from Table 10, an average echnology firm in Table 12 is significanly smaller han an average non-echnology firm. The reason is ha Table 10 and Table 12 conain differen samples: Table 12 has 2,101 non-echnology 29

31 and 298 echnology observaions due o he missing informaion on labor and relaed expenses (daa iem 42) in COMPUSTAT, while Table 10 has 8,527 non-echnology and 2,345 echnology observaions from S&P 1,500 firms. Inser Table 12 here Table 13 presens he coefficiens and sandard errors obained from our OLS regressions of average employee pay for non-echnology and echnology firms. We find ha he leverage raio has a posiive and significan effec on average employee pay for non-echnology firms. For echnology firms, he coefficien on leverage is no saisically significan. We use a Wald es o examine wheher he coefficien on leverage is saisically differen across he wo groups. Wald ess sugges ha alernaive marke and book leverage raios have differenial effecs on average employee pay in echnology versus non-echnology firms. Inser Table 13 here The resuls in his secion demonsrae ha leverage has a greaer effec on boh CEO compensaion and average employee pay for non-echnology firms han for echnology firms, consisen wih our Hypohesis 5. This is because employees in non-echnology firms are more enrenched han hose in echnology firms. Faced wih a greaer degree of enrenchmen, employees or CEOs in non-echnology firms are more fearful of a poenial bankrupcy. Therefore, in equilibrium, heir compensaion is more sensiive o heir firm s leverage raio. 7. Addiional robusness ess 7.1. Issues relaing o he use of panel daa and he use of alernaive measures In our empirical analysis, we use panel daa ses, wih a significan number of firms appearing in muliple years. When we esimae a linear model on panel daa, he sandard OLS assumpion of independence among he observaions is very likely violaed. Therefore, we need o consider boh a firm effec and a ime effec in our regressions. As defined by Peersen (2009), he firm effec refers o he correlaion wihin he same firm across differen years, and he ime effec is he cross- 30

32 secional correlaion among differen firms in he same year. Peersen (2009) compares differen approaches in esimaing sandard errors using financial panel daa. He finds ha, in he presence of a firm effec only, clusering by firms generaes unbiased esimaes of sandard errors. In he presence of boh firm and ime effecs, clusering by firms afer including ime dummies yields unbiased esimaes of sandard errors. Consisen wih Peersen (2009), we find ha he sandard errors of he esimaed coefficiens are significanly smaller if we do no cluser hem by firm. The difference in sandard errors srongly indicaes he exisence of a firm effec. We conrol for boh firm and ime effecs in our empirical analysis. In our mulivariae regressions, we include year dummy variables and cluser he sandard errors by he firm. The sandard errors are also robus o heeroskedasiciy. Anoher way o conrol for boh firm and ime effecs would be clusering by boh year and firm. As a robusness es, we repea our analysis by adoping such an approach and find ha he resuls are very similar o hose we have repored in earlier secions. In our mulivariae analysis of average employee pay, we do no include operaing income volailiy as one of he independen variables. Is compuaion requires five years of daa, which reduces our sample size. As a robusness es, we now include his variable in our analysis, and find ha our resuls remain qualiaively he same, and ha he coefficien of operaing income volailiy is insignifican. Furher, in our analysis of CEO pay, we have been using he oal compensaion including he value of opions exercised. Our resuls remain qualiaively he same if, as anoher robusness es, he oal compensaion including he value of opions graned (raher han opions exercised) is used in he above analysis Leverage and he Z-score An imporan assumpion underlying he Timan (1984) and he BSZ (2010) models is ha higher leverage is associaed wih a greaer probabiliy of bankrupcy, resuling in firms wih higher leverage having o compensae employees for he effec of his increased probabiliy of bankrupcy on heir human capial. Consequenly, o furher undersand he role of leverage on labor coss, we 31

33 examine he correlaion of leverage wih he Alman Z-score, which is a measure of a firm s bankrupcy probabiliy. In he sample of average employee pay, he Pearson correlaion coefficien beween he Z-score and alernaive book leverage is -0.54; in he sample of CEO compensaion, he Pearson correlaion coefficien beween he Z-score and alernaive book leverage is As an addiional robusness es, we replace leverage wih he Alman Z-score in our regressions, and find ha he Z-score has a negaive and significan impac on average employee pay (a he 1% level). The Z-score also affecs he cash and oal compensaion of CEOs negaively and significanly, alhough he effec of he Z-score on CEOs equiy-based pay is no significan. These resuls are available upon reques. The resuls of he above robusness ess confirm ha leverage and bankrupcy probabiliy are posiively correlaed, and ha he probabiliy of bankrupcy indeed affecs boh average employee pay and CEO compensaion. They add addiional suppor o he idea ha he prospec of bankrupcy has an imporan influence on he human capial coss incurred by firms. 8. Conclusion Given he poenially large ax benefis of deb, why do firms adop a low level of leverage? The exising lieraure has shown ha direc bankrupcy coss are no large enough o jusify he empirically observed low leverage raios of firms. Timan (1984) and Berk, Sanon, and Zechner (2010) argue heoreically ha a paricular form of indirec bankrupcy cos, namely, he incremenal employee pay associaed wih an increase in deb, is large enough o preven firms from increasing heir leverage raios. In his paper, we empirically es his predicion. Specifically, we answer he following quesions: Firs, does higher leverage resul in greaer employee compensaion? Second, are he addiional labor coss associaed wih higher leverage large enough o offse he incremenal ax benefis of deb? We conduc our empirical analysis using wo measures of employee compensaion: he magniude of CEO compensaion and average employee pay. We find ha he effec of leverage on 32

34 he magniude of CEO compensaion is economically and saisically significan. For he whole sample, leverage has a posiive effec on cash, equiy-based, and oal compensaion of CEOs in our mulivariae regressions. To esablish causaliy, we also sudy he relaionship beween he compensaion of newly appoined CEOs who are hired from ouside and firm leverage in he year before heir appoinmen. We find ha leverage has a significan effec on he magniude of he compensaion of a new CEO. An increase of one sandard deviaion in leverage corresponds o a 19% increase in he oal compensaion of a new CEO. In boh OLS and insrumenal variable regressions, we find ha leverage also influences average employee pay posiively and significanly. Furher, we show ha, for a firm wih he median level of leverage, he incremenal ax benefis arising from increased leverage are offse by he addiional labor coss associaed wih such an increase. The effec of leverage on average employee pay is posiive and significan for financially safe firms, bu he impac is insignifican for financially disressed firms. We also find ha, while leverage has a posiive and significan influence on CEOs cash, equiy-based, and oal compensaion in non-echnology firms, i does no have a significan influence on CEOs oal or equiy-based compensaion in echnology firms. The effec of he leverage raio on average employee pay is also greaer in non-echnology firms han in echnology firms. Since employees in non-echnology firms can be viewed as more enrenched (in he sense of BSZ, 2010), his provides addiional suppor for he Timan-BSZ predicion. Our insrumenal variable analysis o conrol for he endogeneiy of leverage has some limiaions. In paricular, one poenial criicism of he insrumen we use, namely, he marginal ax rae, is ha he independen variaion in his variable can arise only from pas losses and relaively recen invesmens wih invesmen ax credis, boh of which could independenly influence wages. However, even hough he marginal ax rae may be an imperfec insrumen, i allows us o address he endogeneiy of leverage o a significan degree, in he absence of a naural experimen ha may have allowed us o accoun for he poenial endogeneiy of leverage unambiguously. 33

35 Appendix A. Annual quis raes by indusry and year The quis rae is obained from he daabase of Job Openings and Labor Turnover Survey (JOLTS) on he websie of U.S. Bureau of Labor Saisics. The daa is available from The qui rae is he number of quis (volunary separaions) during he enire year as a percen of annual average employmen. The indusry classificaion is based on Norh American Indusry Classificaion Sysem (NAICS) Mining and logging Consrucion Manufacuring Durable goods Nondurable goods Trade, ransporaion and uiliies Wholesale rade Reail rade Transporaion, warehousing, and uiliies Informaion Financial aciviies Finance and insurance Real esae and renal and leasing Professional and business services Educaion and healh services Educaional services Healh care and social assisance Leisure and hospialiy Ars, enerainmen, and recreaion Accommodaion and food services Oher services Toal

36 References Alman, E., Financial raios, discriminan analysis and he predicion of corporae bankrupcy. Journal of Finance, Anderson, M., Banker, R., Ravindran, S., Execuive compensaion in he informaion echnology indusry. Managemen Science 46, Baldwin, C., Produciviy and labor unions: An applicaion of he heory of self-enforcing conracs. Journal of Business 56, Benmelech, E., Bergman, N., Enriquez, R., Negoiaing wih labor under financial disress. Unpublished working paper. Harvard Universiy. Berk, J., Sanon, R., Zechner, J., Human capial, bankrupcy and capial srucure. Journal of Finance 65, Berkovich, E., Israel, R., Spiegel, Y., Managerial compensaion and capial srucure. Journal of Economics & Managemen Sraegy 9, Bronars, S., Deere, D., The hrea of unionizaion, he use of deb, and he preservaion of shareholder wealh. Quarerly Journal of Economics 106, Brown, D., Fee, E., Thomas, S., Financial leverage and bargaining power wih suppliers: Evidence from leveraged buyous. Journal of Corporae Finance 15, Brown, C., Medoff, J., The employer size-wage effec. Journal of Poliical Economy 97, Chen, L., Zhao, X., On he relaion beween he marke-o-book raio, growh opporuniies, and leverage raio. Finance Research Leers 3, Cronqvis, H., Heyman, F., Nilsson, M., Svaleryd, H., Vlachos, J., Do enrenched managers pay heir workers more? Journal of Finance 64, Dasgupa, S., Sengupa, K., Sunk invesmen, bargaining and choice of capial srucure. Inernaional Economic Review 34, Fama, E., MacBeh, J., Risk, reurn, and equilibrium: Empirical ess. Journal of Poliical Economy 81, Fama, E., Schwer, W., Human capial and capial marke equilibrium. Journal of Financial Economics 4, Frydman, C., Jener, D., CEO Compensaion. Unpublished working Paper. Massachuses Insiue of Technology. Gompers, P., Ishi J., Merick, A., Corporae governance and equiy prices. Quarerly Journal of Economics 118, Graham, J., 1996a. Deb and he marginal ax rae. Journal of Financial Economics 41, Graham, J., 1996b. Proxies for he corporae marginal ax rae. Journal of Financial Economics 42, Graham, J., Lemmon, M., Schallheim, J., Deb, leases, axes, and he endogeneiy of corporae ax saus. Journal of Finance 53, Graham, J., Tucker, A., Tax shelers and corporae deb policy. Journal of Financial Economics 81, Had A., Idenifying muliple ouliers in mulivariae daa. Journal of he Royal Saisical Sociey, Series B54, Had A., A modificaion of a mehod for he deecion of ouliers in mulivariae samples. Journal of he Royal Saisical Sociey, Series B56, Hanka, G., Deb and he erms of employmen. Journal of Financial Economics 48, Heckman, J., Sample selecion bias as a specificaion error. Economerica 47, Hennessy, C., Livdan, D., Deb, bargaining and credibiliy in firm-supplier relaionships. Journal of Financial Economics 93, Hovakimian, A., Kayhan, A., Timan, S., Are corporae defaul probabiliies consisen wih he saic rade-off heory? Review of Financial Sudies 25,

37 Iner, C., Lamber, R., Larcker, D., The srucure and performance consequences of equiy grans o employees of new economy firms. Journal of Accouning and Economics 34, Jaggia, P., Thakor, A., Firm-specific human capial and opimal capial srucure. Inernaional Economic Review 35, Leary, M., Robers, M., The pecking order, deb capaciy, and informaion asymmery. Journal of Financial Economics 95, Masa, D., Capial srucure as a sraegic variable: Evidence from collecive bargaining. Journal of Finance 65, Molina, C., Are firms underleveraged? An examinaion of he effec of leverage on defaul probabiliies. Journal of Finance 60, Murphy, K., Execuive compensaion. Orley Ashenfeler and David Card (eds.), Handbook of Labor Economics, Vol. 3, Norh Holland. Murphy, K., Sock-based pay in new economy firms. Journal of Accouning & Economics 34, Parsons, C., Timan, S., Empirical capial srucure: A review. Foundaions and Trends in Finance 3, Pero E., Spier, K., Capial srucure as a bargaining ool: The role of leverage in conrac renegoiaion. American Economic Review 83, Peersen, M., Esimaing sandard errors in finance panel daa ses: comparing approaches. Review of Financial Sudies 22, Sock, J., Wrigh, J., Yogo, M., A survey of weak insrumens and weak idenificaion in generalized mehod of momens. Journal of Business & Economics Saisics 20, Timan, S., The effec of capial srucure on a firm s liquidaion decision. Journal of Financial Economics 13, Timan, S., Wessels, R., The deerminans of capial srucure choice. Journal of Finance 43, Welch, I., Two common problems in capial srucure research: he financial-deb-o-asses raio and issuing aciviy versus leverage changes. Inernaional Review of Finance 11, Wooldridge, J., Economeric analysis of cross secion and panel daa. MIT Press, Cambridge.

38 Table 1 Summary saisics of variables used in he analysis of CEO compensaion Table 1 summarizes he variables used in he analysis of CEO compensaion. Cash (salary plus bonus) and oal compensaions are provided by Execucomp. We compue equiy-based compensaion as he oal compensaion minus salary, bonus, oher annual pay, and long-erm incenive plan. Marke leverage is compued as (debs in curren liabiliies + long-erm deb)/( debs in curren liabiliies + long-erm deb + marke value of equiy). Alernaive marke leverage is compued as (debs due in 1 year + long-erm deb)/( debs due in 1 year + long-erm deb + marke value of equiy). Alernaive book leverage is compued as (debs due in 1 year + long-erm deb)/( debs due in 1 year + long-erm deb + book value of equiy). Marke capializaion is compued as he sock price muliplied by he number of shares ousanding as of he end of a fiscal year. Marke-o-book raio is he marke capializaion divided by he book value of equiy. All coninuous variables excep leverage are winsorized a he 1 s and 99 h perceniles. All dollar amouns are adjused o 1992 dollars using he consumer price index (CPI). N Mean Median Sd. Dev. 1% cuoff 99% cuoff Cash (Salary + bonus) (in housands) 14, ,531 Equiy-based compensaion (in housands) 14,891 1, , ,656 Toal compensaion including opions exercised (in housands) 14,891 2, , , ,099 Marke leverage 14, Alernaive marke leverage 14, Alernaive book leverage 14, Marke capializaion (in millions) 14,891 4, , ,204 Marke-o-book raio 14, One-year reurn o shareholders (%) 14, CEO age 14, CEO enure (Years as CEO in he firm) 14, CEO is male 14, CEO is also he Chairman 14, G-index 11,

39 Table 2 Ordinary leas square regressions of CEO compensaion Table 2 repors he coefficiens and sandard errors obained from OLS esimaion of he following model: CEOPay = γ + γ Size + γ Leverage + γ MTB + γ RET + γ Age 6 + γ Tenure γ Chair γ MALE + ε. CEOPay is measured in hree ways: he log of CEO oal compensaion, he log of CEO cash compensaion (salary plus bonus), and he log of equiy-based compensaion. Size is he log of marke capializaion of firm i as of year. Leverage, is he leverage raio of firm i as of year. Marke leverage, Alernaive marke i leverage, and Alernaive book leverage are defined he same as in Table 1. MTB i, is he marke-o-book raio of firm i as of year. RET is he reurn o he shareholders of firm i in year. Age is he age of he CEO of firm i as of year ; Tenure is he number of years he execuive has been acing as CEO in firm i prior o year ; Chair is one if he CEO is also he Chairman, and zero oherwise; MALE is one if he CEO of firm i as of year is male, and zero oherwise. Numbers in he parenheses are he sandard errors. The sandard errors are clusered by firm and are also robus o heeroskedasiciy. Regressions in Panel A use marke leverage, regressions in Panel B use alernaive marke leverage, and regressions in Panel C use alernaive book leverage. ***, **, and * indicae significance a he 1%, 5%, and 10% levels, respecively. Panel A: Marke Leverage oal compensaion (1) cash compensaion (2) equiy-based compensaion (3) 3 oal compensaion (4) 4 5 cash compensaion (5) equiy-based compensaion (6) Marke leverage 0.42*** 0.59*** 0.42** 0.43*** 0.55*** 0.45** (0.07) (0.05) (0.18) (0.07) (0.06) (0.20) Firm size 0.41*** 0.29*** 0.66*** 0.41*** 0.28*** 0.66*** (0.01) (0.01) (0.03) (0.01) (0.01) (0.03) Marke-o-book raio (0.005) -0.02*** (0.003) 0.023** (0.010) 0.011** (0.005) -0.01** (0.004) 0.02** (0.01) One-year reurn o shareholders (0.0002) (0.0001) 0.002*** (0.0004) 0.002*** (0.0002) (0.0001) 0.003*** (0.001) CEO age 0.006*** 0.006*** 0.018*** 0.006*** 0.004** 0.019*** (0.002) (0.001) (0.005) (0.002) (0.002) (0.005) CEO enure ** (0.003) (0.002) (0.006) (0.003) (0.002) (0.01) CEO is also he Chairman 0.18*** (0.03) 0.14*** (0.02) 0.24*** (0.07) 0.18*** (0.03) 0.14*** (0.02) 0.24*** (0.07) CEO is male (0.13) (0.08) (0.31) (0.13) (0.09) (0.31) G-Index *** 0.02*** 0.05*** (0.005) (0.004) (0.01) Year effecs Yes Yes Yes Yes Yes Yes Indusry effecs Yes Yes Yes Yes Yes Yes Inercep 3.15*** 3.73*** -1.57** 2.91*** 3.68*** -1.86** (0.23) (0.21) (0.62) (0.31) (0.30) (0.74) Number of observaions 14,891 14,891 14,891 11,527 11,527 11,527 R-squared

40 Panel B: Alernaive Marke Leverage oal compensaion (1) cash compensaion (2) Alernaive marke 0.41*** 0.58*** leverage (0.07) (0.05) Firm size 0.41*** 0.29*** (0.01) (0.01) Marke-o-book raio *** equiy-based compensaion (3) oal compensaion (4) cash compensaion (5) equiy-based compensaion (6) 0.40** (0.18) 0.43*** (0.08) 0.54*** (0.06) 0.45** (0.20) 0.66*** 0.41*** 0.28*** 0.66*** (0.03) (0.01) (0.01) (0.03) 0.023** 0.011** -0.01*** 0.02** (0.005) (0.003) (0.010) (0.005) (0.003) (0.01) One-year reurn o 0.002*** 0.002*** 0.003*** shareholders (%) (0.0002) (0.0001) (0.0004) (0.0002) (0.0001) (0.001) CEO age 0.006*** 0.006*** 0.018*** 0.006*** 0.004** 0.018*** (0.002) (0.001) (0.005) (0.002) (0.002) (0.005) CEO enure ** (0.003) (0.002) (0.006) (0.003) (0.002) (0.01) CEO is also he 0.18*** 0.15*** 0.24*** 0.18*** 0.14*** 0.25*** 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*** 0.05*** (0.005) (0.004) (0.01) Year effecs Yes Yes Yes Yes Yes Yes Indusry effecs Yes Yes Yes Yes Yes Yes Inercep 3.16*** 3.74*** -1.56** 2.92*** 3.69*** -1.85** (0.23) (0.21) (0.62) (0.31) (0.30) (0.74) Number of observaions 14,891 14,891 14,891 11,527 11,527 11,527 R-squared Panel C: Alernaive Book Leverage oal compensaion (1) cash compensaion (2) Alernaive book 0.25*** 0.49*** leverage (0.06) (0.04) Firm size 0.40*** 0.28*** (0.01) (0.01) Marke-o-book *** raio One-year reurn o shareholders (%) CEO age CEO enure CEO is also he Chairman CEO is male (0.005) (0.0002) 0.006*** (0.002) (0.002) 0.18*** (0.03) (0.13) (0.003) (0.0001) 0.005*** (0.001) (0.002) 0.14*** (0.02) (0.08) equiy-based compensaion (3) 0.29** (0.15) 0.65*** (0.02) 0.01 (0.01) 0.002*** (0.0004) 0.018*** (0.005) ** (0.006) 0.24*** (0.07) (0.31) G-Index oal compensaion (4) 0.26*** (0.07) 0.40*** (0.01) (0.006) 0.002*** (0.0002) 0.006*** (0.002) (0.003) 0.18*** (0.03) (0.13) 0.02*** (0.005) cash compensaion (5) 0.44*** (0.05) 0.27*** (0.01) -0.02*** (0.004) (0.0001) 0.004** (0.002) (0.002) 0.14*** (0.02) (0.09) 0.02*** (0.004) equiy-based compensaion (6) 0.34** (0.17) 0.65*** (0.03) 0.01 (0.01) 0.002*** (0.0005) 0.02*** (0.005) (0.01) 0.24*** (0.07) (0.31) 0.05*** (0.01) Year effecs Yes Yes Yes Yes Yes Yes Indusry effecs Yes Yes Yes Yes Yes Yes Inercep 3.23*** 3.83*** -1.49** 3.02*** 3.80*** -1.76** (0.22) (0.20) (0.62) (0.29) (0.29) (0.74) Number of observaions 14,891 14,891 14,891 11,527 11,527 11,527 R-squared

41 Table 3 Fama-MacBeh analysis of CEO compensaion Table 3 repors he coefficien of leverage obained from OLS regression of CEO pay in each fiscal year during : CEOPay = γ + γ Size + γ Leverage + γ MTB + γ RET + γ Age γ Tenure γ Chair γ MALE + ε. CEOPay is measured in hree ways: he log of CEO oal compensaion, he log of CEO cash compensaion (salary plus bonus), and he log of equiy-based compensaion. Size is he log of marke capializaion of firm i as of year. Leverage, is he leverage raio of firm i as of year. Marke leverage, Alernaive marke i leverage, and Alernaive book leverage are defined he same as in Table 1. MTB i, is he marke-o-book raio of firm i as of year. RET is he reurn o he shareholders of firm i in year. Age is he age of he CEO of firm i as of year ; Tenure is he number of years he execuive has been acing as CEO in firm i prior o year ; Chair is one if he CEO is also he Chairman, and zero oherwise; MALE is one if he CEO of firm i as of year is male, and zero oherwise. 4 Year compensaion measure compensaion measure compensaion measure oal cash equiy-based oal cash equiy-based oal cash equiy-based using marke leverage using alernaive marke leverage using alernaive book leverage Mean (-sa) 0.46 (15.73) 0.58 (19.39) 0.49 (4.84) 0.45 (17.33) 0.56 (18.48) 0.48 (4.65) 0.29 (9.69) 0.47 (19.55) 0.35 (4.11)

42 Table 4 Ordinary leas square regressions of he compensaion of newly-appoined CEOs who are hired from ouside Table 4 presens he coefficiens and sandard errors obained from esimaion of he following model of CEO compensaion on he sample of newly-appoined CEOs who are hired from ouside: CEOPay γ + γ Size + γ Leverage + γ MTB + γ RET + γ Age + γ Chair + γ MALE + ε. = CEOPay is measured in hree ways: he naural log of oal compensaion, he naural log of cash compensaion (salary plus bonus), and he naural log of equiy-based compensaion of he newly appoined CEO of firm i in year who is hired from ouside. Size -1 is he log of marke capializaion of firm i in year - 1; Leverage -1 is he leverage raio of firm i in year -1; Marke leverage, Alernaive marke leverage, and Alernaive book leverage are defined he same as in Table 1; MTB -1 is he marke-o-book raio of firm i in year -1; RET is he reurn o he shareholders of firm i in year ; Age is he age of he CEO of firm i as of year ; Chair is one if he CEO is also he Chairman of firm i in year, and zero oherwise; MALE is one if he CEO of firm i as of year is male, and zero oherwise. The numbers in parenheses are he sandard errors. The sandard errors are robus o heeroskedasiciy and are clusered by firm. *** and ** indicae significance a he 1% and 5% levels, respecively. Log of oal compensaion Log of cash compensaion Log of equiybased compensaion 1.11*** (0.41) Log of oal compensaion Log of cash compensaion Log of equiybased compensaion Log of oal compensaion Log of cash compensaion Log of equiybased compensaion Marke 0.75** 0.81*** leverage (0.30) (0.25) Alenaive 0.64** 0.71*** 0.97** marke (0.31) (0.26) (0.42) leverage Alernaive 0.62** 0.62*** 0.92** book leverage (0.30) (0.23) (0.41) Firm size 0.50*** 0.34*** 0.62*** 0.50*** 0.34*** 0.62*** 0.49*** 0.33*** 0.60*** (0.04) (0.03) (0.06) (0.04) (0.03) (0.06) (0.03) (0.03) (0.06) Marke-obook -0.07*** -0.04** -0.12*** -0.07*** -0.04** -0.13*** -0.08*** -0.05*** -0.14*** raio (0.02) (0.02) (0.03) (0.02) (0.02) (0.03) (0.02) (0.01) (0.03) One-year 0.002** * 0.004*** 0.002** * 0.004*** * 0.005*** reurn o (0.001) (0.0009) (0.001) (0.001) (0.0009) (0.001) (0.0012) (0.0009) (0.001) shareholders CEO age * * * (0.01) (0.01) (0.011) (0.01) (0.01) (0.011) (0.01) (0.01) (0.011) CEO is also he Chairman (0.13) (0.10) (0.18) (0.13) (0.10) (0.18) (0.13) (0.10) (0.19) CEO is male (0.63) (0.35) (0.90) (0.64) (0.37) (0.92) (0.64) (0.38) (0.91) Year effecs Yes Yes Yes Yes Yes Yes Yes Yes Yes Indusry effecs Yes Yes Yes Yes Yes Yes Yes Yes Yes Inercep 4.67*** (0.96) 4.79*** (0.59) 3.38** (1.44) 4.78*** (0.97) Number of observaions R-squared *** (0.60) 3.54** (1.45) 4.93*** (0.98) 5.10*** (0.61) 3.77** (1.46)

43 Table 5 Summary saisics of variables used in he analysis of average employee pay Table 5 summarizes he variables used in he analysis of average employee pay. Average employee pay is compued as labor expenses (daa 42) divided by he number of employees (daa 29). Marke leverage, Alernaive marke leverage, and Alernaive book leverage are defined he same as in Table 1. Marke capializaion is he sock price muliplied by he number of shares ousanding as of he fiscal year end. Average sales per employee is he amoun of oal sales (daa 12) divided by he number of employees. Marke-o-book raio is he marke capializaion divided by he book value of equiy. Physical capial inensiy is compued as gross propery, plan, and equipmen scaled by oal asses. Marginal ax rae (MTRB) is he marginal ax rae based on income before he deducion of ineres expenses. All coninuous variables excep leverage are winsorized a he 1 s and 99 h perceniles. All dollar amouns are adjused o 1992 dollars using he consumer price index (CPI). Number of observaions Mean Median Sandard deviaion 1% cuoff 99% cuoff Average Employee Pay (in housands) 5, Marke leverage 5, Alernaive marke leverage 5, Alernaive book leverage 5, Marke capializaion (in millions) 5,269 5, , ,827 Average sales per employee (in housands) 5, ,253 Marke-o-book raio 5, Physical capial inensiy 5, GIndex 1, Marginal ax rae (MTRB) 2,

44 Table 6 Ordinary leas square regressions of average employee pay Table 6 presens he coefficiens and sandard errors obained from he OLS regression of he following model of average employee pay: AEP = β 0 + β1sizei, + β2leveragei, + β3avgsalei, + β4mtbi, + β5pci + ε, where AEP is he log of average employee pay of firm i in fiscal year, and i is calculaed as he log of he oal labor expenses (daa 42) divided by he number of employees (daa 29); Size is he log of marke capializaion of firm i in year ; Leverage is he leverage raio of firm i in year. Marke leverage, Alernaive marke leverage, and Alernaive book leverage are defined he same as in Table 1. AvgSale is he average sales (in housand dollars) per employee, i.e., he amoun of oal sales divided by he number of employees; MTB is he marke-o-book raio of firm i in year ; and PCI is he physical capial inensiy of firm i in year, compued as gross propery, plan, and equipmen scaled by oal asses. Numbers in he parenheses are he sandard errors. The sandard errors are robus o heeroskedasiciy and are clusered by firm. In Panel A, we examine all firms. In Panel B, we sudy financially safe and disressed firms separaely. In Panel C, we add quis raes o he regressions. ***, **, and * indicae significance a he 1%, 5%, and 10% level, respecively. Panel A: All firms All firms All firms All firms All firms All firms All firms Marke leverage 0.23*** 0.28*** (0.08) (0.09) Alernaive marke 0.29** 0.28*** leverage (0.08) (0.10) -- Alernaive book 0.22*** 0.21** leverage (0.07) (0.09) Firm size 0.08*** 0.08*** 0.07*** 0.04** 0.04** 0.030* (0.01) (0.01) (0.01) (0.02) (0.02) (0.017) Average sales per employee (0.0002) (0.0001) (0.0001) (0.0002) (0.0002) (0.0002) Marke-o-book raio (0.003) (0.004) ** (0.004) (0.005) (0.005) (0.006) Physical capial inensiy 0.05 (0.06) 0.04 (0.06) 0.04 (0.06) (0.09) (0.09) (0.09) GIndex (0.01) (0.01) (0.01) Year effecs Yes Yes Yes Yes Yes Yes Indusry effecs Yes Yes Yes Yes Yes Yes Inercep 1.53*** 1.53*** 1.57*** 2.88*** 2.89*** 2.94*** (0.07) (0.06) (0.06) (0.18) (0.18) (0.18) Number of observaions 5,269 5,269 5,269 1,326 1,326 1,326 R-squared

45 Panel B: Safe versus disressed firms Disressed firms Disressed firms Disressed firms Safe firms Safe firms Safe firms Marke leverage ** (0.12) (0.10) Alernaive marke *** leverage (0.11) (0.11) -- Alernaive book ** leverage (0.10) (0.07) Firm size 0.08*** 0.08*** 0.08*** 0.07*** 0.07*** 0.07*** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Average sales per employee 0.002*** (0.0003) 0.002*** (0.0003) 0.002*** (0.0003) (0.0001) (0.0001) (0.0001) Marke-o-book raio (0.005) (0.005) (0.006) * (0.004) * (0.004) ** (0.004) Physical capial inensiy (0.07) (0.07) (0.07) (0.05) (0.05) (0.05) Year effecs Yes Yes Yes Yes Yes Yes Indusry effecs Yes Yes Yes Yes Yes Yes Inercep 2.16*** 2.09*** 2.17*** 2.48*** 2.49*** 2.50*** (0.17) (0.16) (0.16) (0.11) (0.11) (0.11) Number of observaions ,197 2,197 2,197 R-squared Panel C: Quis raes All firms All firms All firms All firms All firms Marke leverage *** (0.13) Alernaive marke 0.55*** leverage (0.13) -- Alernaive book 0.44*** leverage (0.11) Firm size *** 0.09*** 0.08*** (0.02) (0.02) (0.02) Average sales per employee (0.0002) (0.0002) (0.0002) Marke-o-book raio (0.007) (0.007) (0.008) Physical capial inensiy (0.09) (0.09) (0.09) Quis rae *** (0.002) (0.008) (0.01) (0.01) (0.01) Year effecs -- Yes Yes Yes Yes Indusry effecs -- Yes Yes Yes Yes Inercep 3.88*** 1.73*** 1.07*** 1.06*** 1.17*** (0.07) (0.23) (0.25) (0.24) (0.24) Number of observaions 1,993 1,993 1,993 1,993 1,993 R-squared

46 Table 7 Fama-MacBeh analysis of average employee pay Table 7 repors he coefficien on leverage obained from he OLS regression of average employee pay for each fiscal year in : AEP = β 0 + β1sizei, + β2leveragei, + β3avgsalei, + β4mtbi, + β5pci + ε, where AEP is he log of average employee pay of firm i in fiscal year, and i is calculaed as he log of he oal labor expenses (daa 42) divided by he number of employees (daa 29); Size is he log of marke capializaion of firm i in year ; Marke leverage, Alernaive marke leverage, and Alernaive book leverage are defined he same as in Table 1; AvgSale is he average sales per employee, i.e., he amoun of oal sales divided by he number of employees; MTB is he marke-o-book raio of firm i in year ; and PCI is he physical capial inensiy of firm i in year, compued as gross propery, plan, and equipmen scaled by oal asses. Year Marke leverage Alernaive marke leverage Alernaive book leverage Mean (-sa) 0.22 (3.96) 0.27 (4.14) 0.20 (4.12)

47 Table 8 Insrumenal variable regressions of average employee pay Table 8 presens he coefficiens and sandard errors obained from he wo-sage insrumenal variable regression: Leverage = α + α MTRB + α Size + α Avgsale + α MTB + α PCI α 6 ( EBIT / TA) + α 7STD( EBIT / TA) + δ, AEP = β + β Size + β Leverage + β AvgSale β 6( EBIT / TA) + β 7STD( EBIT / TA) + ε. Leverage in he second sage is insrumened by marginal ax rae based on income before ineres expense has been deduced (MTRB). EBIT/TA is earnings before depreciaion, ineres, and axes divided by oal asses, and STD (EBIT/TA) is he sandard deviaion of EBIT/TA in he pas five years. Marke leverage, Alernaive marke leverage, and Alernaive book leverage are defined he same as in Table 1. Numbers in he parenheses are he sandard errors. The sandard errors are robus o heeroskedasiciy and are clusered by firm. ***, **, and * indicae significance a he 1%, 5%, and 10% level, respecively. Firs sage: Leverage is he dependen variable Variable Marginal ax rae (MTRB) Firm size Average sales per employee Marke-o-book raio Physical capial inensiy EBIT/TA STD(EBIT/TA) Marke leverage 0.11*** (0.04) *** (0.002) ( ) *** (0.001) 0.101*** (0.015) -0.80*** (0.05) 0.01 (0.01) β MTB Alernaive marke leverage 0.14*** (0.04) *** (0.002) ( ) *** (0.001) 0.122*** (0.015) -0.76*** (0.05) 0.01 (0.01) β PCI Alernaive book leverage 0.20*** (0.05) (0.002) ( ) 0.018*** (0.001) 0.152*** (0.018) -0.96*** (0.06) 0.01 (0.02) Year effecs Yes Yes Yes Indusry effecs Yes Yes Yes Inercep 0.38*** (0.04) 0.29*** (0.04) 0.19*** (0.04) Number of observaions 2,902 2,902 2,902 R-squared Parial F-es of MTRB: F-saisic (p-value) (0.0005) (0.0011) (0.0001)

48 Second sage: Average employee pay is he dependen variable Variable Average employee pay Average employee pay Average employee pay Marke leverage (insrumened) Alernaive marke leverage (insrumened) Alernaive book leverage (insrumened) Firm size Average sales per employee Marke-o-book raio Physical capial inensiy EBIT/TA STD(EBIT/TA) 2.77* (1.49) ** (1.02) *** (0.03) (0.0001) 0.01 (0.01) 0.01 (0.16) 1.48 (1.13) 0.02 (0.05) 0.11*** (0.02) (0.0001) 0.01 (0.01) 0.02 (0.13) 0.90 (0.72) 0.02 (0.05) ** (0.70) 0.07*** (0.01) (0.0001) -0.01** (0.01) 0.05 (0.12) 0.75 (0.63) 0.02 (0.04) Year effecs Yes Yes Yes Indusry effecs Yes Yes Yes Inercep Number of observaions 1.39** (0.62) 1.82*** (0.35) 2.14*** (0.35) 2,902 2,902 2,902 R-squared

49 Table 9 Heckman wo-sep analysis of average employee pay Table 9 repors he coefficiens and sandard errors obained from a Heckman wo-sep analysis of average employee pay. In he firs sep, we esimae a probi model of wheher a firm repors labor expenses. The dependen variable is one if he daa on labor expenses is non-missing, and zero oherwise; he independen variables include he dummies of he firm s lising exchange, in addiion o oher firm characerisics. In he second sep, we examine he impac of leverage on average employee pay. The inverse mills raio (Lambda) derived from he selecion model is included in he second sep as a regressor. Numbers in he parenheses are he sandard errors. The sandard errors are robus o heeroskedasiciy and are clusered by firm. Marke leverage, Alernaive marke leverage, and Alernaive book leverage are defined he same as in Table 1. *** and ** indicae significance a he 1% and 5% level, respecively. Firs sage: Probi model of firms reporing informaion on labor expenses Variable Firm size Marke-o-book raio Marke leverage Coefficien (Sandard error) 0.26*** (0.01) -0.03*** (0.003) 0.30*** (0.05) Alernaive marke leverage -- Coefficien (Sandard error) 0.26*** (0.01) -0.03*** (0.003) Coefficien (Sandard error) 0.26*** (0.01) -0.03*** (0.003) *** (0.05) Alernaive book leverage Average sales per employee Physical capial inensiy *** (0.0001) 0.48*** (0.03) *** (0.0001) 0.49*** (0.03) *** (0.04) *** (0.0001) 0.49*** (0.03) Exchange dummies Joinly significan Joinly significan Joinly significan Year effecs Yes Yes Yes Indusry effecs Yes Yes Yes

50 Second sage: Average employee pay is he dependen variable Variable Marke leverage Coefficien (Sandard error) 0.20*** (0.04) Alernaive marke leverage -- Coefficien (Sandard error) Coefficien (Sandard error) *** (0.04) Alernaive book leverage Firm size Average sales per employee Marke-o-book raio Physical capial inensiy 0.08*** (0.004) (0.0001) (0.003) 0.047* (0.027) 0.08*** (0.004) (0.0001) (0.003) (0.027) *** (0.04) 0.07*** (0.004) (0.0001) ** (0.003) (0.027) Year effecs Yes Yes Yes Indusry effecs Yes Yes Yes Inercep Inverse mills raio (Lambda) 1.56*** (0.33) *** (0.012) 1.55*** (0.33) *** (0.012) 1.59*** (0.33) *** (0.012) Number of observaions 49,357 49,357 49,357 Censored observaions 44,088 44,088 44,088 Uncensored observaions 5,269 5,269 5,269 Wald chi-square (p-value) 5, (0.0000) 5, (0.0000) 5, (0.0000)

51 Table 10 CEO compensaion in echnology and non-echnology firms: univariae ess Table 10 summarizes and compares CEO compensaion and firm characerisics in echnology and nonechnology firms. Technology firms are defined as companies wih primary SIC designaions of 3570, 3571, 3572, 3576, 3577, 3661, 3674, 4812, 4813, 5045, 5961, 7370, 7371, 7372, and Non-echnology firms are firms wih SIC codes less han 4000 no oherwise caegorized as echnology firms. Toal and Cash (salary plus bonus) compensaions are provided direcly by Execucomp. We compue equiy-based compensaion as he oal compensaion minus salary, bonus, oher annual pay, and long-erm incenive plan. Marke leverage, Alernaive marke leverage, and Alernaive book leverage are defined he same as in Table 1. Marke capializaion is he sock price muliplied by he number of shares ousanding as of he fiscal year end. Marke-o-book raio is he marke capializaion divided by he book value of equiy. All coninuous variables excep leverage are winsorized a he 1 s and 99 h perceniles. All dollar amouns are adjused o 1992 dollars using he consumer price index (CPI). Non-echnology firms Mean (Median) Technology firms Mean (Median) Number of observaions 8,527 2,345 -es -sa (p-value) Kruskal-Wallis es Chi-square (p-value) Toal compensaion (in housands) 2,690 (1,259) 3,310 (988) Cash compensaion (in housands) 1,025 (791) 780 (539) Equiy-based compensaion (in housands) Marke leverage 1,512 (202) 0.21 (0.17) 2,439 (171) 0.08 (0.01) (0.01) 1, Alernaive marke leverage 0.20 (0.15) 0.07 (0.01) , Alernaive book leverage 0.32 (0.32) 0.13 (0.02) , Marke capializaion (in millions) 4,585 (972) 6,413 (905) 3.42 (0.0006) 5.48 (0.02) Marke-o-book raio 3.30 (2.40) 4.36 (3.06) One-year reurn o shareholders (%) (10.22) (10.02) (0.82) Years as CEO in he firm 5.99 (4.00) 6.10 (4.00) 0.67 (0.50) 1.97 (0.16) CEO age (67.00) (60.00) CEO is also he Chairman CEO is male 0.67 (1.00) 0.99 (1.00) 0.51 (1.00) 0.97 (1.00)

52 Table 11 OLS regressions of CEO compensaion in echnology and non-echnology firms Panel A and Panel B repor he esimaed coefficiens and sandard errors obained from OLS regressions of CEO compensaion in echnology and nonechnology firms, respecively. CEOPay = γ + γ Size + γ Leverage + γ MTB + γ RET + γ Age + γ Tenure + γ Chair + γ MALE + ε Marke leverage, Alernaive marke leverage, and Alernaive book leverage are defined he same as in Table 1. The numbers in he parenheses are he sandard errors. The sandard errors are robus o heeroskedasiciy and are clusered by firm. We use Wald es o examine wheher he coefficien on leverage is saisically differen in he regressions of he wo groups. ***, **, and * indicae significance a he 1%, 5%, and 10% levels, respecively. Panel A: CEO compensaion in non-echnology firms Log of oal compensaion (1) Log of cash compensaion (2) Log of equiy-based compensaion (3) Marke leverage 0.51*** (0.09) 0.61*** (0.07) 0.51*** (0.23) Log of oal compensaion (4) Log of cash compensaion (5) Log of equiy-based compensaion (6) Log of oal compensaion (7) Log of cash compensaion (8) Log of equiy-based compensaion (9) Alernaive marke leverage *** (0.09) 0.61*** (0.07) 0.53** (0.23) Alernaive book leverage Firm size 0.41*** 0.31*** 0.64*** 0.41*** 0.31*** 0.65*** 0.40*** 0.30*** 0.64*** (0.01) (0.01) (0.03) (0.01) (0.01) (0.03) (0.01) (0.01) (0.03) Marke-o-book raio 0.013** -0.01*** 0.04*** 0.010** -0.01*** 0.04*** *** 0.028** (0.005) (0.003) (0.01) (0.005) (0.003) (0.01) (0.005) (0.003) (0.013) One-year reurn o shareholders (%) (0.0002) (0.0001) 0.002*** (0.0006) (0.0002) (0.0001) 0.002*** (0.0006) (0.0002) (0.0001) 0.002*** (0.0006) Years as CEO in he firm (0.003) (0.002) (0.01) (0.003) (0.002) (0.01) (0.003) (0.002) (0.01) CEO age 0.005** *** 0.005** *** 0.005** *** (0.002) (0.002) (0.006) (0.002) (0.002) (0.006) (0.002) (0.002) (0.006) CEO is also he Chairman 0.20*** 0.16*** 0.26*** 0.21*** 0.16*** 0.26*** 0.21*** 0.16*** 0.26*** (0.03) (0.02) (0.08) (0.03) (0.02) (0.08) (0.03) (0.02) (0.08) CEO is male (0.18) (0.14) (0.42) (0.18) (0.14) (0.42) (0.19) (0.14) (0.42) Year effecs Yes Yes Yes Yes Yes Yes Yes Yes Yes Indusry effecs Yes Yes Yes Yes Yes Yes Yes Yes Yes Inercep 2.97*** 3.62*** -1.71** 2.97*** 3.62*** -1.72** 3.07*** 3.72*** -1.62** (0.27) (0.23) (0.71) (0.27) (0.23) (0.72) (0.27) (0.23) (0.71) Number of observaions 8,527 8,527 8,527 8,527 8,527 8,527 8,527 8,527 8,527 R-squared *** (0.08) 0.46*** (0.05) 0.37** (0.19)

53 Table 11 (coninued) Panel B: CEO compensaion in echnology firms Log of oal Log of cash compensaion compensaion (1) (2) Marke leverage (0.24) 0.56*** (0.16) Log of equiy-based compensaion (3) (0.66) Log of oal compensaion (4) Log of cash compensaion (5) Log of equiy-based compensaion (6) Log of oal compensaion (7) Log of cash compensaion (8) Log of equiy-based compensaion (9) Alernaive marke leverage (0.25) 0.47*** (0.17) (0.67) Alernaive book leverage (0.18) Firm size 0.40*** 0.26*** 0.65*** 0.40*** 0.26*** 0.65*** 0.40*** 0.26*** 0.65*** (0.03) (0.02) (0.07) (0.03) (0.02) (0.07) (0.03) (0.02) (0.07) Marke-o-book raio *** *** *** (0.01) (0.01) (0.03) (0.01) (0.01) (0.03) (0.01) (0.01) (0.03) One-year reurn o shareholders (%) * (0.0005) (0.0003) (0.001) * (0.0004) (0.0003) (0.001) * (0.0005) (0.0003) (0.001) Years as CEO in he firm (0.007) (0.005) (0.017) (0.007) (0.005) (0.017) (0.007) (0.005) (0.017) CEO age 0.010* 0.012*** 0.03*** 0.010* 0.011*** 0.03*** 0.011* 0.011*** 0.032** (0.006) (0.004) (0.01) (0.006) (0.004) (0.01) (0.006) (0.004) (0.013) CEO is also he Chairman 0.13* 0.12** * 0.12** * 0.12** 0.14 (0.07) (0.05) (0.17) (0.07) (0.05) (0.17) (0.07) (0.05) (0.17) CEO is male (0.30) (0.10) (0.66) (0.30) (0.10) (0.67) (0.30) (0.10) (0.67) Year effecs Yes Yes Yes Yes Yes Yes Yes Yes Yes Indusry effecs Yes Yes Yes Yes Yes Yes Yes Yes Yes Inercep 3.61*** 4.36*** -1.93* 3.60*** 4.37*** -1.95* 3.58*** 4.41*** (0.47) (0.28) (1.12) (0.47) (0.28) (1.12) (0.47) (0.28) (1.13) Number of observaions 2,345 2,345 2,345 2,345 2,345 2,345 2,345 2,345 2,345 R-squared Wald es wheher he coefficien of leverage is differen beween echnology and non-echnology firms: Chi-square (p-value) (0.0000) 1.24 (0.26) (0.0010) (0.0000) 1.39 (0.24) (0.0015) (0.0001) 0.38*** (0.13) 1.15 (0.28) (0.46) 6.99 (0.0082)

54 Table 12 Comparison of average employee pay in echnology and non-echnology firms: univariae ess Table 12 summarizes and compares average employee pay and financial characerisics in echnology and non-echnology firms. Technology firms are defined as companies wih primary SIC designaions of 3570, 3571, 3572, 3576, 3577, 3661, 3674, 4812, 4813, 5045, 5961, 7370, 7371, 7372, and Non-echnology firms are firms wih SIC codes less han 4000 no oherwise caegorized as echnology firms. Average employee pay is compued as labor expenses (daa 42) divided by he number of employees (daa 29). Marke leverage, Alernaive marke leverage, and Alernaive book leverage are defined he same as in Table 1. Marke capializaion is he sock price muliplied by he number of shares ousanding as of he fiscal year end. Average sales per employee is he amoun of oal sales (daa 12) divided by he number of employees (daa 29). Marke-o-book raio is he marke capializaion divided by he book value of equiy. Physical capial inensiy is compued as gross propery, plan, and equipmen scaled by oal asses. All coninuous variables excep leverage are winsorized a he 1 s and 99 h perceniles. All dollar amouns are adjused o 1992 dollars using he consumer price index (CPI). Non-echnology firms Mean (Median) Technology firms Mean (Median) -es -sa (p-value) Kruskal-Wallis es Chi-square (p-value) Number of observaions 2, Average employee pay (in housands) (40.01) (34.56) 1.55 (0.12) 8.62 (0.003) Marke leverage 0.23 (0.19) 0.11 (0.17) Alernaive marke leverage 0.20 (0.16) 0.09 (0.01) Alernaive book leverage 0.33 (0.32) 0.14 (0.03) Marke capializaion (in millions) 10,298 (2,631) 2,303 (187) Average sales per employee (in housands) (174.55) (130.40) Marke-o-book raio 3.17 (2.24) 3.41 (2.16) 0.99 (0.32) 1.02 (0.31) Physical capial inensiy 0.72 (0.67) 0.34 (0.25)

55 Table 13 Leverage and average employee pay: OLS regressions for echnology and non-echnology firms This able repors esimaed coefficiens and sandard errors obained from OLS regressions of average employee pay for echnology and non-echnology firms. AEP = β 0 + β1sizei, + β2leveragei, + β3avgsalei, + β4mtbi, + β5pci + ε. Marke leverage, Alernaive marke leverage, and Alernaive book leverage are defined he same as in Table 1. The numbers in parenheses are he sandard errors. The sandard errors are robus o heeroskedasiciy and are clusered by firm. We use Wald es o examine wheher he coefficien on leverage is saisically differen across he wo groups. *** and ** indicae significance a he 1% and 5% level, respecively. Log of average employee pay in non-echnology firms (1) Log of average employee pay in echnology firms (2) Log of average employee pay in non-echnology firms (3) Log of average employee pay in echnology firms (4) Log of average employee pay in non-echnology firms (5) Log of average employee pay in echnology firms (6) Marke leverage Alernaive marke leverage Alernaive book leverage Firm size Average sales per employee Marke-o-book raio Physical capial inensiy 0.29* (0.18) (0.53) *** (0.20) (0.64) *** (0.02) (0.0003) (0.005) (0.13) 0.23*** (0.06) 0.002** (0.001) -0.06*** (0.02) 0.03 (0.31) 0.10*** (0.02) (0.0003) (0.005) (0.13) 0.23*** (0.06) 0.002** (0.001) -0.06*** (0.02) 0.03 (0.30) *** (0.17) 0.09*** (0.02) (0.0002) (0.006) (0.13) (0.48) 0.23*** (0.06) 0.002** (0.001) -0.06*** (0.02) 0.03 (0.31) Year effecs Yes Yes Yes Yes Yes Yes Indusry effecs Yes Yes Yes Yes Yes Yes Inercep 1.59*** (0.13) 2.40** (0.38) 1.55*** (0.12) 1.81** (0.75) 1.59*** (0.11) 1.86** (0.76) Number of observaions 2, , , R-squared Wald es wheher he coefficien of leverage is differen beween echnology and non-echnology firms: Chi-square (p-value) 2.00 (0.16) 3.87 (0.05) 4.07 (0.04)

56 Fig. 1. Changes in oal labor expenses and ax benefi of deb as leverage increases. This graph plos he changes in oal labor expenses and ax benefi of deb (in million $) for a firm wih median values of leverage raio, average employee pay, oal labor expenses, and oal deb in he sample of Panel A, Table 6. We sar wih he median value of marke leverage, 0.20, and increase i by 0.48 (more han wo sandard deviaions), by incremens of 0.04.