Opton Reprcng and Incentve Realgnment Jeffrey L. Coles Department of Fnance W. P. Carey School of Busness Arzona State Unversty Jeffrey.Coles@asu.edu Tel: (480) 965-4475 Naveen D. Danel Department of Fnance Robnson College of Busness Georga State Unversty nav@gsu.edu Tel: (404) 651-2691 Laltha Naveen Department of Fnance Robnson College of Busness Georga State Unversty laltha@gsu.edu Tel: (404) 651-2632 Frst Draft: September 23, 2002 Ths verson: June 8, 2004 Correspondng author s Jeffrey L. Coles, Department of Fnance, Arzona State Unversty, P. O. Box 873906, Tempe, AZ 85287. Offce: (480) 965-4475; Fax: (480) 965-8539. We thank semnar partcpants at Atlanta Fnance Workshop for helpful comments.
Opton Reprcng and Incentve Realgnment Abstract We provde evdence that frms reprce out-of-the-money executve stock optons n order to realgn manageral ncentves. A sharp declne n stock prce, by reducng the senstvty of executve pay to frm performance (delta) and, n many cases, ncreasng senstvty of executve pay to stock-return volatlty (vega), can cause manageral ncentves to depart from optmal or target levels. Our results suggest that ncreasng delta does not appear to be a strong motvaton for reprcng. Rather, we fnd strong evdence that frms reprce executve optons to reduce rsk-takng ncentves (vega) toward target levels.
Opton Reprcng and Incentve Realgnment 1. Introducton The practce of re-settng the exercse prce of employee stock optons has attracted consderable crtcsm from nsttutonal nvestors and the popular press. Crtcs contend that opton reprcng s symptomatc of agency problems n the frm. Snce reprcng typcally follows a perod of poor stock-prce performance, the clam s that managers n fact are beng rewarded for performng poorly. Reprcng frms, on the other hand, clam that they do so n order to retan valuable members of the management team and to restore manageral ncentves followng stock prce declnes. Whle these explanatons for reprcng are not mutually exclusve, the evdence we provde n ths study s consstent wth the noton that reprcng represents, at least n part, an attempt to realgn manageral ncentves, partcularly the manageral ncentve to take rsk. The use of equty-based compensaton, n the form of stock and optons, has exploded n recent years (Murphy, 1999, and Perry and Zenner, 2000). One effect has been a substantal ncrease n the senstvty of CEO wealth to stock prce (compare Hall and Lebman, 1998, to Jensen and Murphy, 1990). The senstvty of CEO wealth to stock prce, or delta, s seen as algnng the ncentves of managers wth the nterests of shareholders. Hgher delta can mean that managers wll work harder or more effectvely because managers share gans and losses wth shareholders (Murphy, 1999). Of course, another effect of ncreased delta s to expose managers to more rsk. To the extent that managers are relatvely undversfed wth respect to frm-specfc wealth, they are more rsk averse compared to dversfed shareholders. Accordngly, t s possble that managers wll forego some postve-npv projects f those projects are very rsky 1
(Amhud and Lev, 1981; Smth and Stulz, 1985). A second aspect of the ncrease n equty-based compensaton potentally offsets ths tendency. Assocated wth the ncrease n opton grants and holdngs has been an ncrease n the senstvty of CEO wealth to stock-return volatlty, or vega. Opton compensaton, by provdng convex payoffs, potentally can reduce averson to rsky value-ncreasng choces. 1 Recent emprcal evdence supports ths noton. For example, Coles, Danel, and Naveen (2003) fnd that hgher vega mplements rsker fnancal and nvestment polcy choces, ncludng relatvely more nvestment n R&D, less nvestment n tangble assets, more focus on fewer lnes of busness, hgher leverage, and hgher volatlty of stock returns. 2 In the absence of sgnfcant mpedments to contractng, shareholders can structure the manageral compensaton scheme to furnsh managers wth valuemaxmzng delta and vega. But a change n stock prce and/or changes n frm characterstcs can cause both delta and vega to depart from ther target values. For nstance, gven that optons typcally are granted at the money (Murphy, 1999), and supposng that new stock and opton grants are ssued to brng manageral ncentves closer to ther target or optmal levels (Core and Guay, 1999; L, 2002), a stock-prce declne subsequent to a grant could move ncentves away from ther target levels. Several studes have argued that vega, or rsk-takng ncentves, ncrease when optons go 1 Numerous papers have argued that convex payoffs can mtgate the effect of CEO rsk averson and provde the CEO wth ncreased ncentves to take on rsky projects. For example, see Jensen and Mecklng (1976), Myers (1977), Haugen and Senbet (1981), Smth and Stulz (1985), Smth and Watts (1992), Gaver and Gaver (1993), Bzjak, Brckley, and Coles (1993), and Guay (1999). Despte the ntutve appeal of ths argument, ts valdty depends on the manageral utlty functon. As llustrated n Guay (1999), convexty of the payoff structure (e.g., from optons) can be more than offset by concavty of the utlty functon of the rsk-averse manager. Along the same lnes, see Ju, Leland, and Senbet (2002) and Ross (2003). Brado and Ferrera (2003) demonstrate that under certan condtons a call opton could be wrtten that makes all managers prefer rsker projects. 2 Also see Cohen, Hall, and Vcera (2000), Rajgopal and Shevln (2002), Knopf, Nam, and Thornton (2002), and Rogers (2002). 2
underwater. 3 Delta, on the other hand, declnes when optons go underwater, although Jn and Muelbrook (2002) show that the declne need not be substantal. Alternatvely, the optmal or target level of ncentves tself could change. Ths could happen, for nstance, f any of the frm characterstcs that affect ncentves change, such as frms nvestment opportuntes. 4 In ether case, manageral ncentves devate from ther deal level. How mght frms return delta and/or vega to target levels? Core and Guay (1999, p. 150), n reference to pay-performance senstvty, or delta, fnd frms use annual grants of optons and restrcted stock to CEOs to manage the optmal level of equty ncentves. 5 L (2002) provdes complementary evdence that, on average, the frm, through equty and opton grants, and the CEO, through portfolo rebalancng, jontly manage delta towards a target level. Resettng the strke prce of executve optons s another possble strategy. For example, Acharya, John, and Sundaram (2000) and Hall and Murphy (2000) suggest that resettng executve optons could be a sutable way to ncrease delta. Panel A of Fgure 1 shows how opton delta ncreases when the strke prce s reset to a lower level whle holdng stock prce constant (thereby ncreasng the prce-to-strke rato). Of course, the frm also can ncrease delta by grantng the executve new optons or stock, so reprcng and any assocated negatve publcty would be unnecessary. 3 See for nstance, Lambert, Larcker, and Verreccha (1991), Parrno, Poteshman and Wesbach (2002), Basak, Pavlova, and Shapro (2003), Carpenter (2000), and Glson and Vetsuypens (1993). 4 See Bzjak, Brckley, and Coles (1993) and Core and Guay (1999) for the effect of nvestment opportuntes on delta, and Guay (1999) for the effect of nvestment opportuntes on vega. 5 Whle stock grants typcally ncrease only delta, opton grants ncrease both delta and vega. Of course, to the extent that stocks are call optons on the frm s assets, they also provde some rsk-takng ncentves, but Guay (1999) shows that ths s generally nsgnfcant he fnds that the vega from an opton s many tmes larger than that from the underlyng stock. 3
On the other hand, t can be more dffcult to reduce vega should t be desrable to do so. An executve, by exercsng optons, can reduce vega (and ncrease delta), but ths may not be feasble followng a stock prce declne as the optons are lkely to be out-ofthe-money. And even f the executve has n-the-money optons, the board cannot force hm to exercse such optons. In many nstances, therefore, reprcng may be the most mmedately vable mechansm to reduce vega. Along these lnes, L (2002), as well as Glson and Vetusypens (1993) and Carpenter (2000), suggest that reprcng followng poor stock performance can reduce excessve rsk-takng ncentves. Panel B of Fgure 1 llustrates ths pont. The fgure shows how vega changes wth strke prce holdng stock prce constant. Pror to reprcng, the prce-to-strke rato s typcally around 0.45 (Carter and Lynch, 2001). Immedately after reprcng, the prce-to-strke rato s typcally 1.0. The downward adjustment of strke prce assocated wth reprcng ncreases the prce-tostrke rato and reduces vega unless prce-to-strke s extremely low (see Panel B). We examne whether frms reprce employee optons n order to manage target manageral ncentves. In partcular, followng all Execucomp frms over the perod 1992-2000, we examne CEO opton reprcng wth explct consderaton of whether t s done n response to a departure of manageral delta and vega from target levels. We adopt a modfed verson of Core and Guay (2002) methodology to estmate actual CEO ncentves, and follow the approach of Core and Guay (1999) to model target ncentves. The orgnal Core and Guay (2002) methodology has one potental lmtaton n that the average exercse prce s understated n the presence of out-of-the-money optons. Ths s an mportant consderaton gven that reprcers wll have underwater optons. Our modfed approach (dscussed n detal n Secton 3) provdes us more precse estmates of opton-moneyness and manageral ncentves. 4
To model target (or optmal) delta and vega, we follow Core and Guay (1999) and Guay (1999), and use the resduals from the model to measure the departure from target levels. A postve (negatve) resdual mples that the ncentve measure s hgher (lower) than the target level. Our central result s that the propensty to reprce employee optons s postvely related to the vega resdual. In contrast, there appears to be very lttle evdence that the propensty to reprce s negatvely related to the delta resdual. In our data, reprcng s more lkely when vega exceeds the target level, but a departure of delta from the target level has essentally no explanatory power. Varous alternatve explanatons have been offered n the lterature for reprcng, such as rent extracton and agency problems, employee retenton, and reprcng as a mechansm for compensatng executves n cashconstraned frms. Our results hold even after controllng for these alternatve explanatons. Thus, our evdence suggests that reprcng represents, at least n part, an attempt to realgn manageral ncentves, specfcally the ncentve to mplement rsky polces. These results contrbute to the lterature n several ways. Frst, our evdence suggests that reprcng s part of the frm s overall compensaton polcy. Our results complement research that suggests frms manage executve ncentves n a ratonal manner (Core and Guay, 1999; L, 2002). Second, by consderng delta and vega separately nstead of opton-moneyness (as n Carter and Lynch, 2001), we solate the specfc dmenson of ncentve realgnment that motvates reprcng. Thrd, n contrast to Rogers (2003), we consder the devaton from target ncentves rather than the level of ncentves. Rogers (2003) provdes weak evdence that reprcng s more lkely n hghvega frms. Our results suggest that t s the devaton from target ncentves that drves 5
reprcng. In partcular, even after controllng for the target levels of ncentves, we fnd that the probablty of reprcng s postvely related to the devaton from target vega. Fourth, whle equty and opton grants appear to be gven to move delta toward a target level (Core and Guay, 1999), the propensty to reprce does not appear to be related to the devaton from target delta. One possble explanaton resdes n Jn and Muelbrook (2001). They fnd that delta does not decrease much even for sgnfcant declnes n stock prce and, thus, there s lttle need to reprce optons n order to ncrease delta. Fnally, our sample sze of over 4700 frm-year observatons s qute large and represents many ndustres. In comparson, the only other study to consder both rsk-takng ncentves and reprcng, Rogers (2003), uses a sample of 26 casnos. The rest of the paper s organzed as follows. Secton 2 provdes addtonal dscusson of the reprcng lterature and our hypotheses. Secton 3 descrbes the data and Secton 4 detals the methodology to estmate the devatons from target level of vega and target level of delta. Secton 5 dscusses our man results. Secton 6 shows the results are robust to varous alternatve specfcatons. Secton 7 concludes. 2. Dscusson of the Lterature, Hypotheses, and Methods Potental reasons for reprcng employee stock optons nclude rent extracton, employee retenton, nformaton sgnalng, and ncentve realgnment. Among others, Brenner, Sundaram, and Yermack (2000), Chance, Kumar, and Todd (2000), Carter and Lynch (2001, 2003a), Gren, Hand, and Klassen (2003), Callaghan, Saly, and Subramanam (2003a), and Chdambaran and Prabhala (2003a, b) all provde evdence on the mportance of agency problems, entrenchment, and rent extracton, though wth 6
dfferng conclusons. 6 Chakraborty, Shekh, and Subramanan (2003), Carter and Lynch (2003b), Chen (2003), Callaghan, Saly, and Subramanam (2003b), and Chdambaran and Prabhala (2003a) examne the connecton between reprcng and ether manageral turnover or retenton. Gren, Hand, and Klassen (2003) fnd lttle evdence that the act of reprcng sgnals nformaton to the market that future performance wll be better than expected. Several studes focus on the potental connecton between reprcng and manageral ncentve realgnment. Acharya, John, and Sundaram (2000) present a model that llustrates how the benefts of reprcng through an ex post ncrease n delta more than offset the ex ante dluton of ncentves arsng from the expectaton of reprcng. On the emprcal sde, the evdence relatng ncentve realgnment and reprcng tends to be ndrect. For example, Carter and Lynch (2001) fnd that the lkelhood of reprcng s postvely related to the extent to whch optons are out-of-the-money, whch can be nterpreted as evdence that reprcng s motvated by the desre to realgn manageral ncentves. Moneyness of the opton, however, whle related to pay-performance senstvty, s an mperfect measure of delta (Fgure 1, Panel A). Moreover, moneyness also s related to vega n a non-monotonc fashon (Fgure 1, Panel B). Chdambaran and Prabhala (2003a) compare CEO ncentve packages across frms that nclude the CEO n the reprcng versus those that do not. They fnd that, n frms that exclude the CEO n the reprcng, the CEO s portfolo contans relatvely more 6 The methods employed n these papers and others nclude event studes (e.g., of reprcng or turnover), cross-sectonal regresson of event abnormal returns on frm characterstcs, estmaton of the reprcng wealth transfer from shareholders to executves, and qualtatve response models of the propensty to reprce or lkelhood of turnover. Rght-hand sde varables nclude moneyness of employee optons, pror stock returns, board structure, leadershp structure, ownershp structure, other governance characterstcs, proxes for agency problems, and turnover probablty. 7
stock than optons. Chdambaran and Prabhala nterpret ths result as weak evdence that ncentve realgnment plays a role n reprcng. On the other hand, Jn and Muelbrook (2001) fnd that delta decreases only margnally even for sgnfcant declnes n stock prce. In reference to delta, they conclude that restorng ncentve-algnment s seldom a good justfcaton for resettng the stock prce or ssung new opton grants. Chakraborty, Shekh, and Subramanan (2003) fnd evdence consstent wth ths proposton. In ther sample, the potental ncrease n delta based on a reprcng has lttle power to explan the probablty that a frm actually does reprce employee optons. Smlarly, Chdambaran and Prabhala (2003b) fnd that changes n delta followng reprcng appear small. Fnally, to our knowledge, Rogers (2003) s the only other paper that attempts to lnk the reprcng decson to rsk-takng ncentves, but hs study has several lmtatons. Frst, Rogers, usng executve-level regressons, fnds no relaton between the reprcng decson and the rato of vega-to-delta. When he measures rsk-takng ncentves as ether the rato of the vega-to-cash-compensaton, or as the product of vega-to-delta rato and vega-to-cash-compensaton rato, he fnds a postve relaton between the reprcng decson and hs measures of rsk-takng ncentves. 7 Thus the Rogers (2003) model consders the level of (scaled) ncentves as beng the prmary determnant of reprcng. We argue, however, that the level of ncentves s not mportant when frms make reprcng decsons; t s the devaton from target ncentves that s more mportant. Second, the level of ncentves as used by Rogers could merely be proxyng for some 7 Rogers also uses frm-level regressons, where the dependent varable s one f the frm reprces stock optons of any of the top 5 executves, and the ndependent varables are based on aggregate ncentves of all the top 5 executves. The frm s decson to reprce any partcular executve s stock optons, however, s lkely to be related to that executve s ncentves, and not necessarly to aggregate top-management ncentves, renderng an economc nterpretaton of these regressons dffcult. 8
frm characterstc, such as growth opportuntes, whch have been shown to affect both vega and delta. Thrd, Rogers follows the Core and Guay (2002) method for computng ncentves. As Core and Guay pont out, ths method must be modfed for underwater optons. Rogers does not address ths ssue, whereas we use a modfed verson of Core and Guay approach to mtgate ths problem. Fnally, hs results are based on a small sample of frms (26 frms) operatng n one dstnctve (casno) ndustry. In contrast, we use all Execucomp frms from 1992-2000. In ths paper, we examne whether reprcng s a response to msalgnment of ncentves provded by the compensaton contract. Our methodology for estmatng devatons from target ncentves s smlar to that of Core and Guay (1999). Frst, we use data on the manager s entre portfolo of stock and optons to obtan a comprehensve measure of the ncentves faced by managers. Specfcally, we estmate vega and delta usng a refnement to the methodology suggested by Core and Guay (2002) to account for underwater optons. Delta s the senstvty of the executve s wealth to stock prce and captures the manageral ncentve to ncrease stock prce. Vega s the senstvty of the executve s wealth to stock volatlty and captures the manageral ncentve to take rsks. Second, we estmate separate cross-sectonal regressons for each of delta and vega, where the ndependent varables are frm and CEO characterstcs dentfed n the lterature as mportant determnants of these ncentves. Our regresson specfcatons are smlar to Core and Guay (1999) and Guay (1999). Thrd, from these regressons, we obtan the predcted values of delta and vega for the CEO, whch proxy for optmal or target ncentves for the CEO (Core and Guay, 1999). The resduals from these regressons, n each case the actual value of the ncentve measure mnus the predcted level, provde an estmate of the extent to whch the CEO s delta and vega are msalgned. 9
A postve (negatve) resdual mples that the CEO has ncentves that are above (below) target levels. Both opton reprcng and addtonal opton and equty grants can ncrease delta to the target level. Therefore, the probablty of reprcng should be negatvelyrelated to the lagged departure from target delta. If vega s too low, then addtonal opton grants can ncrease vega, but f vega s too hgh, opton reprcng s lkely to be the most vable mechansm for correctng the msalgnment of rsk-takng ncentves. Thus, the propensty to reprce should be postvely related to the lagged departure from target vega. In the followng analyss, our specfcaton and results suggest that t s resdual vega, rather than resdual delta, that has sgnfcant power to explan the reprcng decson. Reprcng appears to mtgate excessve rsk-takng ncentves. 3. Data descrpton and summary statstcs We use the October 2001 verson of Standard & Poor s Execucomp database to obtan data on CEO compensaton. Most studes on reprcng such as Brenner et al. (2000), Chdamabaran and Prabhala (2003a, 2003b), Callaghan et al. (2003a, 2003b), Chakraborty et al. (2003) use data from Execucomp. Execucomp provdes CEO data on salary, bonus, and annual grants and holdngs of stocks and optons. The data are provded for frms n the S&P 500, S&P Mdcap 400, and S&P Smallcap 600 for the perod 1992-2000. We start wth executves who are dentfed by Execucomp as CEOs. We verfy the Execucomp nformaton usng frms proxy statements (from Lexs-Nexs). Execucomp ndcates the dates when the CEO assumed offce and when the CEO qut offce. In some cases, Execucomp fals to dentfy an executve as the CEO even though he or she appears to be the CEO based on these dates. We classfy these ndvduals also 10
as CEOs. Execucomp also ndcates whether a frm reprces stock optons n a gven year. We then rely on the proxy statements to dentfy whether t s the CEO s optons that were reprced n a gven year and, f so, we classfy these observatons as reprcers. All others are classfed as non-reprcers. Consstent wth the pror lterature, we elmnate fnance frms and utlty frms. We obtan other frm-specfc nformaton from Compustat and stock return nformaton from CRSP. 3.1. Modfcaton of Core and Guay(2002) methodology to compute ncentves Our methodology to calculate delta and vega s based on a modfed verson of Core and Guay (2002). Core and Guay (2002) suggest a one-year approxmaton method for calculatng ncentves. Usng ths method, the vega and delta for the executve s entre portfolo of stock and optons can be estmated usng aggregated data from the most recent year s proxy. They fnd that delta and vega calculated usng ther method has a 99% correlaton wth the true underlyng values. One potental lmtaton of ther methodology (as the authors themselves pont out) s that the average exercse prce s understated n the presence of out-of-the-money optons. Whle ths may be an mportant consderaton n any study of reprcng, gven that reprcers wll have underwater optons, ts mpact on our study s lmted for the followng reasons. Frst, Core and Guay suggest that the extent of the bas s less mportant f the study deals wth cross-sectonal varaton n ncentves, rather than focusng on magntude of ncentves (page 624). Snce, n our study, we are nterested n the cross-sectonal varaton n ncentves, the potental for bas s offset to a large extent. Second, they suggest that controllng for the observed prce-to-strke rato n the regresson wll reduce the extent of the bas. Our results are robust to ncludng prce-tostrke rato as an addtonal control varable. Fnally, n our study, we are nterested n 11
devaton from target ncentves (measured as actual mnus target ncentves), and t s not clear that ths s based even f the actual ncentves themselves are based. Nevertheless, to remove any potental for bas, we refne the Core and Guay (2002) approach usng a methodology smlar to that suggested n ther paper (page 624). Specfcally, we use data on opton grants awarded n the most recent three years to more precsely estmate the average exercse prce. The advantage of ths method s that we know the exact strke prce for optons granted n the last three years. In the orgnal Core and Guay method, exact strke prce s known only for grants n the most recent year, and the strke prce for the portfolo of remanng optons s nferred from the proxy data. In dong so, ther method assumes strke prce equals market prce for any underwater optons n the portfolo. Snce we use three years of opton-grant data, we know the exact strke prce for these three years, and we only have to nfer the strke prce for any remanng opton holdngs. 8 Ths mnmzes the error n estmatng the strke prce and hence the ncentves. Incentves calculated usng our method have an over 95% correlaton wth that calculated usng the orgnal Core and Guay (2002) method. Ths correlaton s the same for both reprcers and non-reprcers. Whle all our results are based on modfed ncentves, they are robust to usng the orgnal Core and Guay (2002) method. Ths s not surprsng gven the hgh correlaton between ncentves computed usng the two methods. 8 The choce of 3-year wndow s based on emprcal research that shows that vestng perod of a typcal opton s between two and three years (Huddart and Lang, 1996; Carpenter, 1998). 12
A complete dscusson of our methodology s provded n Appendx 1. 9 We defne delta as the change n the dollar value of the CEO s wealth for a one percentage pont change n stock prce and vega as the change n dollar value of the CEO s wealth for a 0.01 change n the annualzed standard devaton of stock returns. 10 3.2. Varable Defntons and Summary Statstcs Table 1 reports the breakdown of reprcng actvty by year. We fnd the overall frequency of reprcng s about 1.3%, whch s smlar to that reported n Brenner, Sundaram, and Yermack (2000). The results ndcate a declne n reprcng actvty after 1998. Ths could be attrbuted to the 1998 FASB rulng (Interpretaton 44: fnal rule July 1, 2000; retroactve to December 15, 1998) that requres companes to expense any stock optons that they reprce (see Carter and Lynch, 2003a, for a more detaled dscusson of the FASB rule and ts mplcatons). Table 2 presents unvarate statstcs for reprcng and non-reprcng frms. The determnants of the reprcng decson and the target levels of ncentves are defned as follows. () We use the logarthm of sales to proxy for frm sze. () Market-to-book s the market value of assets scaled by the book value of assets, s our proxy for nvestment opportuntes. () R&D s research and development expendture scaled by book value of assets. If R&D expenses are mssng, we defne R&D as zero, as n Bzjak, Brckley, and Coles (1993), and other studes. (v) CAPEX s captal expendture less sale of 9 As explaned n the appendx, delta and vega are computed based on the Black-Scholes opton valuaton model. Ths s consstent wth numerous recent papers such as Yermack (1995); Hall and Lebman (1998); Aggarwal and Samwck (1999); Core and Guay (1999, 2002); Guay (1999); Cohen, Hall, and Vcera (2000); Datta, Iskander-Datta, and Raman (2001); and Rajagopal and Shevln (2002), among others. There s some recent debate, however, as to whether the Black-Scholes model s sutable for valung employee stock optons snce these are non-transferrable and employees are rsk-averse and typcally do not hold the optons untl expraton (see Carpenter, 1998; Hall and Murphy, 2002; Ingersoll, 2002; Betts, Bzjak, and Lemmon, 2002). 10 Guay (1999) shows that opton vega s many tmes hgher than stock vega. Consequently, n ths study, we use the vega of the opton portfolo to measure the total vega of the stock and opton portfolo. Both Knopf et al. (2002) and Rajgopal and Shevln (2002) adopt the same approxmaton. 13
property, plant, and equpment scaled by book value of assets. (v) Leverage s total book debt scaled by book value of assets. (v) Volatlty s the logarthm of daly stock return varance. (v) Surplus cash s the amount of cash avalable to fnance new projects, scaled by assets (see Rchardson, 2002). (v) ROA s return on assets, defned as EBITDA scaled by book value of assets. (x) Stock return and lagged stock return correspond to annual return over the fscal year. (x) Free cash flow s operatng cash flow less common and preferred dvdends scaled by book value of assets. (x) Frm age s the logarthm of the number of years snce the frst tradng date on CRSP. (x) Cash compensaton s the sum of CEO s salary and bonus. (x) CEO and Charman s an ndcator varable that takes the value 1 f the CEO s also the Charman of the board, and 0 otherwse. See Appendx 2 for a complete descrpton of varables. From Table 2 t can be seen that compared to non-reprcers, n the year pror to reprcng, reprcers tend to be sgnfcantly smaller (medan sales of reprcers s $333 mllon versus $877 mllon for non-reprcers) and more volatle (medan standard devaton of daly returns s 3.53% versus 2.38%). The rato of R&D expense to assets s hgher for reprcers (medan = 5.9%) than non-reprcers (medan = 0.3%). Ths s consstent wth the fndng of Carter and Lynch (2001) that reprcers tend to operate n hgh-tech ndustres. Reprcers also tend to exhbt poorer performance n the year pror to reprcng; the medan lagged ROA of reprcers s 11.8% compared to 15.2% for nonreprcers and the medan lagged stock returns are -14.9% for reprcers compared to 14.1% for non-reprcers. The medan lagged ndustry-adjusted performance measures are also sgnfcantly lower for reprcers compared to non-reprcers. These results are consstent wth the pror studes on reprcng mentoned above. 14
4. Determnants of target vega and target delta Our man hypothess s that reprcng serves to realgn ncentves. To estmate the magntude of ncentve msalgnment, we follow Core and Guay (1999). 4.1. Estmatng target delta We regress delta on frm and CEO characterstcs and use the predcted value as the proxy for the target level of delta for each frm-year observaton. We report robust t- statstcs n parentheses beneath the coeffcent estmates. Delta = 0.440 + 0.168 * Frm sze + 0.101 * Market-to-book 0.273 * R&D ( 1.9) (25.4) (10.3) ( 2.0) + 0.376 * CAPEX 0.158 * Leverage + 0.113 * Volatlty (2.7) ( 3.1) (8.7) +0.088 * Surplus cash + 0.029 * CEO tenure (0.8) (22.1) + 2-dgt SIC dummes + Year dummes Number of observatons = 5638, R 2 = 37% Our explanatory varables are based on Core and Guay (1999) and Coles, Danel, and Naveen (2003) and the results are generally smlar to these papers. 11 In our sample, delta s postvely related to frm sze and to market-to-book, a commonly used proxy for growth opportuntes. Delta s postvely related to CAPEX, and negatvely related to R&D and leverage. Ths s consstent wth the evdence n Coles, Danel, and Naveen (2003), who fnd that delta s related nversely to rskness of polcy choces. We fnd a 11 See Demsetz and Lehn (1985), Smth and Watts (1992), Bzjak, Brckley, and Coles (1993), Gaver and Gaver (1993), Mehran (1995), Aggarwal and Samwck (1999), Hmmelberg, Hubbard, and Pala (1999), Pala (2001), Baker and Hall (2002), and Core and Guay (2002) for addtonal studes on the determnants of delta. 15
postve relaton between delta and the volatlty of stock returns, whch s consstent wth frms wth greater dffculty n montorng beng gven hgher delta (Demsetz and Lehn, 1985). Fnally, we fnd that delta ncreases n CEO tenure, whch s consstent wth greater accumulaton of stock and optons by CEOs and mtgaton of horzon problems (see Gbbons and Murphy, 1992; Brckley, Coles, and Lnck, 1999). The resdual from the above regresson gves the devaton from target delta. 4.2. Estmatng target vega To estmate target vega, we estmate a regresson specfcaton smlar to those n Guay (1999) and Coles, Danel, and Naveen (2003). Agan, our results are generally smlar to both these papers. Our regresson results are reported below wth robust t- statstcs gven n parentheses beneath the coeffcent estmates. Vega = 0.110 + 0.014 * Frm sze + 0.004 * Market-to-book + 0.059 * R&D (-14.5) (20.9) (8.7) (5.3) + 0.006 * CAPEX 0.004 * Leverage + 0.042 * Cash compensaton (0.6) ( 1.1) (24.9) + 0.009 * Volatlty + 2-dgt SIC dummes + Year dummes (8.9) Number of observatons = 7049, R 2 = 49% Guay (1999) argues that vega should be hgher n frms wth hgh growth opportuntes so as to provde rsk-averse managers ncentves to undertake rsky but postve-npv projects. Consstent wth ths, we fnd that vega s postvely related to market-to-book. Vega s postvely related to rsker polcy choces such as R&D and leverage as n Coles, Danel, and Naveen (2003). Vega ncreases n frm sze, as n Guay 16
(1999) and Coles, Danel, and Naveen (2003). Consstent wth the latter paper, we also fnd that vega ncreases wth cash compensaton and wth volatlty. The resdual from ths equaton gves the devaton from target vega. 5. Results on reprcng decsons and ncentve algnment In ths secton, we provde detaled evdence on the role of reprcng n realgnng manageral ncentves. A postve (negatve) resdual from ether the delta or the vega regresson mples that the CEO has hgher (lower) than target ncentves. As we outlne n Secton 1, f the CEO s rsk-takng ncentves, or vega, s below target, then t s easy to ncrease t by provdng more stock optons. On the other hand, f the CEO s vega s above target, then reprcng s one of the few vable mechansms by whch vega can be reduced. In ths case, we would expect the probablty of reprcng to be postvely related to the vega resdual as of the begnnng of the year. Smlarly, a negatve resdual from the delta regresson ndcates that the senstvty of CEO wealth to shareholder value (delta) s lower than target levels. Reprcng s one of many mechansms by whch frms can ncrease delta. If an mportant motvaton for reprcng s to ncrease delta, we would expect the probablty of reprcng to be negatvely related to the delta resdual as of the begnnng of the year. 5.1. Incentves around reprcng event Table 3 presents the summary statstcs for CEO ncentves around the reprcng event (year 0). An examnaton of Table 3 ndcates that n the two years pror to reprcng, unadjusted vega for reprcers s lower than for non-reprcers. Also, vega contnues to ncrease after reprcng for both reprcers and non-reprcers. These results are nconsstent wth Rogers (2003). Resdual vega (devaton from target vega), 17
however, s postve and sgnfcantly hgher (p < 0.01) for reprcers compared to nonreprcers n the year pror to reprcng. The resdual vega for reprcers n the year pror to reprcng s $17,419 whch also appears to be economcally sgnfcant relatve to average vega ($49,343). In the year after reprcng, resdual vega decreases to $2,571 and s not sgnfcantly dfferent (p = 0.73) from the resdual vega for non-reprcers. These unvarate results are consstent wth our hypothess that excess vega s assocated wth ncreased lkelhood of reprcng. Resdual delta of reprcers s postve (hgher than target) n the years pror to reprcng and becomes negatve subsequent to reprcng. Moreover, the resdual delta for reprcers s not sgnfcantly dfferent from that of non-reprcers (p > 0.24) n any of the years. These fndngs are nconsstent wth the hypothess that frms reprce to ncrease delta towards ts target level. 5.2. Logstc regresson results Table 4 presents logstc regressons of the reprcng decson. The dependent varable s one f the CEO s optons are reprced n a gven year, and zero otherwse. We use lagged values of both resdual vega and resdual delta as our prmary explanatory varables. Model 1 ncludes only the two resdual ncentve varables, whle Models 2 and 3 also nclude control varables that have prevously been used to explan the probablty of reprcng. In all three models, we nclude 18 ndustry dummes, formed as n Chdambaran and Prabhala (2003a), and also year dummes. 12 We use R&D to proxy 12 We fnd that 48% of reprcers are concentrated n three ndustres categorzed as computer and electronc parts, software and hgh technology, and botechnology ndustres. Ths s consstent wth Chdambaran and Prabhala (2003a), who report a 37% concentraton of reprcers n these three ndustres. 18
for hgh-tech ndustres. 13 Also, Acharya, John, and Sundaram (2000) suggest that reprcng s approprate for frms when management can sgnfcantly affect the dstrbuton of frm returns. Managers are lkely to be more mportant n frms wth large growth optons. We use R&D and market-to-book as proxes. 14 Reprcng s also more lkely when the CEO has more power (Acharya, John, and Sundaram, 2000). We use CEO tenure to proxy for CEO power. In all three specfcatons, the parameter estmates on lagged resdual vega are postve and sgnfcant at 5% or better (p = 0.002, p = 0.013, and p = 0.013, respectvely). Ths s consstent wth our hypothess that frms reprce n order to reduce excessve rsktakng ncentves. In contrast, n all three specfcatons, the coeffcent on resdual delta s negatve but statstcally nsgnfcant at conventonal levels. These results for both vega and delta are consstent wth the unvarate results tabulated n Table 3. In terms of economc sgnfcance, devaton from target vega appears to be an mportant determnant of the lkelhood of reprcng. For example, n the computer and electronc parts ndustry n 1998 (the ndustry and year wth most reprcngs), an ncrease n resdual vega from ts 25 th to ts 75 th percentle (keepng all other varables at ther medan values), ncreases the probablty of reprcng by 30% relatve to the reprcng probablty of the medan frm. Devatons from target delta, on the other hand, appear to be less mportant. For the same ndustry-year combnaton, an ncrease n the resdual delta from ts 25 th to ts 75 th percentle decreases the probablty of reprcng by 12%. 13 Carter and Lynch (2001) use a hgh-tech dummy for frms that are ncluded n the 1998 CorpTech Drectory of Technology. They fnd that R&D for hgh-tech frms s sgnfcantly hgher than for non-hghtech frms. 14 Chance, Kumar, and Todd (2000) also use market-to-book n the logstc regresson. 19
For the control varables, consstent wth pror lterature, we fnd that reprcng s more lkely n smaller frms, more volatle frms, and frms that have had poor stock returns n the past. The coeffcent on market-to-book, CEO tenure, and R&D appear to have no explanatory power. In summary, our prmary result s that frms reprce to realgn one mportant dmenson of manageral ncentves. In partcular, the evdence suggests that frms reprce n order to reduce rsk-takng ncentves. In contrast, there s lttle evdence that frms reprce to realgn pay-performance senstvty. 5.3. Testng alternatve explanatons for reprcng Alternatve explanatons for reprcng have been suggested n the lterature. In ths secton, we demonstrate that our results reman after controllng for these alternatve explanatons. Frst, when labor markets are tght, frms may reprce to retan executves (Carter and Lynch, 2001). Chdambaran and Prabhala (2003a) argue that management turnover may be costler n younger frms as they do not have well-developed lnes of successon, and therefore such frms may be more lkely to reprce to retan CEOs. Consequently we control for frm age n our regressons. The results are shown n Model 1 of Table 5. The coeffcent on frm age s sgnfcantly negatve, ndcatng that younger frms are more lkely to reprce (consstent wth Chdambaran and Prabhala, 2003a and Carter and Lynch, 2001). Second, Dechow, Hutton, and Sloan (1996) and Yermack (1995) suggest that frms may use stock and optons as substtutes for cash compensaton. Thus, cashconstraned frms may be more lkely to reprce to adjust dollar compensaton levels, rather than to algn ncentves. We therefore use CEO cash compensaton as an 20
addtonal control varable n Model 2. We fnd that the coeffcent of cash compensaton s not sgnfcantly related to the reprcng decson. Rogers (2003) uses the level of vega (scaled by cash compensaton, or by delta) as a proxy for rsk-takng ncentves. As we argue throughout our paper, t s the devaton from target ncentves, and not the level of target ncentves that s lkely to affect the reprcng decson. The (weakly) postve relaton between vega and the reprcng decson documented by Rogers could be due to hs falure to control for market-to-book rato n hs regressons (whch other studes such as Chance et al. (2000) have shown to be mportant). Vega, therefore, could be merely proxyng for growth opportuntes. Nevertheless, n Model 3, we control for target vega and target delta (predcted values from ndvdual regressons of vega and delta). We fnd that target delta s postvely related to the probablty of reprcng, whle target vega does not seem to mpact reprcng. Fnally, agency problems have been proposed as a potental explanaton for reprcng (Chance et al., 2000; Brenner et al., 2000; Chdambaran and Prabhala, 2003a). We therefore nclude free cash flow and an ndcator varable that takes the value 1 f the CEO s also the charman of the board as proxes for agency problems. The results are shown n Models 4 and 5. We fnd that nether of these varables s sgnfcantly related to reprcng. An alternate proxy for agency problems n the frm s excess CEO compensaton. Core, Holthausen, and Larcker (1999) fnd that CEOs are pad more n frms wth poor governance structures. We therefore use excess total drect CEO compensaton nstead of free cash flow n the regressons (where total drect compensaton s defned as the sum of salary, bonus, value of opton grants, value of restrcted stock grants, value of long-term ncentve payouts, and all other annual compensaton). Excess total drect compensaton s the resdual from a regresson of total 21
drect compensaton on frm sze, market-to-book, stock and accountng performance, CEO tenure and ndustry and year dummes. We fnd that the coeffcent on excess CEO compensaton s postvely related to reprcng, whch suggests that reprcng s more lkely n frms wth more agency problems, but the coeffcent s not statstcally sgnfcant (p = 0.25). The devaton from resdual vega, however, contnues to be statstcally sgnfcant (p = 0.01). Model 6 consders all of the above varables ncluded n the same regresson. As before we fnd frm age and target delta to be sgnfcantly related to reprcng. In all models, the coeffcent on the devaton from target vega contnues to be statstcally sgnfcant at 5% or better. Ths suggests that realgnng rsk-takng ncentves s a sgnfcant determnant of the reprcng decson. The coeffcent on the devaton from target delta s always negatve but has no explanatory power. 5.4. Results usng matched sample of reprcers The regresson results n Tables 4 and 5 are based on all non-reprcng frm-years as a control sample for the reprcers. Whle ths s consstent wth the methodology n Brenner, Sundaram, and Yermack (2000), other papers n the lterature use control samples matched on varous crtera. The dea s to fnd a control frm wth a smlar ncentve to reprce, rather than controllng for the varables deemed to affect reprcng on the rght-hand sde n a multvarate specfcaton. For example, the control frm n Carter and Lynch (2001) s a randomly selected frm wth underwater executve optons. Chance, Kumar, and Todd (2000) choose a control frm wthn the same 4-dgt SIC code as the reprcng frm and wth a percentage declne n stock prce smlar to the percentage declne of the reprcng frm. 22
Whle havng underwater optons, or havng negatve stock returns, appears to be a necessary condton for reprcng, nether of these two events mply that the executve s vega has rsen above or that delta has fallen below target levels. For nstance, we fnd that the correlaton between return n a gven year and the change n resdual vega durng the year s only -9% whle the correlaton between returns and change n resdual delta s only +8%. Nevertheless, we choose three dfferent control samples based on the exstng lterature. The frst control sample conssts of a subset of frms that have underwater optons n the year pror to reprcng. 15 The second control sample conssts of a matched set of frms from the same ndustry, wth a smlar prce declne n the year pror to reprcng. The thrd control sample conssts of a matched set of frms, agan from the same ndustry, where the CEO s opton portfolo has a smlar prce-to-strke rato n the year pror to reprcng. Table 6 reports logstc regresson results for the control samples descrbed above. The frst column n each Panel corresponds to the most comprehensve model of Table 4 (Model 3) and the second column n each Panel corresponds to the most comprehensve model of Table 5 (Model 6). Panel A of Table 6 reports the logstc regresson results usng the subsample of non-reprcers wth underwater optons. In Panel B, the match s based on lagged prce-tostrke rato and n Panel C, the match s based on lagged stock-return. The results n all sx specfcatons ndcate that the probablty of reprcng s sgnfcantly postvely related (p 0.032) to the devaton from target vega. In contrast, we fnd that the 15 As mentoned earler, we use the latest three years proxy data on opton grants to compute ncentves. We know the exact exercse prces of the opton grants n each of these three years and hence are able to correctly dentfy whether any of these optons are underwater. In the case of optons granted pror to these three years whch are stll unexercsed by the CEO, we are unable to dentfy whether they are underwater due to data lmtatons (see Appendx 1 for more detals). Therefore every frm n our control sample has some underwater optons, but not every frm wth underwater optons s ncluded n the sample. 23
coeffcent on resdual delta contnues to be negatve but s statstcally sgnfcant only n Panel A. Ths result s (weakly) consstent wth frms reprcng to ncrease the delta towards target levels. None of the control varables has sgnfcant explanatory power across all specfcatons. In three of sx specfcatons, the coeffcent on volatlty s sgnfcantly postve whle that on frm sze s sgnfcantly negatve. In two of the three specfcatons, the coeffcent on target delta s sgnfcantly postve whle that on frm age s sgnfcantly negatve. These results are consstent wth earler tables. In summary, our results based on matched control samples support our earler conclusons that reprcng s postvely related to devaton from target vega but only weakly (at best) related to devaton from target delta. 6. Robustness In ths secton we consder the robustness of the reprcng results to alternatve specfcatons for the logstc regresson predctng reprcng, and alternatve specfcatons for target delta and vega. 6.1. Senstvty of logt regressons of reprcng decson to dfferent specfcatons We re-estmate the most comprehensve model of Table 4 (Model 3) and the most comprehensve model of Table 5 (Model 6) usng varous alternatve specfcatons and/or addtonal control varables. We nclude the average prce-to-strke rato of the opton portfolo n the year pror to reprcng to capture moneyness of the opton portfolo as n Carter and Lynch (2001). Also, as dscussed n Secton 2, Core and Guay (2002) suggest that ncludng the prce-to-strke rato wll reduce the measurement error n ncentves. 24
Also, nstead of usng logarthm of sales, we use dfferent proxes for frm sze, such as logarthm of book value of assets (Carter and Lynch, 2001), logarthm of market value of equty (Chance, Kumar, and Todd, 2002), and logarthm of market value of assets (Guay, 1999). Instead of frm returns, we use the medan ndustry return and the frm return net of medan ndustry return (Carter and Lynch, 2001). We also nclude logarthm of CEO age as a proxy for rsk averson (Guay, 1999). Addtonally, we use sales growth and change n ROA, both estmated over the prevous two years (as n Chdambaran and Prabhala, 2003a). Of the control varables, alternatve proxes for both frm sze and returns are sgnfcantly negatve, as documented n the lterature. The coeffcents on CEO age and sales growth are nsgnfcant, whle change n ROA s weakly negatve. In terms of the prmary varables of nterest, our man nferences are robust to these alternatve specfcatons. As n Table 4 and Table 5, the coeffcent on resdual vega s postve and statstcally sgnfcant and the coeffcent on resdual delta s negatve but statstcally nsgnfcant. We use 2-dgt SIC dummes rather than the 18 ndustry groups used by Chdambaran and Prabhala (2003a). Because reprcers are not present n all 2-dgt SIC ndustres, we lose many observatons. Further, we re-estmate the regresson coeffcents and the t-statstcs for the logstc regresson usng a Fama-Macbeth approach. 16 Agan, we lose observatons when we do the regressons year-by-year, as requred for ths approach, snce reprcng frms are not present n all 18 ndustry groups n every year. 16 We estmate the devatons from optmal delta and optmal vega usng annual regressons (nstead of a pooled regresson) so as to ensure the devatons sum to zero n each year. 25
All nferences concernng manageral ncentve realgnment reman unchanged except that resdual delta comes n sgnfcantly negatve n one specfcaton. 6.2. Senstvty of reprcng results to alternatve specfcatons of target delta and target vega regressons We next consder the robustness of our results to alternatve specfcatons of target delta and target vega and the calculaton of the correspondng resdual values. That s, we re-estmate the resduals from the alternatve delta and vega equatons dscussed below. In the target delta regresson, we use logarthm of market value of equty to proxy for sze (nstead of logarthm of sales), nclude logarthm of CEO age as an addtonal control, and use an alternatve defnton of free cash flow as n Core and Guay (1999). In the target vega equaton, as n Guay (1999), we nclude free cash flow and an ndcator varable for tax-loss-carry forwards to proxy for ncentves to hedge. In all cases, our man nferences are the same. We also use the same varables as n Core and Guay (1999) and Guay (1999) except that we use lagged values nstead of 3-year averages n the case of free-cash-flow and unsystematc rsk. 17 As n Baber, Janakraman, and Kang (1996), Gaver and Gaver (1993), and Guay (1999), we use factor analyss to construct a sngle varable - that s, a factor score whch captures varaton common to market-to-book rato, R&D, and CAPEX. In addton, we use log(vega) nstead of vega and log(delta) nstead of delta as the dependent varables. We use the 18 ndustry dummes (as n Chdambaran and 17 Core and Guay (1999) regress log(delta) on log(market value of equty), book-to-market rato, log(unsystematc rsk), log(ceo tenure), 3-year average of free cash flow scaled by book value of assets, ndustry dummes, and year dummes. Guay (1999) regresses vega on log(market value of assets), book-tomarket rato, the rato of R&D expenses to market value of assets, and the rato of nvestment expendtures (captal expendture + acqustons) to market value of assets, and delta. In some specfcatons, addtonally Guay (1999) uses salary and bonus, log(delta), and CEO age. 26
Prabhala, 2003a) n the target vega and target delta equatons nstead of 2-dgt SIC dummes. Agan, our results on resdual vega and resdual delta reman unaltered. In sum, these results as well as the evdence presented n Sectons 6.1 show that our nferences are robust to a wde varety of alternatve specfcatons. Our emprcal results provde strong evdence that frms reprce executve optons n order to realgn manageral ncentves for rsk-takng. 7. Conclusons Potental motves for reprcng nclude rent extracton, executve retenton, nformaton sgnalng, and ncentve algnment. We follow the approach of Core and Guay (1999) to examne whether frms reprce employee optons n order to manage optmal manageral ncentves. In partcular, the ncentves we consder are the senstvty of CEO wealth to frm performance (delta) and the senstvty of CEO wealth to stock-return volatlty (vega). We compute delta and vega usng a modfcaton to the Core and Guay (2002) approach, whch allows us to more precsely estmate ncentves n the presence of underwater optons. Followng all Execucomp frms over the perod 1992-2000, we examne employee opton reprcng wth explct consderaton of whether t s done n response to a departure of manageral delta and vega from target levels. As do Core and Guay (1999), we model target or optmal delta and vega for CEOs and use the resduals from the model to measure the departure from target levels. Our central result s that the propensty to reprce employee optons s postvely related to the vega resdual. Ths relaton s both statstcally and economcally sgnfcant. Ths s consstent wth the noton that reprcng s done to reduce the 27