Optimal CEO Incentives and Industry Dynamics

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Optimal CEO Incentives and Industry Dynamics Antonio Falato and Dalida Kadyrzhanova 2012-78 NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Optimal CEO Incentives and Industry Dynamics Antonio Falato Dalida Kadyrzhanova Federal Reserve Board R. H. Smith School of Business University of Maryland1 This Draft: October 2012 1Corresponding author: Antonio Falato, Federal Reserve Board - Division of Research and Statistics, Washington DC. Phone: (202) 452-2861. Email: antonio.falato@frb.gov. Conversations with John DonaldsonandMikeRiordanstimulatedourinterestinthetopicoftheequilibriumdeterminantsofmanagerial incentives. We acknowledge their encouragement at the early stages of the project. Comments from Tom Cooley, Xavier Gabaix, Kose John, Arvind Krishnamurthy, Pete Kyle, Vojislav Maksimovic, Gordon Phillips, Vincenzo Quadrini, Lemma Senbet, Raven Saks, and seminar participants at 2006 Meetings of the Society for Economic Dynamics (Vancouver), 2007 North American Summer Meetings of the Econometric Society (Duke University), Columbia University, University of Maryland (Smith School of Business), University of Montreal (HEC School of Business), University of Toronto, Federal Reserve Board, and FDIC are gratefully acknowledged. All reamining errors are ours. The analysis, conclusions, and discussion in this paper are those of the authors and do not indicate concurrence by othermembersoftheresearchsta⁄ortheBoardofGovernorsoftheFederalReserveSystem. References to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the authors to protect the tentative character of these papers.

Abstract This paper develops a competitive equilibrium model of CEO compensation and industry dynamics. CEOs make product pricing and product improvement decisions subject to shareholders(cid:146)compensation choices and idiosyncratic shocks to product quality. The choice of highpowered incentives optimally trades o⁄the bene(cid:133)ts from expected product improvements and the associated agency costs. In market equilibrium, the interaction between CEO pay and product market decisions a⁄ects the stationary distribution of (cid:133)rms. We characterize a dynamic feedback e⁄ect of industry structure on CEO incentives. As a result of this e⁄ect, we predict that the performance-based component of CEO pay should be higher, (i) across industries, when the degree of heterogeneity of industry structure is lower; (ii) within industries, when (cid:133)rms are laggards with respect to their industry peers. We empirically estimate payperformance sensitivity for a large sample of U.S. CEOs and other top executives over the 1993 to 2004 period and (cid:133)nd strong support for our theory. Our results o⁄er a novel product market rationale for the increased reliance of CEO pay on bonuses and stock options over the 1990s.

1 Introduction There are large di⁄erences in executive pay practices among (cid:133)rms (see Murphy (1999) for a comprehensive discussion of this fact). This paper argues, both theoretically and empirically, that dynamic competition - i.e., competition for industry leadership - can help us to make progress on the question of which fundamental economic forces drive these di⁄erences. To this end, we develop and test the cross-sectional implications of an industry equilibrium model of optimal CEO incentives. In a large sample of U.S. CEOs and other top executives between 1993 and 2004, we (cid:133)nd strong evidence in support of our theory, in that we document a robust negative relation between pay-performance sensitivity and, (i) across industries, the degree of heterogeneity of industry structure; (ii) within industries, (cid:133)rm position with respect to its peers. TheincentivestructureofCEOcompensationisacontroversialtopicthatattractsattention of both academic researchers and popular press. The classical view of CEO pay as an agency problem(Holmstrom(1979),HolmstromandMilgrom(1987))emphasizesthetrade-o⁄between incentives and insurance. According to this view, shareholders can ensure that CEOs take optimal actions by tying CEO pay to the performance of their (cid:133)rms, that is, by providing high-powered incentives for CEOs to maximize the returns to shareholders. However, the empirical literature has found a puzzling lack of evidence of high-powered incentives, which is typically interpreted to imply that CEOs are not given strong enough incentives to maximize the returns to shareholders (see Bebchuk and Fried (2004) for a forceful statement of this view). While recent studies (e.g., Hall and Liebman (1998)) document a general upward trend in high-powered incentives over the 1990s, compensation specialists and boards of directors remain vocal in decrying the weak link between executive pay and (cid:133)rm 1

performance in U.S. corporations. Wearguethatthis(cid:148)paywithoutperformance(cid:148)interpretationoftheevidenceisnotgranted. In particular, we propose an equilibrium interpretation of the observed di⁄erences in CEO incentive pay across (cid:133)rms. We propose that CEO compensation contracts are endogenously determinedbythecontractingenvironment,whichlikelydi⁄ersacross(cid:133)rmsandindustries. We show that relatively low pay-performance sensitivities emerge as the optimal incentive arrangement whenever returns to CEO e⁄ort are low. By emphasizing that the value of CEO e⁄ort is determined endogenously in industry equilibrium, we do not deny the importance of agency problems between stockholders and managers. Rather, our study complements the agency perspective by studying the role of equilibrium factors in shaping the contracting environment across (cid:133)rms. Our model formalizes the link between CEO pay and industry structure. We introduce a standard optimal CEO compensation problem (Holmstrom (1979), Holmstrom and Milgrom (1987)) into a dynamic industry equilibrium model with di⁄erentiated Bertrand competition (e.g., Ericson and Pakes (1995), Maskin and Tirole (2000)). Shareholders choose CEO pay, while CEOs make product market and e⁄ort choices. The key innovation of our analysis is that we use a structural model with heterogeneous (cid:133)rms, which di⁄er in the quality of their products. By working harder, CEOs can improve product quality (slowly) over time. The resulting competition is dynamic since CEOs of (cid:133)rms that are at early stages of product development must (cid:133)rst catch-up with the leading edge CEOs before battling for leadership in the future. Our focus is on characterizing the way the dynamic interaction between competitors a⁄ects shareholders(cid:146)optimal choice of CEO compensation contracts. Since industry structure ulti- 2

mately results from this dynamic interaction between competitors, the model allows for the simultaneous determination of CEO pay and industry structure. Thekeycross-sectionalpredictionofthemodelisaninverserelationbetweenthemagnitude of the performance-based component of optimal CEO contracts and (cid:133)rm competitive position within its industry. In particular, we predict that industry leaders have lower pay-performance sensitivity than laggards. To see the intuition for this result, consider that we depart from the commonassumptionofexogenousreturnstoCEOe⁄ort. Inourmodel,shareholdersvalueCEO e⁄ort since it increases expected future returns - i.e., it enables (cid:133)rms to grow. As (cid:133)rms climb the product quality-ladder over time, the value of growth opportunities falls. Consequently, leading edge CEOs are optimally given weaker incentives. By contrast, laggards with low quality products value growth opportunities the most and, thus, give stronger incentives to their CEOs. A second important prediction of the model is an inverse relation between the magnitude of the performance-based component of optimal CEO pay and the degree of heterogeneity of industry structure. In particular, we predict that pay-performance sensitivity will be lower in moreheterogeneousindustries, i.e. industriescharacterizedbyafringeanddominant(cid:133)rms. To seetheintuitionforthisresultcontrastthefollowingtwoscenarios(whichresultasequilibrium outcomes of our model): (cid:133)rst, an asymmetric industry, where a dominant (cid:133)rm enjoys a local monopoly power; second, a symmetric industry, where rivals engage in a neck-and-neck battle for leadership. It is exactly in the latter scenario that CEO e⁄ort is most valuable, since by growing faster (cid:133)rms can pull ahead of their rivals. Consequently, optimal pay-performance sensitivityishigherinsymmetricindustries. Finally,sinceanequilibriumoutcomeofourmodel is that symmetric industries tend to exhibit higher rates of growth, we can use the model to 3

derive a (cid:133)nal dynamic prediction: pay-performance sensitivity is higher in growing industries. WetestthesepredictionsempiricallyinalargepanelofU.S.CEOsandothertopexecutives between 1993 and 2005. We link two standard sources of data. Our compensation data is from ExecuComp and our (cid:133)rm data is from Compustat. We build indicators of (cid:133)rm position within its industry based on the ratio of (cid:133)rm sales to median industry sales. We de(cid:133)ne as leaders (cid:133)rms that are in the highest quartile of the distribution of the ratio, and laggards (cid:133)rms that are in the lowest quartile of the distribution of the ratio. Finally, we construct measures of industry turbulence and heterogeneity based on average job turnover within in industry and average distance of (cid:133)rm sales from median industry sales. These measures have been previously employed, although with a di⁄erent motivation, respectively in job turnover (see Davis, Haltiwanger, and Schuh (1996)) and capital structure studies (see Titman (1984), MacKay and Phillips (2005)). Consistent with our industry model of optimal CEO pay, we (cid:133)nd strong evidence of an inverse relationship between pay-performance sensitivity and, (i) across industries, the degree of heterogeneity of industry structure; (ii) within industries, (cid:133)rm position with respect to its peers. These (cid:133)ndings are robust to controlling for industry (cid:133)xed e⁄ects and other variables, such as (cid:133)rm size and industry concentration, that have been found to a⁄ect CEO incentives in previous studies (respectively, Schaefer (1998) and Baker and Hall (2004) for (cid:133)rm size and Aggarwal and Samwick (1999) for concentration). Importantly, the (cid:133)ndings are robust to using measures of incentives based on either (cid:135)ow compensation or the CEO(cid:146)s portfolio price sensitivity (PPS). This is important since it is well-known that the bulk of incentives comes from appreciation or depreciation in the value of outstandinggrants(HallandLiebman(1998)). Asitisstandardintheliterature,wede(cid:133)nethis 4

measureasthechangeinthevalueoftheCEO(cid:146)sstockandoptionportfolioduetoa1%increase in the price of the (cid:133)rm(cid:146)s common stock. Because details on the exercise prices and maturities of CEO options are not fully disclosed in annual statements, we follow Core and Guay(cid:146)s (2002) approximation method. Finally, we (cid:133)nd reliable evidence supporting the (cid:133)ner prediction of the model that the e⁄ect of industry heterogeneity and turbulence on pay-performance sensitivity is stronger for industry laggards than for leaders. While our study of the link between CEO incentives and industry structure within an explicit dynamic equilibrium setting is, to the best of our knowledge, novel to corporate (cid:133)nance, there are various important literatures related to our work. We detail our contribution to these literatures in turn. First, we establish an equilibrium rationale for high-powered incentives and show that dynamic interaction between competitors enriches the set of cross-sectional determinants of CEO pay. By doing so, we contribute to the large literature that seeks to understand why pay-performance sensitivity and, in general, incentive pay practices, di⁄er across (cid:133)rms (Jensen and Murphy (1990), and Gibbons and Murphy (1990) are seminal contributions, Hall and Liebman (1998) is a more recent important study, and Murphy (1999) is a comprehensive survey). At the theoretical level, our study extends optimal contracting models to an industry equilibrium setting (see Bernardo and Chowdhry (2002), Maksimovic and Zechner (1991) and GomesandLivdan(2004)forotherdynamicindustryequilibriummodelsincorporate(cid:133)nance), henceadvancingourunderstandingoftheroleofequilibriumfactorsinshapingthecontracting environment across (cid:133)rms. Second,ourstudycontributestotheliteratureonCEOincentivesandproductmarketcompetition (see Fershtman and Judd (1987), Sklivas (1987), Scharfstein (1988), Schmidt (1997), and Raith (2003) for theoretical contributions, and Aggarwal and Samwick (1999), and Kedia 5

(2003) for empirical work) by bringing this class of models closer to the data. Theoretical work in this area has traditionally taken a static approach and empirical tests have been hampered by the notorious di¢ culty to (cid:133)nd empirical proxies for the intensity of competition. The novelty of our approach is to use a structural model of dynamic competition among heterogeneous (cid:133)rms, which enables us to tightly link pay-performance sensitivity to a rich set of observable industry and (cid:133)rm characteristics, such as, for example, position within the industry. Third, we contribute to the recent literature in industrial organization (e.g. Ericson and Pakes (1995), Pakes and McGuire (1994, 2001), Doraszelski and Satterthwaite (2003), and Besanko and Doraszelski (2004)) that uses dynamic oligopoly models to study the evolution of industry structure. This literature abstracts from corporate control issues and assumes no separation of ownership and control. In contrast, we explicitly model such separation of ownershipandcontrol, whichenablesustostudythee⁄ectsofCEOincentivesontheevolution of industry structure. Finally, our paper joins a small, but growing literature in corporate (cid:133)nance that studies simulatedpanelsbasedonstructuralmodels(GomesandLivdan(2004), HennessyandWhited (2005, 2006) and Strebulaev (2006)). The structural approach provides a useful solution to the endogeneity problems embedded within most empirical studies, which, as shown by Coles, Lemmon, and Meschke (2003), are di¢ cult to correct by using the standard econometric methods. This literature assumes perfect competition among (cid:133)rms, thus ruling out the possibility of strategic interaction among (cid:133)rms. Our contribution to the literature is to allow for strategic interaction and pursue a computational approach to the Markov-perfect Nash industry equilibrium (see Maskin and Tirole (1988, 2000) for a theoretical treatment of this solution concept, and Grenadier (2002) and Novy-Marx (2007) for applications in (cid:133)nance). 6

The remainder of the paper is organized as follows. Section 2 outlines our industry equilibrium model of optimal CEO incentives. Section 3 develops the key cross-sectional implications of the model. Section 4 introduces our data and tests the model(cid:146)s predictions. Section 5 concludes. Proofs and details on the computation of industry equilibrium are contained in Appendix A and Appendix B, respectively. 2 An Industry Model of Optimal CEO Pay To formalize the link between CEO incentive pay and industry characteristics we introduce an optimal compensation problem along the lines of Holmstrom (1979) and Holmstrom and Milgrom (1987) into a dynamic industry equilibrium model with di⁄erentiated Bertrand competition. This section outlines the model. To ease exposition, we consider an industry without entry and exit (Appendix B outlines the more general model with endogenous entry and exit we study through numeric simulations in Section 3). The distinguishing feature of our approach with respect to the prior literature (see Murphy (1999) for a comprehensive survey) is that our framework is consistent with the empirical evidence of substantial (cid:133)rm heterogeneity across a number of characteristics such as size and growth, as well as inventive pay practices. In other words, our model is structural in that we can produce a well-de(cid:133)ned cross-sectional distribution of (cid:133)rms and test whether it provides a reasonable description of the data. Our theoretical approach is based on an industry equilibrium environment with heterogeneous (cid:133)rms, along the lines of Ericson and Pakes (1995) and Besanko and Doraszelski (2004). Our model is an in(cid:133)nite-horizon dynamic game in an industry that comprises two (cid:133)rms, indexed by i = 1;2 : Each (cid:133)rm consists of a risk-neutral shareholder and a risk-neutral CEO. f g 7

The shareholder can in(cid:135)uence (cid:133)rm pro(cid:133)tability only through his choice of CEO compensation, as product market and e⁄ort decisions are delegated to the CEO. The discount rate is r 1 (0;1): (cid:0) 2 Timing and Demand We consider an empirically plausible source of (cid:133)rm heterogeneity: in our model (cid:133)rms di⁄er in the quality of their products, indexed by ! 1;:::;Z (cid:4);i = i 2 f g (cid:17) 1;2 and Z < : Technically, ! represents the (cid:133)rm(cid:146)s individual state. The distribution of i f g 1 product qualities, ! = (! ;! ) (cid:4)2; fully describes the industry at each point of time. A 1 2 (cid:17) convenient feature of our setup is that (cid:133)rms(cid:146)individual states lend themselves to a particularly straightforward interpretation: whenever ! ! ; (cid:133)rm i is the current industry leader and i i (cid:21) (cid:0) (cid:133)rm i is the laggard. (cid:0) The model(cid:146)s primitives, as well as the (cid:133)rm(cid:146)s own state, ! ; and the state of the industry, i !; are common knowledge. At the beginning of each period, (cid:133)rms learn about the current state, !. Once the state is realized, shareholders choose executive compensation and, given compensation, managers choose e⁄ort, x; and compete in the product market, with realized pro(cid:133)ts (cid:25): Product market pro(cid:133)ts, (cid:25); are the outcome if a standard di⁄erentiated Bertrand duopoly. ThereareDconsumers. Consumer(cid:19)whochoosesgoodiobtainsutilityU = g(! )+(y p )+ (cid:19)i i (cid:19) i (cid:0) e ; where ! indexes product quality, g(! ) is the mean utility of consumers choosing good i; (cid:19)i i i p is its price, and y is consumer (cid:19)(cid:146)s income. Each consumer makes the choice that maximizes i (cid:19) his utility. As shown in Pakes and McGuire (1994), the expected fraction of consumers who choose good i; (cid:6)(!;p), is given by exp(g(!i) (cid:0) pi) : Hence, with constant marginal cost, c, 1+ z q=1 (g(!q) (cid:0) pq) P (cid:133)rm i pro(cid:133)ts are given by (cid:25) (cid:25) (!;p) = D(cid:6)(!;p)(p c): Every period, managers optimally i i (cid:17) (cid:0) choose the price, p, to maximize pro(cid:133)ts. Figure 1 summarizes the timing of events within each 8

period. CEOs and E⁄ort By working harder, CEOs in our model can improve the quality of their products. However, consistent with a well documented empirical property of product improvement(see,forexample,Halletal. (1986)andLachandSchankerman(1988),andCohen(1995) for a survey), CEOs in our model face substantial uncertainty over the outcome of their e⁄ort. Thus, we assume that product quality is stochastically increasing in CEO e⁄ort, in the sense that although higher e⁄ort increases the chances of success, it does not guarantee success. Technically, the evolution of product quality for (cid:133)rm i is governed by the following law of motion ! = ! +(cid:23) (cid:24) (1) 0i i i (cid:0) where! is(cid:133)rmi(cid:146)sproductqualityinthenextperiod, (cid:23) 0;1 is(cid:133)rm-speci(cid:133)candrepresents 0i i 2 f g productimprovement,and(cid:24) 0;1 iscommontoall(cid:133)rmsandrepresentsanadverseindustry- 2 f g wide shock. If (cid:23) = 1, managers are successful at increasing product quality. An amount x i of e⁄ort increases the chances of success, i.e. P ((cid:23) = 1 ! ;! ;x ) = xi . Notice that the i j i (cid:0) i i 1+xi chance of success is a concave function of CEO e⁄ort, a property which, as we show in the next subsection, turns out to be key to obtain a unique solution to the problem of the optimal compensation choice. Finally, we require (cid:23) = 0 with probability one if x = 0; i.e. there can be no product improvement without at least some e⁄ort, and P (0 x) = 0 for all x. Industry wide j shocks are exogenous and iid over time, i.e. P ((cid:24) = 1 ! ;! ;x ) = P ((cid:24)) = (cid:14). i i i j (cid:0) Denoting total CEO compensation by w, CEO(cid:146)s preferences are given by a standard additively separable utility function Eu(w) = E(w) r (x) (cid:0) 9

whereu(w)isCEO(cid:146)speriodutility, whichweassumeisalinearfunctionoftotalCEOcompensation (risk-neutrality) and the disutility of CEO e⁄ort, (x): We assume a linear disutility of e⁄ort, i.e. (x) = x:1 Finally, as standard in incentive provision problems, we assume that every period the CEO has a reservation utility, O; which represents the utility he could potentially derive from outside employment opportunities: Importantly, to capture the idea that successful managers have better outside opportunities, we assume that outside utility depends positively on the outcome of product improvement. In particular, CEO outside utility is given by O((cid:23) ) = (cid:23) ; such that O(1) > O(0): i i Shareholders and Incentive Pay Following Grossman and Hart (1986) and Hart and Moore (1990) we assume that CEO e⁄ort is not contractible. Thus, the central problem for shareholders is to design a compensation package to motivate the CEO to exert e⁄ort. As is standardinthetheoreticalliteratureonexecutivecompensation(e.g. HolmstromandMilgrom (1987), HellwigandSchmidt(2002)), inourbaselinemodelwestudylinearCEOcompensation contracts and later (Section 2.2) generalize the model to a broader set of realistic non-linear contracts such as, for example, stock options. Thus, we consider "share"-contracts that specify the CEO claim as a linear function of the stock-market value of the (cid:133)rm: w = s+(cid:11)V ! 0 (cid:0) (cid:1) wheresisthe(cid:148)base-salary(cid:148)componentofCEOcompensation,whichisnon-performancebased, and (cid:11)V ( ) is the performance-based component of CEO compensation. The (cid:148)piece-rate,(cid:148)(cid:11); (cid:1) representsthepercentageofsharesgrantedandV ( )representingthestock-marketvalueofthe (cid:1) (cid:133)rm. Although, loosely speaking, (cid:11) is the key determinant of pay-performance sensitivity, in 10

takingthepredictionsofthemodeltothedatawewillpaycarefulattentiontomapping(cid:11)intoa speci(cid:133)c empirical measure of pay-performance sensitivity. In particular, it is straightforward to show that (cid:11) measures dollar-dollar sensitivity in the sense of changes of dollar pay for changes in dollar (cid:133)rm value, a measure commonly employed in the empirical literature at least since Jensen and Murphy (1990). Importantly, our model makes predictions also for an alternative measure of sensitivity, dollar-log, which looks at changes of dollar pay for percentage changes in (cid:133)rm value. Given the probability distribution of !, shareholders choose CEO compensation (through the board of directors, or the compensation committee) to maximize their expected pro(cid:133)ts net of payments to the CEO subject to satisfying the CEO participation and incentive constraints. CEO compensation decisions are rational in the sense that shareholders correctly anticipate the ensuing product market equilibrium. Formally, shareholders(cid:146)problem is given by: 1 maxV = E (cid:12)t (cid:25) r 1E (w ) ; s:t: (2) 0 0 t (cid:0) t t st;(cid:11)t (cid:0) t=0 X (cid:0) (cid:1) maxE (w ) r (x ) E O ; t t t t t t xt (cid:0) (cid:21) 8 There are several noteworthy features of our model. First, our chosen speci(cid:133)cation of CEO compensation structure is entirely standard in corporate (cid:133)nance since the seminal contribution of Holmstrom and Milgrom (1987). Further, it is worth emphasizing that although we study CEO compensation in an industry setting, we willingly abstract from issues of strategic provision of incentives such as studied in Fershtman and Judd (1987) and Sklivas (1987). Second, our CEO compensation framework is in line with the recent optimal delegation literature that studies the optimal degree of delegation in organizations (see, for example, 11

Dessein(2002))andtheoptimalseparationofownershipandcontrol(see,forexample,Burkart, Gromb and Panunzi (1997), Gomes and Novaes (2004)). Finally, pay-performance sensitivity choices, (cid:11); measure the extent to which shareholders induce the CEO to exert e⁄ort. In our setting, this standard incentive provision problem gains a dynamic component which is novel to the literature. More concretely, as it will become clear from our equilibrium analysis, in our model shareholders value CEO e⁄ort most when it allows them to gain a competitive hedge, i.e. to catch up with or pull ahead of their industry rivals. 2.1 Industry Equilibrium At every point of time, industry structure is fully summarized by the current state of the industry, i.e. the distribution of product qualities, (! ;! ); which essentially determine which i i (cid:0) (cid:133)rmsarerelativelyaheadandwhich(cid:133)rmsarerelativelybehindwithrespecttotheirrivals. The evolution of the state of the industry is driven by CEO e⁄ort, given the stochastic transition rule (1): We solve for equilibrium in two steps: (cid:133)rst, for any given market structure, (! ;! ); i i (cid:0) we solve for the unique CEO e⁄ort and pricing choices, x (! ;! ) and p (! ;! ); and the (cid:3) i i (cid:3) i i (cid:0) (cid:0) resulting pro(cid:133)ts, (cid:25) (! ;! ); second, we employ the equilibrium pro(cid:133)ts obtained in the (cid:133)rst (cid:3) i i (cid:0) step to solve for shareholders(cid:146)optimal compensation choices and the resulting equilibrium industry structure, i.e. the constellation of Markov Perfect equilibrium (MPE) long-run states of the industry. We start with a characterization of the equilibrium CEO e⁄ort and pricing choices. In the product market stage subgame, the Bertrand-Nash equilibrium in CEO pricing strategies is characterized by the set of (cid:133)rst-order conditions @(cid:25)i(!i;! i;pi;p i) = 0; i = 1;2: For any given @(cid:0)pi (cid:0) 8 stateoftheindustry,(! ;! );equilibriumpro(cid:133)tsare(cid:25) (! ;! ) = (cid:25) ! ;! ;p ;p :Given i i (cid:3) i i i i (cid:3)i (cid:3)i (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:1) pro(cid:133)ts, CEOs choose e⁄ort to maximize their expected utility, i.e. max E (w ) r (x ): It xt t t (cid:0) t 12

is straightforward to show that the set of (cid:133)rst-order condition characterizing CEOs(cid:146)choice of e⁄ort, x ((cid:11)); for any given compensation, s;(cid:11) ; is r = (cid:11)@E V ! ;! =@x: (cid:3) f g x 0i 0 i (cid:0) (cid:0) (cid:1) Shareholders choose CEO compensation optimally based on (2): Shareholders(cid:146)maximization problem can be conveniently written in recursive form using the stock market value of the (cid:133)rm or value function, V (! ;! ), which is de(cid:133)ned by the following Bellman equation i i (cid:0) V (! ;! ) = max (cid:25) (! ;! ) r 1E (w )+r 1E V ! ;! ; s:t: (3) i i (cid:3) i i (cid:0) x i (cid:0) x 0i 0 i (cid:0) si;(cid:11)i (cid:0) (cid:0) (cid:0) (cid:8) (cid:0) (cid:1)(cid:9) E (w ) r (x ((cid:11) )) E O x i (cid:3) i x i0 (cid:0) (cid:21) where w = s+(cid:11)V (! ); and the expected value of future pro(cid:133)ts to the shareholder of (cid:133)rm i i 0 givenstate!isde(cid:133)nedbyE V ! ;! = V ! ;! p ! ;! ! ;! ;x ;x . x 0i 0 (cid:0) i (! 0i ;! 0 (cid:0) i ) 2 (cid:4)2 0i 0 (cid:0) i 0i 0 (cid:0) ij i (cid:0) i (cid:3)i (cid:3) (cid:0) i (cid:0) (cid:1) P (cid:0) (cid:1) (cid:0) (cid:1) Denoting the return function of (cid:133)rm i(cid:146)s shareholder by G (!;(cid:11)(!);V ) = (cid:25) (! ;! ) i i (cid:3) i i (cid:0) (cid:0) r 1E (w )+r 1E V ! ;! , we can rewrite the Bellman equation more compactly as fol- (cid:0) x i (cid:0) x 0i 0 i (cid:0) (cid:0) (cid:1) lows: V (! ;! ) = max G (!;(cid:11)(!);V ): Note that the transition probability function i i si;(cid:11)i i i (cid:0) P ( ) is continuous, which implies that G( ) is a continuous function of (cid:11)(!) and V for all ! i (cid:1) (cid:1) and i: A compensation strategy, (cid:11) (!); that attains the maximum given (cid:11) (!) is said to be i i (cid:0) optimal given (cid:11) (!). The boundedness and continuity of G( ) ensures that the objective is i (cid:0) (cid:1) well-de(cid:133)ned and that optimal compensation strategies exist. Equilibrium industry structure is determined jointly by shareholders(cid:146)choice of CEO compensation and by CEO pricing and e⁄ort strategies. Our solution concept for industry structure is Markov perfect equilibrium (MPE). This is subgame perfect equilibrium in Markov strategies, i.e. strategies that depend only on the (cid:148)payo⁄-relevant(cid:148)(Maskin and Tirole (1988, 1995)) state of the game, ! = (! ;! ). Further, our model implies a symmetric pro(cid:133)t func- 1 2 tion, i.e., (cid:25)(! ;! ) = (cid:25) (! ;! ) and (cid:25)(! ;! ) = (cid:25) (! ;! ); we can restrict attention i i i i i i i i i i (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) 13

to symmetric MPE. This implies symmetry in value functions, V (! ;! ) = V (! ;! ) and i i i i i (cid:0) (cid:0) V (! ;! ) = V (! ;! );andinpolicyfunctions,(cid:11)(! ;! ) = (cid:11) (! ;! )and(cid:11)(! ;! ) = i i i i i i i i i i i i (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:11) (! ;! ): Formally, we de(cid:133)ne an MPE as follows i i i (cid:0) (cid:0) De(cid:133)nition 1 A vector of strategies, (cid:11) (!) = (cid:11) ;(cid:11) [0;(cid:11)]2 is an MPE if for any (cid:133)rm (cid:3) (cid:3)i (cid:3)i 2 (cid:0) (cid:0) (cid:1) i, any state !; and any shareholder(cid:146)s compensation strategy (cid:11)~(!) = (cid:11)~ ;(cid:11) [0;a]2; i (cid:3)i 2 (cid:0) (cid:0) (cid:1) G (!;a (!);V ) G (!;(cid:11)~(!);V ): i (cid:3) i i i (cid:21) In words, an MPE is simply a vector of shareholder(cid:146)s compensation strategies such that each strategy is optimal given the rival(cid:146)s strategy, starting from any state. Appendix A shows our model satis(cid:133)es the boundedness, continuity, and uniqueness requirements in Proposition 4 in Doraszelski and Satterthwaite (2003), which allows us to establish the following: Theorem 1 There exists a unique symmetric MPE in pure CEO compensation strategies to the game satisfying (3) with the following properties: V (!) = (cid:25) (!) x((cid:11) (!)) r 1E O +r 1E V ! (4) (cid:3) (cid:3) (cid:0) x 0 (cid:0) x 0 (cid:0) (cid:0) @ @ (cid:0) (cid:1) x((cid:11) (!))+r 1E O = r 1 E V ! (5) (cid:3) (cid:0) x 0 (cid:0) x 0 @(cid:11) @(cid:11) (cid:2) (cid:3) (cid:0) (cid:1) where ! = (! ;! ): i i (cid:0) Proof. See Appendix A. The left hand side of equation (5) represents the marginal cost of high-power incentives for shareholders: stronger reliance on incentive pay, i.e. higher pay-performance sensitivity, (cid:11) ; i increases the cost of CEO e⁄ort and CEO outside option. These costs, however, are traded-o⁄against expected productivity gains, as represented by the right hand side of equation (5): The key contribution of our dynamic model is to highlight 14

a novel source of bene(cid:133)t of high-power incentives: shareholders incentivize CEOs to induce them to pursue growth strategies, i.e. to increase future pro(cid:133)tability, r 1EV (! ); through (cid:0) 0 product improvements. It is immediate from (5) and CEO (cid:133)rst-order conditions that Corollary 2 ! ;! (cid:4)2; the piece-rate, (cid:11) (!); given by the optimal CEO contract is: i i (cid:3) 8 (cid:0) 2 1 (cid:11) (!) = 1 (cid:3) (cid:0) (cid:1)V (!) where ! = (! ;! ) and (cid:1)V (!) = V (! +1;! ) V (! ;! ): i i i i i i (cid:0) (cid:0) (cid:0) (cid:0) There are several noteworthy features of this result. First, there is a feedback e⁄ect of industry structure, !; on optimal incentives. In other words, incentives depend both on own and rival(cid:146)s competitive position. Second, we can take our model to the data since it allows us to sign the feedback e⁄ect. In particular, the model predicts that shareholders should rely more heavily on performance-based pay whenever growth opportunities are more valuable. Second, and importantly, our model fully recognizes that the value of growth opportunities in turn depends on own and rival(cid:146)s CEO compensation strategies. Thus, in the next section we need to solve for V (!) (and x (!)) and (cid:11) (!) jointly and analyze the link between (cid:11) (!) and (cid:3) (cid:3) (cid:3) the industry structure, !; that emerges in equilibrium. This enables us to make testable predictions about which (cid:133)rm in which industries value growth the most, and, thus, link observable (cid:133)rm and industry characteristics to CEO incentives. Third, although (cid:11) (!) depends on industry structure, realized CEO compensation is not (cid:3) indexed to the market and CEOs may end up being rewarded for luck. Thus, the well documented evidence that this is actually the case in the data (see, for example, Bertrand and 15

Mullainathan (2001)), while puzzling for standard agency models, is fully consistent with the optimal compensation contract that emerges from our model. 2.2 Extensions In the baseline model we have consider so far, incentive contracts are limited to stock grants. This section show that it is straightforward to generalize our framework to a broader set of observed incentive contracts, with a speci(cid:133)c focus on stock options. To this end, we need to modify our baseline contract to specify the CEO(cid:146)s claim as: w = s+(cid:11)f V ! 0 (cid:2) (cid:0) (cid:1)(cid:3) wheresisagain(cid:148)base-salary(cid:148)andf[V ( )]istheperiodpayo⁄ofthesecuritytheCEOreceives. (cid:1) Inparticular,forthecaseofastockoptionwithstrikepriceK;f[V ( )] = max V (! ) K;0 : 0 (cid:1) f (cid:0) g In the case of at-the-money options, we have f[V ( )] = max V (! ) V (!);0 : This mod- 0 (cid:1) f (cid:0) g i(cid:133)cation essentially changes the problem of the CEO, who now chooses e⁄ort to maximize E (w ) = s+(cid:11)E max V (! ) V (!);0 : Thus, the set of (cid:133)rst-order condition characterx i x 0 f (cid:0) g izing CEO(cid:146)s choice of e⁄ort, x ((cid:11)); for any given compensation, s;(cid:11) ; is now given by (cid:3) f g r = (cid:11)@E max V (! ) V (!);0 =@x. Given this new set of e⁄ort choice functions, we can x 0 f (cid:0) g solve for shareholders(cid:146)optimal compensation choices and the resulting equilibrium industry structure in complete analogy with the baseline model. 3 Comparative Dynamics and Empirical Implications In this section we develop several empirical implications of our model. To derive testable crosssectional predictions about the performance-based component of optimal CEO compensation 16

contracts, we solve numerically for industry equilibrium and study compensation and product qualitiesthatemergeasjointoutcomesoftheequilibriumoftheindustry. Weincorporateentry and exit into the basic model presented in Section 2 (details on the full mode with entry and exit are in Appendix B) so as to allow for an endogenous determination of industry structure. Solving for equilibrium e⁄ort, compensation, pricing, and entry and exit functions numerically allows us to compute the Markov process that determines equilibrium industry structure, i.e. the distribution of (cid:133)rms over industry states in the long-run implied by (1). To compute the symmetric MPE, we use a variant of the iterative algorithm of Pakes and McGuire (1994) which we detail in Appendix B. We then use the equilibrium distributions to simulate the model for 10,000 periods. This allows us to derive a synthetic panel whose key unconditional moments we can match to our empirical sample in order to discipline our choice of parameter values. Finally, we show that under reasonable parametric choices several novel cross-sectional relations between pay-performance sensitivity and the characteristics of (cid:133)rm competitive environmentemergeasanequilibriumoutcomeofourindustry. Inparticular,ourmodelpredictsan inverse relation between the magnitude of the performance-based component of CEO pay and, (i) across industries, the degree of heterogeneity of industry structure; (ii) within industries, (cid:133)rm position with respect to its peers. 3.1 Calibration Sinceourdataisatanannualfrequency,weassumethatatimeperiodinthemodelcorresponds to 1 year. The model primitives we need to parametrize are r; (cid:25)( ); x ; (cid:30); and P, i.e. the e (cid:1) discount rate, demand, and technological opportunities. We choose r to match an annual interest rate of 4%. The remaining four parameters are chosen so that the model is able to 17

approximate the unconditional moments of our empirical panel, which essentially comprises S&P 1500 (cid:133)rms from ExecuComp (see next section for a detailed description of the data). Demand determines the pro(cid:133)t function, (cid:25)( ): We choose the market size parameter, D; so (cid:1) as to have on average six active (cid:133)rms, since six is the median number of (cid:133)rms per industry (at the four-digit SIC level) in our empirical panel To allow for entry, we set the maximum number of active (cid:133)rms in the industry, N; to seven.2 Technological opportunities are fully described by the properties of the stochastic process thatgovernsthelawofmotionbetweenstates,P,togetherwiththescrapvalue,(cid:30);andthesunk entry cost, x : We choose the transition probability parameter, (cid:14); to match the unconditional e average of industry concentration in the empirical panel. Sunk entry cost and scrap value are chosen to match median (cid:133)rm life-cycle (age) and its standard deviation in the empirical panel. Table 1 contains a summary of parameter values and compares the key summary statistics generated by the stationary equilibrium of the model, i.e. our arti(cid:133)cial panel, with those of our empirical panel. 3.2 Cross-Sectional Results Our model(cid:146)s most direct predictions are about the link between optimal CEO incentive pay, (cid:11); and industry structure. Empirically (cid:11) measures dollar-dollar sensitivity of CEO pay to (cid:133)rm performance in the sense of changes of dollar pay for changes in dollar (cid:133)rm value, a measure commonlyemployedintheempiricalliteratureatleastsinceJensenandMurphy(1990). Thus, the predictions we derive in this section focus on dollar-dollar sensitivity. Importantly, our model makes predictions also for an alternative measure of sensitivity, dollar-log, which looks atchangesofdollarpayforpercentagechangesin(cid:133)rmvalue(HallandLiebman(1998)). Atthe 18

end of the section, we discuss preliminary results involving this alternative empirical measure of sensitivity. Our (cid:133)rst result is about variation in pay-performance sensitivity across-(cid:133)rms. It can be summarized as follows: Result 1 (Position E⁄ect) Industry leaders have lower pay-performance sensitivity than laggards, i.e. in all asymmetric states such that ! > ! + L; then (cid:11) (! ;! +L) < i i (cid:3) i i (cid:0) (cid:0) (cid:11) (! +L;! );with L > 0. (cid:3) i i (cid:0) Panel A of Figure 2 illustrates the result by plotting, for the case when there are two (cid:133)rms active in the industry, the optimal pay-performance sensitivity implied by the model, (cid:11) (! ;! );asafunctionofown(! )andrival(! )states. Ourmodelimpliesstrongvariation (cid:3) i i i i (cid:0) (cid:0) in sensitivity across di⁄erent (cid:133)rms. In particular, moving upward from state 1 to state 19 for any given state of the rival (States S1-S19) corresponds to upward movements in the productquality ladder. Thus, in our model entrants produce at relatively low qualities and over time can climb the quality-ladder and improve quality through CEO e⁄ort. It is apparent from Panel A of Figure 2 that the optimal pay-performance sensitivity, (cid:11) (! ;! ); is a decreasing function of own state. However, the key prediction of our model (cid:3) i i (cid:0) is that not only own, but also rivals(cid:146)states matter. In fact, pay-performance sensitivities are relatively higher for states above the diagonal, i.e. for laggards who produce lower quality products than their rivals, compared to states below the diagonal, i.e. for leaders who produce higher quality products than their rivals. Panel B of Figure 2 provides further insight into the result by plotting the stock market valueofthe(cid:133)rm(leftpanel)andtheequilibriumstructureoftheindustry(rightpanel)implied by the model. Recall form Corollary 2 that, loosely speaking, optimal incentives in our model 19

re(cid:135)ect the slope of the value of growth opportunities, i.e. of the maximized net present value of pro(cid:133)ts. Intuitively, incentives are high when the value of the growth opportunity associated with product improvement is high. Thus, the importance of Result 1 is that it allows us to identify laggards and close neckto-neck competitors as exactly the (cid:133)rms that value growth strategies the most and, thus, optimally choose to give stronger incentives to their CEOs. Since in industry equilibrium relatively asymmetric states do emerge, in a cross-section of (cid:133)rms we expect to see higher pay-performance sensitivity among laggards vis a vis leaders. Our model also has testable implications for the variation of pay-performance sensitivity across industries. We derive these implications by performing comparative dynamics exercises, i.e. by computing industry equilibria under di⁄erent values of key parameters. In particular, we predict that: Result 2 (Symmetry and Growth E⁄ects) Homogenous industries have higher payperformance sensitivity than heterogeneous industries. Growth industries have higher payperformance sensitivity than declining industries. Panel A of Table 1.2 documents this important equilibrium implication of our model by reporting the result of a simple comparative dynamics exercise. The idea of the exercise is to vary the parameter (cid:14); which measures the likelihood of adverse industry-wide shocks: higher valuesof(cid:14) correspondtorelativelyharsherindustryconditions, i.e. toindustrieswhereadverse shocksaremorelikely. Thefrequencydistributionofsymmetricindustrystatesandthegrowth rate of industry output for di⁄erent values of (cid:14) reported in Panel A of Table 1.2 show that in our model these industries tend to be relatively more heterogeneous in the sense of having the mass of the probability distribution of equilibrium industry states concentrated around states 20

where (cid:133)rms are neck-to-neck competitors. In addition, these industries tend to be more fast gorwing. The last column of Panel A shows that there is a negative relation between industry heterogeneity and CEO incentives since industries that in equilibrium are more heterogeneous tendtohavelowerpay-performancesensitivity. Inaddition, thereisapositiverelationbetween industry growth and CEO incentives since industries that in equilibrium grow at a faster rate tend to have higher pay-performance sensitivity. The intuition for this result highlights the central mechanism at work in our model, which is in essence a selection e⁄ect. Harsher industry conditions intensify competition particularly among neck-to-neck rivals since they make exit more likely for (cid:133)rms that fall behind. As shown inPanelBofFigure3,theselectione⁄ectisthat,inequilibrium,heightenedcompetitionamong closecompetitorsmakesheterogeneousstates-i.e., statesaway fromthediagonal-morelikely. Since, as shown in Panel A of Figure 3, in such states optimal incentives are lower on average, our result obtains. An analogous selection e⁄ect is at play for the next prediction: Result 3 (Turbulence E⁄ect) High turnover industries have higher pay-performance sensitivity than low turnover industries. Panel B of Table 1.2 documents the result by again reporting the result of a simple comparative dynamics exercise. The parameter we now vary is the sunk entry cost, x ;3 and higher e values of x correspond to relatively less turbulent industries, i.e. to industries where entry e is more costly and, thus, in equilibrium we observe lower turnover rates. The last column of Panel B shows that there is a positive relation between turnover and CEO incentives since industries that in equilibrium display higher turnover rates tend to have higher pay-performance sensitivity. 21

The intuition for this result again highlights the central selection e⁄ect at work in our model. If entry is more costly, in equilibrium entry only happens if also incumbents are at relatively low levels of their quality-ladder. Thus, higher entry costs make competition among neck-to-neck rivals more intense. The selection e⁄ect is now that, in equilibrium, there is tougher competition among close competitors, which makes heterogeneous states - i.e., states away from the diagonal - more likely. Since, as shown in Panel A of Figure 3, in such states optimal incentives are lower on average, our result obtains. So far, we have derived predictions about the variation in CEO incentives either across (cid:133)rms (Result 1) or across industries (Results 2-3). Our model also makes a (cid:133)ner conditional prediction about the (cid:133)rms among which we expect our industry e⁄ect to be stronger: Result 4 (Interaction E⁄ects)Symmetry and turbulence e⁄ects are stronger for industry laggards than for leaders. Table 1.3 reports summary statistics for this implication of our model by essentially replicating the simple comparative dynamics exercises in Table 1.2 by subsamples -i.e., looking at averages conditional on (cid:133)rms being industry leaders orlaggards. Again, wevarythe parameter the likelihood of adverse industry-wide shocks ((cid:14); Panel A) and the sunk entry cost (x ; Panel e B) In both panels, the last column shows that CEO incentives vary with industry conditions both for leaders and laggards. However, the bulk of variation is among laggards. Again, the selection e⁄ect is at work here. Changes in industry structure - i.e. shifts in the massoftheequilibriumprobabilitydistributionofindustrystatesclearlya⁄ectbothleadersand laggards. However, as Panel A of Figure 2 shows, optimal incentives as a function of the state, (cid:11) (!); are much steeper for laggards than for leaders. Thus, any change in industry states is (cid:3) likely to have a larger quantitative e⁄ect on the pay-performance sensitivity of laggards, as our 22

result states. 4 Data and Empirical Results In this section we implement empirical tests of our model. In particular, after describing our panel data set, we specify an empirical model relating pay-performance sensitivity to the (cid:133)rm and industrycharacteristics that, based on ourmodel, weexpect to be importantdeterminants of incentive pay. We experiment with a number of speci(cid:133)cations and include a variety of controls for other e⁄ects that are recognized in the empirical literature (see Murphy (1999) for a comprehensive survey). In a large panel of U.S. CEOs and other top executives between 1993 and 2004, we (cid:133)nd strong support for our model. 4.1 Data We combine data from two separate sources. Our data on CEO and other top executive compensation are drawn from the ExecuComp database. Our data on product markets are mainly drawn from the Censuses of Manufacturers conducted by the Commerce department. This section describes each of these data sources in turn. 4.1.1 Compensation Data Our executive compensation data is from the ExecuComp dataset compiled by Standard and Poors. This dataset includes data on total compensation for the top (cid:133)ve executives (ranked annually by salary and bonus) at each of the (cid:133)rms in the S&P 500, S&P Midcap 400, and S&P SmallCap 600. In addition to measures of short-term compensation such as salary and bonus, ExecuComp contains data on components of long-term compensation such as long-term incentiveplans,restrictedstock,andstockappreciationrights. Weuseavailabledatafrom1993 23

to2004. RelativetothedatasetsusedinthestudiesbyJensenandMurphy(1990)andGibbons and Murphy (1990), the advantages of the ExecuComp data are that its sample encompasses the largest 1500 (cid:133)rms each year and is not restricted to just chief executive o¢ cers. Table 2 presents descriptive statistics on the components of executive compensation for all executives in the ExecuComp sample between 1993 and 2004 for whom complete data on total compensation is available. The top panel of the table pertains to the 8,320 executives who are identi(cid:133)ed as the chief executive o¢ cer of the (cid:133)rm. The bottom panel describes the other 38,544 executives in the sample. Our measure of total compensation can be divided into short-term compensation and long-term compensation as standard in the literature (see, for example, Gibbons and Murphy (1990) and Aggarwal and Samwick (1999)). Short-term compensationconsistsofsalary,bonus,andotherannualpayments(e.g.,gross-upsfortaxliabilities, perquisites, preferential discounts on stock purchases). Annual short-term compensation averages $1,217,000 for the CEOs and $490,000 for the Non-CEOs. Long-term compensation includes the value of restricted stock granted, stock options granted, payouts from long-term incentiveplans, andallothercompensation(e.g., contributionstobene(cid:133)tplans, severancepayments). Thesampleaveragesoflong-termcompensationare$3,097,000forCEOsand$922,000 for Non-CEOs. Stock options granted are by far the most important component of long-term compensation, accounting for a sample average value of $2,508,000 for CEOs and $727,000 for Non-CEOs. In additional to these measures of (cid:135)ow compensation, we also consider a measure of incentives based on the CEO(cid:146)s portfolio price sensitivity (PPS), in order to take into account the well-known argument that the bulk of incentives comes from appreciation or depreciation in the value of outstanding grants (Hall and Liebman (1998)). As it is standard in the literature, 24

we de(cid:133)ne this measure as the change in the value of the CEO(cid:146)s stock and option portfolio due to a 1% increase in the price of the (cid:133)rm(cid:146)s common stock. Because details on the exercise prices and maturities of CEO options are not fully disclosed in annual statements, we follow Core and Guay(cid:146)s (2002) approximation method. Details of the computation are in Appendix C. As shown in Table 2, the median CEO PPS in our sample is about 1.4, while the average PPS is 4.2, suggesting that the measure has a signi(cid:133)cant right skew. 4.1.2 Firm Data Weincludeinourpanelcontrolsfor(cid:133)rmcharacteristicswhoserelationshipwitpay-performance sensitivity has been documented in previous studies. Firm characteristics are from the Compustat. Outliers are removed by winsorizing the extreme observations in the 1% left or right tail of the distribution. We measure capital as property, plants, and equipment (item 8). Tobin(cid:146)s Q is the ratio of market value of assets to book value of assets. Market value of assets is de- (cid:133)ned as total assets (item6) plus market equity minus book equity. Market equity is de(cid:133)ned as common shares outstanding (item 25) times (cid:133)scal-year closing price (item 199). Book equity is calculated as stockholders equity (item 216) [or the (cid:133)rst available of common equity (item 60) plus preferred stock par value (item 130) or total assets (item 6) minus total liabilities (item 181)] minus preferred stock liquidating value (item 10) [or the (cid:133)rst available of redemption value (item 56) or par value (item 130)] plus balance sheet deferred taxes and investment tax credit (item 35) when available minus post retirement assets (item 336) when available. Book value of assets is total assets (item 6). We measure return on equity (ROE) as the ratio of earnings to average equity for the prior (cid:133)scal year (item 20/(item 60+ item 60 )). t 1 (cid:0) 25

4.1.3 Industry Data We use several sources for industry data. For comparability with previous studies We limit our sample to the manufacturing sector, which contains twenty 2-digit standard industrial classi(cid:133)cation (SIC) codes from 20 to 39, and, within these 2-digit SICs, 458 separate fourdigit SICs (ranging from 2001 to 3999). Financing (cid:133)rms (SICs 6000-6999), and regulated utilities (SICs 4900-4999) are excluded. We use four-digit SIC classi(cid:133)cations to de(cid:133)ne industry membership. In unreported tables we replicate our (cid:133)ndings at the three-digit level with no qualitatively di⁄erent results. To proxy for industry turbulence and (cid:133)rm heterogeneity within an industry we use average jobturnoverwithininindustryandaveragedistanceof(cid:133)rmsfrommedianindustrysales. These measures have been previously employed, although with a di⁄erent motivation, respectively in job turnover (see Davis, Haltiwanger, and Schuh (1996)) and capital structure studies (see Titman(1984), MacKayandPhillips(2005)). Thismeasureallowsustoproxyfortheintensity of managerial e⁄ort directed toward product improvement. To control for standard measures of product market competition used in previous studies, we include in our panel concentration ratios from the Census of Manufactures, conducted by the Bureau of the Census as part of the quinquennial Economic Censuses. Our measure of concentration is the ratio of the sales of the top four (cid:133)rms in the industry to total industry sales. 4.2 Empirical Speci(cid:133)cation and Results To test the empirical predictions of our industry equilibrium model of CEO pay, we extend the standard econometric framework that estimates the sensitivity of pay to performance (see 26

Murphy (1999) for a careful description of this approach) by allowing pay-performance sensitivity to vary in proportion to our measures of product di⁄erentiation. Accordingly, for (cid:135)ow measures of compensation we estimate the following equation: w = (cid:11) (cid:25) +(cid:11) D (cid:25) +(cid:11) D +(cid:11) X +" (6) ijt 1 jt 2 jt jt 3 jt 4 jit jit where the executive i works at (cid:133)rm j in year t. The dependent variable, w ; is dollar comijt pensation, andtheindependentvariablesaredollar(cid:133)rmperformance, (cid:25) ;aloneandinteracted jt with our measures of competitive position (Result 1), industry symmetry (Result 2) and turbulence (Result 3), D . We also include as controls D itself and other variables, such as (cid:133)rm j j size and industry concentration, that control for e⁄ects found in previous studies (respectively, Schaefer (1998) and Baker and Hall (2004) for (cid:133)rm size and Aggarwal and Samwick (1999) for concentration). We follow Jensen and Murphy (1990) and use as our measure of (cid:133)rm performance, (cid:25) ; the jt total dollar returns to shareholders including capital gains and dividends but net of in(cid:135)ation on their holdings at the beginning of the period. We emphasize that this choice is motivated by the fact that dollar-dollar sensitivity is the relevant measure with respect to which our empirical predictions where derived. Finally, we include year- and 2-digit SIC industry-(cid:133)xed e⁄ects. The inclusion of these industry (cid:133)xed-e⁄ect ensures that it is not the variation in the average pay-performance sensitivities between 2-digit industry groups but the variation in the pay-performance sensitivity within those groups that identi(cid:133)es the estimated coe¢ cient. Including the industry e⁄ects also controls for any other factor such as a macroeconomic shock that varies across broad industry groups but not within the narrow industries that comprise them. The null hypothesis is that 27

(cid:11) , the coe¢ cient on the interaction of performance and product di⁄erentiation, is equal to 2 zero. We also report results using the stock measure of compensation, CEO(cid:146)s portfolio price sensitivity (PPS), in which case the above speci(cid:133)cation simpli(cid:133)es to the following: PPS = ijt (cid:11) D +(cid:11) X +" : 1 jt 2 jit jit 4.2.1 Regression Results Figure 4 shows graphically that the data lines up with comfortably with our Result 1. In particular, we estimate equation (6) with total CEO compensation (Panel A) and CEO wealth deltas (Panel B) as dependent variables separately within subsample splits based on a measure of (cid:133)rm position within its industry. We measure competitive position as the ratio of the (cid:133)rm sales to industry median sales in the beginning of the year. Consistent with our position e⁄ect, as we move toward higher competitive position deciles - i.e., for (cid:133)rms that are relatively ahead in their industry - the estimates of pay-performance sensitivity fall by an order of magnitude. These results are con(cid:133)rmed by Table 3, where we estimate equation (6) using competitive position as our key explanatory variable, D :Clearly, (cid:133)rms that are relatively more ahead in j their industry tend to have lower equity incentives, a result which is robust to measuring incentives based on either (cid:135)ow comepnsation or CEO(cid:146)s portfolio price sensitivity (PPS). Symmetry E⁄ect and Interaction To test Result 2, the left panel of Table 4 presents the estimates of equation (6) with total compensation (Panel A) and CEO wealth deltas (Panel B) as dependent variables and a measure of industry-wide heterogeneity built along the lines of MacKay and Phillips (2005). In particular, since our model links the degree of within industry (cid:133)rm heterogeneity to pay-performance sensitivity, we construct a measure of industry (cid:148)Sym- 28

metry,(cid:148)de(cid:133)ned as the average proximity of (cid:133)rm sales to median industry sales. Technically, our Symmetry variable is the inverse of the average distance of (cid:133)rm sales from industry median sales in the beginning of the year. Based on Result 2, we predict a positive relation between Symmetry and pay-performance sensitivity. In all speci(cid:133)cations, executive compensation is denominated in thousands and (cid:133)rm performance is denominated in millions of dollars. We report results for three baseline regressions: (1) with no additional controls, then (2) including industry-(cid:133)xed e⁄ects, and (cid:133)nally (3) including industry-(cid:133)xed e⁄ects as well as controls. In all speci(cid:133)cations, consistent with Result 2, we (cid:133)nd a positive and highly signi(cid:133)cant coe¢ cient on the interaction of industry homogeneity and (cid:133)rm performance: industries with more homogeneous (cid:133)rm sale distribution have higher payperformance sensitivity than industries with a more skewed (cid:133)rm sale distribution. Columns (2)-(3) of Table 4 show that the result is robust to adding (cid:133)xed e⁄ects and controlling for (cid:133)rm size and industry concentration. To test Result 4, we re-estimate equation (6) with total compensation as the dependent variableandthesamplenowsplitbasedon(cid:133)rmpositionwithinitsindustry. Again,wemeasure (cid:133)rmcompetitivepositionastheratioofthe(cid:133)rmsalestoindustrymediansalesinthebeginning of the year. We de(cid:133)ne as Leaders the (cid:133)rms that are in the highest quartile of the distribution of the ratio, and Laggards the (cid:133)rms that are in the lowest quartile of the distribution of the ratio. The center panel of Table 4 reports the results. Consistent with Result 4, the e⁄ect of industrysymmetryonpay-performancesensitivitydependson(cid:133)rmcompetitivepositionwithin its (4-digit SIC) industry. In fact, Column 5 shows that for industry laggards, total executive compensation increases by up to about 31 cents for every thousand dollars of incremental 29

shareholderwealthperyearinheterogeneousindustries-i.e.,industriesinwhichourSymmetry measure is close to zero. Column 4 contrasts this estimate with the case of industry leaders: now the pay-performance sensitivity is about 23 cents per thousand. Again, consistent with Result4,theSymmetrye⁄ectismuchstrongerforlaggardsthanforleadersaspay-performance sensitivity depends on Symmetry for laggards, but not for leaders. The right panel of Table 4 presents the estimates of equation (6) using short-term compensation as the dependent variable. The results are qualitatively very similar to those in the left panel, although magnitudes are much smaller. This is to be expected given the welldocumented fact that short-term incentives, such as, for example, bonuses, have much lower power than long-term incentives. In Table 4 we have constrained pay-performance sensitivity to be equal for CEOs and non-CEOs. In Table 5 we relax this arguably questionable assumption and re-estimate the same set of regressions based on equation (6) restricting the sample to only CEOs. Panel A reports results for (cid:135)ow measures of comepnsation, while Panel B reports results for CEO(cid:146)s portfoliopricesensitivity(PPS).Asexpected, themagnitudeofallpay-performancesensitivity coe¢ cientsismuchlarger, whichisconsistentwiththefactthatCEOsbearmoreresponsibility for decisions that a⁄ect pro(cid:133)ts. Qualitatively, however, our results are unchanged. Finally, the resultsarerobusttomeasuringincentivesbasedoneither(cid:135)owcomepnsationorCEO(cid:146)sportfolio price sensitivity (PPS). Turbulence E⁄ect and Interaction To test Result 3, the left panel of Table 6 presents the estimates of equation (6) with total compensation as dependent variables and a measure of industry turbulence, gross job turnover, based on Davis, Haltiwanger, and Schuh (1996). In particular, we measure turbulence as industry gross job turnover. Based on Result 3, we 30

predict a positive relation between turnover and pay-performance sensitivity. Again, executive compensation is denominated in thousands, (cid:133)rm performance is denominated in millions of dollars, and we report results for three baseline regressions: (1) with no additional controls, then (2) including industry-(cid:133)xed e⁄ects, and (cid:133)nally (3) including industry- (cid:133)xede⁄ectsaswellascontrols. Inallspeci(cid:133)cations, consistentwithResult3, we(cid:133)ndapositive andhighlysigni(cid:133)cantcoe¢ cientontheinteractionofturnoverand(cid:133)rmperformance. Columns (2)-(3) of Table 6 show that the result is robust to adding (cid:133)xed e⁄ects and controlling for (cid:133)rm size and industry concentration. The center panel of Table 6 reports the results of our test of Result 4. As we did for Symmetry, we re-estimate equation (6) with total compensation as the dependent variable and the sample now split based on (cid:133)rm position within its industry. Again, we measure (cid:133)rm competitive position as the ratio of the (cid:133)rm sales to industry median sales in the beginning of the year. We de(cid:133)ne as Leaders the (cid:133)rms that are in the highest quartile of the distribution of the ratio, and Laggards the (cid:133)rms that are in the lowest quartile of the distribution of the ratio. Consistent with Result 4, the e⁄ect of industry turnover on pay-performance sensitivity depends on (cid:133)rm competitive position within its (4-digit SIC) industry. In fact, Column 5 shows that for industry laggards, total executive compensation increases by up to about 39 cents for every thousand dollars of incremental shareholder wealth per year in heterogeneous industries - i.e., industries in which our Turnover measure is close to zero. Column 4 contrasts this estimate with the case of industry leaders: now the pay-performance sensitivity is about 22 cents per thousand. Again, consistent with Result 4, the Turnover e⁄ect is much stronger for laggards than for leaders as pay-performance sensitivity depends strongly on Turnover for 31

laggards, but only weakly for leaders. The right panel of Table 6 presents the estimates of equation (6) using short-term compensation as the dependent variable. The results are qualitatively very similar to those in the left panel, although magnitudes are much smaller. This is to be expected given the welldocumented fact that short-term incentives, such as, for example, bonuses, have much lower power than long-term incentives. Table 7 shows that when we estimate the same set of regressions only for CEOs, the results are qualitatively the same as in the full sample of executives, with larger magnitudes on all the coe¢ cients. Industry Dynamics E⁄ect and Interaction Table 8 presents the estimates of equation (6)withtotalcompensation(PanelA)andCEOwealthdeltas(PanelB)asdependentvariables and a measure of industry growth as the main explanatory variable. In particular, we measure growh as average change in industry output in the sample period. Based on Result 2, we predict a positive relation between industry growth and pay-performance sensitivity. Again, executive compensation is denominated in thousands, (cid:133)rm performance is denominated in millions of dollars, and we report results for three baseline regressions: (1) with no additional controls, then (2) including industry-(cid:133)xed e⁄ects, and (cid:133)nally (3) including industry- (cid:133)xede⁄ectsaswellascontrols. Inallspeci(cid:133)cations, consistentwithResult2, we(cid:133)ndapositive and highly signi(cid:133)cant coe¢ cient on the interaction of growth and (cid:133)rm performance. In addition, we have a reliably negative coe¢ cient on the interaction of a dummy for industry decline. This dummy takes the value of one for industries in the bottom quartile of output growth. Columns (2)-(3) of Table 8 show that the result is robust to adding (cid:133)xed e⁄ects and controlling for (cid:133)rm size and industry concentration. 32

Panel B of Table 8 reports results for CEO(cid:146)s portfolio price sensitivity (PPS). Clearly, the resultsarerobusttomeasuringincentivesbasedoneither(cid:135)owcomepnsationorCEO(cid:146)sportfolio price sensitivity (PPS). Robustness Throughout the paper, we have limited our measure of total compensation to the annual (cid:135)ow of resources that the shareholders could have kept for themselves had they not used it to compensate the executive. In practice, an executive also receives incentives from the e⁄ect of her actions on the value of her stock holdings. If an executive owns stock in her (cid:133)rm, then the total increment in her wealth due to the performance of her (cid:133)rm will include not only the extra pay she receives as part of the pay-performance sensitivity built into her compensation but the appreciation on her personal stock holdings. Recognizing this, the shareholders of her (cid:133)rm will incorporate a lower pay-performance sensitivity into her contract. Hence, the optimal compensation contract becomes a function of both industry structure and executives(cid:146)stock holdings. Conditional on a particular allocation of the executives personal wealth, however, the relationship between industry structure and pay-performance sensitivity is unchanged. To check for robustness of our results to the incentives provided by inside ownership, in Table 9 we control for the executives holdings of her (cid:133)rm. Column 1 reports estimates of equation (6) using total compensation as the dependent variable. Although the estimated coe¢ cient of insider ownership is negative and signi(cid:133)cant, all our previous (cid:133)ndings stand. 5 Conclusion This paper develops an industry equilibrium model of optimal CEO incentives. We tested the key predictions of the model empirically in a large panel of U.S. executives between 1993 and 33

2004 and, consistent with the model, found strong evidence of an inverse relationship between pay-performance sensitivity and, (i) across industries, the degree of heterogeneity of industry structure; (ii) within industries, (cid:133)rm position with respect to its peers. In particular, we found strong evidence that (cid:133)rm competitive position is an important determinant of CEO incentives, in that industry leaders have reliably weaker pay-performance sensitivity than laggards. Agency models of CEO pay emphasize the trade-o⁄between incentives and insurance but are silent on the sources of value of CEO e⁄ort, thus leaving the important question of the link between economic fundamentals and CEO incentives essentially unanswered. Our model and empirical tests emphasize the importance of economic fundamentals for CEO incentives. Moreover, they provide a novel product market rationale for the otherwise puzzling infrequent use of high-powered incentives. 34

References [1] Aggarwal, Rajesh K., and Andrew A. Samwick, 1999, (cid:148)The Other Side of the Tradeo⁄: the Impact of Risk on Executive Compensation,(cid:148)Journal of Political Economy 107, 65-105. [2] Aggarwal, Rajesh K. and Andrew A. Samwick, 1999, (cid:148)Executive Compensation, Strategic Competition and Relative Performance Evaluation: Theory and Evidence(cid:148), Journal of Finance, 54(6), 1999-2043 [3] Anderson, Simon P., Andre de Palma, and Jacques-Francois Thisse, 1992, Discrete Choie Theory of Product Di⁄erentiation, Cambridge, MIT Press [4] Bagwell, Kyle, Garey Ramey and Dan Spulber, 1997, (cid:148)Dynamic Retail Price and Investment Competition(cid:148), Rand Journal of Economics, 28(2), 207-227. [5] Baker, George P., and Brian J. Hall, 2004, (cid:148)CEO Incentives and Firm Size,(cid:148)Journal of Labor Economics, 767-798. [6] Bebchuk, Lucian and Jesse M. Fried, 2004, Pay without Performance: the Unful(cid:133)lled Promise of Executive Compensation, Harvard University Press. [7] Bebchuk, Lucian, Jesse M. Fried and David Walker, 2002, (cid:148)Managerial Power and Rent Extraction in the Design of Executive Compensation, University of Chicago Law Review 69, 751-846. [8] Bebchuk, Lucian and Oren Bar-Gill, 2003, (cid:148)The Costs of Permitting Managers to Sell Their Shares,(cid:148)Working paper, Harvard Law School. [9] Bebchuk, Lucian and Yaniv Grinstein, 2004, (cid:148)The Growth of US Executive Pay,(cid:148)mimeo, Harvard University. [10] Bebchuk, Lucian and Yaniv Grinstein, 2005, (cid:148)Executive Pay and Firm Size,(cid:148) mimeo, Harvard University. [11] Bergman, Nittai and Dirk Jenter, 2003, (cid:148)Employee Sentiment and Stock Option Compensation,(cid:148)Working paper, MIT. [12] Bergstresser, Dan and Thomas Philippon, 2002, (cid:148)CEO Incentives and Earnings Management,(cid:148)Journal of Financial Economics, forthcoming. [13] Bernard, Andrew B. and J. Bradford Jensen, and Peter K. Schott, 2003, (cid:148)Falling Trade Costs, Heterogeneous Firms, and Industry Dynamics,(cid:148)mimeo, Yale SOM. [14] Bernardo, Antonio E. and Bhagwan Chowdhry, 2002, "Resources, Real Options, and Corporate Strategy," Journal of Financial Economics, 63, pp.211(cid:150)234 [15] Bertrand, Marianne and Sendhil Mullainathan, 2001, (cid:148)Are CEOs Rewarded for Luck? The Ones Without Principals Are,(cid:148)Quarterly Journal of Economics 116, 901-32. [16] Bertrand, M. and Mullainathan, S., 2003, (cid:147)Enjoying the Quiet Life? Corporate Governance and Managerial Preferences(cid:148), Journal of Political Economy, 111, 1043-1075. 35

[17] Besanko, David , and Ulrich, Doraszelski, 2004, (cid:147)Capacity Dynamics and Endogenous Asymmetries in Firm Size,(cid:148)Rand Journal of Economics, 35(1): 23-49 [18] Bolton, Patrick, and David, Scharfstein ,1990, (cid:147)A Theory of Predation Based on Agency Problems in Financial Contracting,(cid:148)American Economic Review 80(1): 93(cid:150)106. [19] Bolton, Patrick, Jose Scheinkman and Wei Xiong, 2004, (cid:148)Pay for Short-Term Performance: Executive Compensation in Speculative Markets,(cid:148)Journal of Corporation Law, forthcoming. [20] Budd,Christopher,ChristopherHarrisandJohnVickers,1993,(cid:148)AModeloftheEvolution of Duopoly: Does the Asymmetry between Firms Tend to Increase or Decrease?,(cid:148)Review of Economic Studies, 60(3), pp. 543-573 [21] Burkart, Mike, Denis Gromb and Fausto Panunzi, 1997, (cid:148)Large Shareholders, Monitoring, and the Value of the Firm(cid:148), Quarterly Journal of Economics, 112(3): 693-728 [22] Caves, R.E., 1998, (cid:148)Industrial Organization and New Findings on the Turnover and Mobility of Firms,(cid:148)Jornal of Economic Literature, 36: 1947-82 [23] Cohen, W.M., 1995, (cid:148)Empirical Studies of Innovative Activity,(cid:148)in P. Stoneman (ed.): Handbook of Innovation and Technological Change, Oxford, Blackwell. [24] Coles, J., M. Lemmon, and J. F. Meschke, 2003, "Structural Models and Endogeneity in Corporate Finance," Working paper. [25] Coles, Je⁄rey L., Naveen Daniel, and Lalitha Naveen, 2005, (cid:148)Managerial Incentives and Risk-Taking,(cid:148)forthcoming, Journal of Financial Economics. [26] Core, John E., and Wayne R. Guay, 2002, (cid:148)Estimating the Value of Employee Stock Option Portfolios and Their Sensitivities to Price and Volatility,(cid:148)Journal of Accounting Research 40, 613-630. [27] Davis, Steven J, John C. Haltiwanger, and Scott Schuh, 1996, Job Creation and Destruction, MIT Press [28] Demsetz, Harold, and Kenneth Lehn, 1985, (cid:148)The Structure of Corporate Ownership: Causes and Consequences,(cid:148)Journal of Political Economy 93, 115-577. [29] Doraszelski, Ulrich and Mark Satterthwaite, 2003, (cid:148)Foundations of Markov-Perfect Industry Dynamics: Existence, Puri(cid:133)cation, and Multiplicity(cid:148), mimeo, Harvard University [30] Dow, James, Gary, Gorton, and Arvind, Krishnamurthy, 2004, (cid:148)Equilibrium Investment and Asset Prices under Imperfect Corporate Control,(cid:148)Forthcoming, American Economic Review [31] Ericson, Richard and Ariel Pakes, 1995, (cid:148)Markov-Perfect Industry Dynamics: a Framework for Empirical Work(cid:148), Review of Economic Studies, 62, 53-82 [32] Fershtman,ChaimandKennethJudd,1987,(cid:148)EquilibriumIncentivesinOligopoly(cid:148),American Economic Review, 77(5), 927-940 36

[33] Gibbons, Robert and Kevin J. Murphy, 1990, (cid:148)Relative Performance Evaluation for Chief Executive O¢ cers,(cid:148)Industrial and Labor Relations Review 43, 30-51. [34] Gibbons,RobertandKevinJ.Murphy,1992,(cid:148)OptimalIncentiveContractsinthePresence of Career Concerns,(cid:148)Journal of Political Economy, 100(3), 468-505. [35] Gomes, JoaoandDimitriLivdan, 2004, (cid:147)OptimalDiversi(cid:133)cation: ReconcilingTheoryand Evidence(cid:148), Journal of Finance, 59, pp.507-535. [36] Gompers, Paul A., Joy L., Ishii, and Andrew, Metrick, 2003, (cid:147)Corporate Governance and Equity Prices,(cid:148)Quarterly Journal of Economics 118(1): 107-155. [37] Gort, M., 1963, (cid:148)Analysisof Stabilityand ChangesinMarketShares,(cid:148)JournalofPolitical Economy, 71: 51-63. [38] Grenadier, S., 2002, (cid:147)Option Exercise Games: An Application to the Equilibrium Investment Strategies of Firms(cid:148), Review of Financial Studies 15, pp.691-721. [39] Hall, B. J. and J. B. Liebman (1998), (cid:147)Are CEOs Really Paid Like Bureaucrats?(cid:148)Quarterly Journal of Economics. 113, August, p. 653-691. [40] Hall, Bronwyn H., Zvi Griliches, and Jerry A. Hausman, 1986, (cid:148)Patents and R&D: Is There a Lag?(cid:148)International Economic Review 27: 265-83. [41] Hart, Oliver D., 1983, (cid:148)The Market Mechanism as an Incentive Scheme,(cid:148)Bell Journal of Economics 14, 366-82. [42] Hart, Oliver D., 2001, (cid:147)Financial Contracting,(cid:148)Journal of Economic Literature, 39(4), 1079-1100 [43] Hart, Oliver D. and John Moore, 1990, (cid:147)Property Rights and the Nature of the Firm,(cid:148) Journal of Political Economy, 98(6), 1119-1158. [44] Hellwig, Martin F., and Klaus M. Schmidt, 2002, (cid:148)Discrete-Time Approximations of the Holmstrom-Milgrom Brownian-Motion Model of Intertemporal Incentive Provision,(cid:148) Econometrica 70, 2225-2264. [45] Hennessy, Christopher, and Toni Whited, 2005, "Debt Dynamics," Journal of Finance 60, pp. 1129-1165. [46] Hennessy, Christopher, and Toni Whited, 2006, "How Costly is External Financing? Evidence from a Structural Estimation," forthcoming, Journal of Finance. [47] Hermalin, Benjamin, 2004, (cid:148)Trends in Corporate Governance,(cid:148)Journal of Finance, forthcoming. [48] Himmelberg, Charles P., R. Glenn Hubbard, and Darius Palia, 1999, (cid:148)Understanding the Determinants of Managerial Ownership and the Link Between Ownership and Performance,(cid:148)Journal of Financial Economics 53, 353-84. [49] Holmstrom, Bengt, 1979, (cid:148)Moral Hazard and Observability,(cid:148)Bell Journal of Economics 10, 74-91. 37

[50] Holmstrom, Bengt, and Paul Milgrom, 1987, (cid:148)Aggregation and Linearity in the Provision of Intertemporal Incentives,(cid:148)Econometrica 55, 303-28. [51] Jensen, Michael, 1986, (cid:147)The Agency Costs of Free Cash Flow: Corporate Finance and Takeovers,(cid:148)American Economic Review, 76(2), 323-330. [52] Jensen, Michael, 2004, (cid:148)Agency Costs of Overvalued Equity,(cid:148)Discussion paper, The European Corporate Governance Institute. [53] Jensen, M., and W. Meckling, 1976, (cid:148)Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure,(cid:148)Journal of Financial Economics, 3, 305-360. [54] Jensen, Michael and Kevin J. Murphy, 1990, (cid:148)Performance Pay and Top Management Incentives,(cid:148)Journal of Political Economy, 98(2), 225-64. [55] Jensen, Michael and Kevin Murphy, 2004, (cid:148)Remuneration: Where We(cid:146)ve Been, How We Got to Here, What are the Problems, and How to Fix Them,(cid:148)Discussion paper, The European Corporate Governance Institute. [56] Jovanovic, Boyan, and Peter L. Rousseau, 2001, "Why Wait? A Century of Life Before IPO," American Economic Review, 91 (2), 336-341. [57] Kedia, Simi, 2002, (cid:147)Product Market Competition and Top Management Compensation,(cid:148) mimeo, Harvard Business School [58] Kole, Stacey and Lehn, Kenneth. (cid:148)Deregulation, the Evolution of Corporate Governance Structure, and Survival.(cid:148)The American Economic Review, 1997, 87(2, Papers and Proceedings of the Hundred and Fourth Annual Meeting of the American Economic Association), pp. 421-25. [59] Kovenock, Dan, andGordonPhillips, 1995, (cid:148)Increased DebtandProduct-MarketRivalry: HowDoWeReconcile TheoryandEvidence,(cid:148)American Economic Review85(2), 403-408. [60] Kovenock, Dan, and Gordon Phillips, 1997, (cid:148)Financial Structure and Product Market Behavior: An Examination of Plant Exit and Investment,(cid:148)Review of Financial Studies 10(3), 767-804. [61] Lach, Saul, and Mark Schankerman, 1988, (cid:148)Dynamics of R&D and Investment in the Scienti(cid:133)c Sector,(cid:148)Journal of Political Economy 97(4): 880-904. [62] MacKay, Peter, and Gordon, Phillips, 2005, (cid:148)How Does Industry A⁄ect Firm Financial Structure?(cid:148)forthcoming, Review of Financial Studies. [63] Maksimovic, Vojislav, 1988, (cid:148)Capital Structure in Repeated Oligopolies,(cid:148)RAND Journal of Economics 19, 389-407. [64] Maksimovic, Vojislav, and Joseph Zechner, 1991, (cid:148)Debt, Agency Costs, and Industry Equilibrium,(cid:148)Journal of Finance 46, 1619-1643. [65] Marris,RobinL.,1964,TheEconomicTheoryofManagerialCapitalism,London,Macmillan. 38

[66] Maskin Eric and Jean Tirole, 1988, (cid:148)A Theory of Dynamic Oligopoly, I & II(cid:148), Econometrica, 56(3), 549-600 [67] Maskin Eric and Jean Tirole, 2000, (cid:148)Markov-Perfect Equilibrium, I: Observable Actions(cid:148), mimeo, Harvard University.. [68] Mehran, Hamid, 1995, (cid:148)Executive Compensation Structure, Ownership, and Firm Performance,(cid:148)Journal of Financial Economics 38, 163-185. [69] Murphy, Kevin J., 1985, (cid:148)Corporate Performance and Managerial Remuneration: An Empirical Analysis(cid:148), Journal of Accounting and Economics 7 (April): 11-42 [70] Murphy, KevinJ., 1999, (cid:148)ExecutiveCompensation,(cid:148)inOrleyAshenfelterandDavidCard (eds.), Handbook of Labor Economics, Vol. 3b, Elsevier Science North Holland. [71] Murphy, Kevin, 2002, (cid:148)ExplainingExecutiveCompensation: ManagerialPowerversusthe Perceived Cost of Stock Options,(cid:148)University of Chicago Law Review, 69, 847-69. [72] Nikaido, H., 1968, Convex Structures and Economic Theory, New York, Academic Press. [73] Novy-Marx, Robert, 2007, "Investment-Cash Flow Sensitivity and the Value Premium," mimeo, Chicago GSB [74] Pakes, Ariel and Paul McGuire, 1994, (cid:148)Computing Markov-perfect Nash Equilibria: Numerical Implications of a Dynamic Di⁄erentiated Product Model(cid:148), RAND Journal of Economics, 25(4), 555-589 [75] Pakes, Ariel and Paul McGuire, 2000, (cid:148)Stochastic Algorithms, Symmetric Markov-perfect Equilibrium and the (cid:146)Curse(cid:146)of Dimensionality(cid:148), mimeo, Harvard University. [76] Palia, Darius, 2001, (cid:148)The Endogeneity of Managerial Compensation in Firm Valuation: A Solution,(cid:148)Review of Financial Studies 14, 735-64. [77] Philippon, Thomas, 2003, (cid:147)Corporate Governance and Aggregate Volatility,(cid:148)Working paper, M.I.T. [78] Prendergast, Canice, 2002, "Uncertainty and Incentives," Journal of Labor Economics, 20(2), pp.115-37. [79] Raith, Michael, 2003, (cid:148)Competition, Risk and Managerial Incentives,(cid:148) American Economic Review 93, 1425-1436. [80] Schaefer, Scott, 1998, (cid:147)The Dependence of Pay-Performance Sensitivity on the Size of the Firm,(cid:148)Review of Economics and Statistics, 80(3), pp.436-443. [81] Scharfstein, DavidS., 1988, (cid:148)ProductMarketCompetitionandManagerialSlack(cid:148), RAND Journal of Economics, 19: 147-155 [82] Schmidt, Klaus M., 1997, (cid:148)Managerial Incentives and Product Market Competition(cid:148), Review of Economic Studies, 64: 91-213 [83] Sklivas, Steven D., 1987, (cid:148)TheStrategicChoiceof ManagerialIncentives(cid:148), RANDJournal of Economics, 18: 452-458 39

[84] Strebulaev, I. A., 2006, (cid:147)Do Tests of Capital Structure Theory Mean What They Say?" Journal of Finance, forthcoming. [85] Titman,Sheridan,andR.Wessels,1988,(cid:148)TheDeterminantsofCapitalStructureChoice,(cid:148) Journal of Finance 43, 1-19. [86] Titman, Sheridan, 1984, (cid:148)The E⁄ect of Capital Structure on a Firm(cid:146)s Liquidation Decision,(cid:148)Journal of Financial Economics 13, 137-151. [87] Vickers, John, 1986, (cid:148)The Evolution of Industry Structure When There Is a Sequence of Innovations(cid:148), Journal of Industrial Economics, 35, 1-12. [88] Williamson, Oliver E., 1964, The Economics of Discretionary Behavior: Managerial Objectives in a Theory of the Firm, Englewood Cli⁄s, N.J.: Prentice-Hall. [89] Yermack, David, 1995, (cid:148)Do Corporations Award CEO Stock Options E⁄ectively?,(cid:148)Journal of Financial Economics 39, 237-69. 40

Appendix A. Proofs Proof of Theorem (1). We (cid:133)rst prove existence of a symmetric, pure strategy MPE by verifying that our model (3) satis(cid:133)es the conditions of Proposition 4 in Doraszelski and Satterthwaite (2005) (DS). Proposition 1 (Doraszelski and Satterthwaite (2005)) Assume that 1. (i) The state space is (cid:133)nite, i.e. N < and M < : (ii) Pro(cid:133)ts are bounded, i.e. there 1 1 exists (cid:25) < s.t. (cid:25) < (cid:25) (!) < (cid:25) for all ! and all i. (iii) Investments is bounded, i.e., i 1 (cid:0) x < and xe < : (iv) The distributions of scrap values F ( ) and setup costs Fe( ) 1 1 (cid:1) (cid:1) have continuous and positive densities and bounded supports, i.e. there exist (cid:30) < and 1 (cid:30) e < s.t. the supports of F ( ) and Fe( ) are contained in the interval (cid:30);(cid:30) and 1 (cid:1) (cid:1) (cid:0) e e (cid:2) (cid:3) (cid:30) ;(cid:30) ; respectively. (v) Firms discount future payo⁄s, i.e., (cid:12) [0;1): (cid:0) 2 h i 2. G (!;u(!);V ) is a continuous function of x(!); (cid:24)(!); and V for all ! and all i; where i i i u(!) = (x(!);(cid:24)(!)) is the vector of (cid:133)rms(cid:146)e⁄ort and cuto⁄ entry/exit strategies. 3. Transition function P ( ) is UIC admissible and x is (cid:133)nite and larger than (cid:12) V(cid:3) V (cid:3) ; (cid:1) (cid:0) S (cid:16) (cid:17) with V i V (cid:3) ;V(cid:3) j j: 2 h i 4. The local income functions are symmetric and exchangeable, i.e. G (! ;:::;! ;! ;! ;:::;! ;u (!);:::;u (!);u (!);u (!);:::;u (!);V ) i 1 i 1 i i+1 N 1 i 1 i i+1 N i (cid:0) (cid:0) = G (! ;:::;! ;! ;! ;:::;! ;u (!);:::;u (!);u (!);u (!);:::;u (!);V ) 1 i i 1 1 i+1 N i i 1 1 i+1 N 1 (cid:0) (cid:0) 41

for all symmetric functions and all i; and G (! ;! ;:::;! ;:::;! ;:::;! ;u (!);u (!);:::;u (!);:::;u (!);:::;u (!);V ) 1 1 2 k l N 1 2 k l N 1 = G (! ;! ;:::;! ;:::;! ;:::;! ;u (!);u (!);:::;u (!);:::;u (!);:::;u (!);V ) 1 1 2 l k N 1 2 l k N 1 for all exchangeable functions, k 2; and all l 2: (cid:21) (cid:21) Under assumptions 1;2; and 3, an equilibrium exists in cuto⁄ entry/exit and pure investment strategies. If, in addition assumption 4 holds, then a symmetric and anonymous equilibrium exists in cuto⁄ entry/exit and pure investment strategies. Lemma 1 There exists a symmetric MPE in pure strategies to the game that satis(cid:133)es (4) (5). (cid:0) Proof. It su¢ ces to verify that the game satis(cid:133)es assumptions 1-4 in Prop A-0. Note that for the basic model in Section 2 without entry and exit we only need to provide arguments for existence and uniqueness of compensation strategies. 1. Our model has N < (cid:133)rms with states ! 1;:::;M and M < : Firms discount i 1 2 f g 1 future payo⁄s using (1+r) 1 (0;1), and we assume that compensation expenditures (cid:0) 2 are bounded (x < ). Boundedness of cost function (assumed functional form for costs 1 implies that c(M +n) = c(M) n) implies that the pro(cid:133)t function (cid:25) (! ;! ) is bounded. (cid:3) i j 8 These boundedness conditions satisfy assumption 1 in (DS). 2. V enters G ( ) only through the expected value of (cid:133)rm i(cid:146)s future cash (cid:135)ows, ensuring i i (cid:1) continuityofG ( )inV forall! andalli. Moreover, currentpro(cid:133)tisadditivelyseparable i i (cid:1) from investment and the transition probability function P ( ) is continuous, which implies (cid:1) 42

that G( ) is a continuous function of x(!) for all ! and i: Continuity of G (!;x(!);V ) i i (cid:1) in x(!) and V satis(cid:133)es assumption 2 in (DS). i 3. OurtransitionprobabilityfunctionP (! ! ;! ;x )satis(cid:133)estheuniqueinvestmentchoice 0ij i (cid:0) i i (UIC) admissibility condition in (DS). We assume, in addition, that x > (cid:12) V(cid:3) V (cid:3) ; (cid:0) S (cid:16) (cid:17) with V i V (cid:3) ;V(cid:3) j j; which ensures that assumption 4 in (DS) holds. 2 h i 4. Our model of product market competition gives rise to symmetric pro(cid:133)t functions, i.e. (cid:25) (! ;! ) = (cid:25) (! ;! );which,togetherwiththefactthatP ! ;! ;! ;! ;x (!);x (!) 1 i j 2 j i 1 0i 0j i j i j (cid:16) (cid:17) = P ! ;! ;! ;! ;x (!);x (!) ; ensures that the local income functions G ( ) are 2 0j 0i j i j i i (cid:1) (cid:16) (cid:17) symmetric and exchangeable, and, thus, satisfy assumption 5 in (DS). Appendix B. Details of Computation This appendix describes the approach used to solve the full model with entry and exit numerically once the parameters of the model are set. Every period there are n N heterogeneous (cid:20) (cid:133)rms active and N n potential entrants. To enter from state !e shareholders must pay a (cid:0) random sunk cost of xe drawn from a distribution Fe( ) independently and identically distribi (cid:1) uted across (cid:133)rms and periods with E((cid:30)e) = (cid:30)e. Setup costs are private information. We let i (cid:31)e(!;(cid:30)e) 0;1 indicate stay out or entry respectively. If a string of unsuccessful outcomes i i 2 f g occurs, shareholders may (cid:133)nd it optimal to exit and liquidate the (cid:133)rm, in which case they get a sell-o⁄ value of (cid:30) dollars, exit in the next period and never re-enter again. Followi ing Doraszelski and Satterthwaite (2003), we assume that scrap values are randomly drawn from a distribution F ( ) with E((cid:30) ) = (cid:30); independently and identically distributed across i (cid:1) 43

(cid:133)rms and periods, and privately observed prior to making exit and e⁄ort decisions. We let (cid:31) (!;(cid:30) ) 0;1 indicate exit or continuation respectively. With respect to our earlier de(cid:133)nii i 2 f g tion in Section 2, the symmetric MPE now comprises also an operating probability, which for an incumbent is given by ’ (!) = (cid:31) (!;(cid:30) )dF ((cid:30) ) and represents the probability that ini i i i R cumbent i remains in the industry; while for a potential entrant is ’e(!) = (cid:31)e(!;(cid:30)e)dF ((cid:30)e) i i i i R and represents the probability that potential entrant i enters the industry. The solution to the problem of the (cid:133)rm is found using value and policy function iteration method along the lines of Pakes and McGuire (1994). It exploits the computational simpli(cid:133)cation entailed by the Markov Perfect assumption combined with the recursivity of the optimization problem. The algorithm iterates on the vector containing value functions, V, and the vector of policies, X, (one for each state !), until the maximum of the element-by-element di⁄erence between successive iterations in these vectors is below a pre-speci(cid:133)ed tolerance level. All computations are carried out in Gauss 3.0. The algorithm iterates on the V and X matrices until the maximum of the element-byelement di⁄erence between successive iterations in these matrices is below a pre-speci(cid:133)ed tolerance level. The calculations in each iteration are performed separately for each row (industry structure) using only the old values of the matrices V and X: If each element of V and X has converged, then we are assured of having computed a MPNE of the dynamic game. We describe the process that provides us with new V and X matrices at every iteration. The computation is done separately for each element of V and X: Thus we describe what the algorithm does to V [!;n] and X[!;n], where ! is the industry vector, and n stands for ! ; i for every [!;n] ((cid:10)n;N): Although we illustrate the updating process for the typical element 2 [!;n]; this process is done to all possible states [!;n] ((cid:10)n;N): 2 44

For a given (!;n), the values of V (!;n) and X(!;n) at each new iteration are calculated as follows: V: the value function at the kth iteration is written as (cid:15) 1 1 1 (cid:30);sup A(!;n) x+(cid:12) ::: Vk 1(!+(cid:28) (cid:23);n) x 0 (cid:0) Vk(!;n) = max8 (cid:21) (cid:0) (cid:28)1=0 (cid:28)N=0(cid:23)=0 (cid:0) (cid:2) 9 > > P P P > > > < p (cid:28) xk 1;(cid:23) ::p((cid:28) x;(cid:23))::p (cid:28) xk 1;(cid:23) p((cid:23)) > = 1 j 1(cid:0) h j N j N(cid:0) > (cid:16) (cid:17) (cid:16) (cid:17) > > > > > : ; Denote the (cid:133)rm(cid:146)s expected discounted value for each of the two possible realizations of its state process, (cid:28), as 1 1 1 1 1 ::: ::: Vk 1(z (cid:23);n)p((cid:23)) (cid:0) (cid:0) (cid:2) CV (z;n) = (cid:12)2 (cid:28)1=0 (cid:28)h 1=0(cid:28)h+1=0 (cid:28)N=0(cid:23)=0 3 P (cid:0)P P P P 6 p (cid:28) xk 1;(cid:23) ::p (cid:28) xk 1;(cid:23) p (cid:28) xk 1;(cid:23) ::p (cid:28) xk 1;(cid:23) 7 6 6 1 j 1(cid:0) h (cid:0) 1 j h(cid:0) (cid:0) 1 h+1 j h(cid:0)+1 N j N(cid:0) 7 7 4 (cid:16) (cid:17) (cid:16) (cid:17) (cid:16) (cid:17) (cid:16) (cid:17) 5 That is, CV ( ) sums over the probability weighted average of the possible states of the (cid:1) future competitors, but not over the investing (cid:133)rm(cid:146)s own future states. Hence, we can rewrite Vk as A(!;n) x+(cid:12) ax CV (!+e(n);n) Vk(!;n) = max8(cid:30);sup2 (cid:0) 1+ax 39 (7) > > < x (cid:21) 0 +(cid:12) 1 CV (!;n) > > = 6 1+ax 7 6 7 > 4 5> > > : ; where e(j) is a vector of zeros except for the jth element which is one. Then, whenever Vk(!) (cid:30) (cid:21) ax 1 Vk(!;n) = sup A(!;n) x+(cid:12) CV (!+e(n);n)+(cid:12) CV (!;n) (cid:0) 1+ax 1+ax x 0 (cid:21) (cid:20) (cid:21) X: denote by xk(!;n) the level that solves (7); and by D the derivative with respect x (cid:15) 45

to x: Assuming that the (cid:133)rm remains active, the optimal x(!;n) solves ax 1 1 = (cid:12) D CV (!+e(n);n)+D CV (!;n) x x 1+ax 1+ax (cid:20) (cid:18) (cid:19) (cid:18) (cid:19) (cid:21) ax ax 1 = (cid:12) D v1 D v2 x x 1+ax (cid:0) 1+ax (cid:20) (cid:18) (cid:19) (cid:18) (cid:19) (cid:21) and v1 CV (!+e(n);n) and v2 CV (!;n): Note that (cid:17) (cid:17) 1 a D = = a[1 p(x)]2 x 1+ax (1+ax)2 (cid:0) (cid:18) (cid:19) when (cid:28) = 1 (and, hence, p(x) = ax ). Thus, x(!;n) solves 1+ax 1 = (cid:12) a[1 p(x)]2v1 a[1 p(x)]2v2 (cid:0) (cid:0) (cid:0) h i 1 = (cid:12)a[1 p(x)]2(v1 v2) (cid:0) (cid:0) 1 [1 p(x)]2 = (cid:0) (cid:12)a(v1 v2) (cid:0) 1 = p(x) = 1 ) (cid:0)s (cid:12)a(v1 v2) (cid:0) p(x) Taking the inverse of p(x); implies x(!;n) = : a ap(x) (cid:0) Finally, we can use the derived formula to update the value function (cid:15) ax(!;n) A(!;n) x(!;n)+(cid:12) CV (!+e(n);n) Vk(!;n) = max8(cid:30);sup2 (cid:0) 1+ax(!;n) 39 > > < x (cid:21) 0 +(cid:12) 1 CV (!;n) > > = 6 1+ax(!;n) 7 6 7 > 4 5> > > : ; 46

Note that if Vk(!;n) = (cid:30), then x is 0 with probability one. Hence, the actual x level is xk(!;n) = Vk(!;n) (cid:30) x(!;n) (cid:21) n o where is the indicator function which takes the value of one when condition inside is f(cid:1)g satis(cid:133)ed, and zero otherwise. 47

Appendix C. Details of Computation of CEO Portfolio Sensitivities (Core and Guay (2002) We de(cid:133)ne the CEO(cid:146)s portfolio price sensitivity (PPS) as the change in the value of the CEO(cid:146)s stock and option portfolio due to a 1% increase in the price of the (cid:133)rm(cid:146)s common stock. Partialderivativesoftheoptionpricewithrespecttostockprice(delta)arebasedontheBlack- Scholes(1973)option-pricingmodeladjustedfordividendsbyMerton(1973),withthefollowing standard parameters: N is the cumulative probability function for the normal distribution; S is the price of the underlying stock; X is the exercise price of the option; (cid:27) is the expected stock return volatility over the life of the option; r is the risk-free interest rate; T is the time to maturity of the option in years; and d is the expected dividend yield over the life of the option. The six variables necessary to compute the delta and vega of an option are the exercise price, time to maturity, volatility, risk-free rate, dividend yield, and stock price. All of these input variables are either directly observable or can be accurately estimated. Because details on the exercise prices and maturities of CEO options are not fully disclosed in annual statements, we follow Core and Guay(cid:146)s (2002) approximation method. They show that their method explains 99% of the actual variation in option portfolio values and sensitivities. WepartitiontheCEO(cid:146)soptionportfoliointothreeparts: (1)optionsfromnewgrants, (2) exercisable options from previous grants, and (3) non-exercisable options from previous grants. ExecuComp provides full information on exercise prices (item EXPRIC in ExecuComp) and times to maturity for new grants (item EXDATE in ExecuComp ), which makes the computation of option delta fairly straightforward. However, for previously granted options, no data are available on exercise prices and times to maturity. To estimate the average exercise price for previously granted options, we use the (cid:147)realizable values(cid:148)as in Core and Guay (2002). The 48

realizable value is the immediate exercise value of the CEO(cid:146)s options. We divide the realizable value of previously granted options by the number of options to (cid:133)nd how much, on average, the stock price is above the exercise price. Subtracting this (cid:133)gure from the stock price yields the exercise price. We follow Core and Guay (2002) when estimating times to maturity for previously granted options (unexercisable and exercisable). First, we assume that the time to maturity of an unexercisable option is one year less than that of a new grant. This assumption is consistent with evidence in Kole (1997) that vesting periods are narrowly bounded between 20 and 28 months, with an average of 24 months. Second, we assume that the time to maturity of an exercisable option is three years less than that of an unexercisable option. Consequently, we set the maturity of an unexercisable (exercisable) option to the new grant(cid:146)s maturity minus one (four). If no options are granted in the current year, we set the time to maturity of an exercisable (unexercisable) option to six (nine) years. Once the delta ((cid:1)) of each option partition are determined, we calculate the CEO(cid:146)s portfolio price sensitivity (PPS) as follows: S PPS = ((cid:1) N +(cid:1) N +(cid:1) N +N ) NG NG PGEX PGEX PGUNEX PGUNEX STOCK 100 where S and N. denote the stock price and the number of options/stocks in hundreds of thousands. The subscripts NG, PGEX, PGUNEX, and STOCK stand for new grants, previously granted exercisable options, previously granted non-exercisable options, and stock holdings, respectively. 49

Appendix D. Figures and Tables Figure 1: Timeline Shareholders Managers Managers Nature Shareholders choose choose choose draws collect Continue s ,a p x ‡ 0; e p it it it it t it Managers w collect t realized w it Exit f t t+1 50

Figure 2: CEO Incentives, Value, and Equilibrium Industry Structure Panel A: CEO Incentives 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 91S 81S 71S 61S 51S 41S 31S 21S 11S 01S 9S 8S 7S 6S 5S 4S 3S 2S 1S 1 0 Firm 2 Firm 1 Panel B: Value Function and Equilibrium Industry Structure 250 0.2 200 150 0.1 100 50 0 0 S16 S16 16 16 Firm 2 S11 11 S11 11 Firm 1 S6 6 S6 6 S11 S11 Panel A plots optimal CEO incentives of Firm 1, (cid:11) (!); as a function of the state of the industry, (cid:3)1 !, for two (cid:133)rms and a given state of the other active (cid:133)rms. Panel B plots the value function of Firm 1, V (!); as a function of the state of the industry, ! (left panel); and the ergodic distribution 1 (frequency) of Markov-Perfect equilibrium industry states ! (right panel). Higher states correspond to higher quality. 51

Figure 3: Average CEO Incentives and Equilibrium Industry Structure Panel A: CEO Incentives - Industry Average 12S 91S 71S 51S 31S 11S 9S 7S 5S 3S 11S 3 5 7 9 11 31 51 71 91 12 1 0.75 0.5 0.25 0 Panel B: Comparative Dynamics of Equilibrium Industry Structure (cid:14) = 0:3 (cid:14) = 0:4 0.2 0.2 0.1 0.1 0 0 S16 S16 16 16 S11 11 S11 11 S6 6 S6 6 S11 S11 (cid:14) = 0:6 (cid:14) = 0:7 0.2 0.2 0.1 0.1 0 0 S16 16 S16 16 S11 11 S11 11 S6 6 S6 6 S11 S11 Panel A plots the average optimal CEO incentives of Firm 1 and Firm 2, (cid:11) (!)+(cid:11) (!), as a (cid:3)1 (cid:3)2 function of the state of the industry, !, for a given state of the other active (cid:133)rms. Panel B plots, for a range of aggregate shock parameters (cid:14), the ergodic distribution (frequency) of Markov-Perfect equilibrium industry states !. Higher states correspond to higher quality. 52

Figure 4: Pay-Performance Sensitivity and Firm Position $1.60 $1.40 $1.20 $1.00 $0.80 $0.60 $0.40 $0.20 $0.00 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Firm position (ratio of sales to median industry sales) eulav mrif ni egnahc 0001$ hcae rof noitasnepmoc latot OEC ni egnahC This (cid:133)gure plots estimated pay-performance sensitivities for CEOs of (cid:133)rms in di⁄erent position deciles. Pay-performance sensitivity is estimated with pooled OLS regressions. The dependent variable is a measure of executive compensation in a particular (cid:133)rm year. The measure of total CEO compensation consists of short-term compensation and long-term compensation. Short-term compensation consistsofsalary, bonus, andotherannualpayments(e.g., gross-upsfortaxliabilities, perquisites, preferential discounts on stock purchases). Long-term compensation includes the value of restricted stock granted, stock options granted, payouts from long-term incentive plans, and all other compensation (e.g., contributions to bene(cid:133)t plans, severance payments). Performance is de(cid:133)ned as the product of the total in(cid:135)ation-adjusted return to shareholders and the beginning of period market value of the (cid:133)rm (Jensen and Murphy (1990)). Industry is de(cid:133)ned by four-digit SIC code. Position is the ratio of the (cid:133)rm(cid:146)s sales to industry median sales in the beginning of the year, winsorized at 1%. Data is annual for 1993-2004, with only manufacturing (SIC 2000-3999) (cid:133)rms included. 53

Table 1: Parameter Values and Summary Statistics Panel A: Parameter Values Parameter Description Benchmark Value D demand 15 (cid:14) aggregate shock 0:6 (cid:30) scrap value 0:1 X entry cost 0:2 e (cid:12) discount rate 0:96 Panel B: Summary Statistics Statistics Panel Data Model Number of (cid:133)rms 6 6 Industry concentration 0.37 0.43 Median (cid:133)rm age 30 36 Standard deviation of (cid:133)rm age 21 18 Summarystatisticsforthemodelarecomputedover15yearperiods,startingatrandomdrawsfrom theergodicdistributionofstates,repeatingtheprocedure1000times. Industryconcentrationisde(cid:133)ned asthemarketshareofthefourlargest(cid:133)rmsintheindustry. Ageisthenumberofperiodssincethe(cid:133)rm (cid:133)rst appeared in the arti(cid:133)cial panel. In the panel data, age is from Jovanovic and Rousseau (2001). 54

Table 1.2: Cross-Sectional Results I Panel A: Symmetry and Growth E⁄ects Parameter Frequency of Symmetric States CEO Incentives (cid:14) = 0:3 64.7% 0.76 (cid:14) = 0:4 50.8% 0.62 (cid:14) = 0:5 50.0% 0.60 (cid:14) = 0:6 32.5% 0.47 (cid:14) = 0:7 25.7% 0.32 (cid:14) = 0:9 11.7% 0.15 Panel B: Turbulence E⁄ect Parameter Turnover Rate CEO Incentives xe = 0:01 29.5% 0.54 xe = 0:1 20.9% 0.49 xe = 0:2 15.6% 0.47 xe = 0:3 7.8% 0.32 xe = 0:4 2.3% 0.20 Panel A reports, for a range of aggregate shock parameters (cid:14), the incidence of symmetric states and average CEO incentives in the Markov-Perfect industry equilibrium. Incidence of symmetric states is computed as percentage of states such that ! = ! L for all i=1,...,7, and L=2. Average i i (cid:0) (cid:6) CEO incentives is computed as the arithmetic average of optimal CEO incentives, (cid:11) , across all (cid:133)rms. (cid:3) Panel B reports, for a range of entry costs xe, the average turnover rate and average CEO incentives in the Markov-Perfect industry equilibrium. Average turnover rate is computed as as {probability of entry+probability of exit-probability of entry and exit}. Average CEO incentives is computed as the arithmetic average of optimal CEO incentives, (cid:11) , across all (cid:133)rms. (cid:3) 55

Table 1.3: Cross-Sectional Results II Panel A: Interaction E⁄ects - Symmetry Parameter Leader Laggard (cid:14) = 0:3 0.17 0.95 (cid:14) = 0:4 0.17 0.83 (cid:14) = 0:5 0.15 0.72 (cid:14) = 0:6 0.14 0.57 (cid:14) = 0:7 0.11 0.41 (cid:14) = 0:9 0.08 0.21 Panel B: Interaction E⁄ects - Turbulence Parameter Leader Laggard xe = 0:01 0.18 0.68 xe = 0:1 0.16 0.62 xe = 0:2 0.14 0.57 xe = 0:3 0.14 0.39 xe = 0:4 0.13 0.26 This table reports, for a range of aggregate shock parameters (cid:14) (Panel A) and for a range of entry costs xe (Panel B), the average CEO incentives of leaders and laggards in the Markov-Perfect industry equilibrium. AverageCEOincentivesiscomputedasthearithmeticaverageofoptimalCEOincentives, (cid:11) , across all (cid:133)rms. For any industry state, each (cid:133)rm(cid:146)s position is calculated as the ratio of its sales to (cid:3) industry median sales. Leaders are (cid:133)rms in the highest quartile of the ratio, laggards are (cid:133)rms in the lowest quartile of the ratio. 56

Table 2: Components of Executive Compensation This Table presents descriptive statistics on the components of executive compensation for all executives in the ExecuComp sample for years 1993-2004 for whom complete data on total compensation is available. The top panel of the table pertains to the executives who are identi(cid:133)ed as the chief executive o¢ cerofthe(cid:133)rm. Thebottompaneldescribestheotherexecutivesinthesample. Themeasureoftotal compensation can be divided into short-term compensation and long-term compensation. Short-term compensation consists of salary, bonus, and other annual payments (e.g., gross-ups for tax liabilities, perquisites, preferential discounts on stock purchases). Long-term compensation includes the value of restricted stock granted, stock options granted, payouts from long-term incentive plans, and all other compensation(e.g.,contributionstobene(cid:133)tplans,severancepayments). Long-termshareistheaverage share of compensation that is long-term, at the individual level. Portfolio wealth delta is the price sensitivity of the CEO(cid:146)s option portfolio calculated as in Core and Guay (2002). Payment Category Mean Median Standard (Thousands of Dollars) Deviation CEOs (N=8320) Total Compensation 4315 2051 12677 Short Term Compensation 1217 893 1180 Salary 599 544 325 Bonus 569 322 937 Other Annual 49 0 250 Long Term Compensation 3097 1011 12405 Restricted Stock Granted 283 0 1633 Stock Options Granted 2508 703 12078 LT Incentive Plan Payout 166 0 813 All Other 138 20 845 Long-Term Share of Total 0.484 0.427 0.264 Portfolio Wealth Delta 4.25 1.41 7.79 Non-CEOs (N=38544) Total Compensation 1442 746 2844 Short Term Compensation 490 365 830 Salary 281 244 164 Bonus 191 105 752 Other Annual 19 0 185 Long Term Compensation 922 310 2516 Restricted Stock Granted 80 0 554 Stock Options Granted 727 209 2191 LT Incentive Plan Payout 50 0 263 All Other 52 9 490 Long-Term Share of Total 0.423 0.444 0.270 57

Table 3: Executive Compensation and Competitive Position - All Executives This table reports pooled OLS regressions of pay-performance sensitivity. The dependent variable is a measure of executive compensation in a particular (cid:133)rm year. The measure of total compensation consists of short-term compensation and long-term compensation. Short-term compensation consists of salary,bonus,andotherannualpayments. Long-termcompensationincludesthevalueofrestrictedstock granted, stock options granted, payouts from long-term incentive plans, and all other compensation. Portfolio wealth delta is the price sensitivity of the CEO(cid:146)s option portfolio calculated as in Core and Guay(2002). Performanceisde(cid:133)nedastheproductofthetotalin(cid:135)ation-adjustedreturntoshareholders and the beginning of period market value of the (cid:133)rm divided by 100 (Jensen and Murphy (1990)). Position within industry is determined by the ratio of the (cid:133)rm(cid:146)s sales to industry median sales in the beginning of the year, winsorized at 1%. Industry is de(cid:133)ned by four-digit SIC code.. Data is annual for 1993-2004, with only manufacturing (SIC 2000-3999) (cid:133)rms included. All regressions include year and industry (cid:133)xed e⁄ects. Industry (cid:133)xed e⁄ects are at the 2-digit SIC level. Standard errors are robust to heteroskedasticity and arbitrary serial correlation within industry-year cells. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. All Firms Panel A: Annual Compensation Performance 0.337 (cid:3)(cid:3)(cid:3) (0.131) Position 437.402 (cid:3)(cid:3)(cid:3) (168.611) Performance* -0.329 (cid:3)(cid:3)(cid:3) Position (0.122) Performance* -0.223 (cid:3)(cid:3)(cid:3) Size (0.030) Observations 13080 Firms 449 Adjusted R2 0.23 Panel B: CEO Portfolio Wealth Delta Position -0.307 (cid:3)(cid:3)(cid:3) (0.060) Observations 5698 Firms 423 Adjusted R2 0.14 Industry (cid:133)xed e⁄ects Yes Year (cid:133)xed e⁄ects Yes 58

Table 4: Executive Compensation and Symmetry - All Executives This table reports pooled OLS regressions of pay-performance sensitivity. The dependent variable is a measure of executive compensation in a particular (cid:133)rm year. The measure of total compensation can be divided into short-term compensation and long-term compensation. Short-term compensation consists of salary, bonus, and other annual payments (e.g., gross-ups for tax liabilities, perquisites, preferential discounts on stock purchases). Long-term compensation includes the value of restricted stockgranted,stockoptionsgranted,payoutsfromlong-termincentiveplans,andallothercompensation (e.g., contributions to bene(cid:133)t plans, severance payments). Performance is de(cid:133)ned as the product of the total in(cid:135)ation-adjusted return to shareholders and the beginning of period market value of the (cid:133)rm (Jensen and Murphy (1990)). Industry is de(cid:133)ned by four-digit SIC code. Symmetry is industry-year average proximity to median sales in the industry. Position within industry is determined by the ratio of the (cid:133)rm(cid:146)s sales to industry median sales in the beginning of the year, winsorized at 1%: leaders are (cid:133)rms in the highest quartile of the ratio, laggards are (cid:133)rms in the lowest quartile of the ratio. The regressions with controls include industry concentration, measured as domestic four-(cid:133)rm concentration ratio, and (cid:133)rm size, measured as assets at the beginning of the year and winsorized at 1%, and their respective interactions with performance. These coe¢ cients are omitted from the table for brevity, and are available upon request. Industry (cid:133)xed e⁄ects are at the 2-digit SIC level. Data is annual for 1993-2004, with only manufacturing (SIC 2000-3999) (cid:133)rms included. Standard errors are robust to heteroskedasticity and arbitrary serial correlation within industry-year cells. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. All Firms By Position: Short Term Leader Laggard Compensation no (cid:133)xed (cid:133)xed (cid:133)xed (cid:133)xed (cid:133)xed (cid:133)xed e⁄ects, no e⁄ects, no e⁄ects, e⁄ects, e⁄ects, e⁄ects, controls controls controls controls controls controls (1) (2) (3) (4) (5) (6) Performance 0.062 0.117 0.344 0.234 0.313 0.001 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (0.003) (0.007) (0.017) (0.031) (0.113) (0.003) Symmetry -366.031 -449.727 46.555 267.468 38.481 -8.244 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3) (47.498) (49.078) (48.918) (157.382) (66.983) (9.884) Performance* 0.027 0.025 0.020 -0.001 0.205 0.001 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3) Symmetry (0.003) (0.004) (0.003) (0.006) (0.076) (0.001) Industry (cid:133)xed e⁄ects No Yes Yes Yes Yes Yes Year (cid:133)xed e⁄ects Yes Yes Yes Yes Yes Yes Observations 30326 30326 30216 8670 8300 30326 Firms 449 449 449 114 112 449 Adjusted R2 0.29 0.36 0.37 0.32 0.25 0.35 59

Table 5: Executive Compensation and Symmetry - CEO Only This table reports pooled OLS regressions of pay-performance sensitivity. The dependent variable is a measure of executive compensation in a particular (cid:133)rm year. The measure of total compensation can be divided into short-term compensation and long-term compensation. Short-term compensation consists of salary, bonus, and other annual payments (e.g., gross-ups for tax liabilities, perquisites, preferential discounts on stock purchases). Long-term compensation includes the value of restricted stockgranted,stockoptionsgranted,payoutsfromlong-termincentiveplans,andallothercompensation (e.g., contributions to bene(cid:133)t plans, severance payments). Portfolio wealth delta is the price sensitivity of the CEO(cid:146)s option portfolio calculated as in Core and Guay (2002). Performance is de(cid:133)ned as the product of the total in(cid:135)ation-adjusted return to shareholders and the beginning of period market value of the (cid:133)rm (Jensen and Murphy (1990)). Industry is de(cid:133)ned by four-digit SIC code. Symmetry is industry-yearaverageproximitytomediansalesintheindustry. Positionwithinindustryisdetermined by the ratio of the (cid:133)rm(cid:146)s sales to industry median sales in the beginning of the year, winsorized at 1%: leaders are (cid:133)rms in the highest quartile of the ratio, laggards are (cid:133)rms in the lowest quartile of the ratio. Theregressionswith controlsinclude industryconcentration, measured asdomestic four-(cid:133)rm concentrationratio,and(cid:133)rmsize,measuredasassetsatthebeginningoftheyearandwinsorizedat1%, and their respective interactions with performance. These coe¢ cients are omitted from the table for brevity,andareavailableuponrequest. Industry(cid:133)xede⁄ectsareatthe2-digitSIClevel. Dataisannual for 1993-2004, with only manufacturing (SIC 2000-3999) (cid:133)rms included. Standard errors are robust to heteroskedasticity and arbitrary serial correlation within industry-year cells. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. All Firms By Position: Leader Laggard no (cid:133)xed (cid:133)xed (cid:133)xed (cid:133)xed (cid:133)xed e⁄ects, no e⁄ects, no e⁄ects, e⁄ects, e⁄ects, controls controls controls controls controls (1) (2) (3) (4) (5) Panel A: Annual Compensation Performance 0.116 0.113 0.608 0.385 0.703 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (0.011) (0.011) (0.025) (0.047) (0.252) Symmetry -925.050 -1202.607 -125.116 -234.478 317.852 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (153.185) (157.654) (78.382) (246.646) (160.425) Performance* 0.025 0.022 0.021 0.006 0.588 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) Symmetry (0.010) (0.011) (0.005) (0.010) (0.186) Observations 5805 5805 5783 1462 1445 Firms 449 449 449 114 112 Adjusted R2 0.38 0.39 0.35 0.38 0.47 Panel B: CEO Portfolio Wealth Delta Symmetry 2.553 2.490 2.391 1.474 4.463 (cid:3)(cid:3) (cid:3)(cid:3) (cid:3)(cid:3) (cid:3)(cid:3) (1.016) (1.002) (1.011) (0.926) (1.868) Observations 5702 5702 5702 1435 1420 Firms 423 423 423 109 106 Adjusted R2 0.11 0.14 0.22 0.18 0.25 Industry (cid:133)xed e⁄ects No Yes Yes Yes Yes Year (cid:133)xed e⁄ects Yes Yes Yes Yes Yes 60

Table 6: Executive Compensation and Job Turnover - All Executives This table reports pooled OLS regressions of pay-performance sensitivity. The dependent variable is a measure of executive compensation in a particular (cid:133)rm year. The measure of total compensation can be divided into short-term compensation and long-term compensation. Short-term compensation consists of salary, bonus, and other annual payments (e.g., gross-ups for tax liabilities, perquisites, preferential discounts on stock purchases). Long-term compensation includes the value of restricted stockgranted,stockoptionsgranted,payoutsfromlong-termincentiveplans,andallothercompensation (e.g., contributions to bene(cid:133)t plans, severance payments). Performance is de(cid:133)ned as the product of the total in(cid:135)ation-adjusted return to shareholders and the beginning of period market value of the (cid:133)rm (Jensen and Murphy (1990)). Industry is de(cid:133)ned by four-digit SIC code. Turnover is industry average annualized job creation and destruction (Davis, Haltiwanger, and Schuh (1998)). Position within industry is determined by the ratio of the (cid:133)rm(cid:146)s sales to industry median sales in the beginning oftheyear, winsorizedat1%: leadersare(cid:133)rmsinthehighestquartileoftheratio, laggardsare(cid:133)rmsin the lowest quartile of the ratio. The regressions with controls include industry concentration, measured asdomesticfour-(cid:133)rmconcentrationratio, and(cid:133)rmsize, measuredasassetsatthebeginningoftheyear andwinsorizedat1%,andtheirrespectiveinteractionswithperformance. Thesecoe¢ cientsareomitted from the table for brevity, and are available upon request. Industry (cid:133)xed e⁄ects are at the 2-digit SIC level. Dataisannualfor1993-2004, withonlymanufacturing(SIC2000-3999)(cid:133)rmsincluded. Standard errorsarerobusttoheteroskedasticityandarbitraryserialcorrelationwithinindustry-yearcells. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. All Firms By Position: Short Term Leader Laggard Compensation no (cid:133)xed (cid:133)xed (cid:133)xed (cid:133)xed (cid:133)xed (cid:133)xed e⁄ects, no e⁄ects, no e⁄ects, e⁄ects, e⁄ects, e⁄ects, controls controls controls controls controls controls (1) (2) (3) (4) (5) (6) Performance 0.051 0.049 0.327 0.219 0.393 0.010 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (0.002) (0.002) (0.018) (0.032) (0.100) (0.004) Turnover -27.816 14.646 -87.192 -240.589 -149.056 -0.725 (18.996) (20.059) (55.453) (185.810) (72.472) (11.213) Performance* 0.010 0.011 0.035 0.020 0.226 0.005 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3) Turnover (0.001) (0.001) (0.006) (0.011) (0.088) (0.001) Industry (cid:133)xed e⁄ects No Yes Yes Yes Yes Yes Year (cid:133)xed e⁄ects Yes Yes Yes Yes Yes Yes Observations 29606 29606 29596 8526 8245 29606 Firms 449 449 449 114 112 449 Adjusted R2 0.18 0.18 0.36 0.31 0.24 0.34 61

Table 7: Executive Compensation and Job Turnover - CEO Only This table reports pooled OLS regressions of pay-performance sensitivity. The dependent variable is a measure of executive compensation in a particular (cid:133)rm year. The measure of total compensation can be divided into short-term compensation and long-term compensation. Short-term compensation consists of salary, bonus, and other annual payments (e.g., gross-ups for tax liabilities, perquisites, preferential discounts on stock purchases). Long-term compensation includes the value of restricted stockgranted,stockoptionsgranted,payoutsfromlong-termincentiveplans,andallothercompensation (e.g., contributions to bene(cid:133)t plans, severance payments). Performance is de(cid:133)ned as the product of the total in(cid:135)ation-adjusted return to shareholders and the beginning of period market value of the (cid:133)rm (Jensen and Murphy (1990)). Industry is de(cid:133)ned by four-digit SIC code. Turnover is industry average annualized job creation and destruction (Davis, Haltiwanger, and Schuh (1998)). Position within industry is determined by the ratio of the (cid:133)rm(cid:146)s sales to industry median sales in the beginning oftheyear, winsorizedat1%: leadersare(cid:133)rmsinthehighestquartileoftheratio, laggardsare(cid:133)rmsin the lowest quartile of the ratio. The regressions with controls include industry concentration, measured asdomesticfour-(cid:133)rmconcentrationratio, and(cid:133)rmsize, measuredasassetsatthebeginningoftheyear andwinsorizedat1%,andtheirrespectiveinteractionswithperformance. Thesecoe¢ cientsareomitted from the table for brevity, and are available upon request. Industry (cid:133)xed e⁄ects are at the 2-digit SIC level. Dataisannualfor1993-2004, withonlymanufacturing(SIC2000-3999)(cid:133)rmsincluded. Standard errorsarerobusttoheteroskedasticityandarbitraryserialcorrelationwithinindustry-yearcells. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. All Firms By Position: Short Term Leader Laggard Compensation no (cid:133)xed (cid:133)xed (cid:133)xed (cid:133)xed (cid:133)xed (cid:133)xed e⁄ects, no e⁄ects, no e⁄ects, e⁄ects, e⁄ects, e⁄ects, controls controls controls controls controls controls (1) (2) (3) (4) (5) (6) Performance 0.108 0.225 0.658 0.555 0.710 0.045 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (0.012) (0.025) (0.051) (0.083) (0.331) (0.010) Turnover -48.854 428.279 -99.542 -41.221 -331.054 28.306 (165.951) (181.388) (164.047) (495.688) (217.116) (36.673) Performance* -0.010 0.004 0.052 0.038 0.522 0.021 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) Turnover (0.011) (0.020) (0.018) (0.029) (0.113) (0.004) Industry (cid:133)xed e⁄ects No Yes Yes Yes Yes Yes Year (cid:133)xed e⁄ects Yes Yes Yes Yes Yes Yes Observations 5651 5651 5651 1462 1445 5651 Firms 449 449 449 114 112 449 Adjusted R2 0.38 0.43 0.54 0.53 0.41 0.54 62

Table 8: Executive Compensation and Industry Dynamics - CEOs only This table reports pooled OLS regressions of pay-performance sensitivity. The dependent variable is a measure of executive compensation in a particular (cid:133)rm year. The measure of total compensation consists of short-term compensation and long-term compensation. Short-term compensation consists of salary,bonus,andotherannualpayments. Long-termcompensationincludesthevalueofrestrictedstock granted, stock options granted, payouts from long-term incentive plans, and all other compensation. Portfolio wealth delta is the price sensitivity of the CEO(cid:146)s option portfolio calculated as in Core and Guay(2002). Performanceisde(cid:133)nedastheproductofthetotalin(cid:135)ation-adjustedreturntoshareholders and the beginning of period market value of the (cid:133)rm divided by 100 (Jensen and Murphy (1990)). Position within industry is determined by the ratio of the (cid:133)rm(cid:146)s sales to industry median sales in the beginning of the year, winsorized at 1%. Industry is de(cid:133)ned by four-digit SIC code.. Data is annual for 1993-2004, with only manufacturing (SIC 2000-3999) (cid:133)rms included. All regressions include year and industry (cid:133)xed e⁄ects. Industry (cid:133)xed e⁄ects are at the 2-digit SIC level. Standard errors are robust to heteroskedasticity and arbitrary serial correlation within industry-year cells. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. All Firms Panel A: Annual Compensation Performance 0.110 (cid:3)(cid:3)(cid:3) (0.015) Growing Industry 1222.526 (cid:3)(cid:3)(cid:3) (318.296) Performance* 0.041 (cid:3)(cid:3)(cid:3) Growing Industry (0.015) Declining Industry -642.385 (cid:3)(cid:3)(cid:3) (300.251) Performance* -0.013 Declining Industry (0.020) Observations 13080 Firms 449 Adjusted R2 0.12 Panel B: CEO Portfolio Wealth Delta Growing Industry 2.220 (cid:3)(cid:3)(cid:3) (0.220) Declining Industry -1.086 (cid:3)(cid:3)(cid:3) (0.225) Observations 5698 Firms 423 Adjusted R2 0.06 Industry (cid:133)xed e⁄ects No Year (cid:133)xed e⁄ects Yes 63

Table 9: Summary and Robustness This table reports pooled OLS regressions of pay-performance sensitivity. The dependent variable is a measure of executive compensation in a particular (cid:133)rm year. The measure of total compensation can be divided into short-term compensation and long-term compensation. Short-term compensation consists of salary, bonus, and other annual payments (e.g., gross-ups for tax liabilities, perquisites, preferentialdiscountsonstockpurchases). Long-termcompensationincludesthevalueofrestrictedstock granted,stockoptionsgranted,payoutsfromlong-termincentiveplans,andallothercompensation(e.g., contributions to bene(cid:133)t plans, severance payments). Performance is de(cid:133)ned as the product of the total in(cid:135)ation-adjusted return to shareholders and the beginning of period market value of the (cid:133)rm (Jensen and Murphy (1990)). Percentage Owned is the percentage of common equity held by the executive through stocks and options. Industry is de(cid:133)ned by four-digit SIC code. Symmetry is industry-year averageproximitytomediansalesintheindustry. Turnoverisindustryaverageannualizedjobcreation and destruction (Davis, Haltiwanger, and Schuh (1998)). Position within industry is determined by the ratio of the (cid:133)rm(cid:146)s sales to industry median sales in the beginning of the year, winsorized at 1%: leaders are (cid:133)rms in the highest quartile of the ratio, laggards are (cid:133)rms in the lowest quartile of the ratio. The regressions with controls include industry concentration, measured as domestic four-(cid:133)rm concentration ratio, and (cid:133)rm size, measured as assets at the beginning of the year and winsorized at 1%, and their respective interactions with performance. These coe¢ cients are omitted from the table for brevity, and are available upon request. Industry (cid:133)xed e⁄ects are at the 2-digit SIC level. Data is annual for 1993-2004, with only manufacturing (SIC 2000-3999) (cid:133)rms included. Standard errors are robust to heteroskedasticity and arbitrary serial correlation within industry-year cells. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. All Executives CEO Only robustness summary robustness summary (ownership) leader laggard (ownership) leader laggard (1) (2) (3) (4) (5) (6) Performance 0.723 0.072 0.578 1.827 0.477 1.505 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (0.059) (0.055) (0.206) (0.214) (0.133) (0.686) Percentage Owned 41.647 544.921 (cid:3) (88.739) (318.458) Performance* -0.578 -1.135 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) Percentage Owned (0.027) (0.117) Performance* 0.181 0.026 0.261 0.416 0.009 1.149 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) Turnover (0.038) (0.035) (0.134) (0.132) (0.081) (0.468) Performance* 0.305 0.002 0.152 0.287 0.005 0.846 (cid:3)(cid:3)(cid:3) (cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3) Symmetry (0.029) (0.007) (0.071) (0.102) (0.018) (0.388) Observations 1848 4712 4185 1143 903 844 Firms 449 114 112 449 114 112 Adjusted R2 0.21 0.38 0.26 0.40 0.63 0.48 64