Managerial Legacies, Entrenchment and Strategic Inertia

Transcription

1 Managerial Legaies, Entrenhment and Strategi Inertia Catherine Casamatta CRG, University of Toulouse I Alexander Guembel Saïd Business Shool and Linoln College University of Oxford July 12, 2006 We would like to thank Bruno Biais, Roberta Dessì, Andrea Eisfeldt, Guido Friebel, Denis Gromb, Thomas Mariotti, Colin Mayer, Meg Meyer, Jean Charles Rohet, Elu von Thadden, Luy White and seminar partiipants at Nuffield College, Mannheim University, Saïd Business Shool, the 4th Corporate Finane workshop at the LSE, the EFMA 2005, ESSFM Gerzensee 2005, the 10th Anniversary Finane Meeting of CEMAF/ISCTE and the AFA 2006 for helpful omments. Guembel would like to thank the IDEI, where part of this researh was arried out, for their kind hospitality and gratefully aknowledges funding reeived by the Mihael Zilkha Trust at Linoln College. All remaining errors are ours. Correspondene address: Alexander Guembel, Linoln College, Oxford, OX1 3DR, UK, Tel.: , alexander.guembel@sbs.ox.a.uk

2 Managerial Legaies, Entrenhment and Strategi Inertia ABSTRACT This paper investigates a firm s deision to retain or fire a poorly performing CEO when the CEO s hoie of strategy has long-term ash flow impliations beyond the CEO s tenure (he leaves behind a legay). Replaing a CEO beomes expensive and entrenhment emerges, beause in the presene of a legay the firm s future performane an only partially be attributed to a newly employed manager. Future performane therefore affets a new CEO s reputation less than the inumbent s. This makes it more expensive to provide inentives to a new manager. Moreover, when the firm hooses to hange its strategy it will optimally do so with a different manager, beause the inumbent has a vested interest in seeing the new strategy fail: Failure improves the inumbent s reputation beause it is regarded as a vindiation of his original hoie. Changing strategy, however, is ostly, beause of the high expliit wages that a new manager needs to be paid. Firms therefore exhibit inertia in their adoption of new strategies. We thus endogenise the link between CEO turnover and strategy hanges. We derive empirial impliations regarding firm s poliies to re-set the strike prie of exeutive stok options, managerial replaement deisions, and apital realloation. Keywords: Reputational onerns, exeutive stok options, entrenhment, strategi inertia, managerial turnover. JEL lassifiation numbers: D82, G30, J33 I. Introdution A ompany s CEO typially takes ations that affet the firm s future performane beyond his own tenure. The CEO leaves behind a legay. How an a firm inentivize a new CEO who inherits the legay? How does a legay affet the firm s deision to dismiss a poorly performing CEO and to adopt new strategies? And what is the nature of the relationship between a CEO and the strategy he stands for? Even though these questions are at the ore of ompany s onerns in the fae of poor performane, existing theory provides little guidane regarding the eonomi mehanisms that link CEOs to strategies or that link suessive CEOs intertemporally. This paper is aimed at addressing these questions. The starting point of this paper is the observation that there is an intertemporal link between a new CEO s inentives and the ations taken by his predeessor. The fat that suh a link exists is probably more reognized in the aounting literature than in eonomis. Pouriau (1993) for example provides evidene that the suessor of a CEO who has left the firm for non-

3 routine reasons, tends to depress aounting earnings during their first year in offie. Pouriau interprets this as an attempt to take a bath while still being able to blame the predeessor. In addition, there is rih anedotal evidene that illustrates the prevalene of the blame game in whih CEO suessors often engage, with the orresponding hallenge for outsiders to attribute responsibility fairly. For example, when Mike Parton, former CEO of Maroni, was asked How muh blame do you aept for Maroni s [...] troubles? he replied that You an t be part of management and just wash your hands of it. However, I was not a board member when the key strategi deisions were made, and it s diffiult to say what my view might have been had I been on the board at the time. 1 This paper formalizes the intertemporal link between the performane of two generations of CEOs and thereby introdues the notion of a managerial legay. When a CEO works for a firm he may initiate a long-term strategy that affets the distribution of the firm s ash flows for several periods into the future. For example, deisions that involve large apital expenditures or that establish a firm s reputation in a partiular market segment typially have long lasting impat. Beause of this, suh long-term strategi deisions may affet the inumbent CEO s future inentives to implement a given strategy as well as the inentives of a potential suessor who inherits the inumbent CEO s legay. By the same token, if the firm an adopt a new strategy, its suess or failure may throw light on the quality of the initial strategi deision made by the CEO. This gives the inumbent CEO a vested interest in seeing his initial strategy sueed that a new manager does not have in the same way. Conversely, an inumbent CEO who has to hange strategy may have a vested interest in seeing that new strategy fail so as to vindiate his original hoie. As a result ontrat design, employment deisions and strategy hoies are intrinsially linked. The aim of this paper is to explore these dynami links when managers an reate legaies in the firms they manage. We onsider a set-up in whih a firm manager an be inentivized both expliitly and impliitly through reputational onerns. The manager needs to take a long-term strategi deision the quality of whih depends on his intrinsi and unknown ability. After a strategy has been hosen it needs to be implemented, whih requires managerial effort over two periods. Cash flows are also generated over two periods and depend both on the initial hoie of strategy and on the implementation effort. The firm an dismiss the inumbent manager after poor performane and hire a new manager to ontinue the projet s implementation. We assume that a new manager is equally well suited to implement the projet. We first determine the optimal wage ontrats and employment poliy for shareholders. We show that in the presene of reputational onerns by managers, the threat of firing after bad performane is time inonsistent, i.e., the inumbent manager is entrenhed. The reason is that it is heaper to employ an inumbent manager after he performed badly than to hire a new manager, beause the new manager knows that his reputation is less sensitive to performane than that of the inumbent. As a result, higher expliit inentives for the new manager have to replae the impliit inentives of the inumbent, whih is ostly to the firm. From an ex ante 1 Institutional Investor, April

4 point of view, entrenhment an be ostly, though. We haraterize under whih onditions entrenhment is detrimental to shareholders. Seond, we explore how the presene of a legay affets the joint deision to hange manager and strategy. We therefore allow for the ase where the firm an hange its strategy at an interim date. We show that in suh a situation, the firm will always hange the manager if it hanges its strategy. This is beause the manager s reputation is linked to his original deision, whih gives him an inentive to sabotage the new strategy. Hene, we predit that there is a lose assoiation between hanges in a firm s strategy and managerial turnover. A diret impliation of this is that firms will exhibit strategi inertia. In the presene of reputational onerns, the firm faes a ost when it hanges strategy: it forgoes the advantage of employing a manager who is inentivized strongly by the desire to prove that his original strategy hoie was orret. The firm may therefore refrain from hanging strategy (and manager) even when it knows that an alternative strategy is more likely to be profitable. To illustrate the ideas developed in this paper, onsider the example of the reent hanges at DaimlerChrysler. Former CEO Jurgen Shrempp was replaed by Dieter Zetshe in January, 2006, after 10 years at the head of the ompany. Shrempp s vision had been to reate a Welt AG (World In), i.e. the world s leading transportation ompany. To do this, he merged Daimler and Chrysler in 1998, and took important stakes in Asian armakers Mitsubishi and Hyundai in Despite subsequent years of poor performane, and against general opinion, Shrempp was not dismissed. Quite unexpetedly, a few months after renewing Shrempp s mandate in early 2005, the board of diretors deided to dismiss him, to promote Dieter Zetshe, and to abandon the firm s worldwide strategy in favor of a refous on the firm s ore ompetenies. Business analysts omments were: The bak-to-basis move, whih aims to streamline management and ut osts by a net $ 1 billion a year is a dramati reversal of the imperial management ulture and high-flying ambitions of predeessor Jurgen Shrempp. 2 This paper offers an explanation for why Shrempp and his World In strategy were maintained so long in spite of poor performane, and why Shrempp s replaement oinided with the Board of Diretors deision to hange strategy. Our argument is not based on the premise that Zetshe was more able than Shrempp at implementing the new strategy, but on the impliit inentives arising from being the initiator of a long term strategy. The same reason that made it preferable to keep Shrempp to implement the initial strategy, required hanging CEO to implement the new strategy. Of ourse, we annot rule out the possibility that Zetshe was onsidered more able than Shrempp at implementing the new strategy. In fat, it is probably part of the reason why he was hosen among the suessor andidates. 3 Still, this argument alone annot explain why it took so long for the board to dismiss Shrempp. There was general agreement on the fat that Shrempp showed exeptional longevity onsidering the finanial troubles endured by DaimlerChrysler throughout his mandate. 2 BusinessWeek online, january 25, Indeed, Zetshe aquired a good reputation at implementing ost-utting strategies by suessfully restoring profits when he was diretor at Chrysler. 3

5 Last, we address the question of the implementation of the optimal wage ontrat, and show that when the inumbent manager is retained, the seond period ompensation ontrat an be interpreted as a stok option sheme with reset exerise pries in response to poor performane. Our paper thus offers a rationale for the observation that CEOs are often retained and exeutive stok options are reset to a lower strike prie in response to poor performane (Brenner, Sundaran and Yermak (2000), Chane, Kumar and Todd (2000)). 4 We identify two distint determinants of the seond period ompensation ontrat and thus two distint reasons for repriing. Firstly, repriing may our in order to refresh the manager s seond period inentives after his existing stok options have beome of little value. 5 Seond, repriing may our to retain the inumbent manager. This happens when his reputational onerns are very high. At that point the manager has a strong inentive to resign, and free-ride on the future manager s effort in order to rehabilitate his reputation. Our paper thus an aount for both rationales for stok option repriing put forward by pratitioners. More generally, our analysis allows to derive preditions regarding exeutive ompensation pakages, employment poliies, and strategi hoies, and diretly speaks to some empirial findings in the literature. First, we show that managerial retention and exeutive stok-option repriing after bad performane are optimal when reputational onerns are strong, and when firms have to make long-term strategi hoies. We would therefore expet entrenhment and stok-option repriing to be more prevalent at the stage of a firm s life when a long-term strategi hoie ontinues to impat on the firm s future performane. This predition is borne out by the empirial evidene in Carter and Lynh (2001) and Chidambaran and Prabhala (2003) who find that repriing is more prevalent in younger firms and firms that are more onentrated in the tehnology setor. Both attributes fit our desription of firms where long-term strategi hoies are partiularly important. Our analysis speifies that one should observe more strategi ontinuity when managers are retained, and stok options are repried. Seond, our results predit both a relutane to hange strategy after bad performane, and a lose assoiation between managerial replaement and major hanges in the firm s strategy. Taken together, these two preditions an help to understand why CEO turnover is signifiantly orrelated with strategi hange, while poor performane alone is not (see for instane Romanelli and Tushman (1994) or Gordon, Stewart, Sweo and Luker (2000)). Finally, we identify reputational onerns as a possible ontratual frition that prevents firms from realloating assets to their more produtive use. Our analysis an shed light on the empirial findings of Eisfeldt and Rampini (2005) who doument that the benefits of apital realloation are underexploited during reessions. To the extent that managerial expertise is more needed, when profits are low and investment opportunities are sare, our model provides 4 Alternative explanations for the observed lenieny of board towards CEOs rely on the idea that top exeutives wield power in setting their own wage (Bebhuk and Fried (2003), Bertrand and Mullainathan (1999, 2001)). We show here that these seemingly sub-optimal ompensation pakages are ompatible with shareholder value maximization. 5 Aharya, John and Sundaram (2000) also highlight this point. They do not, however, onsider the possibility of replaing the manager after poor performane. Our paper expliitly addresses this question and integrates the analysis of optimal inentive ontrats and employment poliy. 4

6 an explanation for the observed lak of apital realloation. II. Relation to the literature Our paper relates to a number of ontributions in the theory of dynami inentives, based on moral hazard issues (see Holmstrom (1979) and Rogerson (1985) for the analysis of, respetively, stati and dynami models of moral hazard). 6 The role of impliit inentives through reputational onerns was pioneered by Holmstrom (1982) and extended to allow for expliit inentives by Gibbons and Murphy (1992). The literature building on this mainly fouses on inentive distortions that arise out of reputational onerns (e.g., Holmstrom and Riart-i-Costa (1986), Jeon (1998), Milbourn, Shokley and Thakor (2001)). We depart from this literature by stressing the fat that managerial talent an influene the hoie of long-term projets. Therefore, eah period ash flow provides information on the quality of the original projet hoie, and affets differently the reputation of the manager who made the initial hoie, ompared to the reputation of a later-oming manager. In most of the literature on areer onerns, managerial ability affets only the urrent period ash flows, and any manager s reputation is updated regardless of his history with the firm. 7 This setting allows us to investigate the relationship between areer onerns, dismissal and expliit inentives. In the above quoted literature, dismissal does not typially arise as an issue, beause managers are in perfet ompetition at eah date, driving wage down to the point where the firm is indifferent between ontinuation with the same or another manager. In our setting this is no longer true, beause managers take an ation that has long lasting impat. While the standard result of areer onerns stresses the fat that old managers have less areer onerns than young ones and should be given more expliit inentives, our model shows that managerial tenure within a firm an inrease reputational onerns, and redue the use of expliit inentives. The role of a termination threat in managerial inentives has reeived a lot of attention in the literature on debt as a ommitment devie (e.g., Diamond (1984), Gale and Hellwig (1985), Bolton and Sharfstein (1990), Aghion and Bolton (1992), Dewatripont and Tirole (1994) and others). In line with this literature, our analysis shows that if the termination threat an bedesirablefromanexantepointofview,itisexpostineffiient, and an be subjet to renegotiation. 8 However, we depart from this literature along several dimensions. First, the ost of termination is not exogenous in our model: it is refleted in the higher inentive ost of hiring a new manager with lower reputational onerns. As a onsequene, ommitting to fire the manager ex ante is not always optimal, while in those models, it is always ex ante optimal to threaten to liquidate. Seond, these papers assume that a manager is perfetly 6 See Prendergast (1999) for a review on inentives in firms, and Abowd and Kaplan (1999) and Murphy (1999) for reviews on inentive ompensations for exeutives. 7 Reputation updating an be asymetri aross managers though, if managers are given more or less power to take deisions, as pointed out by Ortega (2003), or if intrafirm ommuniation is limited as in Friebel and Raith (2004). 8 In that sense, our analysis is also related to the theoretial literature on soft budget onstraint, initiated by Dewatripont and Maskin (1995). 5

7 entrenhed and the only way to provide a termination threat is to liquidate the firm, i.e., it is impossible to fire a manager, hire a new one and ontinue the firm. 9 This assumption is typially justified with referene to asset (or strategy) speifi skills that render a manager indispensable (e.g., Shleifer and Vishny (1989)). While we believe that suh speifi skillsdo play a role in pratie, we also believe that the relationship between a manager and a firm s hosen strategy goes beyond speifi skills. We therefore make the (extreme) assumption that managers have no inherent produtivity advantages in implementing a given strategy, and an be replaed by equally effiient managers. Entrenhment then arises endogenously from the inentive impliations of managerial legaies. Almazan and Suarez (2003) show that managerial entrenhment may arise as an optimal arrangement when the board of diretors is weak. This is beause entrenhment redues the amount of expliit inentives neessary to pay a manager. In their paper severane payments emerge as a way to provide optimal ex ante inentives for effort hoie. 10 Almazan and Suarez (2003) use quite a different notion of entrenhment from ours. Entrenhment for them means that a manager is retained even though it is ex post ineffiient, beause it is ex ante optimal to do so. We use entrenhment to desribe the ex ante ineffiient retention of a manager due to the ex post optimality of retaining him. 11 There is little literature that expliitly addresses the dynamis of strategy hange. Dow and Raposo (2005) argue that firms strategi hoies are suboptimal in that managers have a preferene for projets that are diffiult to implement. Their model, however, fouses on a stati set-up where a manager has onsiderable lee-way in the initial strategy formulation, but one a strategy has been hosen the issue of hange does not arise. Our paper is quite different in that we explore the dynamis of wage, employment and strategy hange. Boot (1992) and Prendergast and Stole (1996) onsider a CEO s inentive to respond to private information about the quality of the firm s strategy (or investment projet). In both papers hanging strategy may reflet badly on the CEO s ability, whih an lead to underreation to private information. Prendergast and Stole (1996) show that CEOs with long (short) tenure may under- (over-) reat to private information. Like them, we identify tenure as an important variable determining willingness to hange strategy. However, in ontrast to both the above papers, we do not assume that CEOs have private information aboutthebestprojet. Wethereforefind that firms may exhibit strategi inertia not only from the perspetive of the manager s private information, but also onditional on publily available information. This observation appears to be onsistent with the anedotal evidene desribed above, whereby firms fail to adopt a new strategy in spite of a widespread understanding that the urrent strategy is not optimal. 9 See, however, Zwiebel (1996), Fluk (1998) and Berkovith, Israel and Spiegel (2000) who assume that managers an be replaed at a ost and derive apital struture impliations. See also Spear and Wang (2005) who extend the standard dynami moral hazard framework to the ase where the agent an be replaed. 10 Severane payments also appear to be an optimal response to managerial inentives in Eisfeldt and Rampini (2004). In that paper, severane payments help eliiting information on the projet s produtivity by ompensating managers for loss of ontrol. 11 In that sense, our paper is loser to Crémer (1995) who shows that firmsmaynotbeableexposttofire poorly performing managers if they an attribute it to bad luk, although it would be ex ante optimal to do it. 6

8 The organization of the paper is the following. Setion III presents the basi model struture. In Setion IV, we solve for the optimal wage ontrats and the firm s firing poliy when the CEO hooses a strategy that annot subsequently be hanged. Setion V explores wage ontrats, and managerial turnover when the firm an swith strategies at the interim date. Setion VI disusses the implementation of wage ontrats and empirial impliations, while Setion VII disusses some model extensions. The Appendix ontains proofs. III. The model There are three dates t =0, 1, 2. Afirm an undertake a projet at the initial date t =0,whih yields an unertain payoff at the two subsequent dates. The projet hoie an be thought of as a long-term strategi deision by the firm, whih will affet the firm s ash flows two periods into the future. The firm hires a manager whose task it is to make this strategi hoie and to implement it subsequently. Assume that the manager s hoie of strategy S an be either good (S = G) or bad (S = B). One a strategy has been hosen it annot be hanged for two periods. Later on we will also onsider the ase where the firm an adopt another strategy at the interim date. There are two types of managers, m {L, H}. High type managers (m = H) always pik the good strategy, whereas low types always pik the bad strategy. 12 The ex ante probability of a manager being a high type is q The manager s type is not known to either the manager or the firm, and both have the same beliefs about the manager s type. Moreover, we assume that neither an observe diretly the quality of the hosen strategy. One a strategy has been hosen, its likelihood of suess depends on the effortthata manager puts into its implementation over the next two periods. The firm then generates payoffs att =1, 2. Denote by e 1 {e 1, e 1 } the effort hoie, that affets payoffs att =1. Similarly, e 2 {e 2, e 2 } is the effort hosen after the first-period payoffs, affeting payoffs atthe final date t =2. Exerting the high effort level e t >e t arues a ost to the manager. Define the inrease in effort levels by e t e t e t. Thehoieofeffort affets payoffs in the following way. The payoff is always low (R l ) when the manager has hosen the bad strategy S = B, i.e., any effort by the manager in either period is wasted in that ase. When he has hosen the good strategy S = G the firm generates a high payoff R t = R h at t =1with probability e 1 and the low payoff R l with probability 1 e 1. Date t =2payoffs areaffeted by effort e 2 in the same way. Figure 1 summarizes the struture of the model. Suppose first that effort is observable, that is, that there is no moral hazard. To make things interesting, we assume that in that ase the firm maximizes its profit by induing effort at any stage of the game. This implies the following onditions. First, the firm wants to hire a manager and to indue effort at t =2, whatever the level of profit realized at the end of the 12 Note that our results go through if we assume that a bad manager an sometimes hoose a good strategy. What matters for our results is that the projet s hoie annot be perfetly inferred from the realization of the profits at eah period. 7

9 t=0 t=1 t=2 q ½ ½ ½> 0 ½ ½½ Q QQQ H e R h 1 µ e 2 µ G Z Z ZZZ~ ZZZ~ 1 e 1 1 e R 2 l R h R l 1 q 0 Q Qs L B - R l - R l Figure 1: Timing and payoffs under symmetri information. first period. This is ensured by the ondition: p i (e 2 R h +(1 e 2 )R l )+ 1 p i R l p i (e 2 R h +(1 e 2 )R l )+ 1 p i R l, where p i prob(s = G R 1 = R i ) and i {l, h}. After manipulations, the above equation boils down to: R h R l p i e 2. (1) Seond, the firm wants to hire a manager and indue him to exert effort at t =1, given that effort will be indued at t =2. This is ensured by the ondition: q 0 [e 1 R h +(1 e 1 )R l + e 2 R h +(1 e 2 )R l ]+(1 q 0 )[R l + R l ] q 0 [e 1 R h + e 2 R h +(1 e 2 )R l ) +(1 e 1 )R l ]+(1 q 0 )[R l + R l ], whih simplifies to: R h R l. (2) q 0 e 1 The firm hires a manager at t = 0 with reputation q 0 and offers a wage ontrat that speifies ompensation payments ontingent on the date 1 and 2 payoffs. Denote by w i,j the wage payment at date t =2ontingent on both performane observations R i,r j at dates 1 and 2 respetively, where i, j {l, h}. The firm an hoose to fire the manager at t =1and hire a new manager from a pool of indistinguishable managers of reputation q 0. The effort level e t, t =1, 2 is hosen by the manager who is employed by the firm at date t 1. Note that the firm an offer different seond period wage ontrats to a new or the inumbent manager. Similarly, we denote date 1 wage payments by w i. Finally, let p i,j prob(s = G R 1 = R i,r 2 = R j ). Managers are risk neutral and have limited liability. We also assume that they are about their reputation when they stop working for the firm (or when the projet stops). Denote by 8

10 f(q) the value that a manager derives from having reputation q. We assume that f(q) is i) inreasing in q, 13 and ii) suh that the manager would prefer to aept to be employed by the firm at the initial date instead of not being employed. Intuitively, when the firm performs well, outside firms will be more eager to hire well-performing managers, and be willing to pay them higher wages. This assumption is also onsistent with reent empirial evidene by Fee and Hadlok (2003) who show that managers who are promoted to CEO positions at new firms ome from firms with better than average past stok prie performane. Denote by q i,j thedate2reputation 14 of the manager onditional on the performane observations R i,r j at date 1 and 2 respetively, where i, j {l, h}. Weansimilarlydefine the date 1 reputation by q i. Moreover, we assume that the manager remains unemployed for one period if he is fired at date 1. After that, his reputation is updated on the basis of the firm s ash flows. The manager s reputation therefore ontinues to evolve even when he no longer works for the firm. Given the struture of the model reputation evolves in a very simple way: if the firm generates high ash flows R h at either date, it beomes lear that the manager must have piked the orret strategy S = G. His reputation then jumps to q =1, beause only high type managers are apable of hoosing a good strategy. This yields the following reputation updates: q h = q h,l = q h,h = q l,h =1. The manager enjoys then a reputation benefit f(1). The only other remaining reputation updates are then q l < 1 after bad performane at t =1,andq l,l <q l after two onseutive bad performanes. At the end of period 2, we an denote the differene in the value of the reputation between q =1and q l,l by f f(1) f(q l,l ). Last, in the remainder of the paper, we make the assumption that effort is not observable, i.e. that there is moral hazard. IV. Optimal wages and firing poliy We now analyze the wage ontrats and the optimal retention poliy of the firm. Throughout the paper we restrit attention to long-term ontrats that are renegotiation proof. In other words, we allow the firm and the manager to renegotiate the employment ontrat if it is mutually profitabletodoso. 15 In our ase this means that the firm annot write a ontrat that ommits it to a partiular employment poliy. And the firm annot fore the manager to work for it if this would violate his partiipation onstraint at the interim date. Hene, we do not allow 13 One an think of f(q) being determined by an expliit alulation of the net present value of future wage payments when the agent has urrent reputation q. For the sake of tratability we will not endogenize f(q), but simply assume that the higher the manager s reputation, the higher the future expeted wage (see Dewatripont, Jewitt and Tirole (1999) for a omprehensive analysis of areer onerns models). 14 In our model, reputation is the probability of the manager being of a good type. Beause we assume a perfet math between types and strategy hoies, the reputation q i,j is equal to the probability of having piked the orret strategy p i,j.bothmaydiffer in a more general setting. To interpret more aurately the eonomi effets at stake, we maintain distint notations for the two variables. 15 This assumption appears to be onsistent with the empirial evidene provided by Chen (2004). He finds that firms that ommit not to reprie options, issue an abnormally large amount of new stok options. This suggests that firms do reward managers after poor performane,evenwhenthey-tryto-ommitnotdoso. 9

11 employment ontrats that i) are ex post detrimental to shareholders, and ii) would deny the manager the right to quit the job if he wishes to do so. Of ourse, sine managers will antiipate this, it will modify their ex ante inentives to work. A. Optimal seond period wages Let us first onsider the firm s employment poliy and wage payments at the interim date. The wage ontrat and the deision to retain or fire the manager at the interim date 1 are determined so as to maximize shareholder wealth, subjet to the manager s inentive ompatibility and partiipation onstraints. The following proposition gives the optimal wage ontrats that maximize shareholders wealth and satisfy both sets of onstraints, depending on whether the inumbent manager is retained or fired and depending on whether date 1 ash flows were high or low. 16 Proposition 1 Suppose the firm earns R h at date 1. Then the optimal date 2 wage ontrat is the same regardless of whether the manager is retained or fired at date 1. The optimal wage ontrat is given by w h,h =, e 2 (3) w h,l = 0. (4) Suppose the firm earns R l at date 1. The optimal date 2 wage ontrat if a new manager is hired is given by w l,h new = w l,l new = 0, and if the original manager is retained it is given by ½ w l,h old = max p l f; e 2 p l e 2, (5) ¾, (6) p l e 2 w l,l old = 0. (7) The wage payment w h,h from performing well in the seond period, after having performed well in the first period is independent of reputational onerns. The reason is straightforward: when the manager has performed well one, it is known that he hose the good strategy initially and therefore he annot improve his reputation further by performing well again. Therefore the same wage is given to an inumbent or a new manager We assume as before that shareholders want to indue effort at the seond period, whatever the first period ash-flows, and we provide below a suffiient ondition for this. 17 Under more general information strutures the inumbent manager may well ontinue to have reputational onerns after good performane, whih will simply provide an additional reason for retaining the inumbent manager after good performane. 10

12 When the inumbent manager is fired after poor interim performane and a new manager is hired, he reeives a wage payment only if he performs well at the subsequent date. In this way the manager is given stronger inentives to exert effort. The total reward required is p l e 2, whih is exatly the payment that satisfies the new manager s inentive ompatibility onstraint. His partiipation onstraint is always satisfied at this wage level. When the inumbent manager is retained, there are two possible values of the wage payment when he performs well, depending on whether the inentive ompatibility or the partiipation onstraint is binding. The inentive ompatibility onstraint requires a bonus payment w l,h p l e 2 old = f. This is similar to the wage payment given to a new manager, exept that now not all inentives have to be provided expliitly. The inumbent manager is onerned about his reputation and therefore some of his inentives are provided impliitly. This redues the neessary expliit wage payment by an amount f orresponding to his reputational onern (see equation (6)). The differene arises beause only the inumbent, but not the new manager is driven by reputational onerns: The latter annot be held responsible for the initial strategi hoie and therefore a failure to implement the strategy suessfully, does not allow the market to update its belief over the new manager s ability to take good strategi deisions. Therefore reputational onerns provide inentives to the inumbent manager, but not the new manager. This result differs from standard results in the literature (e.g., Holmstrom s (1982), Gibbons and Murphy (1992)), where a manager s ompensation ontrat is determined only on the basis of his urrent reputation and his remaining reputational onerns. We show that managers reeive very different ompensation ontrats solely on the basis of whether or not they have worked in the firm before, and thereby had an opportunity to affet the distribution of the firm s ash flows beyond the time horizon of their own tenure. When the reputational onerns beome very strong, the inumbent manager s inentive ompatibility onstraint an be satisfied without any expliit wage payment. The manager is willing to exert effort merely in order to improve his reputation, and does not need a monetary gain. But, this raises the possibility for the inumbent manager to leave the firm and free ride on the effort exerted by a new manager: he would then obtain the reputation benefit without inurring the effort ost. If the firm wishes to retain the inumbent manager it has to give him a positive wage even though his inentive ompatibility onstraint is slak. This wage is determined by the binding partiipation onstraint of the manager, and is equal to p l e 2. Corollary 1 If f > e 2 e 2 p l (8) e 2 then the seond period wage ontrat is determined so as to ensure that the inumbent manager an be retained. Otherwise the seond period wage ontrat is determined so as to render the high effort level inentive ompatible. Our model aptures two realisti features of dynami wage ontrating in a simple setup. Wage ontrats have to satisfy an inentive ompatibility onstraint, whih is the usual 11

13 onstraint that muh of the theoretial literature on ageny problems has foused on. 18 At the same time wage ontrats have to satisfy a non-trivial partiipation onstraint in order to deter the manager from leaving the firm with a view to free-riding on effort exerted by another manager in the future. This effet is not aptured in previous models of CEO ompensation, even though managerial retention is one of the main onerns stated for repriing by pratitioners. In most models of CEO ompensation it is assumed that managers take ations whose impat expires after one period. 19 The utility of a manager who quits a firm is therefore independent of that firm s future performane. This is not the ase in our model, beause the manager s reputation may be affeted by the firm s future performane even after he stopped working for that firm. Let us now to turn to the firm s optimal employment poliy. For sake of onsisteny, we need to make an assumption on the model parameters to ensure that, given the wages determined in Proposition 1, shareholders want to indue effortatthefirst period. Assumption 1 R h R l e 2 e 2 p l. (9) e 2 Assumption 1 provides a suffiient ondition for shareholders to indue high seond period effort, for any first-period payoff. Note that this ondition is stronger than the firstbestondi- tion (1). This is beause inentivizing managers is ostly: the differene in returns between the high and low states must be high enough to over the ageny rent. Proposition 2 If the manager has any reputational onerns, the only renegotiation proof employment poliy is to retain the manager at date 1, regardless of his performane. This statement follows straightforwardly from the omparison of the seond period wage payments given in equations (5) and (6). Suppose the firm observes poor interim performane R l. At that stage it has the hoie of retaining the manager and paying wage w l,h old in ase of subsequent suess, or hiring a new manager and paying wnew l,h >w l,h old. Sine the identity of the manager has no impat on performane the firm learly prefers to retain the manager. Moreover, the manager also prefers to remain with the firm, beause it allows him to earn the ageny rent in the next period. Note that at the interim stage, the firm is indifferent between retaining and firing the manager after good performane. However, we show below that ommitting to retain the manager after good interim performane is ex ante optimal. The optimal renegotiation-proof ontrat will thus entail retaining the manager after good interim performane. 18 Oyer (2004) is an exeption to this. He fouses on the role of absolute performane based ompensation in adjusting wage to the CEO s hanging outside opportunities. 19 An exeption to this is Gorton and Grundy (1997) who point to a related free rider problem in a model where managers need to be given expliit long-term wage payments. Hene, managers in their model may free ride beause they benefit via expliit wage payments if the firm s future performane is strong even if they quit the firm. 12

14 Proposition 2 ontains the key result on entrenhment. It states that a poliy of firing the manager after poor interim performane is time inonsistent beause employing a new manager is more expensive than retaining the inumbent manager, without yielding any benefit tothe firm after date 1. Note that this result holds even though the inumbent has no produtivity advantage over a newly employed manager, whih is what usually underlies stories of entrenhment (e.g., Shleifer and Vishny (1989) or Crémer (1995)). B. Optimal first period wages Let us now omplete the haraterization of wage ontrats by solving for the optimal first period ompensation sheme. Sine the manager is always retained at the interim date, it is easy to write down his first period inentive ompatibility onstraint. The resulting optimal firstperiodpaymentsaregiveninthefollowinglemma. Lemma 1 Under the optimal renegotiation proof ontrat, the manager s first period wage is given by w h = ³ max[0, e 2 w h,h w l,h q 0 e old (1 e 2 ) f], 1 w l = 0, where w h,h and w l,h old are given in Proposition 1. The expliit first period inentive payment w h depends on two kinds of inentives: the size of the seond period ageny rent and reputational onerns. Sine the manager is retained after poor performane, he will earn an ageny rent in the next period even though he performs badly. This affets negatively his first period inentives and inreases w h. On the other hand, the ageny rent he earns when being retained after good first period performane provides an additional inentive at the first date to exert effort: w h dereases with w h,h. For this reason, it is optimal ex ante to ommit to retain the manager after good performane, even if shareholders are ex post indifferent between retaining the inumbent manager and hiring a new one. 20 The seond type of inentive stems from reputational onerns. If the manager ahieves a high ash flow at the first date he improves his reputation and this by itself provides an impliit inentive to exert effort. But this effet is attenuated by the manager s opportunity to restore his reputation after bad performane. C. The ost of entrenhment An important question that arises one we know the full solution to the ontrating problem is whether the type of entrenhment identified above atually arries any ost to the firm. Entrenhment arises beause the firm is unable to ommit redibly to a poliy of firing the 20 The fat that future rents enhane present inentives to effortiswell-knownindynamimoralhazardmodels, and goes bak to Rogerson (1985). Our model brings new insight by explaining why future rents an have ountervailing effets on inentives beause shareholders also prefer to retain managers after bad performane. 13

15 manager at the interim date. It would always want to renegotiate suh a ontrat and retain the manager instead, even if he performed poorly. As a result it annot redibly use the threat of dismissal to provide first period inentives to the manager. Note that this ontrasts with other results in the literature on dynami ontrating, e.g., v. Thadden (1995) and Almazan and Suarez (2003). In those papers the firm s interim effiient deision may be to liquidate the firm (v. Thadden) or fire the manager (Almazan and Suarez), and this may redue the manager s ex ante effort inentives. In those papers there is therefore a role for long-term ontrats that bind the firm to retaining the manager at the interim date. In our model suh ontrats do not help, beause the firm s interim effiient deision already involves retaining the manager more often than may be ex ante desirable. We now identify what would be the gain in firm value if shareholders ould ommit ex ante to an employment poliy (e.g. to firing the manager, if it is optimal to do so). Proposition 3 If the shareholders an ommit to an employment poliy, the ex ante expeted firm value inreases by: h i q 0 max e 1 (w h ŵ h ) (1 e 1 )e 2 (wnew l,h w l,h old ); 0, (10) where wnew l,h and w l,h old are defined in Proposition 1, wh inlemma1and ŵ h ³ =max[0, e 2 w h,h (1 e 2 ) f]. q 0 e 1 Proposition 3 states that the ability to ommit ex ante to an employment poliy is not neessarily valuable for shareholders. It is valuable only if (10) is positive. Put differently, it is sometimes optimal from an ex ante point of view to retain the manager after bad performane. Expression (10), where ŵ h represents the first period wage of the initial manager, if he is fired after bad performane, illustrates learly the trade-off faed by shareholders. On the one hand, ommitting to firing the poorly performing original manager inreases the seond period wage expenses, beause it is more ostly to inentivize a new manager, with no reputational onerns. This extra ost is measured by (1 e 1 )e 2 (wnew l,h w l,h old ). On the other hand, when the original manager is fired if he performs badly, it is easier to inentivize him at the first period, whih redues expeted wage expenses by e 1 (w h ŵ h ). Whether ommitting to firing the original manager is benefiial to shareholders depends on whih effet dominates. In our model where inentives an be given by reputational onerns, wages might turn out to be equal to zero. Intuitively, this will be the ase if reputational onerns are large. To highlight the interation between impliit and expliit inentives, and to fous on the most realisti ases where managers need also positive wages to be inentivized, we assume that firstperiod wages are stritly positive. Sine ŵ h <w h, we simply need to ensure that ŵ h 0, whih is true under assumption 2: Assumption 2 f ³ 1 q 0 e 1 e 2 e 2 1 e 2. (11) 14

16 Using assumption 2, as well as the wage values determined before, one an determine the net benefit of ommitting to an employment poliy. Corollary 2 Under assumption 2, if the inumbent manager s wage w l,h old is determined by his binding partiipation onstraint, the net benefit ofommittingtoanemploymentpoliyisequal to: q 0 h i p l max e 2 (1 e 1 p l ) e 2 (1 e 1 ); 0. (12) e 2 If w l,h old is determined by the inumbent manager s binding inentive ompatibility onstraint, it is equal to: q 0 max e 1 e 2 p l e 2 f e 1 ;0. (13) e 2 Suppose the seond period wage when the manager is retained is determined by the binding partiipation onstraint. Then from (12) we an see that the ost of entrenhment depends on the likelihood p l with whih the good strategy S = G is being operated onditional on having observed poor performane R l. Note that the manager is indifferent between being retained and being fired onditional on the realization R l. One might therefore expet firing, and the threat thereof, to do nothing to inentives. This, however, is not the ase, beause the manager s atual payoff differs depending on whether the operating strategy is G or B. In partiular, if S = G then the manager stritly prefers to be retained, even if his partiipation onstraint is binding. Conversely, he would stritly prefer to quit if he knew that S = B. His atual wage is set to make him indifferent onditional on R l only and is thus a probability weighted average between an expeted payoff onditional on S = G and S = B. Sinehisfirst period inentives are only affeted by his payoffs when the strategy is S = G the threat of losing the rent when R l ours provides an impliit inentive to exert effortatthefirstdate. Thelowerp l themorehebenefits from being retained onditional on S = G and the stronger are the inentive impliations from athreatoffiring. The benefit tothefirm from being able to ommit to firing the manager after poor performane is therefore higher when p l is lower. Suppose instead that the seond period wage of the retained manager is determined by the binding inentive ompatibility onstraint. Then the firm s ex ante value inreases when using the threat of dismissal only if reputational onerns are not too high. Consider the benhmark ase when there are no reputational onerns, i.e., f =0. Inthatasethefirm would ertainly want to make use of the threat of dismissal, to mitigate the managerial moral hazard problem. From expression (13), one an see that the net benefit of ommitting to firing the manager is then positive. But when the manager ares a lot about his reputation, the redued ost of first period wages may be outweighed by the additional ost of replaing the manager at the seond date. The firm would then not benefit fromommittingtofire a poorly performing manager. 15

17 V. Interim Strategy Change In the above analysis it has been assumed that the strategy hoie ould not be altered throughout the lifetime of the projet. We now extend the analysis to the ase where the firm an hoose to abandon the strategy that was started in favour of the alternative, untested strategy. By onstrution, the suess or failure of the new strategy throws some light on whether the original strategy hoie was good or bad. This new set-up extends our earlier notion of a managerial legay. The manager s reputation is affeted in the longer term by an earlier strategi hoie, even if the strategy has been hanged in the meantime, and even if the manager no longer works for the firm. The kind of situation we have in mind is one where a firm faes a major strategi deision and a CEO takes the initiative to make that hoie. If the firm s subsequent performane is poor, it may wish to abandon this strategy and instead go for the obvious alternative strategy that had been rejeted initially. Note that the deision to hange strategy is now relatively mehanial and per se does not require a CEO to develop a new strategi vision, although it requires effort to implement the alternative strategy suessfully. In the example of DaimlerChrysler developed in the introdution, while the World In strategy was really Shrempp s vision, the deision to abandon the worldwide expansion, and to refous on the firm s ore ompetenies (making high quality ars) was taken by the board (for example, the deision to retreat from Asia was taken while Shrempp was still in plae, against his advie). 21 To investigate the joint deision to hange strategy and possibly manager, we now assume that the firm an swith strategy at the interim date. First period ash flows are distributed exatly as desribed in Setion IV. Seond period ash-flows still depend on whether the good or the bad strategy is implemented. Therefore, if the firm stiks to the original strategy, the seond period ash flow distributions are exatly as desribed in Setion IV. If the firm swithes, the ash flow distribution depends on the quality of the strategy that is being implemented in the seond period. If the firm implements the bad strategy at the seond date, its payoff is R l with ertainty. If it implements the good strategy it is R h with probability e 2 and R l with probability 1 e 2. To derive the optimal strategy/manager hoie, we first determine the optimal employment poliy for eah interim strategy hoie, and then onlude on the optimal strategy hoie. A. Choie of manager given the hoie of strategy The firm s interim deision problem is now more ompliated, beause it needs to deide simultaneously (i) whih strategy to pursue, (ii) with whih manager to pursue it, and (iii) what wage ontrat the hosen manager should be offered. To aount for all these possibilities, the wage ontrats will be generially written: w i,j hoie,manager, where the variable hoie an be stik 21 Boot (1992) and Prendergast and Stole (1996) impliitly provide another reason for leaving this deision to the board and not the CEO: If the latter makes a strategy reommendation based on private information, he may have an inentive to report in favor of the strategy originally hosen by him. We assume, however, that the manager has no superior information about the quality of the strategy, and therefore this issue does not arise. 16

18 or swith and the variable manager an be old or new. For example, w l,h stik,old represents the wage offered to the original manager if the firm stiks to the initial strategy, retains the original manager, and generates ash-flows R l at t =1and R h at t =2. The following proposition establishes the optimal employment poliy, given the firm s deision to swith strategy. Proposition 4 If the firm wants to swith strategy at the interim date, it is optimal to hire a new manager, i.e. w l,h swith,old wl,h swith,new. AordingtoProposition4thefirm will never pursue a new strategy with the inumbent manager. The intuition for this result is as follows. The inumbent manager s reputation is tied to the suess of the strategy that he hose. Suess of the new strategy therefore implies that the manager s original strategy hoie was poor. This gives the manager an inherent interest in making the new strategy fail, whih would be interpreted as a vindiation of his original hoie. As a result the inumbent manager has an inentive not to put effort into implementing the new strategy. The firm an overome this bias if it provides suffiiently strong expliit inentives. This, however, is more expensive than simply hiring a new manager, who has no vested interest in the firm s hosen strategy. Thus, the same reputational onerns that made it heap to employ the inumbent manager to implement the original strategy, now make it expensive to employ him to implement a new strategy. The only ase where the inumbent manager is equally well suited as a new manager to implement the new strategy, is when he has no reputational onerns at all. This result is important, beause it haraterizes the reason why in pratie there is frequently a lose assoiation between a partiular strategy and the manager who implements it. We depart from the standard assumption that managers have an inherent produtivity advantage with regards to a partiular strategy (e.g., Shleifer and Vishny, 1989). Managers in our model have no suh produtivity advantage, but they happen to implement only those strategies that they have hosen originally. This is beause an assoiation between the ost of inentivizing a manager and the type of strategy he implements arises endogenously from the manager s reputational onerns. It would in priniple be straightforward to require from the agent that he aept a strategy hange after poor performane and implement the new strategy. This, however, would make it diffiult for the manager to be genuinely interested in seeing the strategy sueed. We now derive the optimal employment poliy when the firm deides to stik to the original strategy. Note that even in that ase, the inumbent manager s wage is affeted by the ability to hange strategy. Indeed, the possibility to swith strategy affets the manager s interim partiipation onstraint ompared to the ase where no suh hange is possible, beause the distribution of the manager s reputation depends on the firm strategy hoie were the inumbent manager to resign. The optimal wage ontrat for the inumbent is derived in the following Lemma. 17