London Business School

Based on Whitley's “national business systems” (NBS) institutional framework, we theorize about and empirically investigate the impact of nation-level institutions on firms’ corporate social performance (CSP). Using a sample of firms from 42 countries spanning seven years, we construct an annual composite CSP index for each firm, based on social and environmental metrics. We find that the political system, followed by the labor and education system, and the cultural system are the most important NBS categories of institutions that impact CSP. Interestingly, the financial system appears to have a relatively less significant impact. We discuss implications for research, practice and policymaking.


Introduction
The problems resulting from separation of ownership and control have long been recognised in the corporate governance and corporate finance literature (Berle and Means (1932), Jensen (1986), Hart (1995), Shleifer and Vishny (1997)). Mintzberg's analysis of the cult of leadership has three steps: (i) companies need to change and adapt; (ii) the CEO, rather than the shareholders, has power to decide on the direction of change (the strategy) and (iii) the CEO may not select the optimal kind of change (for shareholder value): it may be overly dramatic. In this paper, we study this idea from the perspective of the corporate governance literature, i.e. using agency theory. Note that much of this literature has used an incomplete contracting approach (Hart (1995)), and each of the three steps in Mintzberg's argument suggests contractual incompleteness. This is the approach we take here.
Standard agency theory takes as given an incentive problem, the principal and agent then agree on a (constrained) optimal contract, and then the agent goes to work on the problem. One feature of CEO compensation that is clearly different to this standard agency model is that the contract is adjusted over time to reflect the evolution of the firm's performance and its strategic direction. There is an annual pay setting round at which options and incentive plans are renegotiated. Since he or she can influence the firm's strategy, the CEO may be able to influence the compensation contract. A dramatic merger, or restructuring of the whole business, can lead to larger options grants.
There are many examples in recent corporate history where radical corporate change went hand in hand with high executive compensation and options grants: Coca-Cola under Roberto Goizueta, the Daimler-Chrysler merger, GE under Jack Welch, Chris Gent and the Vodafone-Mannesmann takeover, Enron, the Glaxo Wellcome -SmithKline Beecham merger. Some of these dramatic changes were successful, others were failures. 1 Our paper is not about disastrous dramatic change. It is about change that is expected positive-NPV at the outset, given the available information, that may end up being either successful or unsuccessful depending on the outcome, but that is overly dramatic in the sense that less radical change would have been higher NPV. We argue that the way executive compensation responds to changes in strategy can lead to management choosing change that is more radical than the shareholders would optimally prefer: the CEO of a regional electricity utility may find it personally more profitable to create a global web-based energy market-maker. We then look at ways the incentive contract can be modified at the outset to anticipate this problem.
This analysis is consistent with (but not identical to) two themes that run through much corporate finance research: free-cash-flow theory, and non-valuecreating mergers. First, there is free-cash-flow theory. Jensen (1986Jensen ( , 2000, building on earlier analyses of managerial empire-building (Baumol (1959), Marris (1967), and Williamson (1964)), has argued that managers of public corpo-rations have a systematic tendency to overinvest. 2 Overinvestment of free-cashflow can be viewed as similar to overly dramatic change. For example, cashgenerating low-growth businesses may tempt their managers to seek growth through excessive diversification. Lang and Litzenberger (1989), Lang, Stulz and Walking (1991), Mann and Sicherman (1991), Blanchard, Lopez-de-Silanes and Shleifer (1994), Kaplan and Zingales (1997), and Lamont (1997) provide empirical evidence supporting free-cash-flow theory.
Second, there is the evidence that many mergers add little or no value to the acquirer (Asquith, Bruner and Mullins (1983), Jarrell, Brickley and Netter (1988), Bradley, Desai, Kim (1988), Jarrell and Poulsen (1989), Bruner (2002)) 3 . Mergers are a clear example of dramatic change and hence, this evidence can help explain why managers may undertake dramatic acquisitions even when this is not shareholder-value maximising.
Of course, change is valuable and necessary: we do not argue that change is bad. In this paper, we argue that firms that do change may sometimes change in the wrong way. This may be particularly relevant because the recent years have been a period with a high rate of corporate change. For example, the 1990's was a decade of mega-mergers. US M&A activity over 1993-1999 amounted to an annual average of 8.4% of GDP, compared with less than 4% in the 1980's and less than 2% in the 1970's (Weston, Siu and Johnson (2001), table 7.4). While in previous decades merger targets were typically about 10% of the size of the acquirers, in the 1990's it became common for companies to acquire targets almost as large as, or sometimes even larger than, themselves (AOL-Time Warner and Vodafone-Mannesmann are examples).
The basic outline of our model is as follows. We study a situation where shareholders are able to set compensation optimally given incentive compatibility constraints, but where top management has the advantage of formulating strategy. In this agency problem, after the CEO has chosen a strategy, incentives are set or adjusted before the CEO proceeds to implementation of the chosen strategy. The CEO's ability to formulate strategy is part of his or her job, so this is a natural assumption. We use a setting with incomplete contractibility 4 and limited liability 5 for the manager, leading to an option-like contract (re-ward for success, no penalty for failure) conditioned on firm value. We assume that change will require substantial effort from the CEO at the implementation stage, while business as usual requires much less. Hence the reward for success must not only induce the CEO to put in the required effort to implement change compared to the alternative of no effort, but also compared to the much easier task of maintaining the status quo. Since implementation of a given strategy is non-contractible, we show that the latter becomes the binding constraint, and the more personally demanding are the alternatives he finds to the status quo, the higher his surplus 6 . By choosing a task whose success is highly dependent on his own performance, the CEO is able to extract higher surplus from his shareholders. We conclude that high-powered incentives can encourage overly dramatic strategies. 7 Anticipating this divergence, what could shareholders do? If they are concerned that the CEO's incentives are not well aligned with their own, they could simply give him enough options or equity at the outset and the conflicts would disappear. In other words, if they are worried that corporate strategy may be distorted by the CEO so as to extract larger incentive packages of equity or options, they could simply hand over the large package at the outset anyway. This scenario, which we call "ex-ante contracting" works but is expensive: there is a trade-off to be made. In very unstable environments, in which it is likely that change (and particularly, dramatic change) is possible, setting such an ex-ante contract might be optimal from the shareholders' perspective. Thus, a commitment to high compensation can reflect strategic discretion, without being directly related to the CEO's effort cost of implementation.
The drawback of this contracting policy is that the shareholders commit in advance to high pay (as a deterrent to overly dramatic change) even though, expost, sometimes dramatic change will turn out not to be an option anyway. We assume that the firm's strategic environment is not completely predictable in advance, and that the CEO has an advantage in being better informed about it than the shareholders. Hence there is a range of possible strategic choices, not all of which are always available. In some states of the world, dramatic change is the only (positive NPV) alternative to the status quo, in other states more moderate change is also available (and is higher NPV than dramatic change), while in other states dramatic change is not an option. In the last case, exante contracting means that the CEO ends up being unnecessarily highly paid to implement a simpler strategy. Therefore in less changeable environments, shareholders may prefer a wait-and-see approach to contracting, setting the this feature, if there is one, would be interesting but is not the subject of this paper. 6 Core and Guay (2002) present empirical evidence that more unstable environments tend to be positively related to CEO pay-performance sensitivity. 7 Note that this is different from two other problems that may occur with high-powered compensation: (i) straight cheating by managers; (ii) managerial rent-seeking through control of the contracting process. Both of these problems are important and relevant: CEO's may use false accounting to overstate profits and inflate the stock price, and instead of representing shareholders' interests, boards may cooperate with management in agreeing excessive compensation. (The recent confrontation between board and shareholders of Glaxo SmithKline is an illustration). However, those are not the problems we study here. compensation package after the strategy is chosen while accepting that this might cause distortion in strategy.
Another way for shareholders to discourage excessively dramatic strategies would be for them to refuse to go along with the whole notion of dramatic change and the associated high compensation, by committing in advance not to pay high salaries. It is not clear that shareholders can credibly commit, since ex-post they may want to renegotiate, but perhaps social norms can provide a mechanism for limiting CEO pay. We therefore extend the analysis to investigate this case also. This will work in dissuading the CEO from unnecessary dramatic change, but again there is a trade-off: it is equally effective in dissuading the CEO from dramatic change in states when it is desirable.
How can the conclusions of our model be used to throw light on the evolution of executive pay in the past 10-15 years? We believe that during this period, shareholders have become much more aware of the need for large public firms to change. Partly this is due to the revolutionary technical changes that have transformed the economic landscape. It may also be that shareholders simply became more aware of this general issue, following a period in the late 1980's when many firms that had become insufficiently focused on value creation were restructured by external means (LBO's and other hostile takeovers). In terms of our model, we could hypothesise that before this period, there were informal political and social conventions that effectively limited CEO pay (pre-commitment to a limit on renegotiation), and this limitation either became suboptimal because of changes in the parameters of the model, or it became unsustainable and broke down as the pressure for change suggested higher and higher CEO pay (ex-ante contracting).
Our analysis can be used to interpret the available empirical evidence on top management pay in the US. Hall and Liebman (1998) document that on average a 1% increase in stock price led to an increase of $124,000 in CEO wealth in 1994, while in 1998 this value exceeded $500,000. Murphy (1999) shows a sharp increase in pay-performance sensitivity over the early 1990's, reaching a level almost double the 0.325% reported in the seminal Jensen and Murphy (1990) article. Conyon and Murphy (2000) report similar findings for the UK, showing that both the level and the pay-performance sensitivity of CEO compensation increased sharply over the 1990's. One can interpret these studies on the 1990's as evidence of a period of high changeability and a commitment to high compensation ("ex ante contracting" in terms of our model).
Core and Guay (2001b) describe the cross-sectional variation of executive compensation, finding that the median large firm has options outstanding that amount to 5.5% of common stock, rising to 10-14% for growth industries (computer, software, pharmaceutical) but lower at only 2-3% for low growth industries such as utilities and petroleum firms. Likewise Murphy (1999) finds pervasive use of stock options in most industry groupings, but not in utilities. One can interpret this, again, as evidence of higher strategic discretion of CEO's and more changeable environments in growth industries. An alternative explanation, in line with standard agency theory, is that higher compensation in high growth industries simply reflects higher agency costs of effort (or private benefits of control). Plausibly however, while managerial jobs in growth industries may be more challenging and more onerous, this difference in effort seems unlikely to be large enough to explain very large divergences in compensation packages. Along the same lines as the findings of the above studies, Demsetz and Lehn (1985) and Core and Guay (2003) find a strong positive relationship between firm risk and CEO pay-performance sensitivity 8 . Core and Larcker (2003) examine firms with "target ownership plans" specifying the minimum amount of stock that must be held by executives and show that non-CEO executives typically hold much less equity (relative to base salary) than the CEO. We can interpret this as evidence that given the special role of the CEO, his or her incentives should be stronger to give good incentives in strategy formulation.
We now comment on the relationship of this paper to some of the existing agency literature. First, there is the multi-tasking version of the principal agent problem in which the agent has two tasks to which he can devote effort, but the output of only one of those tasks is measurable (Holmstrom and Milgrom, 1991). Making incentives more high-powered relative to that output measure can be counterproductive, reducing the agent's effort on the other task. There is a similarity with our model in that incentives for one variable (effort, in our model) can distort another variable (strategy choice). However, in our model both variables contribute to the same measured output (firm value). In our model, the analysis is driven by the incomplete nature of the contracting, the sequencing of the agents' choices, and the opportunities for renegotiation of the contract, which are absent from the multi-tasking model.
Free cash flow theory (Jensen (1986)) suggests that challenging strategies can be beneficial for CEOs in terms of private benefits. The private benefits are not explicitly modelled: they are exogenous. Rather than direct private benefits, we consider indirect benefits arising from the effect of managerial activity on compensation. Hence, while free cash flow theory is rather broader than our analysis, our paper could be viewed as compatible with free cash flow theory and as offering a rationalization of how private benefits can arise. Shleifer and Vishny (1989) argue that managers have an incentive to entrench themselves by making investments that create specific human capital. They will favour projects that they alone can operate and cannot easily be transferred to another manager, much like a computer programmer who writes a deliberately cryptic code that makes him or her indispensable and able to extract rents. In our paper, there is no specific human capital. Shleifer and Vishny's effects rely on the manager manipulating his position in the labour market, while the effects in our paper rely on manipulating the agency problem. Prendergast (2002) considers a setting where the agent has discretion in deciding how to solve the problem. He starts by noting that the majority of empirical studies find a positive relationship between the risk of an agency problem and the strength of the agent's incentives, contrary to the predictions of the standard model. This empirical evidence is compatible with our analysis. He then points out that in reality agents can decide how to go about solving a problem, which is similar to our point that CEO's decide strategy as well as implementing it. He goes on to address different issues to the ones we address here, concluding that if a problem is uncertain, the principal may not know how to solve it, so he will let the agent decide that and motivate him with strong incentive pay, while if a problem is predictable the principal will know how to solve it and will directly monitor the agent's actions instead.
The paper is structured as follows. Section 2 presents the model and the main results of the paper. In section 3 we consider an extension in terms of the contracting mechanism, allowing for the firm to pre-commit to a ceiling in compensation. Section 4 presents brief concluding remarks.

Modelling strategy formulation and implementation
In this section we start by presenting the main ingredients and structure of the model. We assume the firm is run by a risk neutral manager with no personal financial resources. There are three main steps for him to take: he formulates a strategy, takes it to the shareholders for approval, and then implements it. The outcome of the strategy formulation process -i.e., the strategy that is proposed by the manager -is influenced both by the manager's choice and by random factors 9 . There are two strategies for change that may, or may not, be feasible. The manager learns whether change is feasible, and if so, which kind of change, and then chooses to formulate one of them. The shareholders do not observe which of these two might be feasible. They are denoted M and D, corresponding to notions of "Moderate change" and "Dramatic change" respectively. In addition the status quo option B "Business-as-usual") remains available. We assume that there is a probability p (known by all) that change will be feasible. With probability 1 − p only strategy B will be available. If change of some nature is feasible, the manager will learn whether it is moderate change (strategy M ) that is possible (with probability q M ), or dramatic change (strategy D) (probability q D ) or both (probability 1 − q M − q D ). In this last case the CEO will have to choose whether to investigate strategy M or strategy D. Whereas the manager observes these realizations (i.e., whether change is feasible, and if so, whether M , D, or both), the shareholders will only learn what the CEO communicates. We assume that the CEO cannot credibly invent feasible strategies when he communicates with the shareholders, but he can conceal them if he wants to.
So, at the time the CEO decides which strategy to formulate, there are four possible combinations of alternatives: In the fourth case, there is potential for the manager to make a choice of which strategy to formulate and this may or may not be in the shareholders' interests. Distortion of the strategy choice means that the manager would choose D while, in first-best, the shareholders would prefer M , or vice versa. In the first, second and third cases, the manager has no discretion.
The next stage is for the CEO to present his strategy to the shareholders for their agreement, and then proceed to implement the chosen strategy. The details of the contracting process between shareholders and CEO are described in the next subsection.
We need to complete the description of the strategies by characterizing them in terms of parameters, and imposing some conditions on these parameters. Strategy i (for i = B, M,or D) is characterized by effort from the CEO costing him e i and probability π i of success, in which case the firm is worth V . Otherwise (no effort, or sufficient effort but bad luck) the firm is worth nothing.
We assume that strategy M ("moderate change") involves substantially more effort than carrying out business as usual (e M > e B ) and also a higher rate of success (π M > π B ). We make a stronger assumption, that e M e B > π M π B . This means that change requires higher effort than the status quo and has a higher chance of success, but proportionately the personal effort increases by more than the chance of success of the firm.
Turning to strategy D -"dramatic change" -this represents a larger and more radical restructuring than M and will require higher managerial effort, so e D > e M . We also assume that π D > π M . However, we suppose that there are diminishing returns to change, so that the relationship between chance of success and effort has a concavity property: The assumption of diminishing returns is key to our analysis: as we shall see, the idea that really dramatic change is very costly to implement is important in deriving our results.
We impose a few more assumptions on the parameters, whose purpose will be made clear in the following sub-sections. We assume that V ≥ e B π B . This means that shareholders are willing to compensate the CEO for the effort in implementing strategy B, rather than just letting the firm collapse. Similarly for M and D we impose: As will become clear below, condition (2) implies that the shareholders will be willing to pay for the CEO to carry out strategy M if it is presented as an alternative to the status quo B. Condition (3) performs the same role for D.
We impose a condition to ensure there is a conflict of interest between the CEO and the shareholders regarding strategy formation.
As will be shown, (4) means that shareholders prefer strategy M over strategy D, whereas (1) implies the opposite for the CEO.

Contracting between the shareholders and the manager
There are three main elements to contracting in our model: the manager's limited liability, the timing, and the assumptions on contractibility. Since the manager has no financial resources, the contract he agrees with the shareholders can stipulate non-negative payments only. Potentially there are two times when contractual arrangements could be made or renegotiated: (1) at the initial stage, when the manager is appointed and before he has had time to formulate a strategy and (2) after he has formulated a strategy but before he has implemented it. We assume that strategies are observable, but not verifiable. Hence, a contract can stipulate a payment that is conditional on firm value, but not on the strategy. We consider contracts that take the form of a positive payment in case the firm is worth V , and no payment if the firm is worth nothing. 10 Two key implications of the contractibility assumption are: (i) it is not possible to make payments conditional on the strategic plans he proposes to the shareholders; (ii) the CEO can agree to implement one strategy, and later decide he prefers to implement another.
We will consider three types of contract, and study conditions for each to be optimal. The first type is what we call "ex-post contracting." In this case, the contract is set at stage (2). This allows the contract to be set in response to the strategy that the manager has formulated. The second type is what we call "ex-ante contracting". The contract is set at stage (1) and can then be renegotiated at stage (2) by mutual agreement. The main part of our analysis considers the trade-off between ex-ante and ex-post contracting, but we then extend the analysis to consider a third type of contracting in which, as part of an ex ante contract, shareholders commit to rule out certain types of stage (2) renegotiation. Both in the initial negotiation and in the event of renegotiation, we maximize the principal's payoff subject to meeting the agent's reservation utility level 11 .

Setting CEO compensation once strategy is decided: ex-post contracting
To start with suppose that CEO compensation is not fixed at the time he is appointed. Instead, the owners wait until the CEO has presented his strategic plans, then they agree on the choice of strategy and set the compensation package (i.e. ex-post contracting). In the simplest case (occurring with probability 1 − p), there is no option for change and the manager's task is simply to implement strategy B. Let m B be the contractual payment in the event the firm is worth V . Incentive compatibility requires: and the solution is 12 : The CEO obtains his reservation utility level (zero) in this case. The second case is when the manager reports that there is an opportunity for moderate change M . Denote the resulting payment m M (in the event the firm is worth V ). If the shareholders agree to implement M , then there are two incentive compatibility conditions. The first requires expected compensation to outweigh the effort cost: and the second requires that the manager really has an incentive to choose M over B. Since the choice of strategy is observable but not contractible, if the payment is inadequate, he may pretend to implement M , but actually stick to B. The condition is: It follows immediately from our assumption that e M e B > π M π B that (8) is the binding constraint, thus the optimal contract "overcompensates" the manager relative to his reservation wage (0). The solution is 13 : 11 This is a standard assumption. 12 So long as V ≥ m B , as we have assumed. In other words B is a positive NPV project. 13 So long as the shareholders are willing to pay that much, i.e. (V −m M )π M ≥ (V −m B )π B , as we have assumed (condition (1)). In other words M has a higher (positive) NPV than B, even allowing for the agency costs of implementing it as an alternative to B.
We denote his expected payoff from strategy M as U M : where the second equation follows by setting equality in (8) and substituting (9). We can use this derivation for strategy M to see the characteristics a CEO likes in a strategy. Perversely the manager prefers strategies with a lower chance of success and a higher effort level: Proposition 1 So long as the parameters of the model satisfy condition (2), the CEO's preferences for strategies are increasing in effort e M and decreasing in the chance of success π M .
Proof. Immediate by inspection of (11). The next case is when the manager reports that there is an opportunity for dramatic change D as the alternative to B. D has higher effort and higher chance of success than M , but a less favourable effort/success ratio. The incentive compatibility constraints are similar, the solution 14 for the payment is m D = e D −e B π D −π B and the CEO's expected payoff is: Again the CEO derives positive surplus (U D > 0). Furthermore, if the CEO has a choice, he prefers to formulate strategy D over M : U D > U M :, since diminishing returns (assumption (1) above) implies e D −e B πD−πB > e M −e B πM −πB . Turning to the shareholders, they prefer both M and D to B, even taking into account the agency costs of implementing them. Both types of change have higher NPV's than B (which is itself positive NPV), by assumptions (2) and (3). Given a straight choice between M and D, they would prefer M so long as: which follows immediately from assumption (4). This condition has a natural interpretation: the left hand side is the extra firm value created by strategy D; the first term on the right is the extra managerial effort cost of implementing D, and the second term represents the additional rents earned by the manager from ex-post contracting to implement D.
For future reference, their expected payoff (averaging over the different possible strategic choices facing the manager) when the CEO chooses M is: On the other hand if the manager were to formulate D instead, their expected payoff would be: Proposition 2 So long as the parameters of the model satisfy conditions (1), (2), (3) and (4), with ex-post contracting the CEO prefers to formulate the dramatic change strategy D, whereas the shareholders would prefer him to investigate the moderate change strategy M .
To summarize the analysis so far: both types of change (M and D) are positive NPV and preferred by the shareholders to the status quo B (even allowing for the agency cost of implementing them). Of the two, the shareholders prefer moderate change, M . Hence ex-post contracting creates a further agency conflict: the CEO's strategy choice is distorted because he has an incentive to seek hard strategies with an unfavorable effort/success ratio, which drive up the required compensation. He will prefer dramatic change D.

Setting compensation to influence strategy formulation: ex-ante contracting
Anticipating this problem with ex-post contracting, shareholders may seek to derive a better alignment between their interests and the CEO's by designing a better compensation package at an earlier stage. Clearly, one way of ensuring better alignment of interests is just to give the CEO a large enough equity share at the outset -if he gets 99% of the equity, this difficulty will probably be eliminated. But giving the CEO a large amount of equity, share options, or an equivalently generous bonus is very expensive. There is a trade-off. In this section, we explore how shareholders can use this kind of device to correct the CEO's incentives. We assume they are able to set the compensation contract ex-ante, before the CEO has started to formulate strategy. Thus the contract consists of a specified payment in the event the firm is worth V , and it can be renegotiated later when when strategy is agreed, if mutually agreeable. 15 Suppose that the initial contract is set at m * (when the firm is worth V ) and that renegotiation results in final compensation levels m * B , m * M , and m * D (in the event the firm is worth V ) in each of the three possible combinations of feasible strategies that the manager may present to the shareholders following the strategy formulation stage. 16 Clearly m * B , m * M , and m * D cannot be less than m * ; otherwise the manager would not agree to the renegotiation. 15 We assume here that agents cannot pre-commit not to renegotiate, an assumption that is discussed and relaxed in section 3 below. 16 Although the set of possible histories is richer, it is clear that the contract could not distinguish between events such as: (i) the manager initially had a choice between M and D, but chose to investigate D,versus (ii) the manager initially had no choice, and D was the only option for change. Thus, the most that one could hope for is three final different payments m * B , m * M , and m * D . Even then, we shall see that not all combinations of three positive real numbers are achievable.
Note that ex-post contracting can be viewed as a special case of ex-ante contracting. If a low payment is set initially (at most m * = m B ) it will simply be renegotiated upwards to where m B , m M , and m D are the ex-post contracts as seen in the previous subsection. Hence, we refer to this case as ex-post contracting, while the term "ex-ante contracting" will be reserved for the case where m * > m B .
Lemma 3 Ex-ante contracting results in a floor m * such that The standard procedure for solving an agency problem requires us to list all the possible actions of the agent, for each action to compute the cheapest way for the principal to induce the action, then to compute the principal's payoff that results. Finally the principal must compare his payoff across actions and pick the optimal one. For this model the action space of the agent is complex because the CEO needs to decide whether to choose M or D, if there is a choice, at the strategy formulation stage; whether to choose B or the alternative, if there is one, at the strategy choice stage; and whether to put in the required effort for the chosen strategy at the implementation stage. However, we can simplify matters by eliminating actions where there is ever a possibility that the CEO does not put in sufficient effort, as proved in the appendix. Given this we can specify the relevant actions by describing the choices of strategy at the formulation stage and at the implementation stage (i.e., ignoring all actions without sufficient effort). The list and full description of the eight such actions is given in the appendix, where we also show that six of these can be eliminated from consideration here 17  we have πD−πB´− (e D − e M ) π M One can verify that m * defined by this formula does not exceed m D , as a consequence of our assumption that e D e M > π D π M (or equivalently m M < m D ). We can now complete the solution of the problem by examining the shareholders' preferences. The shareholders will prefer to induce choice of Moderate Change Favoured using ex-ante contracting with m * as just derived, instead of accepting using ex-post contracting to induce choice of Dramatic Change Favoured, if: 17) or, equivalently: The shareholders will use ex-ante contracting with a guaranteed minimum m * (paid in the event of success) to induce the CEO to prefer Moderate Change Favoured if This ex-ante contracted payment m * will be renegotiated upwards to m D in case the CEO has no choice over available strategies and presents D to the shareholders.
If the inequality is reversed, they will prefer the ex-post contract implementing Dramatic Change Favoured.
This expression has a simple intuitive explanation. The term on the left-hand side represents the shareholders' payoff when the CEO chooses to formulate M over D, which is the benefit of a regime of ex-ante contracting. For this to be desirable it must be greater than the payoff when the CEO makes the opposite choice (the first term on the right hand side). But it must also compensate the shareholders for two drawbacks associated with ex-ante contracting: (i) the shareholders will have to overpay the CEO (relative to the ex-post levels) for implementing strategy B when there are no alternatives to it (the second term on the RHS); (ii) and they will also overpay the CEO (relative to the ex-post levels) for implementing strategy M , even when dramatic change D is not a real threat (third term on the RHS).
So, when is ex-ante contracting desirable? Clearly, if change is likely (p is big) then ex-ante contracting is helpful to the shareholders, because one of its drawbacks, the overpayment for strategy B, is less relevant. Also, if moderate change, M , is not particularly likely (low q M ) ex-ante contracting is also more favourable because the overpayment for strategy M (the third term on the RHS) is less important. One can interpret both of these cases (bigger p and low q M ) as corresponding to a more unstable, changeable business environment.
The conclusion is that in a highly changeable environment, shareholders might wish to pre-commit to give the CEO an apparently overgenerous package of incentives at the outset. Although sometimes overgenerous relative to the immediate task at hand, it would improve his incentives for strategic decision making.

Extension: Setting a ceiling on compensation
In this section we consider the case where the shareholders may be able to make pre-commitments about the ex-post renegotiation of the CEO's compensation contract, specifically when they set an upper limit to compensation. The benefit of such a compensation scheme is the possibility of preventing the CEO from choosing to formulate his preferred strategy D over the shareholders' preferred strategy M . If he knows he will never be paid enough to implement an overambitious strategy, the CEO will lose interest in such a plan. Establishing ex ante a credible ceiling on compensation will also have a drawback: even when dramatic change is the only possible alternative to the status quo it will be disregarded.
The issue of pre-commitment not to renegotiate is open to debate. Clearly, companies and CEO's often do renegotiate compensation. The typical arrangement consists of an annual salary combined with a stock option grant, so each year the salary is renegotiated and the CEO also adds more options to an existing portfolio. Hence, pre-commitment may seem unrealistic. From a theoretical point of view also, it is hard to see how commitment could be enforced -perhaps by a third party, but then, the design of the contract with this third party would have to be complex and might violate the spirit of our prior assumptions on contractibility between CEO and shareholders, and it would certainly be unrealistic.
However, perhaps social norms or individual ethics could be used as a precommitment device. For example, in Sweden it would probably be possible for a company to pre-commit never to pay the CEO over US $1bn (as was paid to Roberto Goizueta of Coca-Cola over a ten-year period). Even within the US, 25 years ago such a payment might have been impossible. Thus pre-commitment may be possible at the social level. At the level of the individual company, it may be possible to pre-commit by appointing individuals who are known to be strongly opposed to high compensation to the board or to the remuneration committee. Having large block shareholders with this view might also have the same effect. For that reason, we believe it is worthwhile to explore the case where pre-commitment is possible. We therefore add the following assumptions: We consider the possibility of pre-committing not to engage in certain types of renegotiation. The only pre-commitment of that type we consider reasonable is setting an upper bound on the level of compensation. 18 Clearly this will not help implementation of the two actions previously analysed ( Moderate Change Favoured and Dramatic Change Favoured ) because they require a high payment for strategy D. The interesting case now is the possibility to implement another action which we call Moderate Change Only 19 . This action is characterized by choosing M whenever it is available and ignoring D even if it is the only alternative to B. The shareholders can correct the manager's preference for D by committing themselves never to pay enough compensation to make the CEO willing to implement that strategy. It therefore requires m * B ≥ m B , and m * M ≥ m M but not m * D ≥ m D . So the cheapest way to implement it is to (credibly) pre-commit never to pay more than m M . The 18 The last assumption rules out pre-commitment of the sort where the board commit never to offer a payment in a set A, and the set A can have an arbitrary shape such as a subset of the real line that is not connected, the set of rational numbers, etc... Recall that strategies are not contractible, hence the contract cannot be made contingent on the CEO's announced strategy. 19 Denoted by action A7 in the appendix. In the appendix we also show that the remaining five actions, A2, A3, A4, A6 and A8 are not optimal. expected payoff for the shareholders is then: We can compare this expected payoff to what shareholders would obtain in case the contracting process involved ex-ante contracting without pre-commitment, and also in case we had an ex-post contracting policy. From these comparisons we reach the following results.

Proposition 7
The shareholders will prefer ex-ante contracting with a ceiling on compensation (inducing the CEO to choose Moderate Change Only) to exante contracting with unrestricted renegotiation if: with m * as defined in proposition 5.
Proof. Direct from comparison of the shareholders' expected payoff with exante contracting with a ceiling (equation (19)) and without an upper bound (LHS of inequality (17)).

Proposition 8
The shareholders would prefer the ex-ante contract with a ceiling (inducing the CEO to choose Moderate Change Only) to ex-post contracting if: Proof. Direct from comparison of the shareholders' expected payoff with exante contracting with a ceiling (equation (19)) to their expected payoff with ex-post contracting (equation (14)). If both conditions (20) and (21) are satisfied, the shareholders would prefer to completely rule out the possibility of the manager formulating strategy D. We can see that setting an upper bound would be optimal if it is unlikely that the strategy for dramatic change, D, is the only one available (low probability q D ) and if the probability of change (p) is also low. This is the scenario in which the drawback of setting an upper bound on compensation -the CEO will not develop strategy D (which is, by assumption, superior to the status quo) when this is the only option available to him -is less relevant.
We can interpret this case (low q D and small p) as a very stable environment: a low chance of any change being required, especially dramatic change. In this case the ceiling on compensation is a cheap way to correct the manager's preference for dramatic change. If this is not the case, then it will be preferable to set either an ex-ante contract without a ceiling, or an ex-post contract.

Conclusion
Our results are simple to summarize. In this paper we analyzed how managerial rent-seeking may distort strategy choice in favour of overambitious change. We showed that, in our model: • In a highly changeable environment (change is more likely to be required) the shareholders should commit to a policy of high pay, offering the CEO a high-powered incentive package at the outset in order to improve his incentives for strategy making, even if sometimes this results in excessive (with hindsight) payment at the implementation stage. This will correct his incentives away from excessively drastic restructuring and towards the kind of change that is preferred by shareholders.
• Another way of curbing CEO's tendency towards dramatic change is to pre-commit never to pay the very high compensation package required to implement dramatic change. It may be difficult to make this precommitment, but if it is possible, then it will be preferred if dramatic change is not likely relative to moderate change, or if change is unlikely overall.
• Finally, in some situations it may be optimal for the shareholders simply to accept some strategic distortion and to negotiate compensation to the required level after the strategy has been selected. This will occur when change is not very likely or when shareholders do not have strong preferences among the possible change options.
• Our analysis can help understand why CEO's typically receive much higher pay and stronger incentives than other senior managers. Their pay reflects strategic discretion more than direct compensation for effort. Our results are also consistent with the empirical evidence of higher incentives in more changeable environments.
• One can also hypothesise that the economic pressures in recent years for firms to transform and reinvent themselves have led to "ex-ante contracting", i.e., a precommitment to high pay and incentives. It is an open question whether mechanisms (perhaps at the social level) to precommit to upper bounds on pay are either possible or optimal.
An alternative perspective on CEO pay is that high pay is not optimal at all for shareholders and simply results from rent extraction in collusion with the board. While we do not deny this possibility, we consider it is worthwhile exploring how high pay can arise in a model with only relatively minor departures from standard agency theory.

Appendix
Remarks on incomplete contractibility The approach we take here applies what is now a standard paradigm, following many papers in the past decade that have studied financial contracting where some variables are observable by both parties, but not contractible. Aghion and Bolton (1991), Hart (1995), and Hart and Moore (1998) are leading examples. In general, non-contractibility can arise because it would be too expensive or too complex to make contracts fully conditional, or because it would be difficult for a third party to verify fulfilment of the conditional clauses in the contract (since contracts depend on third parties such as courts or private arbitrators for enforcement). For example, Hart (1995) and Hart and Moore (1998) assume that investment is not contractible. They argue that even if third parties were able to verify monetary expenditures on investment, they would be unable to tell if the funds were applied properly to the right kinds of projects. This is similar to our assumption that strategy is non-contractible. Only "hard" variables are contractible upon.
To motivate our analysis, consider the following example. Suppose that a professor is given responsibility for a degree programme. He or she can choose to run the programme on a business-as-usual basis, i.e. attending the usual committee meetings, checking the lecturing performance of his or her colleagues, monitoring the performance of the programme office in regard to admissions and records, and taking an appropriate interest in student welfare. This is a moderately time consuming, but manageable task and the professor would be entitled to expect some measure of compensation in the form of a reduced teaching load or moderate salary supplement. Now suppose the professor is tasked with restructuring the course, with a freshly-thought out structure and syllabus that reflects current student demand and is fully up to date. The first observation we make is that engineering such a change is extremely demanding in terms of time and energy. Many proposed changes, however trivial, will meet with resistance from entrenched lecturers. Even colleagues who are enthusiastic about change will respond by making a wide variety of counter-proposals that makes coordination on an improved outcome difficult. So change from the status quo can be disproportionately costly and the professor, unless he obtains intrinsic satisfaction from the task, will not be keen to take on this job without substantial additional compensation (generally in universities, such extra compensation is unlikely!). Of course, firms may not be quite as resistant to change as some universities but we argue that the same broad feature applies.
The second observation we make is that it may be easy to spot whether the changes made by a colleague are profound or superficial, but it may be hard to prove that assessment in a way that is credible to an outsider. A professor may feel that the course director's "relaunched" programme is just a minor variant of the old one, but, if he were challenged on this view, it might be hard for him to convince a colleague in another department, a student, or the course director's lawyer. So it would be impossible to condition compensation on the successful implementation of a truly improved degree programme. We argue that in businesses generally, similar reasons are likely to make it impossible to condition pay directly on strategy. However in businesses, unlike universities, pay can be conditioned on share price.
To summarise the two key points of the example: first, change can require disproportionately high effort to implement, relative to the status quo. Second, writing a contract conditional on "satisfactory change" is likely to be impossible.
In our model we have in mind a variety of corporate strategies, of which a possible simple example would be the introduction of a comprehensive costcutting programme. At first sight, it might seem that costs can be verified easily from the company accounts. However, this would be a crude way of monitoring the successful implementation of a genuinely shareholder-value increasing rationalisation plan. It could likely be quite easy for the manager to slash costs with a poorly executed programme of cutbacks that would actually damage shareholder value in the longer term (through employee morale, quality control problems, disruption of supplier relationships, etc). Just because a strategy has implications for an easily quantifiable variable, it does not mean that controlling that variable will actually verify implementation of the strategy.
Holmström and Milgrom (1991) give a good illustration of this kind of problem. A teacher in South Carolina was found to have boosted her class exam performance, and hence her own performance rating, by passing the children answers to the statewide geography test. Another egregious example is given in the Financial Times of 4 February 1994: the recently appointed chairman of Audi discovered that the previous year's sales figures had been "bolstered by an old trick. Audi France officials confirmed yesterday that 'several tens of thousands' of cars had been parked with French distributors [and hence recorded as sales], only to be shipped back to Germany last year. Since many of them lacked airbags and ABS braking systems -regarded as essentials in the Germany quality car market -selling them was no easy task." This shows why even apparently "hard" data such as company accounts may not be of much help in writing conditional contracts.
Lemmas and proofs Proof of Lemma 3. We show that m * B = max{m * , m B }; a similar argument holds for the other two cases BM and BD. Let the ex-ante contract specify a payment m * in the event the firm succeeds and is worth V . First note that, if renegotiation occurs, it will occur upwards only, otherwise the CEO will reject the proposal. Hence m * B ≥ m * . Next, note that the renegotiated payment cannot exceed the ex-post contracting level, otherwise the shareholders could offer less and get the same behaviour from the CEO.
Suppose first that m * ≥ m B , then it is clear we cannot have renegotiation and m * B = m * . Next suppose that m * < m B , then it is clear (from the derivation of m B ) that renegotiation is mutually beneficial and will lead to a contractual payment m * B = m B .

Lemma 9
Under an optimal contract, the CEO always puts in the required effort for the chosen strategy.
Proof. Consider first the case of ex-post contracting and the possibility that in case B is the only available option, the CEO does not put in the required effort e B (implying m * B < m B ). By our assumption that V > m B , this is suboptimal because the shareholders would prefer to pay m B and induce effort. The cases BM and BD are similar given our assumptions that the parameters satisfy conditions (2) and (3). In the case of ex-ante contracting the same reasoning applies.
In the case of a ceiling on compensation we know that the ceiling would never be below m B , by our assumption that V > m B . Therefore if the manager moves away from implementing B with effort, it must be in order to implement another strategy with effort. To explain the notation, take the first action on the list, (D, M BM , D BD ), which represents the outcome of ex-post contracting. The first symbol, D, means that when the manager has a choice between M and D at the strategy formulation stage (with probability p (1 − q M − q D )), he picks D. The second symbol, M BM , means that when the CEO presents strategy M to the shareholders as an alternative to B, they pick M (to avoid lengthy circumlocutions, we do not distinguish between the shareholders picking the strategy at the implementation stage and the CEO picking the strategy. Obviously, since the manager can always cheat if wants to, the shareholders knows which strategy he is going to pick so we may as well assume they both pick the same one.) Note that even if he picks D in preference to M at the strategy formulation stage when he has the choice, there will sometimes (probability q M ) be occasions when M is the only option. The third symbol, D BD , means that when the manager presents option D to the shareholders as an alternative to B, they pick D.
The three actions that are referred to in the text are A1 (Dramatic Change Favored), A5 (Moderate Change Favored), and A7 (Moderate Change Only).
For the purposes of the analysis in Section 2 (i.e. the main analysis on ex-post contracting and ex-ante contracting) we can rule out actions A2 to A4 and A6 to A8 as suboptimal, because renegotiation would always take place to implement the change strategy (M or D) rather than the status quo B, by our assumptions on the exogenous parameters satisfying conditions (2) and (3). The argument is similar to the previous proposition. That leaves action A5 (Moderate Change Favored, which corresponds to the manager choosing to investigate M rather than the more dramatic D) in addition to A1 (Dramatic Change Favored).
For the extension to the analysis of pre-commitment in section 3, we cannot exclude actions simply because they are ex-post suboptimal -after all the purpose of precommitment is precisely to allow ex-post inefficient actions in some outcomes, so as to improve ex-ante incentives. However, one can readily see that A2 and A6 are inferior to A1, A3 is inferior to A7, while A4 and A8 are inferior to all three actions A1, A5 and A7. That leaves A1, A5, and A7 as analyzed in the text.