Journal of Economic Behaviorand Organization ETSEVIER JOURNALOF Economii Behavior & Organization Vol. 28 (1995) 417-442 The strategic implications of sunk costs: A behavioral perspective Roth Parayre ’ Southern Methodist Universiiy. Cox School of Business, Dallas 7X 75275, USA Received28 June 1993;revised 30 November 1994 Abstract This paper examines some of the strategic implications of the sunk cost phenomenon in sequential allocation decisions. Drawing from psychology and behavioral decision theory, we first present a taxonomy of possible causes for the ‘sunk cost effect’, the tendency of many managers to throw good money after bad. We then present the analysis of some implications of this behavior in strategic situations. A two-period strategic game is developed and analyzed to derive optimal allocations as a function of one player’s sunk cost behavior. We establish when this behavior can be used as a successful precommitment strategy by the sunk cost player, and when it is exploitable by an opponent. Welfare implications are also explored. The sunk cost effect constitutes a form of strategic preference manipulation, in which a credible preference change can be induced to provide a strategic advantage to one of the players. JEL classification: C72; DSl; D92 Keywords: Sunk cost; Escalation of commitment; Preference change; Duopoly; Signailing game ‘[People] are irrational, that’s all there is to that! Their heads are full of cotton, hay and rags!’ - Alan Jay Lerner, My fair lady. ’ The comments of James Brander, Kenneth MacCrimmon, Donald Wehrung, Colin &meter, two anonymous referees and the Editor are gratefully acknowledged. The usual disclaimer applies. Detailed mathematical proofs are available from the author, upon request. 0167-2681/95/$09.50 0 1995 Elsevier Science B.V. All rights reserved SSDI 0167-2681(95)00045-3 418 R. Parayre / J. of Economic Behavior & Org. 28 (19951417-442 . . . between the thin theory of the rational and the full theory of the true and the good there is room and need for a broad theory of the rational.’ - Jon Elster, 1983, Sour grapes: Studies in the subversion of rationality. 1. Introduction One of the more perplexing problems facing managers is that of deciding on the allocation of resources to a project that requires a sequence of repeated investments before any or all of the benefits arise. Once some resources have been allocated to the project, further information may become available, and the manager may find out that he is better off by NOT investing further in the project. But there is a dilemma: by letting go of the project, it will be difficult for the manager to escape the feeling that the resources previously committed have been ‘wasted’ - even though, before the additional information became available, it may have been perfectly rational for him to invest in the project’s early stages. This problem, and the associated tendency to ‘throw good money after bad’, is commonly known as a sunk cost problem, where the manager is torn between ‘cutting his losses’ part-way through the project or persisting with it in the hope that, against the odds or the better judgment of a financial analysis, additional investment will bring the project to a successful fruition. Economic analysis provides a clear-cut recommendation to the manager facing the sunk cost problem. If the objective is to maximize profits, then the allocation of funds at any point in time should be based exclusively on future costs and benefits. The amounts invested in the past are sunk costs; neither they nor their amortization are relevant to today’s decisions, Those who violate this rule are said to be ‘throwing good money after bad’, leading to the popular label of the sunk cast fallacy. It is hard to dispute this line of reasoning. Yet, the apparent fallacious behavior of throwing good money after bad persists, which justifies its study in economic settings. Empirical evidence from the stock market’s response to divisional terminations (Statman and Sepe, 1989; Parayre, 1993) gives strong indications of the presence of a sunk cost effect. The market rewards firms with an average positive abnormal return of 3% to 4% when they finally ‘pull the plug’ on divisions known to have been struggling in the past, supporting the proposition that managers persist with projects longer than the economics of the project would dictate. For example, after accumulating losses for years on its L-101 1 commercial aircraft, Lockheed announced in December 1981 that it would finally kill the project. Lockheed stock jumped 18% the following day. Similarly, following sustained losses, Texas Instruments announced in October 1983 that it would quit the home computer business. The company’s stock price jumped 22% the next day. R. Parayre/ J. of Economic Behavior & Org. 28 (1995) 417-442 419 If the stock market accounts for management’s sunk cost bias, we can also expect competitors to adjust their strategies in light of the knowledge of an opponent’s escalating commitment to a losing project. Can one player exploit an opponent’s sunk cost effect to gain a strategic advantage and improve his position? Conversely, is it always strategically undesirable, in economic terms, for a player to display a sunk cost effect? In short, when is (or isn’t) the ‘sunk cost fallacy’ strategically destructive? This paper focuses on such strategic implications of the sunk cost phenomenon, by formalizing these intuitions in illustrative models of duopolistic competition. We show that: - in spite of different behavioral assumptions, formal micro-economic and game theory models, based on utility-maximization, can be used to derive the implications of sunk cost behavior; - even though escalating commitment may weaken the escalating firm, a psychological commitment to sunk costs can under some conditions be to one’s advantage. The next section presents a taxonomy of causes for the sunk cost phenomenon. This taxonomy presents a number of psychological causes of the sunk cost effect that complement conventional rationality arguments proposed in the economic literature. In section 3, we develop a model of a two-person strategic game, and examine how the presence of the sunk cost phenomenon affects the outcomes in competitive situations. Section 4 presents conclusions and discusses future research directions. 2. A taxonomy of causes of escalating commitment Sunk cost behavior may be modeled as rational, even though it sometimes violates strict economic optimality, by incorporating additional factors (economic or non-economic) into a utility maximization framework. Modeling behavior in this way was done as early as thirty years ago by Williamson (1963) in Cyert and March’s Behavioral Theory of the Firm. Consider the following examples. First, escalating commitment to a project can be optimal for a firm as a result of moving down an experience curve, because of synergies in the firm’s portfolio, if switching costs are high or if sunk capital costs make a marginal additional investment worthwhile. Escalating the commitment is justified on purely economic grounds, based on marginal costs and benefits, thus maximizing the expected net present value of the firm. ’ ’ The sunk cost phenomenon, as we have defined it, entails psychological exceed purely economic switching costs. costs of switching that 420 R. Parayre/ J. of Economic Behavior & Org. 28 (1995) 417-442 The sunk cost effect can also result from inappropriate managerial incentives, when the manager’s self-interest 3 clashes with the firm’s economic objectives as they apply to the committed project. As pointed out by Staw and Ross (1987a), perseverance in the face of adversity is a quality highly valued by society, and may contribute positively to the manager’s reputation as a decision maker especially under asymmetric information, where the desirability of abandoning a project is information privately held by the manager. Under these conditions, the manager’s reputation (which influences his earning potential) becomes an important component of his decisions. A utility-maximizing manager may well display the sunk cost effect and persist with a committed project beyond the point that is financially optimal for the firm. From the firm’s position, a sunk cost effect results. 4 Kanodia et al. (1989) examine the sunk cost phenomenon in this light, as an effect driven by information asymmetries between principals and agents that includes reputation building on the part of the manager. 5 As a further example, consider the dilemma you face on a Sunday afternoon: you need to work on a paper, but would like to watch the first quarter of a football game on television. Your prior preferences are ordered as follows: watching one quarter 5 watching no football $ watching the whole game. In accordance with these preferences, you settle down to watch one quarter of football. The problem you face is that after watching the first quarter, your preferences change so that you want to watch ‘just a few more minutes’ of the game. After those few minutes, the process repeats itself, and you eventually become trapped. You end up watching the whole game instead of only one quarter, and experience regret for having wasted away the afternoon. The change in preferences may be the result of ‘addiction’ or habit formation rather than of justifying past investments of time. But just as in a sunk cost situation where a manager wishes to ‘recoup’ sunk investments, a (momentary) preference change takes place which influences the decision to persist with a committed course of action. The topic of preference change and the role of commitment in stabilizing one’s behavior have been discussed by several authors, and compiled in Loewenstein and Elster (1992). ’ Differing incentives are not new to economic theory: they are at the heart of the principal-agent problem. Here, reputation effects enter the manager’s utility function as they influence the net present value of his compensation. These effects may include Iong-term reputation as well as short-term compensation. 4 Of course, the firm may never know the (ex ante) economic merit of the decision to escalate. Because of information asymmetries, it can only observe the increased investment in the committed project. 5 Staw and Ross (1987a) (Staw and Ross, 1987b) discuss some of the organizational solutions, including appropriate managerial incentives, that can help mitigate the occurrence of escalation. However, optimal managerial incentives to reduce escalation behavior have not been formally established. This remains a potentially rich problem for agency theory research. R. Parayre / .I. of Economic Behavior & Org. 28 Cl995) 417-442 421 Finally, choices may also result from ‘wrong’ or fallacious subjective parameters, distorted by the sunk cost. Substantial empirical evidence supports such cases of wishful thinking or the illusion of control (Knox and Inkster, 1968; Arkes and Blumer, 1985). These examples, presented in decreasing order of economic rationality, offer some very interesting strategic implications. 3. The sunk cost phenomenon in duopolistic competition 3.1. A generic model Firms must often build up capacity before entering an industry. After doing so, overcapacity may result because of a subsequent shock in the environment which reduces total industry demand, because of additional entrants, or because of over-optimistic projections about capacity requirements. An ‘escalating’ firm will want to use some of its overcapacity despite the fact that it should leave it idle. In general, the more expensive the capacity, the more a lirrn will want to use it. Sample industries of recent overcapacity include banks, DRAM chip manufacturing, many chemicals, air transport, as well as firms with high fixed costs such as extractives (aluminum, steel or soda ash). The wish to recover fixed costs that are sunk will often lead to the sunk cost ‘fallacy’. Consider therefore a situation involving two firms (or players), competing as duopolists in a product market. There are two time periods. At time t,, Player A must decide on the amount s to be invested (sunk) in a project, into (say) plant or R&D. At time t,, Player A must determine his play a, simultaneously with Player B determining her play b. 6 The plays can be investment allocations, production capacities, quantities produced, or prices. This joint play produces revenues for each player received at the end of t,, resulting in profits rrA(s,a,b) and 7rB(s,a,b). Each player is seeking to maximize his/her own utility function. The key departure from a traditional neo-classical economic analysis of the problem lies in the arguments that may enter Player A’s utility function. In an analysis of the sunk cost phenomenon, considerations other than just the impact of marginal profits (or future profits, exclusive of the sunk cost) must be part of Player A’s utility function. Since Player A’s commitment to the project stems from the cost s that he has sunk into it, the sunk cost s must therefore become an explicit argument in Player A’s utility function, along with marginal profits. 6 This analysis extends to other situations where the two players must allocate funds to a project common to both players, such as a joint venture or a competitive project. In this case, the players’ decision at t, is to determine how many additional funds to allocate to the project. Moreover, as each player may also have alternative investment opportunities, the profits from this common project am net of all the other activities each player may be involved in. 422 R. Parayre / .I. of Economic Behavior & Org. 28 (1995) 417-442 In this two-period situation, we can think of s as determining which of several Nash games will be played at t,. Player B has no control over the amount s sunk by Player A at t,. Because of the recursive determination of A’s optimal s * from the Nash solution at t,, Player B can only indirectly influence A’s choice of s through her allocation decision at t,, using her knowledge of A’s failure to ignore sunk costs in determining that allocation. Because B cannot precommit or signal any departure from her Nash allocation prior to her decision at t 1, she will have no choice but to play her Nash move in the game chosen by A through A’s choice of s *. Player A therefore enjoys a clear first-mover advantage, even though final outcomes ultimately depend on the utility functions of both players. We can examine the situation at t,, in which Player A has already sunk s at t,. The optimization problem at t, is: for Player A: max u*(a, b *, SX # n*(a, b *, s) if the sunk cost effect applies) B for Player B: max 7rB(a*, b, s), b where a [and b] represents Player A’s [and B’s] play at t,. With s being Player A’s sunk investment, and if a is quantity produced (or production capacity), escalation of commitment can be expressed as u,“,> 0. That is, as the sunk cost s in a project increases, the marginal value of escalating commitment in that project also increases. ’ This commitment, which may stem from behavioral considerations discussed in section 2, enters the manager’s utility function at time t,. 8 A comparative statics analysis of the Nash game played at t, provides insight into the impact of Player A’s sunk cost effect on the resulting equilibrium allocations, and on the strategic advantages or disadvantages that emerge. Situations where the decision variables are quantities produced, production capacity, or further investments in a project, usually exhibit downward sloping reaction functions. Taking first-order conditions, differentiating them totally with respect to s and using information from the second-order conditions 9, for downward sloping reaction functions we get: da* db’ ds > 0 and ds <o. : For our purposes, this need only hold locally, around the equilibrium. In many cases, this characterization may be based on purely economic considerations - for example, under significant inteltemporal economies of scale in project #l. Under intertemporal economies, one is better off investing smaller amounts sequentially in the project, either because learning occurs or because congestion diseconomies make a huger one-time investment in the project at t, uneconomical. Similarly, sinking s at t, may reduce A’s marginal cost of production at t,. ’ Second-order conditions require that at the Nash equilibrium, u& < 0, atb < 0 and I&T& u$qi > 0. In addition, downward sloping reaction functions require u& naF < 0. R. Parayre / J. of Economic Behavior & Org. 28 (1995) 417-442 423 Thus as Player A’s sunk costs in project 1 increase, A’s tendency to persist with project 1 also increases - accompanied by a shift in his reaction function. B’s profit function, on the other hand, and hence her reaction function, are unaffected by a change in s (except through the induced change in a’). An increase in a* therefore leads to a decrease in b’ as we move southeast along B’s downward sloping reaction function. When the direction of increasing profits is towards the monopoly solution, as in most Cournot duopolies (where firms compete on quantities), then the following holds at the Nash equilibrium: As A’s sunk cost s increases, 7~A (excluding s) increases, while r* decreases. Hence, it is not to B’s advantage for A to increase his sunk cost commitment s, as A’s escalation effect makes Player B worse off. This result lies at the core of our signalling application, which we now develop. 3.2. Sinking costs as a precommitment strategy: using your sunk cost effect to your advantage Spence (19771, Spence (19791, Dixit (1979) and Dixit (1980) demonstrated the economic benefits of sinking costs into idle production capacity as a means of deterring entry. Ghemawat (1991) also stresses the broader economic importance of commitment as a strategic tool. While economic deterrence is quite compelling, we argue that there is also economic value in sinking costs as a psychological deterrent. We therefore model the implications of sunk cost behavior resulting from psychological causes - as opposed to earlier economic explanations or to reputation concerns in repeated games played by profit-maximizing players. lo We examine the psychological importance of commitment in strategic situations, which in turn will influence the economic outcomes of the firm making the commitment. We illustrate this via a simple Cournot duopoly signalling example I’, which builds on the intuition from the preceding comparative statics. Consider the following situation. Player A has sunk s into the development of a product which, at t ,, he will sell in a duopoly with Player B. Moreover, Player A can be of one of two types: Type N: Displays 120sunk cost effect Type S: Displays a sunk cost effect. Type being information private to Player A, Player B starts off by assuming a probability distribution about A’s type, based on a population base-rate or on Player A’s reputation for having escalated commitment in the past. B then ‘” We do however preserve ‘internal consistency’ - where at any point in time, decisions are those which maximize. one’s utility function at that time. We also preserve the recursive logic of dynamic programming. ” For the fundamentals on incomplete-information games of this type, see Harsanyi (1967-68). 424 R. Parayre / J. of Economic Behavior & Org. 28 (1995) 417-442 observes the amount s that A sinks into the development of the product, and with this information updates her beliefs about A’s type. Using this updated (posterior) distribution, B then goes on to play a simultaneous Nash (Cournot) output game with A. The price of the (jointly produced) product is: p=-y-6(a+b), (y,E>O). With little loss of generality, costless production is assumed (for simplicity), as is no time discounting between t, and t, (as would be the case for very short time periods). We can solve for the equilibrium solution(s) of this game recursively, in three distinct steps. First, we solve B’s output game. Let p be p(s), B’s posterior about A’s type ( p also known to A). Then B’s output decision is to My[y - 8(a + b)]b s.t. A plays argmax[ y - S(a + b)] a and plays argkax[r a (a(s) da/ds>O) 20; with probability ( 1 - p) - S(a + b)]a + a( with probability p. That is, B maximizes rrB under the condition that A of type N maximizes rr A and A of type S maximizes uA = +r~’+ o(s)a. The term a(s>a is a particular parametrization of escalation of commitment - where an increase in s encourages increasing a, thus leading to u:~ > 0 in this Coumot case. This is our fundamental characterization of a ‘sunk cost effect’, and is intended to capture the increased psychological commitment associated with an increase in s. B’s optimal production is found by solving the first-order condition for B simultaneously with the first-order condition for A. Given the uncertainty about Player A’s type, Player B must take Player A’s (uncertain) production to be a* = (1 - p)ah + pa: I This is shown in Fig. 1. This graph shows the optimal reaction functions for Player A of Type N, of Type S, and for Player B, and the resulting optimal production for B (point P *> given the uncertainty about A’s type. Point P * is located at (1 - p)P, + pPs. We obtain b* = Y - pa(s) 36 . R. Parayre / J. of Economic Behavior & Org. 28 (1995) 417-442 425 b*= a Fig. 1. Player B’s output decision via reaction functions. Player A’s production will then be a function of his own type. Solving A’s first-order condition simultaneously with b’, we find that A of type N will produce * _ aN - 2Y + Pa 66 with u;=nN-- TB= * _ 4Y2 + 4PW(S) + p2a2(s) 366 2y* - pya!(s) - pW(s) 186 Similarly, A of type S will produce * _ 2Y + (3 + p)a(s) as 66 R. Parayre / J. of Economic Behavior & Org. 28 (1995) 417-442 426 yielding * _ 4y2 + (12 + 4p)ya(s) + (p + 3)2aZ(s) us 363 na _- 2Y2 - (3 + P)V(S) + (3P - p2)a2(s) 188 The interesting result at t, concerns r*, the marginal profit part of u*: * _ 4Y2 + 4PYQ) =s - + P2a2w 366 -- ~2(s) 46 . In this case, any a(s) < 4py/(9 - p2> will increase rt. I2 Hence, Player A can actually be belter ofSfinan&lly at t, than in the case where A is known to be maximizing rr * only, even though he displays the sunk cost effect - a positive by-product of B’s uncertainty about A’s type. This occurs because B reduces her Nash production to account for the possible sunk cost effect of A. This result addresses the impact of the sunk cost effect as of t,, once s has already been committed. But should Player A deliberately sink s at t, to begin with (assuming it is not necessary), in order to then use his sunk cost effect at t, to his competitive advantage? Can Player A be made better off by such a strategy even at t,? We can find out by comparing the profit function at t, with and without a sunk investment s. With no discounting between periods, Player A’s profit function at t,, when investing s, equals: I3 4y2 + 4pyc!L(s) + p%?(s) ForAofTypeN: n$= ForAofTypeS: 7~:= 368 472 + 4pyo!(s) + p%?(s) 366 -S 02(s) -4s-s The optimal amount s* depends upon the specific values of the problem parameters. Characterizing this equilibrium requires that we consider 3 possible parameter scenarios. These are illustrated in Fig. 2. Case I TN” and rt are both maximized at .sA= si = 0. This is the equilibrium. Case 2 rri is maximized at sl; > 0 and nt is maximized at si = 0. Then any s > 0 signals with probability 1 that A is of Type N, so that p = 0. Hence both types of A are best off playing s * = 0. This is again the equilibrium. I2 Non-negativity conditions for b constrain the allowable parameter range to be y 2 pa(s). I3 Note that for any positive s and the associated positive (I(S), ?T: > w[. When s = 0 and hence (Y(s) = 0, ?r; = lrs” = y*/9s. l R. Parayre / J. of Economic Behavior & Org. 28 (1995) 417-442 427 428 R. Parayre /J. of Economic Behauiar & Org. 28 CI995) 417-442 Case 3 rr: is maximized at sl > 0 (and because ri > ,rr.$‘Vs, rr$ is maximized sk > 0). This is the richer, more interesting case. It occurs when 4PW(S) + p202(s) 368 02(s) - 46 at -s>O(forTypeS) (and hence, automatically 4PYdS) + p2a2(s) 366 -s>O(forTypeN)). Here, both A Types N and S have an incentive to increase p, which yields greater profits. Hence, it is to A’s economic advantage (in terms of purely economic profits), regardless of his type, to put himself in a sunk cost situation at t, - a first-mover advantage. A then benefits from an increased a* in equilibrium, under the threat that he may be maximizing a utility function at t, that includes a sunk cost effect component. This advantage to a player A of type S thus stems from the output adjustment it triggers in B; an aa’ditionaf advantage to a player A of type N stems from B’s belief that A may be of type S. Intentionally sinking (and losing) costs thus becomes beneficial because it convinces an opponent that you may not be a perfectly rational adversary (from a narrow rA perspective) in the next time period - which is something you may be able to exploit. The following results characterize the equilibrium of this signalling game: Non-existence of a separating equilibrium Because 7~; * > rt * > y2/9S (the latter holding when s = 0), there exists no s > 0 that would make Type S better off than s = 0 while making Type N worse off than s = 0 (see Fig. 2, Case 3). Therefore, a Type S can never shake himself free of a Type N trying to mimic him - which a Type N will always want to do (to increase p>, as any sN # sg would signal that A is of Type N and would push p to 0. Uniqueness of the pooling equilibrium Because si maximizes n=c given p, it also maximizes p (it signals that the amount s being played is the best move for a player of Type S). From the preceding paragraph, si is also the best move of a Type N. Hence, the unique optimal s for both types of A is si. This is a non-informative allocation, so that Player B’s posterior about Player A’s type = p = Player B’s prior about Player A’s type. Hence, p is an exogenous parameter which, in conjunction with the particular values of y. S and cr(s), will determine which Case (1, 2, or 3) is being played. R. Parayre / J. of Economic Behavior & Org. 28 (1995) 417-442 429 A’s sunk cost, regardless of A’s type, will be and = 0 = si in Cases 1 and 2, in Case 3. The subsequent equilibrium output allocations of A and B will then be made, as described above. Note that a nonzero s * invested (in Case 3) by Player A of either type makes him better off, while making Player B worse off than if s = 0. The sunk cost effect thus provides Player A with an interesting means by which to gain a strategic edge. By sinking money into the project at t,, Player A of Type S precommits himself to displaying a sunk cost effect in the ensuing game, and in effect, to behaving ‘irrationally’ (from a pure profit perspective) at t, - in essence, strategically manipulating his own preferences, in order to be better off financially in the end. To quote Schelling (1960): ‘It may be perfectly rational to wish oneself not altogether rational ’ (p. 18). While Schelling made that statement in the context of making threats credible, it also applies to our case of sunk costs. The model just developed formalizes that intuition. By sinking costs at t,, Player A signals the possibility that he will maximize uA rather than g A at t,. If Player A can successfully convince Player B that he is subject to a sunk cost effect and play an apparently - from Player B’s perspective - irrational strategy, then A can gain a strategic advantage. To do so, A must precommit himself to sinking s into the product at t, and must convince B that the effect of this sunk cost will be (Y(S)in his utility function. Of course, in order to convince B it may be necessary for A (of Type S) to actually fall prey to a sunk cost effect - a strategy in the spirit of true precommitment. In that case, no misrepresentation occurs: the advantage to A results because of, and in spite of, his own sunk cost effect. Frank (1988) presents a compelling, comprehensive discussion of the adaptive value of what appear to be ‘irrational’ choices, and also addresses the problem of mimicry (e.g., a Type N trying to pass off as Type S). This subtle finding of a ‘rational irrationality’ requires further elaboration. In a two-period model, the problem of bridge-burning, as an example, assumes a constancy of purpose over both periods. Indeed, you burn your bridges at t, to ensure that your choice at t, maximizes the utility function of t,. Similarly, Ulysses asked to be bound to the mast of his ship to ensure that his ‘irrational’ self at t, did not undo what his ‘rational’ self wanted at t,. By constraining the choice set available at t, , a strategic precommitment thus ensures consistency of actions with the preferences of t,. Subgame perfection is thus imposed by restricting flexibility at t, to choices which are optimal from the ‘rational’ perspective of t,. In contrast, our sunk cost player is allowed, depending on his type, to display his ‘irrationality’ (of maximizing u instead of rr) at t,, rather than constraining that possibility away. B must then adjust her output to account for that possibility. Given that adjustment, A would always be better off maximizing 7~ at that point. 430 R. Parayre / J. of Economic Behavior & Org. 28 f 1995) 417-442 But under this strictly ‘rational’ subgame perfect solution, B would never adjust her output. The advantage to A occurs because A cannot consciously choose his type at t,. The possibility of being ‘irrational’ (i.e., sub-optimal) at t, is what opens the door (in some cases) to the possibility of greater profits for A of either type. Thus to maximize rr over two periods, you may need to force yourself, or leave yourself vulnerable to, maximizing u (# n-) in the second period. This result differs from earlier rational explanations or implications of sunk costs. Sunk costs as precommitment thus take on a very different, more behavioral, connotation here. It is legitimate to ask whether a manager’s sunk cost effect is a credible commitment for firm A to be of Type S. Why, for example, shouldn’t firm A fire a Type S manager once the deterrence effect is known to B? Two possible answers come to mind. If firm A establishes a reputation for firing Type S managers before t, (and always playing aA at t,), it will quickly lose the benefit of this deterrence effect in future similar encounters against other competitors (in a multiple-encounter world). Alternatively, as demonstrated in Kanodia et al. (1989), escalation of commitment can arise out of information asymmetries and agency concerns on the part of the manager. Firm A shareholders may therefore be unable to know the manager’s type before t, any better than the probabilistic knowledge which firm B possesses in a single-encounter world. In either case, sinking costs can constitute a credible commitment to Type S behavior in the output game at t,. Arkes and Blumer (1985) recognized that the sunk cost effect can be used to one’s advantage. They quote an argument made by Dowie (1981) in the context of nuclear energy. He suggested that if construction of power plants can be secretly initiated, sunk cost arguments can then be brought up when the public finds out, to argue that it is too late for construction to be stopped. While somewhat cynical, this example does point out the potential for Player A to impose a sunk cost either to himself or to Player B - as a strategic move that permits him to get his way. Self-imposing sunk costs demonstrates resolve for persisting with the project. Imposing sunk costs on someone else, which would be the case if public funds were used in construction of the power plants, transfers the sunk cost effect to others. l4 This recognizes sunk costs as an important strategic tool in delegation or principal/agent situations. This is an important problem, with significant potential for analysis. A more general formulation of sunk costs as precommitment is represented generically as a game in extensive form, in the game tree in Fig. 3. Player A must first decide how much of a cost s, if any, to sink into a project. Some uncertainty then gets resolved: If successful, Player A can then reap I4 Prior investments may in fact make persisting alternative at that point. with construction of the plants the most desirable R. Parayre / .I. of Economic Behmior monopoly profits 4” & Org. 28 (1995) 417-442 431 U-Cl A A A maximizes AA (u*, nBla;,b*) (x*. xBla;,ti) Fig. 3. Detailed game tree for sunk costs as precommianent. monopoly profits - having pre-empted Player B’s involvement in the project. If unsuccessful, Player A loses the initial allocation of s and then enters a simultaneous Nash game with Player B. It is the unsuccessful resolution of the uncertainty which drives the psychological motivations leading to a sunk cost effect for Player A - and the resulting change in his utility function. Player A, which may not have perfectly anticipated his reaction to sunk costs (hence his prior q* ), ends up in one of two utility states (i.e., with one of two utility functions) following these sunk costs. Player B does not know the utility state of Player A, as this is A’s privately held information. She only has a prior probability estimate, qB. A and B then proceed to play the Nash game of the second period along the lines indicated in the bottom half of the figure, yielding the equilibrium allocations. The payoffs at the bottom of the figure reflect A’s different utility functions. l5 432 R. Parayre / J. of Economic Behavior & Org. 28 (1995) 417-442 3.3. Price competition In the case of Bertrand (price) competition, usually associated with heterogeneous products, having sunk s (into plant, say) can motivate the manager to try and operate as close to full capacity as possible (so as not to admit that the plant may be oversized). So while the decision at t, is one of pricing, the sunk cost effect will reward increasing production, and hence the manager wants to decrease price. So when the decision variables a and b are prices rather than quantities, escalating commitment can therefore be expressed as u$ < 0. I6 Price competition generally produces upward sloping reaction functions. For upward sloping reaction functions resulting from price competition, we get “: db’ da’ ds < 0 and ds <o. A decrease in a* thus leads to a decrease in b ’ as we move down B’s upward sloping reaction function. ‘* Determining which player ends up better off because of A’s sunk cost effect, or as the sunk amount s increases, depends on the direction of increasing profits along the players’ reaction functions, which in turn depends on the elasticity of the demand curves. This requires additional structure to be imposed on the problem. Consider for example the following simple parametrization of a Bertrand duopoly, which has each firm’s demand linear in both firms’ prices: qa = y - 6p, + a(p, - p,) demand for As product qb = y - 6p, + o(p, - pb) demand for B’s product where pa, p,, are the prices for A and B’s products. In this case, Player A will r9 max uA = 7~~ + a(s)qa = q,p, + o(s)q, Pa = [Y - SP, + o(p, - P,)](P, + o(s)) (4s) 2 O;do/ds ’ 0) *’ A similar albeit more extensive version of this tree could be used to describe. escalation of commitment in the context of the ‘dollar auction’ game. I6 The case of price competition may be less compelling and possibly less prone to sunk cost behavior, as a pricing decision is not literally one of escalating commitment. ” Upward sloping reaction functions require u$ , ?r,f:> 0. The results are then derived in a manner similar to that described in section 3.1. ‘s Milgrom and Roberts (1990) suggested different economic examples of upward-sloping reaction functions. R&D races, oil drilling and even arms races display upward-sloping reaction functions. In these cases, where the decision variables are.quantities, the sunk cost effect is expressed as u$ > 0, and the comparative statics cannot be. signed. I9 Here. cr(s)q, leads to u$~< 0, capturing escalation of commitment in this Bertrand case. R. Parayre/ J. of EconomicBehavior& Org. 28 (1995) 417-442 433 and Player B will maxnB = qbpb = [Y - ‘Pb + +a Pb)]Pb- - Pb Taking first-order conditions and solving simultaneously, the Nash equilibrium prices are computed to be -2a.(s)s* + 2&y - 4a(s)6a P: = + 3ya - 2a(s)a2 46* + 86a + 3u2 pb’ = - -26y+a(s)6u-3yu+a(s)u* (26 + a)(2S + 30) yielding uA_ (8+u)]2u(s)6*+26y+4a(s)8u+3yu+a(s)u*]* (26 + a)*(26 + 3u)* 7FA= - (6 +a) (26 + a)*(26 + 3u)* . [j’ - k* + 3ja(s) u* + ka(s)u* + 2a2(s)u4] wherej=(2S2+4&r)a(s)>Oandk=y(26+3g)>0; yrB _ (6 + u)[ -280 + a(s)& - 3yu + a(s)u’]’ (26 + a)*(26 + 3u)* While u* is increasing in a(s), we find both yr* and 7rB to be decreasing in a(s). Therefore, in this simple Bertrand case, both A and B lower their prices and both end up worse off because of A’s sunk cost effect. If s has already been sunk (in anticipation of greater demand than actually obtained), A’s sunk cost effect makes him worse off. If s has not already been sunk, it is not to A’s advantage to intentionally sink s in order to signal that he may be of Type S. 3.4. Exploiting an opponent’s sunk cost effect 2o We now turn our attention to the problem of exploiting an opponent’s sunk costs. Under what conditions will one player be better off because of an opponent’s 2oThere is an ongoing debate concerning the (lack 00 dynamic consistency of non-expected utility agents, and how they can be exploited in economic decisions under uncertainty.For a thoroughreview of this debate, see Machina (1989). While not concerned with the same type of behavior, our present context of changing preferences following sunk costs raises some similar philosophical issues. 434 R. Parayre/ J. of Economic Behavior & Org. 28 (1995) 417-442 sunk cost effect? The impact of one player’s sunk cost effect on an opponent in general will depend on the particular economic problem being studied. It is natural that under the conventional single-product Coumot assumptions of section 3.1, B cannot profit from A’s increased commitment to the product. But by extending the situation to a Cournot duopoly with rwo distinct products at t , , there now is the possibility for B to exploit A’s increased commitment in product #I (and away from product #2) by focusing more of her investment on product #2. 21 The presence of a second product in which the duopolists also compete thus provides an opportunity for B to benefit from A’s sunk cost effect. We can work out an example of such multipoint Cournot competition. As before, Player A has sunk s into the development of product # 1 which, at t ), he will sell in a duopoly along with Player B. But there is also another product, whose demand is independent of the first, which A and B also sell in a duopoly. At t,, Player A has an amount A to invest and Player B has B. 22 Player A puts a into product #I and A - a into product #2; Player B puts b into product #I and B - b into product #2. The price of product # 1 is: p = y - S(a + b), (y, S > 0). The price of product #2 is: q = 0 - a[(A - a) + (B - b)], (0,~ > 0). Once again, no costs of production are assumed, so that the optimization problem is: Player A: max[y - 6(a + b>]a + [f3 - a(A + B - a - b)l(A - a) + o(s>a, B ((Y(s) 2 0) Player B:max[y - &(a + b)]b + [fl - a(A + B - a - b)l(B - b). b Derivation of the Nash equilibrium allocations and profits is once again obtained by solving the two players’ first-order conditions simultaneously. The resulting profits (at t, 1 are: +9B( 68 + yo - &[A + B])] *’ The issue of multipoint competition has been studied by Bulow et al. (1985) in their economic analysis of multimarket oligopolies. *’ This budget constraint on Player A is necessary to introduce a dependence between the allocations on the two projects. Without some form of constraint, Player A’s sunk cost effect on product #l would not affect his profits and optimal allocation on the independent product #2. The results of any sunk cost effect in this case would be driven purely by competition in market #I, replicating the results just derived in the single-product case. In the extreme, in the absence of such a budget constraint Player A could sink money into both projects, and benefit from his own sunk cost effect in both markets! But in many organizational contexts, it is quite reasonable to talk of a budget constraint, as divisions are often allocated fixed pots of money and divisional managers ate usually limited to those fixed amounts especially when the time periods (and the associated windows of opportunity for investment) am short, and managers have no time to raise additional funds. R. Parayre/ J. of EconomicBehavior & Org. 28 (1995) 417-442 435 a(s)<0 not allowed The shaded areas constitute the feasible regions for producing non-negative quantities. -3aA - There are 3 possible regions: Region 1 : n*? in a(s) and T? 1 in a(s) Region II : 24 in a(s) and n?J in a(s) Region III : K*J. in a(s) and # tin a(s) Fig. 4. Regions of exploitability for the deterministic two-product example. n*= g(s~u)[-2ui(s)cu(s)(Y-8)+(y-~)2 +9A(60 -I- ya - 6a[A + B])] Fig. 4 shows the different regions that make each player better off or worse off at t, because of player A’s sunk cost effect. In this deterministic 2-product case in which player A is known to both parties to fall prey to a sunk cost effect, the benefits of the effect can accrue either to A or to B (but never to both), or to neither player. 23 For example, the case in which y= 8, i.e., when the demand curves for both products have the same intercept, we are in region III of Fig. 4. Player B’s profit R. Parayre / J. of Economic Behauior & Org. 28 (1995) 417-442 436 increases in CX(S)while A’s decreases, so that B is made better off by Player A’s sunk cost effect. In this situation, Player A suffers from a first-mover disadvantage, which Player B can then exploit. So, unlike in our previous examples, we cannot draw unequivocal conclusions about the (dis-ladvantages of the sunk cost effect. Yet even in this simplest of two-product situations, with costless production and linear demand, the sunk cost effect may offer potential for exploitation by one of the two parties. Extending this game to a signalling one yields much the same conclusion as our single-product signalling example, derived in much the same way. Profits at t, in the two-product signalling game are: ~2(4 A==A_ TS N 4(S+a) A will now sink s * > 0 only when the parameters are in the region which makes both types N and S better off (Case 3 in section 3.21, and under those conditions will sink s = sS . For any other parameter set, A will sink s * = 0, and no sunk cost effect will take place. For example, if both products have identical demand curves (i.e., y = 0 and 6 = v ), the profits at t, are given by: mi=yA-G(A+B); + p202 (s) which is increasing in o(s) and in p 726 a,A=yA-G(A+B)$ + (P’ - 9)02(s) 726 which is decreasing in o(s) t/p. These equations show A of type S to be made worse off (rr,” is decreasing in (Y(S)Vcu(s)) because of his sunk cost effect - regardless of B’s belief p about A’s Type. Therefore, while A of Type N benefits financially from p > 0 and (Y(S)> 0, A of type S never has an incentive to sink any s at t, (or to ‘bum money’, as it were 24>,thus making the equilibrium fully revealing: A of type S always sinks s = 0, and the normal economic Nash equilibrium holds at t,. Any s > 0 tells B that A is a Type N hoping to pass off as Type S (i.e., trying to misrepresent his type), which has for effect to push p to 0, again resulting in the economic Nash equilibrium at t,. Hence, A (regardless of type) can never be made better off by his sunk costs. Any s > 0 (including s sunk inadvertently) will be lost, and any A of type S will only be made worse off financially by his sunk cost effect. 23 This graph is based on S > (r and A > B. The three regions look slightly different but the conclusions remain unchanged when 6 < u and/or A < B. 24 Van Damme (1989) addresses the strategic role of ‘burning money’ as a means of selecting an equilibrium via forward induction - a problem quite different from the one addressed here. R. Parayre / .I. of Economic Behavior& Org. 28 (1995) 417-442 437 3.5. Welfare implications If A and B were playing a cooperative game, the first-best outcome for them would obviously be to operate jointly as a monopoly with some sort of sharing rule. But this solution is ruled out in a non-cooperative environment, where the duopoly solution prevails. Yet given the possibility for one player to be made better off by a sunk cost effect, we can ask whether joint welfare can be improved if one party has a sunk cost predilection. In the single-product signalling game in section 3.2, joint welfare can indeed be greater than in the purely ‘rational’ duopoly setting. When B adjusts her output to account for the possibility of A’s sunk cost effect and A turns out being of the ‘economic’ Type N, joint profits r* + 7~~ are 2y2/96 + p&)(27 p&1)/36& which exceed the total duopoly profits under no sunk cost effect of 2 y2/96. 25 If A is the ‘escalating’ Type S, joint profits are 2y2/96 + [(2p 6)ya(s) + (6~ - p2 - 9>(r 2(s>]/36S, which is always less than in the purely rational duopoly case. A similar conclusion arises in the situation in which both A and B can display a sunk cost effect. In this case, where both players sink costs in the first time period and Player B also has some chance of being of the ‘escalating’ type, there is an economic pressure for each player to decrease output (to adjust for a possible escalating opponent), and a psychological pressure to increase output if one is of the escalating type. This situation now contains four different reaction functions: for A,, As, B, and Bs, with the resulting equilibria (ai,bG), (ai,b,‘), <ai,bi), (ai ,bG 1. See Fig. 5. Analytical results can be derived in a manner which closely parallels section 3.2. Assume that both players share an equal probability p of being of Type S and to display a sunk cost effect of magnitude (Y(S).Under these conditions, (ah,bA) is located southwest of the purely economic duopoly solution located at the intersection of A, and B, (point e in Fig. 51, and has greater joint welfare than point e if pa(s) < y/2 (i.e., if there is a small chance of a player being Type S and/or the sunk cost effect is weak). Otherwise joint welfare will be less than at point e. Note that A of Type N now has to decrease output to account for the possibility of B being of Type S. Thus the threat of a Type S opponent in effect disciplines players into reducing their output levels. Under the right conditions, 26 this brings them closer to the first best solution of joint monopoly output, making them both better off than if they both knew each other to be ‘perfectly’ rational! The other three equilibrium allocations (ai ,bl 1, (ai ,bi > and (al ,bi > can be shown to result in ” The development in section 3.2 required that y > PLY(S)as a non-negativity condition, leading to this general conclusion. 26 If an opponent’s expected sunk cost effect is strong, players may reduce their output levels below the joint monopoly output, and may actually be worse off than at the economic duopoply level. 438 R. Parayre/ J. of Economic Behavior & Org. 28 (1995) 417-442 a Fig. 5. Cournot signalling game with two players of unknown type. welfare losses when compared to point e, as they all entail greater joint production, and thus produce smaller joint profits. In the two-product case of section 3.4, joint welfare increases when (Y(S)< -(ye> -- the re gion to the southwest of the line of slope - 1 bisecting the bottom-right quadrant in Fig. 4. Anything to the north of that line results in reduced joint welfare. joint 3.6. The strategic manipulation of an opponent’s preferences and other types of commitment situations In the preceding examples, any advantages to Player B are purely by-products of A’s sunk cost effect, as B has no control over the amount s sunk by A at t,. 27 But A’s sunk cost effect can present yet another set of opportunities to Player B. By moving first, B may be able to strategically manipulate A’s preferences to her own advantage by actively influencing A’s choice of a sunk amount s. This will allow B to actively encourage and exploit Player A’s sunk cost effect. Examples of such strategic manipulations abound. Just as an economic (sunk cost) commitment to a project results in greater psychological commitment to it (in 27 unless B too can mis-represent her true preferences at t, - a situation which we do not address here. R. Parayre/ J. of EconomicBehauior& Org. 28 (I 995) 417-442 439 effect, increasing the utility of a positive outcome from that project), psychological commitments will often have the same effect. In essence, past commitments, be they economic or psychological, often increase future willingness-to-pay. 28 Consuming heroin increases an addict’s need (and utility) for future consumption of the drug, hence increasing his willingness-to-pay. Car dealers low-ball (Cialdini et al., 1978) customers by having them agree to a base purchase price on a car, only to come back with a higher price (or with high-priced options) after ‘checking with the manager’ who would not agree to the initial low price. Charitable organizations have you commit to giving a small amount, and then escalate their request, with great success - a strategy known as the foot-in-the-door technique (Freedman and Fraser, 1966). Stores offer huge discounts on some products to catalyze high-priced tie-in sales - or loss-leader pricing. Or they advertise a ‘super-special while quantities last’ to bring customers to the store, and quickly run out in order to sell the product or a substitute at much higher prices a bait and switch strategy. All of these situations stress the influence of psychological commitment on increasing consumers’ willingness-to-pay. In each of these cases, Player B moves first and tries to have A commit psychologically to a product or idea. If successful, this strategy can increase A’s willingness-to-pay. Each of these situations can be modeled as a three-period game, where B moves first, tries to induce commitment from A, and follows up that commitment with a higher-priced offer. Each situation offers the potential for true strategic preference manipulation on the part of an opponent. These games are a topic of ongoing work. 4. Conclusion This paper has addressed a problem which, despite its economic importance, has been largely ignored in economics and strategy. The sunk cost phenomenon the tendency to overinvest in previously committed projects - has long been one of the most persistent problems in decision making. The frequent warning that ‘sunk costs shouldn’t matter’ in determining future allocations, made in just about every managerial accounting or finance textbook, is an acknowledgment of the pervasive nature of this phenomenon. Yet there have been few attempts made at systematically modeling its causes or its implications. This paper has taken the first steps toward a formal, systematic analysis of the implications of the sunk cost phenomenon. We have modeled some strategic effects of the sunk cost phenomenon in order to make behavioral predictions and provide prescriptive advice. We have demonstrated that even in the simplest of *s Economic models of addiction have been proposed by Becker and Murphy (1988), and discussed in Loewenstein aad Elster (1992). 440 R. Parayre / J. of Economic Behavior & Org. 28 (1995) 417-442 competitive settings, such as two-period duopolies, a sunk cost effect could lead to some interesting and somewhat counter-intuitive insights. Enrichments such as additional dynamics, more uncertainty, and population ecology considerations could be introduced. These added degrees of freedom would increase the range of potential strategic implications of sunk cost behavior. These can be explored in future research. Market arguments could be used to contend that the presence of strategic rivals would impose economic rationality on otherwise escalation-prone agents - ‘irrational’ firms being driven out of the market, thus constraining sunk cost behavior to very isolated cases. It is indeed legitimate to ask whether managers engaged in intensely competitive environments display reduced, the same or greater sunk cost tendencies as do individuals facing personal decisions, or laboratory subjects responding to hypothetical questions (as in Arkes and Blumer, 1985). Some evidence (Brockner and Rubin, 1985) suggests that competitive situations actually aggravate the sunk cost tendency - the concept of ‘winning at all cost’ overriding economic rationality. When combined with the managerial reputation and self-presentation motives that exist in organizational settings, the potential for sunk cost behavior among competing firms is great indeed. Teger (1980) has presented empirical results on the ‘dollar auction’, and has pointed out the importance of sunk cost considerations in the escalation of warfare. The sunk cost phenomenon may even have value in single-agent situations. For example, the knowledge of one’s own escalation tendencies forces one to consider the initial investment decision very seriously. Prelec and Herrnstein (1991) also argue that adherence to specific decision ‘principles’ is valuable because of the inherent difficulty in continually making cost-benefit calculations. The vow of marriage, for exampIe, can be viewed as a public declaration of adherence to the sunk cost fallacy, which eliminates the need for future cost-benefit analyses. The modeling approach in this paper can extend beyond the sunk cost problem to analyze other behavioral departures from conventional economic rationality, in problems located at the interface between psychology and economics. In particular, the issue of preference change is a very rich, yet largely unexplored topic in economics. (For a notable exception, see Loewenstein and Elster, 1992.) In this paper, it was assumed that one could perfectly anticipate one’s own future change in preferences that follows sunk costs, and by so doing could mitigate the negative implications of the sunk cost effect (or even extract positive ones) by adjusting one’s sunk costs at t,. Yet evidence from social psychology (Nisbett and Wilson, 1977) suggests that people will likely be poor predictors of their own future behaviors. What’s more, as discussed by Akerlof (1991) in a paper on procrastination and obedience, many situations involve changes in preferences that were not foreseen as of t,, resulting in pathological or even destructive outcomes. This is an important problem, which could be addressed within our present framework. Fundamental questions about the pervasiveness and importance of sunk cost behavior in competitive settings also remain open. Is an individual’s sunk cost R. Parayre /J. of Economic Behavior & Org. 28 (1995) 417-442 441 behavior a generic tendency or is it strongly situation-specific? Do managers, in practice, exploit (expectations of) sunk cost behavior to increase profits? And what is the impact of competitors and of the intensity of competition on sunk cost behavior? These are all important questions which should be subject to empirical investigation. References Akerlof, George A., 199 I, Procrastination and obedience, Richard T. Ely Lecture, American Economic Review 81 (2), I-19. Arkes, Hal R., and Catherine Blumer, 1985, The psychology of sunk cost, Organizational Behavior and Human Decision Processes 35, 124-140. Becker, Gary, and Kevin Murphy, 1988, A theory of rational addiction, Journal of Political Economy 96,675-700. Brockner, Joel, and Jeffrey 2. Rubin, 1985, Entrapment in escalating conflicts, New York, SpringerVerlag. Bulow, Jeremy I., John D. Geanakoplos and Paul D. Klempercr, 1985, Multimarket oligopoly: Strategic substitutes and complements, Journal of Political Economy 93, 488-511. Cialdini, Robett B., Rodney Bassett, John T. Cacioppo, and John A. Miller, 1978, Low-ball procedure for producing compliance: Commitment then cost, Journal of Personality and Social Psychology 36, 463-476. Dixit, Avinash, 1979, A model of duopoly suggesting a theory of entry barriers, Bell Journal of Economics 10, 20-32. Dixit, Avinash, 1980, The role of investment in entry deterrence, Economic Journal 90, 95-106. Dowie, M., 1981, Atomic psyche-out, Mother Jones 6, 21-23, 47-55. Elster, Jon, 1983, Sour grapes: Studies in the subversion of rationality, New York, Cambridge University Press. Frank, Robert H., 1988, Passions within reason, New York, Norton. Freedman, Jonathan L., and Scott C. Fraser, 1966, Compliance without pressure: The foot-in-the-door technique, Journal of Personality and Social Psychology 4, 195-202. Ghemawat, Pankaj, 1991, Commitment: The dynamic of strategy, New York, Free Press. Harsanyi, John C., 1967-68, Games with incomplete information played by ‘Bayesian’ players, Management Science 14, 159-182, 320-334, 486-502. Kanodia, Chandra, Robert Bushman, and John Dickhaut, 1989, Escalation errors and the sunk cost effect: An explanation based on reputation and information asymmetries, Journal of Accounting Research 27, 59-77. Knox, Robert E., and James A. lnkster, 1968, Post-decision dissonance at post time, Journal of Personality and Social Psychology 8, 319-323. Loewenstein, George, and Jon Elster, 1992, Choice over time, New York, Russell Sage Foundation. Machina, Mark J., 1989, Dynamic consistency and non-expected utility models of choice under uncertainty, Journal of Economic Literature 27, 1622-1668. Milgrom, Paul, and John Roberts, 1990, Rationalizability, learning and equilibrium in games with strategic complementarities, Econometrica 58, 1255- 1277. Nisbett, Richard E., and Timothy D. Wilson, 1977, Telling mote than we can know: Verbal reports on mental processes, Psychological Review 84, 23 l-259. Parayre, Roth, 1993, Escalating commitment and the market impact of discontinuation decisions - An empirical study, SMU Cox School of Business working paper. 442 R. Parayre / J. of Economic Behavior & Org. 28 (1995) 417-442 Prelec, Dmzen, and R.J. Herrnstein, 1991, Preferences or principles: Alternative guidelines for choice, in Richard J. Zeckhauser (ed.), Strategy and choice, Cambridge, MA, MIT Press. Schelling, Thomas C., 1960, The strategy of conflict, Cambridge, MA, Harvard University Press. Spence, A. Michael, 1977, Entry, capacity, investment and oligopolistic pricing, Bell Journal of Economics 8, 534-544. Spence, A. Michael, 1979, Investment strategy and growth in a new market, Bell Journal of Economics IO, I-19. Statman, Meir, and James F. Sepe, 1989, Project termination announcements and the market value of the firm, Financial Management Winter 1989, 74-8 1. Staw, Barry M., and Jerry Ross, 1987a. Knowing when to pull the plug, Harvard Business Review 65, (21, 68-74. Staw, Barry M., and Jerry Ross, 1987b, Behavior in escalation situations: Antecedents, prototypes, and solutions, In L.L. Cummings and Barry M. Staw @Is.), Research in organizational behavior, vol. 9. Greenwich, CT, JAI Press. Teger, Allan I., 1980, Too much invested to quit, New York, Pergamon Press. Van Damme, Eric, 1989, Stable equilibria and forward induction, Journal of Economic Theory 48, 476-496. Williamson, Oliver E., 1963, A model of rational managerial behavior, in Richard M. Cyert and James G. March, A Behavioral theory of the firm, Englewood-Cliffs, NJ, Prentice-Hall, Ch.9, 237-252.