The author would like to thank the participants in the Law and Economics workshop at the University of Chicago Law School for helpful comments. He gratefully acknowledges funding from the John M. Olin foundation and the Center for the Study of the Economy and the State, Graduate School of Business, University of Chicago.

¹ Restatement (Second) of Contracts [[section]] 344(a),(b) (1981). The reliance rule is so called because the seller is compensated for costs, such as the cost of producing customized goods which cannot be resold, incurred because he was relying on the buyer to accept what he had ordered.

²Thus, for instance, the third sentence of Goetz and Scott, Measuring Sellers' Damages: The Lost-Profits Puzzle, 31 Stanford Law Review (1979) reads "The answer seems simple: The seller should be awarded damages sufficient to place it in the same economic position it would have enjoyed had the buyer performed the contract." The most careful and detailed analysis of the case for expectation damages that I have seen is Steven Shavell, Damage Measures for Breach of Contract, 11 Bell J. Econ. 466 (1980).

³For instance, Goetz & Scott, supra note 2.

⁴This is one of the central issues of Shavell, supra note 2. Consider a situation where the seller is choosing between two alternative production technologies, one of which results in lower production costs than the other if the goods are produced but higher reliance costs if the order is cancelled. If the seller knows that, in case of breach, he will be compensated for any reliance costs, he will choose his optimal technology as if the probability of breach were zero; the result is an inefficiently high level of reliance.

⁵The first reference to this point that I have come across is in Fuller and Perdue, The Reliance Interest in Contract Damages:1, 46 Yale Law Journal 52 (1936). See also Goetz & Scott, supra note 2. Cooter & Eisenberg, in Damages for Breach of Contract, 73 California Law Review 1434 (1985) point out that the argument that there are no profits applies only to a marginal sale (p. 1449). They argue that expectation and reliance damages become equal (under perfect competition) only as the probability of breach approaches zero. Their argument is based not on the fact that in competitive equilibrium there are no profits to be lost but on the fact that in competitive equilibrium the demand curve faced by a firm is perfectly elastic, if the firm did not make one sale (the one that is going to be breached) it could have made another. The argument may be summarized as follows:

Seller contracts to produce a good at a price P. He produces it, but buyer refuses to take delivery. In a competitive market, if seller had not made that contract he would have made an identical contract instead. That contract would have been fulfilled with a probability 1-p, where p is the probability of breach. As p approaches zero, the result of reliance damages (make seller as well off as if he had made another contract with probability 1-p of fulfillment) approach those of expectation damages (make seller as well off as if he had made this contract and it had been fulfilled). As long as p>0, reliance damages are less than expectation damages.

The mistake in this is that it ignores the damage payment that seller would receive if he made another contract and it was breached. Let D be the damages for breach of contract, whether for the actual contract or the hypothetical replacement contract, and [[pi]] the net gain to a firm of making a contract and having it fulfilled. If the firm had made another contract, it would have had a probability 1-p of fulfillment with gain [[pi]] and a probability p of breach, with gain D, so reliance damages require that:

D=(1-p)[[pi]] +pD

\ D-pD=(1-p)D= (1-p)[[pi]]

\ D=[[pi]]

So the seller is entitled to recover his lost profits, just as with the expectation rule, even if p!= 0.

⁶This paper is concerned with damage rules in situations where reliance and expectation lead to different results. The distinction between reliance and expectation rules might also be of interest in situations where the two rules, properly applied, lead to the same result, but one is easier to apply than the other. It might, for instance, be easier to measure the price of a good and the cost of completing it and calculate expectation damages as the difference between the two than to calculate the cost of reliance.

⁷This argument appears in verbal form in Cooter & Eisenberg supra note 5, and more formally in Shavell supra note 2. Its first statement may be in John H. Barton, The Economic Basis of Damages for Breach of Contract, 1 J. Legal Studies 277 (1972).

⁸I have been able to find only two articles in which the effect of the damage rule on the formation of the initial contract plays an important role. One is Diamond & Masking, An equilibrium analysis of Search and Breach of Contract, I: Steady States, 10 Bell Journal of Economics 282 (1979). It analyses a very different problem from the one discussed here. Potential contract partners are involved in a stochastic search process; breach occurs when a searcher finds someone with whom he can form a better contract than the one he already has. The other is Barton supra note 7. On pages 296-299 he analyzes, if I correctly understand him, a situation which combines bilateral monopoly and asymmetric information. He does not assume rational expectations--the worse informed party appears to act on an assumption which is wrong on average as well as in the particular case.

⁹An example of a real firm which comes fairly close to fitting the description of this section would be an elite university such as Harvard or MIT. It has a considerable degree of monopoly since, as any (Harvard) man will gladly explain, there is no substitute for a Harvard education. Many students drop out before receiving their degrees for personal reasons which, from the University's standpoint, may be viewed as random events. Unlike the firms analyzed here, however, universities engage in extensive discriminatory pricing. There is also some question as to whether they behave like profit maximizing firms.

¹⁰More generally, R represents whatever costs the producer bears as a result of a consumer ordering and then cancelling. These might include modifying a customized unit to resell, finding a customer for a unit that was not customized, or simply adjusting to the additional uncertainty due to one consumer being willing to breach one more unit if the event occurs. If the good is not customized and the number of customers is large, this last case should involve a very small R, due to the law of large numbers.

¹¹Here and elsewhere, benefits and costs are expected values, since both consumers and producers are assumed risk neutral.

¹²Provided that the marginal value curve is monotonic down, as we have assumed. When I say that a demand curve equals a marginal value curve, I mean that they are the same line. Considered as functions, D shows quantity as a function of price, MV shows marginal value as a function of quantity.

¹³D* is the (average) demand curve for one customer; for N customers the demand curve should be D* multiplied N times in the horizontal direction. Profit is then also multiplied by N. Similarly, consumer surplus should be multiplied by N, since there are N identical consumers. Conclusions about what rule maximizes the sum of profit and consumer surplus are unaffected, so we can ignore these points and proceed as if N were equal to one with no loss of generality.

¹⁴This is not quite correct; since a marginal value curve only exists for non-negative values of Q, it cannot be a straight line from -* to +*. For sufficiently high values of P, we have a horizontal average of the high demand curve and zero, which is not equal to a vertical average of the high and low demand curves. The implications of this are worked out in the appendix.

¹⁵This assumes the transformed problem where R=0. The condition for efficient division between the two situations is more complicated for R>0, but the reliance rule still satisfies it--as it must, given that any problem with R>0 has an equivalent problem with R=0, as shown above.

¹⁶A. C. Pigou, The Economics of Welfare, part II, Chapter XVII (fourth ed. 1950). First degree price discrimination is perfect price discrimination; every unit of the good is sold at the highest price the consumer of that unit will pay, leaving no consumer surplus. Under second degree price discrimination, the monopolist sets n different prices and arranges things so that every unit is sold for the highest of those prices at which the consumer is willing to buy it. Under third order price discrimination, the monopolist separates his customers into a number of different groups and charges a different price to the members of each different group.

¹⁷Here again, the discussion applies directly to the transformed problem with R=0, and must be modified to apply to the case R>0.

¹⁸Shavell supra note 2, at 482, Proposition 5. This is demonstrated in Shavell's "First Case"--with one party deciding on reliance and the other on breach. As the next quote shows, he reaches an even stronger result when the same party decides both. In his concluding remarks he qualifies the conclusion somewhat in cases involving risk aversion, and mentions the possibility that the court may lack the information necessary to apply one or the other damage measures, but neither of these points has anything to do with the central argument of this paper.

¹⁹Shavell supra note 2, at 485.

²⁰Shavell supra note 2, at 469, fn 13. Similarly, in S. Shavell, The Design of Contracts and Remedies for Breach, 99 QJE 123 (1984), Shavell writes "The parties are assumed already to have met ... they will make a contract themselves if doing so would result in a higher expected utility for each than not making any contract, and this will generally be presumed to be the case."

²¹This is also true of the quite different analysis of the problem by William P. Rogerson in his article Efficient Reliance and Damage Measures for Breach of Contract, 15 Rand Journal of Economics 39 (1984). He eliminates the problem of inefficient breach by assuming that it will always be solved by ex post negotiations between the two parties. His analysis hinges on the effect that the expectation of such renegotiation has on the level of reliance chosen by one party; he concludes, like Shavell, that expectation damages are superior to reliance damages. Also like Shavell, he ignores in his analysis any effect that the damage rule will have on whether the initial contract occurs and at what price.

²²The proof is contained in a longer version of this article available from the author. By describing an outcome as efficient, I mean that no better outcome can be produced given the information available when the relevant decisions are made. I am still assuming a fixed level of reliance; with variable reliance the outcome may be inefficient, for reasons explored in Shavell supra note 2.

²³Here again, a university is a good real-world example; both a student and his parents are likely to have some private information about the probability that he will drop out.

²⁴It might seem more natural to consider situations in which the buyer's special knowledge concerned probability of breach rather than cost of production. But if cost of production is known ex ante then there are no profits ex post, so reliance and expectation rules imply the same damages. With uncertain costs but symmetrical knowledge about them, the expectation rule is superior even if the buyer has special information about the probability of breach, since under the expectation rule the seller does not care whether or not the buyer breaches and is therefore uninterested in the probability of breach.

²⁵This is the same problem known as adverse selection in the context of insurance markets. In that case the problem is made worse by the fact that the good is more valuable to high cost (i.e. high risk) customers and less valuable to low cost customers, with the result that if the seller charges the average cost high cost customers buy more than low cost customers, raising the average cost.

²⁶This case is discussed briefly in Appendix II. Under single price monopoly, all-or-nothing breach results in identical outcomes under both damage rules.

²⁷Here and in Appendix II, although I am analyzing the transformed problem (reliance=0), I drop the asterisk and write MV1,2, instead of MV*1,2 for purposes of simplicity.

²⁸ The analysis does not apply to MC>A-pE, but the conclusion does. Expectation leads to no output, no consumption, no profit and no surplus, so reliance, which leads to positive output, is superior.

²⁹ Negative damages could occur through bargaining. The buyer who wants to breach realizes that doing so will benefit the seller, so he threatens to take delivery unless paid not to. That situation would not arise if expectation damages applied in both directions; the seller could initiate breach and pay damages of zero, since the buyer would not be injured. But it might arise if that solution was blocked by a contract specifying liquidated damages or by a performance rule applied to the seller, or if the court was unable to observe the event that made the buyer wish to breach. In such situations one might observe an extreme version of the problem of inefficient ordering under asymmetric information, with the high cost buyer ordering units he knows he will not want in order to be bribed not to take delivery.