Sequential Common Agency: The Revelation Principle Anastasia Kartasheva (cid:3) July 18, 2006 Abstract The paperextends the Revelation Principletosequentialcommon agencygames under asymmetric information. Each period a principal contracts with a common agent. An implemented allocation is observed by other principals. Depending on whether the message reported by the agent to a principal is observed by other principals, we distinguish between private and public communication. Under pri- vate communication, the Revelation Principle applies, but optimal contracts are stochastic. However, the dimension of the support of an equilibrium contract does not exceed the number of types that achieve this stage with a positive probability. Under public communication, the reporting strategy of agent is stochastic, but the true type is reported with a positive probability. We demonstrate that the two regimes are equivalent in that they result in the same distribution of allocations. The results hold when the agent(cid:146)s type is not persistent, or the outcome of the contract is observed with noise. Keywords: common agency, sequential mechanisms, dymanic contracts, adverse selection. Revelation Principle. JEL Codes: C73, D82. 1 Introduction InthispaperwederivetheRevelationPrincipleforsequential commonagencygames. We analyze a situation in which a number of principals contract sequentially with an agent ThispaperisadaptedfromChapter1ofmydoctoraldissertationattheUniversityofToulouse. Iam (cid:3) indebted to Jean-Jacques La⁄ont and Patrick Rey for their encouragement and supervision of my work. I am grateful to Bruno Jullien, David Martimort, Ilya Segal and Jean Tirole, and seminar participants in Toulouse, ECARES, Brussels, EEA Meeting 2005 for helpful comments. Correspondence address: Department of Risk Management and Insurance, J.Mack Robinson College of Business, Georgia State University, P.O. Box 4036, Atlanta, GA 30302 - 4036, USA. Email: [email protected]. Home page: http://www.rmi.gsu.edu/rmi/faculty/kartasheva.htm 1 under asymmetric information about agent(cid:146)s preferences. At each stage one principal o⁄ers a contract to the agent. The implementation of the contract results in allocation which is observed by the other principals. An allocation is payo⁄ relevant for all the principals and the agent, and may a⁄ect the feasible sets of allocations of subsequent principals. The framework we study applies to many economic environments. Examples include: (i) non-exclusive credit, where information sharing between creditors a⁄ects the amount of credit and the probability of debt repayment (Padilla and Pagano (1997), Pagano and Jappelli (1993), Sharpe (1990)); (ii) interaction of public and private health insurance programs, which a⁄ects the choice of the insured and the terms of contract of the private insurer (Culter and Gruber (1996)); (iii) retail industries such as supermarkets, airlines, credit cards, where the information about the purchase history of customers allows sellers to o⁄er personalized deals (Acquisti and Varian (2002), Chen and Zhang (2001), Taylor (2002), Villas-Boas (1999)); and (iv) certi(cid:133)cation intermediaries, where information dis- closure by an intermediary a⁄ects the size and the distribution of the surplus between the buyer and the seller (Lizzeri (1999), Peyrache and Quesada (2004)), (v) interaction between (cid:133)rm(cid:146)s (cid:133)nancing and production decisions, where the choice of (cid:133)nancial structure a⁄ects (cid:133)rm(cid:146)s position vis-a-vis its competitors (Gertner, Gibbons and Scharfstein (1988), Bhattacharya and Ritter (1983)). Althoughtheaboveapplicationsprovideusefulinsights, inmostofthemeitherthereis nostrategicroleforprincipals,duetocompetitionassumption,forexample,ortheanalysis is restricted to very particular institutional arrangements such as linear contracts. As a result, the predictions are very sensitive to assumptions of a particular model. One possible reason for the lack of a uni(cid:133)ed, general approach is that the Revelation Principle1 widely used to study contractual relationships under asymmetric information, may not be valid in the environments with more than one principal. The standard Rev- elation Principle states that when a principal contracts with agents under asymmetric information about agents(cid:146)preferences, any contract can be described by a direct incentive compatible mechanism in which the terms of the contracts are based on the agent(cid:146)s report on its private information, and the agent has incentives to report the information thruth- 1The revelation principle for the environment in which a principal contracts with agents who have private information has been developed by Gibbard (1973), Green and La⁄ont (1977), Dasgupta, Ham- mond and Maskin (1979), and Myerson (1979). Myerson (1982) extends the revelation principle to the situationswhentheprincipalalsofacesmoralhazardproblems,inadditiontoasymmetriesofinformation. 2 fully. The practical application of this result is that the optimal contract can be found by the means of optimal programming subject to incentive compatibility constraints. Extending the Revelation Principle to games with many principals can be problem- atic. The literature started by analyzing common agency games in which the principals o⁄er contracts simultaneously to the agent(s). Epstein and Peters (1999) characterize the universal message space that can be used to characterize any indirect mechanism. How- ever, this message space may not be practical. The reason is that in such situation each principal would like to make its contract dependent on the contracts o⁄ered by the other principals, leading to the problem of in(cid:133)nite regress. As a result the simplest message space needed to describe the mechanisms must be rather rich, and is hard to work with. When principals o⁄er contracts in a sequential manner, the problem of in(cid:133)nite regress does not arise: Once a principal and an agent implement a contract, the principal cannot improve her payo⁄by requesting information about subsequent o⁄ers. Thispaperisrelatedtotwootherrecentpapersintheliterature. Whentheoutcomeof contractingbetweenaprincipalandanagentisnotobservedbytheotherprincipals,Pavan and Calzolari (2006) show that the equilibria can be described within the message space thatincludesagent(cid:146)stypesandtheallocationsimplementedwiththeprecedingprincipals. On the contrary, we consider sequential common agency games with public contracts in whichthecontractandtheoutcomeofthecontract, thatis, theallocationimplementedat each stage, are observable by the other principals. This framework is a better description of economic environments mentioned above. Also it leads to a di⁄erent type of externality that a contract between a principal and an agent exerts on the other principals: The implemented allocation itself becomes a signal about agent(cid:146)s private information. In this respect the paper is closely related to work on dynamic principal - agent relationships under imperfect commitment of Bester and Strausz (2001). We show that techniques developed in Bester and Strausz can be applied to study a dynamic contracting problem with many principals, or any combination of single principal - agent relationships under imperfect commitment and multiprincipal relationships. In this respect we provide a generalization of Bester and Strausz result. We study general communication mechanisms in which, without loss of generality, a contract is composed of a message space and a decision rule. The agent sends a message from the speci(cid:133)ed message set and, based on the message, a principal commits to a (possibly stochastic) contract that implements a feasible allocation. All principals are free to choose the message spaces and decision rules, which may in particular be state 3 dependant. In general there are three elements that can signal agent(cid:146)s private information to subsequent principals: the mechanism, its outcome, and the message reported by the agent. Thus we distinguish between the cases of public and private communication. Under private communication, the message reported by the agent to one principal is not observedbytheotherprincipals. Underpubliccommunication, theprincipalsalsoobserve the information submitted by the agent to preceding principals. The main result of the paper is that the set of equilibrium allocations of the sequential common agency game can be characterized within the type space. Naturally, the agent(cid:146)s incentives to report private information to a principal depend crucially on whether this report is observed by the other principals. Underprivatecommunication, theagentisnotconcernedthathisreporttooneprinci- pal may a⁄ect the contracting choices with the other principals. As a result, the standard version of the Revelation Principle applies at each stage game. At the same time, de- terministic contracts are suboptimal. Indeed, assigning a distinct allocation to each type implies that the outcome of the contract is a perfect signal about agent(cid:146)s type to the other principals. By o⁄ering a lottery a principal can limit the information about the agent(cid:146)s type revealed by the outcome of the contract. Thus, an optimal contract of a principal is a menu of lotteries designed for each type of agent. However, this result does not imply anyrestrictions onthe structure of the lottery. Next, we studythe structure of anoptimal lottery and show that the dimension of its support does not exceed the number of types that reach a given stage with a positive probability. Therefore, an optimal contract can be characterized as a menu of lotteries over a (cid:133)nite support. Consequently, an optimal contract can be found as a solution to an optimization problem. Under public communication, the revelation of private information by the agent may be costly for both the principal and the agent. The reason is that this information can be used by other principals in the subsequent stages. A similar problem arises when a single principal contracts with an agent over a number of periods. As new information about the agent becomes available during the relationship, the principal and the agent may (cid:133)nd it mutually bene(cid:133)cial to renegotiate the initial long term contract. However, anticipating the renegotiation of the initial contract, the agent may become less prompt to reveal its private information. This ultimately increases the cost for the principal of inducing a truthful report. As a result, the principal may prefer to decrease the informativeness of the agent(cid:146)s report about its private information. Bester and Strausz (2001) analyze this 4 situation and establish that the equilibria of the game can be characterized using direct mechanisms in which it is an optimal strategy for the agent to report its type. In contrast with the standard Revelation Principal, however, the agent does not necessarily reveal its private information with probability one: It may be bene(cid:133)cial for both parties that the agent randomizes and misreports its information with some probability. In this paper we show that the technique developed by Bester and Strausz (2001) can be extended to sequential common agency games with public communication. This is because at each contracting stage, the principal, be it the same or a di⁄erent one, with possibly a di⁄erent objective at each state, is constrained to o⁄er allocations that belong to the Perfect Bayesian equilibria of the continuation game. The main di⁄erence between our framework and that of imperfect commitment is that a principal does not internalize the impact of its contract on subsequent stages, and out-of-equilibrium messages may be needed to preserve the equilibrium outcome. However, we show that these messages can be preserved by the means of a direct mechanism. We compare the set of equilibria under private and public communication. We show that the two sets are equivalent in a sense that for each type of agent they induce the same probability distribution of allocations. The main idea of this result is that a prin- cipal can generate the same belief about agent(cid:146)s type for subsequent principals either by designing a lottery on the set of the messages, as under public communication, or the set of implemented allocation, as under private communication. The characterization results of the paper also apply to situations when the type of agent is not persistent over time, and when the messages or allocations are observed with some exogenous noise. In these case this information must be incorporated in de(cid:133)nition of the Bayes rule, but the equilibria can still be studied within the type space. OurresultisweakerthantheRevelationPrincipleofEpsteinandPetersasitholdsonly for equilibrium mechanisms while the later paper constructs a language to characterize all the mechanisms of the game. However, we believe that it provides a useful tool to study many applications. A useful practical feature of our result is that, like in single principal- agent mechanism design problems under asymmetric information, it allows to state the sequential contracting problem as a sequence of programming problems in each of which a principal maximizes its expected payo⁄under incentive compatibility constraints. In the next section we present an example that illustrates the main features of the result. In section 3 we set up the model of the sequential common agency. Then in sections 5 and 4 we establish the Revelation Principle for the case of private and public 5 communication. Section 8 concludes. 2 Example In this section we present a simple example of sequential contracting that illustrates the issues addressed in the paper. Two manufacturers, P and P , contract sequentially with 1 2 a common supplier A for provision of an essential input. The manufacturers produce two goods that they sell on di⁄erent downstream markets. The contracting game lasts for two periods. Each period i = 1;2 P contracts with A for the provision of a quantity of input i q 0 at price t that allows him to produce at most q units of a (cid:133)nal good. Denote i i i (cid:21) C = (q ;t ) the contract between the P and A. The outcome of the contract between P i i i i 1 and A is observed by P who makes an o⁄er in the beginning of second stage. The inverse 2 demand function in P (cid:146)s downstream market is P(q ) = 1 q : i i i (cid:0) The supplier produces an input at constant marginal cost (cid:18) which is her private infor- mation. It may be a low cost (cid:18) with probability (cid:23) or a high cost (cid:18) with probability 1 (cid:23), (cid:0) where 0 < (cid:23) < 1 and (cid:1)(cid:18) (cid:18) (cid:18) > 0. (cid:17) (cid:0) The pro(cid:133)t made with P equals u = t (cid:18)q , so A(cid:146)s total pro(cid:133)t from serving the two i i i i (cid:0) manufactures is u = u +u . The pro(cid:133)t of each manufacturer is v = (1 q )q t . 1 2 i i i i (cid:0) (cid:0) In the absence of information asymmetries about the cost the contracting decisions of manufacturers are independent. Each P captures all the joint surplus with A by o⁄ering i a contract t = (cid:18)q and producing an e¢ cient quantity i i q ((cid:18)) = 1(1 (cid:18)): (cid:3) 2 (cid:0) P thus obtains v ((cid:18)) = 1(1 (cid:18))2, and A gets no rent, u = u = 0. i i 4 (cid:0) 1 2 When the cost of the supplier is not known to the manufacturers, each manufacturer wouldliketoscreenthesupplierinordertobasehismarketstrategyonthecost. However, in an otherwise symmetric situation, by observing the outcome of contracting between P and A, P receives an additional signal about the supplier resulting in updated beliefs 1 2 (cid:22) = Pr((cid:18) C ): This information is valuable for P but may be disadvantageous for A. By 1 2 j decreasing the uncertainty of P about A, P allows P to extract a bigger share of their 2 1 2 joint surplus. Thus, to reveal any information to P , A has to be compensated for the 1 loss of the information advantage with P . This information externality between the two 2 contracts increases the cost of information revelation for P . It thus a⁄ects P (cid:146)s trade-o⁄ 1 1 6 between e¢ ciency and the informational rents. This leads to the question: How much information P would like to acquire from A? 1 Forexample, P mayabstainfromrevealing(andlearning)anyinformationbyo⁄ering 1 a single contract for both types of cost. As a result, P does not incur the cost of 1 information revelation, but ignores the supplier(cid:146)s cost. To be accepted by both types, P 1 must pay at least t = (cid:18)q. Then, regardless the type, P produces the e¢ cient quantity 1 for high cost q and leaves to a low cost A a positive rent u = (cid:1)(cid:18) q . The pro(cid:133)t of P (cid:3) (cid:3) (cid:3) 1 under this contract is vP = 1(1 (cid:18))2; which corresponds to the pro(cid:133)t of dealing with a 1 4 (cid:0) high cost A under full information. Obviously, this contract has high e¢ ciency costs due to the low production level asked from a low cost supplier. Under this contract, the outcome of the (cid:133)rst stage provides no new information to P , 2 who thus contracts with A under the prior belief (cid:22) = (cid:23): The best contract for P is a 2 screening mechanism that makes the production contingent on the value of the cost. To induce the low cost type to reveal her information, P has to leave her the rent that she 2 can obtain by overstating the cost. The contract of P must therefore satisfy the following 2 incentive compatibility condition: u u +(cid:1)(cid:18)q : (1) 2 (cid:21) 2 2 The optimal trade o⁄between rent extraction and e¢ ciency (see La⁄ont and Martimort (2002)) results in a downward distortion in the output q : 2 (cid:23) q = q = 1(1 (cid:18) (cid:1)(cid:18)); 2 (cid:3)(cid:3) 2 (cid:0) (cid:0) 1 (cid:23) (cid:0) and a positive rent for the low cost A: u = (cid:1)(cid:18)q . The total rent of the low cost A is (cid:3)(cid:3) (cid:3)(cid:3) thus uP = u +u , where P stands for pooling. (cid:3) (cid:3)(cid:3) Alternatively, P can design a contract with distinct outcomes for each type of A. 1 Then the outcome of the (cid:133)rst stage allows P to infer perfectly the type of A, implying 2 that A gains no rent at the second stage. To induce her to reveal the information, P 1 must therefore compensate A for the rent she could obtain by overstating the cost in each of the two stages. If a low cost A selects the contract designed for the high cost type, P is persuaded that he is facing a high cost supplier. This strategy allows A to gain u 2 (cid:3) at the second stage, and u at the (cid:133)rst stage. Therefore, the total cost of information (cid:3)(cid:3) revelation to P under separating contract equals to uS = u +u . 1 (cid:3) (cid:3)(cid:3) The bene(cid:133)t of information for P is the e¢ ciency gains of production: P produces an 1 1 e¢ cient quantity q when dealing with a low cost supplier, and a conditionally e¢ cient (cid:3) 7 quantity q when dealing with a high cost supplier. The pro(cid:133)t of P is vS = (cid:23)[1(1 (cid:3)(cid:3) 1 1 4 (cid:0) (cid:18))2 (u +u )]+(1 (cid:23))[1(1 (cid:18))2 ( (cid:23) (cid:1)(cid:18))2]: (cid:0) (cid:3) (cid:3)(cid:3) (cid:0) 4 (cid:0) (cid:0) 1 (cid:23) (cid:0) The two examples of contracts presented illustrate how the information externality a⁄ects the equilibria of the game. Also it suggest that the ability of P to alter the 1 information transmitted to P a⁄ects the incentives of A to reveal it to P . Then the 2 1 natural questions are: What is the optimal degree of revelation? and What is the best strategy for P ; in the absence of any ad hoc restrictions on the class of contracts from 1 which he can choose. 3 The model WeconsideradynamicgamebetweenN principals,P ;:::;P ,andasingleagent,A.There 1 N are N stages. At each stage P contracts with A over an allocation x X . The outcome i i i 2 of the contracting game de(cid:133)nes an allocation x = (x ;:::;x ) X = X ::: X ; 1 N 1 N 2 (cid:2) (cid:2) where all X , i = 1;:::;N are assumed to be metric spaces. Denote x (x ;:::;x ) i (cid:0)i (cid:17) 1 i the outcomes of contracting up to period i and x+ (x ;:::;x ) the outcomes of i+1 (cid:17) i+1 N contracting from period i + 1 till period N: The decisions of principals P ;:::,P may 1 i restrict the feasible choice of principal P . To account for this feature we assume that i+1 onceP implementsallocationx ,thefeasiblechoiceofP isrestrictedtoF (x );where i i i+1 i+1 (cid:0)i F ( ) is a correspondence F : X X : The agent has ex-ante private information i+1 i+1 i i+1 (cid:1) ) about its type (cid:18) (cid:2) = ((cid:18) ;:::;(cid:18) ), where 2 T < that is persistent through N 1 T 2 (cid:20) 1 stages. In Section 7 we show how the results extend to the case of non-persistent private information. The prior distribution of types, (cid:13) = ((cid:13) ;:::;(cid:13) ), with (cid:13) > 0 for t = 1;:::;T 1 T t and (cid:13) = 1; is common knowledge. Denote (cid:2) the set of types that play at stage i with t i t a positive probability. P We consider communication mechanisms (contracts) which are functions from mes- sages to probability distributions over feasible allocations: A mechanism of principal P , i (cid:0) ; consists of a message space M and a decision rule (cid:12) ( ): For each message m M ; i i i i i (cid:1) 2 P commits to a decision (cid:12) (m) (cid:1) ; where (cid:1) is the set of probability distributions over i i i i 2 F (x ). M is assumed to be a metric space, and denotes the Borel (cid:27) algebra on i (cid:0)i 1 i Mi (cid:0) (cid:0) M : The decision is a measurable mapping (cid:12) : M (cid:1) : Denote (cid:12) ((cid:12) ;:::;(cid:12) ) the i i i ! i (cid:0)i (cid:17) 1 i mechanisms proposed by principals P ;:::,P . 1 i The strategy of P is the choice of a mechanism (cid:0) . Contracts are incomplete in i i that P cannot contingent its contract on the decisions taken by the other principals i 8 P . The agent(cid:146)s strategy at stage i is a message to P . Formally it is described by a i i (cid:0) mapping from the type-contract space to the space S of probability measures over ; i i M (cid:27) : (cid:2) (cid:0) ::: (cid:0) S . Denote (cid:27) (cid:6)(cid:13) (cid:27) and note that (cid:27) S : Also denote i 1 i 1 i i t i;t i i (cid:2) (cid:2) (cid:2) (cid:0) ! (cid:17) 2 (cid:12) = (cid:6)(cid:13) (cid:12) and note that (cid:12) (cid:1) : i t i;t i i 2 The payo⁄ of P depends on the allocation x and on the type of agent (cid:18) . Denote i t v (x ) the payo⁄of P at stage j when the agent is of type (cid:18) . It should be emphasized i;t (cid:0)j i t that P controls directly only the allocation x . However, x may have an indirect impact i i i on allocations x+ through two channels: by a⁄ecting the feasible choice of P+ and by i+1 i+1 changing the perception of P+ about the agent(cid:146)s type. The payo⁄of P at stage i equals i+1 i to v (x ), and his overall payo⁄is i;t (cid:0)i N v (x) = v (x ): t i;t (cid:0)i i=1 X Similarly, the payo⁄of the agent (cid:18) at stage i is u (x ), so its overall payo⁄is given by t i;t (cid:0)i N u (x) = u (x ): t i;t (cid:0)i i=1 X The functions v ( ) and u ( ); i = 1;:::;N are continuous and bounded on their domains. i i (cid:1) (cid:1) The timing of the game is the following: The agent learns its type (cid:18) (cid:2): (cid:15) 2 At each stage i; i = 1;:::;N, P o⁄ers A to play (cid:0) ; that results in allocation x : i i i (cid:15) Once (cid:0) is played, principal P observes information I on past contracting activi- i i+1 i (cid:15) ties (I is speci(cid:133)ed below) and updates beliefs about the agent(cid:146)s private information i to p . i At stage N +1 the game ends. (cid:15) The information of the agent at stage i consists of its type (cid:18), the mechanisms (cid:12) (cid:0)i o⁄ered by P , the pro(cid:133)le of messages m (m ;:::;m ) sent to P ; and the pro(cid:133)le (cid:0)i (cid:0)i(cid:0)1 (cid:17) 1 i(cid:0)1 (cid:0)i(cid:0)1 of outcomes x realized at stages 1;:::;i 1: Denote hA the history of the game for the (cid:0)i 1 (cid:0) i (cid:0) agent, where hA = (m ;(cid:12) ;x ;(cid:12) ;(cid:18)): i (cid:0)i 1 (cid:0)i 1 (cid:0)i 1 i (cid:0) (cid:0) (cid:0) 9 For example, at i = 2; h = (m ;(cid:12) (m );x ;(cid:12) ;(cid:18)): i 1 1 1 1 2 The information of P at stage i; I ; has at most three components. The (cid:133)rst one is the i i sequence of mechanisms (cid:0) ((cid:0) ;:::;(cid:0) ) that were o⁄ered by the preceding principals (cid:0)i (cid:17) 1 i P . The second one is the sequence of allocations that resulted from these mechanisms, (cid:0)i x . Finally, the third one is the sequence of messages that were communicated by the (cid:0)i agent to the preceding principals P , m . We assume that (cid:0) and x become common (cid:0)i 1 (cid:0)i (cid:0)i (cid:0)i (cid:0) knowledge at stage i. For the sequence of messages we distinguish between public and private communication. Under public communication, P observes all the messages that i has been sent by the agent to the preceding principals P . In this case the history of the (cid:0)i game for P is i h (P ; public) = (m ;(cid:0) ;x ): i i (cid:0)i 1 (cid:0)i 1 (cid:0)i 1 (cid:0) (cid:0) (cid:0) Given that P observes the sequence of messages m , it can also infer the decisions (cid:12) i (cid:0)i 1 (cid:0)i 1 (cid:0) (cid:0) that were implemented by P : Under private communication, the message reported by (cid:0)i 1 (cid:0) A to P is their private information: Then the history of P is i i h (P , private) = ((cid:0) ;x ); i i (cid:0)i 1 (cid:0)i 1 (cid:0) (cid:0) thatis, itiscomposedonlyofthepro(cid:133)leofmechanismso⁄eredbytheprecedingprincipals and the pro(cid:133)le of the realized allocations. Denote ((cid:12);(cid:27)) ((cid:12) ;(cid:27) )N the strategy pro(cid:133)le (cid:17) i i i=1 for the principals and the agent in the game (cid:0) = (cid:0) ;:::;(cid:0) , ((cid:12)+;(cid:27)+) ((cid:12) ;(cid:27) )N the f 1 Ng i i (cid:17) k k k=i strategy pro(cid:133)le at stages k = i;:::;N, and ((cid:12) ;(cid:27) ) ((cid:12) ;(cid:27) )i the strategy pro(cid:133)le at (cid:0)i (cid:0)i (cid:17) k k k=1 stages k = 1;:::;i. The observed history h results in updating of beliefs concerning the type of the agent. i The posterior belief of Pi is a measurable mapping pi : Ii 1 Pi, where Pi = p Rj+(cid:2)ij (cid:0) ! f 2 j p > 0; (cid:6)p = 1 is the set of probability distributions over (cid:2) . j j i j g At stage i, P faces a history h (P ) and a state (x ;p ); and o⁄ers a mechanism i i i (cid:0)i 1 i (cid:0) (cid:0) : The outcome of the mechanism (cid:0) determines the history h (P ) and the beliefs i i i+1 i+1 p : I P, resulting in the subsequent state (x ;p ): i+1 i ! (cid:0)i i+1 We study the Perfect Bayesian Equilibria of the game. De(cid:133)nition 1 A Perfect Bayesian Equilibrium of the game (cid:0) consists of a pro(cid:133)le ((cid:12);(cid:27)) and beliefs ((cid:13);p ;:::;p ) that satisfy the following three conditions: 1 N 1 (cid:0) 1. Optimality of (cid:12): The mechanism of P , i = 1;:::;N is optimal, anticipating its i 10