In the mechanism design problem with multiple agents, Che and Kim (2006) has proposed a way to deal with collusion among agents who are privately informed of their types.
Specifically, they show that, even in the face of collusive agents, a principa ...
In the mechanism design problem with multiple agents, Che and Kim (2006) has proposed a way to deal with collusion among agents who are privately informed of their types.
Specifically, they show that, even in the face of collusive agents, a principal can attain any surplus level that she can achieve
without collusion. Their notion of collusion-proofness, however, does rule out the possibility that some agent refuses to participate in the principal's mechanism anticipating the followng scenario: If he participates, then he will be forced to join the collusive ring and accept a low payoff since otherwise other agents would be colluding against him. This possibility will undermine the implementation of the desired outcome for the principal.
In this research, we suggest a way to dramatically strengthen the Che and Kim's notion of collusion proofness to treat the above problem. In fact, the notion we suggest here, called 'strongly collusion-proof implementation', will require no restriction on collusive behavior, except that collusion occurs after the contract acceptance
decision - a key assumption of Che and Kim and Laffont and Martimort (1997, 2000) preserved here. First, we do not require agents to collude always: Collusion may involve only some agents and may occur only sometimes
when ``the condition is right.'' We therefore relax the assumption often made implicitly in the literature including CK that
the collusive proposal need to be accepted by all types of agents.
Second, there is no restriction on how many coalitions there are, whom each coalition comprises, how they operate, and who proposes what kind of side contracts. A side contract can be proposed by a third party or by one of the informed agents, or can even be a consequence of negotiation among several
agents. Further, the design of the collusion-proof mechanism requires no knowledge about any of these matters. Lastly, we do not restrict the agents' out-of-equilibrium beliefs.
Under fairly general environment, we are able to construct a mechanism that is strongly collusion-proof while implementing the desired outcome for the principal. We do so in both cases of independent and correlated types. In the independent types case, the desired outcome is exactly implementable while it is only approximately implementable in the correlated types case.