Detecting Profitable Deviations

In this paper I offer necessary and sufficient conditions for implementability in a quasi-linear principal-agent model with arbitrary type spaces. I extend Rochet’s Theorem by allowing the principal to observe information that may be correlated with the agent’s type. By viewing the agent’s gains from misreporting as payments in a hypothetical zero-sum game, I show that an allocation is implementable if and only if every infinitesimally detectable deviation is at most infinitesimally profitable. This leads to generalizations of existing results, such as revenue equivalence and implementation with moral hazard, as well as new results altogether, such as budget balanced implementation (interim and ex post), existence of bargaining solutions and revealed stochastic preference. JEL Classification: D21, D23, D82.