Deterministic versus stochastic mechanisms in principal-agent models

This paper shows that, contrary to what is generally believed, decreasing concavity of the agent's utility function with respect to the screening variable is not sufficient to ensure that stochastic mechanisms are suboptimal. The paper demonstrates, however, that they are suboptimal whenever the optimal deterministic mechanism exhibits no bunching. This is the case for most applications of the theory and therefore validates the literature's usual focus on deterministic mechanisms.