Justifying the First-Order Approach to Principal-Agent Problems

It is of interest to know when the incentive compatibility conditio n in principal-agent problems can be replaced by the condition that the agent's expected utility be stationary in effort. The Mirrlees-Rogerson conditions do not work if the principal can observe more than one observable statistic. Also, the Mirrlees-Rogerson assumption that the distribution function of output is convex in the agent's action is unsatisfactory even in the context of the basi c model; it is too restrictive. The paper presents a suite of conditio ns that are applicable to the multistatistic case and replaces the objectionable convex distribution function assumption.