Optimal Incentives Under Moral Hazard: from Theory to Practice
Abstract: This paper addresses the following practical question: given an existing incentive contract, what information must a manager acquire to determine how to improve upon that contract? We use a canonical principal-agent framework under moral hazard and assume the principal has productivity data corresponding to some status quo contract. Our main result shows that if the principal has a priori information about the agent’s marginal utility function, and she carries out an experiment in which she perturbs the existing contract, then she can estimate how the agent will respond to a change in his marginal incentives, as well as how the agent’s marginal incentives will respond to any other perturbed contract. The information provided by such an experiment, therefore, serves as a sufficient statistic for the question of how to locally improve upon the existing contract optimally. The same informational requirements hold, and an analogous sufficient statistic result is obtained, when the principal is restricted to choosing from a lower-dimensional parametric class of contracts; e.g., linear contracts. We also describe the informational requirements for assessing global optimality, where local information like the type described above is insufficient.