Data-Driven Contract Design

This paper proposes a prior-free model of moral hazard wherein the principal's beliefs about the agent's production technology are characterized by revealed preference data. The principal and the agent are each financially risk neutral and the agent's preferences are understood to be quasilinear in effort. Prior to contracting with the agent, the principal observes the outcome of a finite number of experiments, each of which consists of (1) a contract and (2) the distribution of output (but not the effort cost) associated with the agent's best response to that contract. She views any technology that rationalizes this data as plausible, and evaluates contracts according to their guaranteed expected payoff against the set of rationalizable technologies. In this environment, robustly optimal contracts augment the contracts in the experimental data with equity payments. Our paper makes no assumptions about the agent's technology beyond rationalizability of the revealed preference data, which itself is characterized by straightforward textbook criteria