Delegation without Priors

Abstract

A principal relies on information held by a biased agent in decision-making. Without a prior over the circumstances that may arise, she designs a delegation rule to minimize the worst-case regret. Optimal delegation combines a default action, a fixed gap above it, and a higher discretionary interval with random boundaries. This structure separates two margins: whether the decision departs from the default and how much discretion the agent receives conditional on doing so. The default and gap limit excessive adjustment when little adaptation is warranted, while the higher interval preserves flexibility when larger adjustments are needed. Randomizing the discretionary interval's floor and ceiling hedges the principal's exposure across potential worst-case states. This rule is optimal among all stochastic incentive-compatible mechanisms.