Dynamic Reward Design

Abstract

We study a dynamic mechanism design problem with limited liability. A principal hires an agent to work over a finite horizon. The agent's costs are i.i.d. across periods and privately known, while his working status is publicly observable. The nonnegative-payment constraint distinguishes our problem from the dynamic screening literature. We identify conditions under which the optimal mechanism takes a simple form: a consecutive-working menu, in which the agent chooses when to start working and is then incentivized to work consecutively until the end. We explain why this consecutive structure is optimal despite i.i.d. costs and additively separable payoffs, shedding light on real-world reward schemes that require consecutive actions.