Sequential Contracts

We study the principal-agent setting, where a principal delegates the execution of a costly project to an agent. In the classical model, the agent chooses an action among a set of available actions. Every action is associated with some cost, and leads to a stochastic outcome for the project. The agent's action is hidden from the principal, who only observes the outcome. The principal incentivizes the agent through a payment scheme (a contract) that maps outcomes to payments, with the objective of finding the optimal contract - the contract maximizing the principal's expected utility. In this work, we introduce a sequential variant of the model, capturing many real-life settings, where the agent engages in multiple attempts, incurring the sum of costs of the actions taken and being compensated for the best realized outcome. We study the contract design problem in this new setting. We first observe that the agent's problem - finding the sequential set of actions that maximizes his utility for a given contract - is equivalent to the well-known Pandora's Box problem. With this insight at hand, we provide algorithms and hardness results for the (principal's) contract design problem, under both independent and correlated actions. For independent actions, we show that the optimal linear contract can be computed in polynomial time. Furthermore, this result extends to the optimal arbitrary contract when the number of outcomes is a constant. For correlated actions we find that approximating the optimal contract within any constant ratio is NP-hard.