Dr Henrique Castro-Pires Dr

This paper studies the optimal joint design of incentives and performance rating scales in a principal-manager-worker hierarchy. The principal wants to motivate the worker to exert unobservable effort at the minimum feasible cost. Given the worker's effort, two signals are realized: public and verifiable output and a private non-verifiable signal known only to the manager. The principal may try to elicit the manager's private information by requiring her to evaluate the worker's performance. Payments may depend on output and the manager's evaluation. I show that the principal can achieve no more than what is feasible with a binary rating scale. I also identify scenarios where subjective evaluations are valuable (non-valuable), reduced transparency is advantageous, and forced ranking outperforms individual evaluations. (JEL: D23, D82, D86, J41)

This paper analyzes the optimal menu of contracts offered by a risk-neutral principal to a risk-averse agent under moral hazard, adverse selection, and limited liability. We show that a limited liability constraint causes pooling of the most efficient agent types. We also find sufficient conditions under which full pooling is optimal, regardless of the agent's risk aversion or type distribution. Our model suggests that offering a single contract is often optimal in environments with moral hazard, adverse selection, and in which the principal faces a limited liability constraint. (JEL: D82, D86)

While many real-world principal-agent problems have both moral hazard and adverse selection, existing tools largely analyze only one at a time. Do the insights from the separate analyses survive when the frictions are combined? We develop a simple method—decoupling—to study both problems at once. When decoupling works, everything we know from the separate analyses carries over, but interesting interactions also arise. We provide simple tests for whether decoupling is valid. We develop and numerically implement an algorithm to calculate the decoupled solution and check its validity. We also provide primitives for decoupling to work and analyze several extensions.