Conditional Delegation and Optimal Supervision

This paper analyzes a simple modification of a standard mechanism in hierarchical centralized structures with hard-information supervision. The supervisor receives a signal about the productive agent's technology. With some probability the supervisor learns the true agent's technology, otherwise she learns nothing. Our design lets the productive agent choose between two competing contracts, a "secure" contract or a grand contract subject to uncertainty. The mechanism eliminates agency costs by providing the productive agent with the possibility of avoiding inspection. When productive agent is risk averse, our mechanism also provides him with an insurance coverage: as a consequence, this mechanism would be worthwhile even abstracting from collusion.