Optimal Monitoring Policies in Agencies

This article considers a principal-agent problem in which the principal has access to a costly monitoring technology that can be used to acquire additional information about the agent's actions subsequent to observing the agent's output. Although randomized monitoring policies are feasible, we show that in a variety of contexts optimal monitoring policies are deterministic and "lower-tailed," that is, there exists some critical level of output such that further investigation of the agent's actions occurs if and only if output falls below this critical level.