Incentivizing Agents through Ratings
I study the optimal design of ratings to motivate an agent's investment in quality when transfers are unavailable. The principal designs a rating scheme that maps quality to a (possibly stochastic) signal. The agent privately knows his ability and chooses a quality level. A competitive market then offers the agent a wage equal to his expected quality given the signal. When restricted to deterministic ratings, lower censorship is optimal if the ability density is single-peaked, and pass/fail tests are optimal as well if the density is increasing. When randomization is allowed, lower censorship remains optimal if the ability density is log-concave or increasing, and pass/fail tests remain optimal if the density is increasing. By contrast, every optimal rating scheme involves randomization if the density is decreasing and sufficiently log-convex---roughly speaking, if intermediate ability is scarce relative to high and low ability.