On incentive compatible designs of forecasting contracts

In the paper the optimal design of forecasting contracts in principal-agent setting is investigated. It is assumed that the principal pays the agent (the forecaster) based on an announced forecast and an event that materializes next. Such a contract is called incentive compatible if the agent maximizes her payoff when she announces her true beliefs. This paper relaxes the assumption present in earlier works on this subject that agent’s beliefs are deterministic by allowing them to be random (i.e. stemming from estimation). It is shown that for binary or nominal events the principal can learn only expected values of agent’s predictions in an incentive compatible way independent of agent’s signal space. Additionally it is proven that incentive compatible payment schemes give the agent a strictly positive incentive to improve the precision of her estimates.