Endogenizing private information: Incentive contracts under learning by doing

This paper investigates the design of incentives in a dynamic adverse selection framework when agents' production technologies display learning effects and agents' rate of learning is private knowledge. In a simple two-period model with full commitment available to the principal, we show that whether learning effects are over- or under-exploited crucially depends on whether learning effects increase or decrease the principal's uncertainty about agents' costs of production. Hence, what drives the over- or under-exploitation of learning effects is whether more efficient agents also learn faster (so costs diverge through learning effects) or whether it is the less efficient agents who learn faster (so costs converge). Furthermore, we show that if divergence in costs through learning effects is strong enough, learning effects will not be exploited at all, in a sense to be made precise.