Disagreement and Learning in a Dynamic Contracting Model

We present a dynamic contracting model in which the principal and agent disagree about the resolution of uncertainty, and we illustrate the contract design in an application with Bayesian learning. The disagreement creates gains from trade that the principal realizes by transferring payment to states that the agent considers relatively more likely, changing incentives. The interaction between incentive provision and learning creates an intertemporal source of "disagreement risk" that changes second-best risk sharing. There is a regime shift between economies with small and large belief differences, and an early shock to beliefs can lead to large persistent differences in variable pay even after beliefs have converged. The principal's value function is convex in the underlying belief differences because the more optimistic the agent relative to the principal, the sharper the incentives and the lower the agent's required compensation; the result is a value to uncertainty and learning. Under risk- neutrality, "selling the firm" to the agent does not implement the first best because it precludes state-contingent trades.