Delegated Learning and Non-Credible Communication

Abstract

We consider a setting in which an impatient agent acquires payoff-relevant information about the true state of the world. The agent endogenously chooses when to stop learning, at which point an uninformed principal takes an action to maximize her own expected payoff. The agent's preferences are biased relative to the principal's, generating misalignment of expected payoffs. When communication is non-credible, the principal can only rely upon the agent's endogenous stopping rule when strategically specifying her course of action. In the no-communication equilibrium, the agent adopts a one-sided stopping rule as a function of her posterior belief that is consistent with the principal's pre-specified course of action at the time of stopping. When the principal has commitment power, relative to the full-communication equilibrium, the agent is always worse off; for intermediate values of prior beliefs, the principal is better off. The one-sided equilibrium stopping rule (and associated action) can switch discretely as a function of prior beliefs, generating dramatic regime changes for arbitrarily small changes in beliefs. When learning is initiated in the no-communication equilibrium there is a non-zero probability of indefinite delay, in which the agent never ceases learning and the principal never takes an action.