Hierarchical experimentation

We consider a bandit problem faced by a team of two heterogeneous players. The team is hierarchical in that one player (the principal) retains the exclusive right to terminate the project while the other player (the agent) focuses strictly on implementing the project assigned. As a key departure from the existing literature, we assume that the principal may be privately informed about the project quality. In contrast to the literature, the belief in our model generally follows a non-monotonic path: while each failure makes the agent less confident in the project, the uninformed principal drops out gradually over time, which partially restores the agent's motivation. We derive explicit solutions for the agent's effort and the principal's exit decisions, which allow us to obtain a full characterization of the equilibrium. Our analysis elucidates how and under what conditions an organization becomes trapped in a stagnant phase in which little action takes place.