Optimal incentive contracts and information cascades

We examine information aggregation regarding industry capital productivity from privately informed managers in a dynamic model with optimal incentive contracts. Information cascades always occur if managers enjoy limited liability: when beliefs regarding productivity become endogenously extreme (optimistic or pessimistic), learning stops. There is no learning if initial beliefs are extreme, or if agency conflicts are severe. In contrast to the literature, cascades occur even when signals have unbounded precision or when there are rich action spaces. Relaxing limited liability constraints is not sufficient to avoid cascades; we provide sufficient conditions for efficient information aggregation through incentive contracts.