Social Learning with Payoff Externality: Why Some Learn While Others Do Not

This paper studies a model of social learning with externalities and possibly with limited observability of past actions. Bayesian agents sequentially choose an action each, after observing the actions of a subset of preceding agents. These actions and a private signal, conditionally independent across agents, supply information about the common state of the world. Each agent’s payoff depends on the state and the actions of the subset of preceding agents (neighbors). With complete observability of past actions, I show that social learning occurs even in the presence of externality. Social learning also occurs with limited observability but in the absence of externality. However with both limited observability of the history of actions and action externalities, social learning is bounded away from full information. When different agents have randomly determined observability, even a small chance of having limited observability can hamper social learning. These results stand in contrast to the existing literature on herding and social learning. I also show that when agents know the extent of observability of previous agents, agents with better observability learn while the rest do not.