Empiricist Learning Rules on Social Networks : Convergence and Quality of Information Aggregation

This paper proposes a novel learning model on social networks that captures settings where individuals interact frequently on multiple, relatively short-lived topics. In this model, each period features a new draw of nature and multiple rounds in which information arrives, gets aggregated and diffuses through network links. The repetitive nature of interactions across periods allows for a separation between learning about the environment and aggregating information about the current state. A class of empiricist learning rules achieve convergence of learning on all networks. On clique trees, these learning rules further achieve strong efficiency in information aggregation. The paper also presents a converse to the positive efficiency result and identifies distinct reasons why efficiency is hard to obtain in general circumstances, despite the fact that convergence of learning holds generally