Learning by Slow Unlearning
39 Pages Posted: Last revised: 26 Jan 2025
Date Written: December 21, 2024
Abstract
This paper examines a sequential social learning model with finite signals, where individuals update beliefs using a power Bayes’ rule that allows heterogeneous weighting of public and private information. We establish that, under certain regularity conditions, a correct informational cascade almost surely emerges in the limit if and only if society gradually reduces its reliance on public information at a rate slower than 1/i, where i denotes the individual’s position in the sequence. If society abandons public information too rapidly, information aggregation fails, and individuals rely solely on their private signals in the limit. Conversely, if public information is never fully discarded, both correct and incorrect cascades may persist
Suggested Citation: Suggested Citation