Publications:
[1] Sequential learning under informational ambiguity, 2026, American Economic Review [Link] [SSRN]
|
Abstract: This paper studies informational robustness in social learning, where individuals face ambiguity and consider a set of possible data-generating processes. I show that, under sufficient ambiguity, an information cascade occurs almost surely, regardless of the underlying DGP.
Supplementary Materials: published version, extended version [SSRN] |
|
Abstract: This paper introduces a biased updating rule under ambiguity. Individuals consider a set of DGPs and update beliefs using those that can best justify their endowed biases. I characterize the resulting learning dynamics and show that biased updating can lead to persistent mislearning and belief polarization.
|
Working papers:
[1] The wisdom of crowds or group irrationality? Non-Bayesian social learning with misspecification, December 2025. R&R @ American Economic Journal: Microeconomics (2nd-round) [SSRN]
|
Abstract: This paper introduces group irrationality: situations in which society fails to learn the truth even though each individual could do so in isolation. I show that group irrationality is prevalent under non-Bayesian social learning rules, but can be avoided when learning rules are insensitive to extreme beliefs or when individuals sufficiently underreact to information.
|
|
Co-author: Yuyi Li (UNC, PhD student)
Abstract: This paper introduces an unlearning algorithm for social learning, in which society gradually reduces the weight on public information. We show that, if this weight decays at a sub-reciprocal rate, a correct information cascade emerges almost surely in the long run. |
[3] If you are NOT so smart, why are you rich? Robust market selection with general recursive preferences, March 2024. [SSRN]
|
Co-author: Pablo Beker (University of Warwick) Abstract: This paper introduces a unified framework of market selection with general recursive preferences. We characterize long-run consumption dynamics and show that the market selection hypothesis is no longer robust in general. In particular, consumers with incorrect beliefs can survive against a large class of preferences and beliefs. |
[4] Schedules and prioritization: A behavioral foundation for multi-armed bandits and stopping problems,
June 2026. [SSRN]
|
Co-author: Can Urgun (UNC)
Abstract: This paper provides a behavioral foundation for bandit index policies. Starting from preferences over local contingent schedules, it derives recursive stopping representation and shows how calendar time can be priced by an index. The resulting index rule nests expected-utility bandits, Bayesian learning, robust preferences, and Pandora-style search. |