"Group Cascades in Strategic Voting", with James Stark and Haozheng Chen, January 2025

Abstract: We bridge the theories of information cascades (Bikhchandani et al. 1992) and strategic voting (Feddersen and Pesendorfer 1998) by studying a sequence of voting groups whose members have common values and independent private information. Each group strategically votes on a decision after observing all earlier groups' decisions. We show that the asymptotic public belief distributions and expected utilities are sensitive to the voting rules. When all groups use the same voting rule, the rule that maximises asymptotic payoff can differ from the one that maximises the first group's information revelation. This difference occurs only when the former rule induces limit beliefs strictly above the threshold that triggers cascades, which is always true for the unanimous rules but not the relatively symmetric rules. We present examples where, under a weak prior belief, the simply majority rule is best for the first group but asymptotically dominated by the unanimous rule. 

"Headlines as truncated experiments", with John Nachbar, Nov 2024

Abstract: Do consumers of online news select news stories that conform to or challenge their existing belief? We rationalize the mixed empirical findings by modeling a Bayesian decision maker who seeks information for its instrumental value and a data generating process that is common in many online environments: a headline precedes the full content. The headline reveals the direction but not the exact information value of the content. We identify the key features of the content distribution that induce a preference for the (a) prior-contradicting headline, (b) prior-conforming headline, and (c) the same headline for all priors.

"Correlated information cascades," October 2021

Abstract: This paper extends the model of information cascade (Bikhchandani, Hirshleifer, and Welch, 1992) by introducing two correlated queues of players. By comparing the equilibrium outcomes of the two-queue game with the benchmark one-queue game, this paper finds out that the observability of correlated players decreases the probability of correct permanent cascades.     (.pptx slides)

"Non-competing persuaders," European Economic Review, August 2020, volume 127, 103454

Abstract: I study Bayesian persuasion games with multiple persuaders in which the persuaders are non-competing: all persuaders want the decision maker to take the same action, regardless of the state. In the case of a single persuader, it is known from previous research that the persuader-optimal information design leaves the decision maker with no surplus. In this paper, I show that with two or more non-competing persuaders and independent tests, there are always equilibria in which the decision maker receives surplus. Moreover, if there is exogenous noise then the decision maker receives surplus in every equilibrium, provided the number of persuaders is sufficiently large; asymptotically, the decision maker learns the true state in every Pareto optimal symmetric equilibrium with infinitely many persuaders. Moreover, with sufficient exogenous noise, having more than one persuader not only improves the welfare of the decision maker but it also improves the welfare of the persuaders.

"Helpful laymen in informational cascades," ​Journal of Economic Behavior and Organization, August 2015, volume 116, pages 407-415

Abstract: This paper extends Bikhchandani, Hirshleifer and Welch’s informational cascade model by introducing two types of players: experts with high signal accuracy and laymen with low signal accuracy. If a small enough fraction of laymen are present in the population, the probability of having a correct cascade is strictly higher than if no laymen are present. This is because the presence of laymen makes experts less eager to follow suit, which increases the amount of private information revealed.