Do We Learn from Mistakes of Others? A Test of Observational Learning in the Bandit Problem
31 Pages Posted: 13 Jan 2016 Last revised: 10 Dec 2016
Date Written: December 9, 2016
I experimentally investigate observational (social) learning in the simple two-armed bandit framework where the models based on Bayesian reasoning and non-Bayesian reasoning (count heuristics) have different predictions. The results contradict the predictions of the Bayesian rationality e.g. Bayesian Nash Equilibrium, Naıve herding model (BRTNI): Subjects follow the choices that contain no information about the state of the world, follow the coinciding choices of others (though it is empirically suboptimal), sustain losses making every first choice and cascade early then Bayesian-based models predict, but not in a random way. In addition, the Quantal Response Equilibrium is tested and the robustness of the theory questioned.
Keywords: Observational learning, information cascade, experiment
JEL Classification: C92, D82, D83
Suggested Citation: Suggested Citation