Bayesian Learning in Unstable Settings: Experimental Evidence Based on the Bandit Problem
We study learning in the “Boardgame,” a bandit problem where the outcome probabilities of six arms switch (“jump”) over time — a restless bandit. In the experiment, optimal Bayesian learning tracks the jumps – through learning of the probability of a jump or direct jump detection – and, once a jump has occurred, re-learns the outcome probabilities. It is much more complex than the natural alternative which learns through trial-and-error (adaptive expectations). Yet, when combined with a partially myopic decision rule, Bayesian learning better matches the behavior observed in the lab. This result suggests that agents may be less limited in their computational capacities than previously thought, and that complexity does not always hamper fully rational learning.
(job market paper). Most recent version: May 2010.
Hierarchical Versus Forgetting Bayes: Probabilistic Learning under Changing Contingencies
Using simulated data, we compare the properties of dynamic Bayesian algorithms that effectively learn shifting outcome probabilities.
With P. Bossaerts.
Knightian Uncertainty: How Does it Affect the Ability to Detect Jumps? Experimental Evidence
We show that proper perception of Unexpected Uncertainty — jump detection — in the Boardgame calls for a nudge whereby the experimenter brings the phenomenon of jumps in the outcome contingencies to the subjects’ attention. In a variant of the Boardgame in which we suppressed such nudge — in it, subjects were initially naive to the phenomenon of jump occurrence in the task — subjects no longer detected jumps. The same subjects next performed the Boardgame and Bayesian learning prevailed like in the original experiment. This result suggests that the mental models (the structures which serve to select and organize the information coming from the environment, on which the learning processes operate) the subject used to predict outcomes in the task, assumed stationarity.
