Cristián Pablo Ugarte, UC Berkeley
We study the problem of estimating the preferences of an agent whose choices violate the Generalized Axiom of Revealed Preferences due to imperfect implementation. We propose an estimator that asymptotically recovers the underlying preferences under mild assumptions. In finite data, the set of estimators is characterized by the subset of directly revealed preferences that are interpreted as incorrect. We show how to generalize our method to a vast class of choice environments.