Learning in Matching

Abstract

This paper studies how learning shapes outcomes in decentralized matching markets with incomplete information. In a one-to-one worker-firm matching environment where workers privately observe a payoff-relevant state, firms learn from matching behavior. Building on recent Bayesian and prior-free theories of stable matching, we experimentally examine whether stability can emerge through decentralized processes and how learning drives convergence. We identify three fundamental learning patterns, learning from conditional evaluation, learning from blocking, and learning from no blocking, each with fully and partially revealing subtypes, as well as compound learning patterns. Our results show that most markets converge to stable outcomes, demonstrating the predictive power of the theory; however, learning from no blocking and full revelation of the state hinder convergence, and our analysis provides evidence on the mechanisms driving these effects.

Keywords: Matching, Incomplete Information, Learning