Learning and Noisy Equilibrium Behavior in an Experimental Study of Imperfect Price Competition*

We consider a duopoly pricing game with a unique Bertrand–Nashequilibrium. The high‐price firm has a nonvanishing market share, however, and intuition suggests that observed prices may be positively related to this market share. This relationship is implied by a model in which players make noisy (logit) best responses to expected payoff differences. The resulting logit equilibrium model was used to design an experiment in which the high‐price firm's market share varies. The model accurately predicts the final‐period price averages. A naive learning model predicts the observed differences in the time paths of average prices.