Information and Learning in Bertrand and Cournot Experimental Duopolies

Abstract

In this paper we report the results from a series of experiments on Cournot (homogeneous and differentiated products) and Bertrand (differentiated products) duopoly markets with no uncertainty, fixed endpoints and random matching. For each set, the experiments are designed with three alternative information set up: I) no information (participants are only informed on their own payoff for the period), 2) average industrial profit (participants are informed on their own performance, as well as on the average profit in all markets), 3) imitation (players are informed, on request, on their rivals' past successful actions). The effect of different information structures on individual behaviour in market experiments is a long debated issue. Recently, using evolutionary arguments, it has been argued that the imitation of successful strategies induces more competitive equilibria in market games (M. Schaffer, 1989; F. Vega-Redondo 1997). By the same token, the information on the industry's average profitability might induce more collusive outcomes, if such markets signals are perceived by players as aspiration levels and if they therefore try new strategies anytime their profits fall below such threshold (F. Palomino and F. Vega-Redondo, 1999; H. Dixon, 2000). Our aim is to test such predictions in duopoly price and quantity games. We find that the imitation learning rule prevails when players have full information on their rivals' previous choices, and such learning behaviour induces more competitive outcomes in the Cournot market designs. As for the aspiration learning rule, the evidence is unclear. Whilst in the majority of the cases, players experiment new strategies when their payoff falls below the average profit, as predicted by the aspiration rule, we find no evidence of convergence to collusion, though in the Cournot experiments, the fraction of players choosing cooperative actions in the last stages of the game significantly increase in the second information setting.