Experiments on Cournot Competition
Experiments on Cournot Duopoly
The experiments on Cournot competition, where sellers choose quantities to produce have focused mainly on two issues: Whether subjects do play the Nash predictions or not and how subjects learn to play given different information conditions.
Holt (1985) and Mason, Phillips, and Redington(1991) found in their experiments that the quantity choices of players typically fall between the Cournot and collusive levels in multi-period duopoly games.
Holt and Villamil (1986) in their four seller sessions gave no demand information to the sellers. They found that the mean and median prices were near the competitive level, far away from the Cournot level.
Binger et al. (1990) ran a forty-period Cournot market experiment with complete demand information. Averaging across sessions and periods, the market quantities approximate the Cournot levels for the two-seller-no-communication sessions, and quantities were midway between competitive and Cournot levels for the five-seller-no-communication sessions.
Cox and Walker (1998) investigates different models of learning using a Cournot market environment. In particular they were interested in whether subjects eventually converge to the predicted Nash equilibrium responses derived for Cournot competition. In addition they were also in the path of play that subjects follow in Cournot duopoly markets. They also looked at the stability issues of the equilibrium. They used different marginal cost structures in their experiment and both fixed and random matching of subjects. The experiments where the learning models predict convergence to the stable interior Nash equilibria, subjects predominantly chose the Nash equilibrium quantities after only a few periods, and the deviations from Nash Equilibrium were very small. In contrast, the cases where the learning models predict convergence to stable boundary Nash equilibrium, subjects seldom chose quantities that were part of the interior or boundary equilibria.
Hück, Normann, and Oechssler (1999) looked at imitation as a possible learning mechanism in Cournot oligopoly. In their experiment four players chose output in the interval [0,100], for forty periods. They compare behavior under four conditions, designed to test learning theories which require different kinds of information. They had the following four information treatments: Under the “BEST” treatment, demand, cost and total industry output was shown to each player; Under “FULL,” in addition to demand and cost, other firms’ chosen quantities and profits were shown; Under “IMIT” only the other firms’ quantities and profits were shown; Finally under the “NOIN,” treatment each player was shown only his own profit. They found that in the “BEST” condition, outputs move towards the Cournot-Nash predictions. In “FULL” outputs go up towards the Walrasian predictions. In “IMIT” they found outputs to have high fluctuations.
Overall, multi-period Cournot quantity-choice experiments show that outcomes fall on both sides of the Cournot prediction, and in some cases show signs of near perfect collusions. When there are more than two sellers, outcomes are found often to be more competitive than the Cournot predictions. In terms of the behavior observed in quantity setting experiments the experiments mainly show signs of best-reply learning and imitation learning suggesting that imitation can be quite an important and practical form of learning.