Exploring Human Behaviour and Learning in Experiment Cournot Settings

Abstract

In this paper we study the results of three well-known experiments on duopoly markets, in which participants played a repeated market game in similar experimental conditions.

We address two different questions: which learning theory is better able to describe the players' decisions in the experimental games and which of these rules better explains the process of equilibrium selection in some representative individual markets. To this aim, we test two alternative learning rules, the reinforcement theory, or "rote" learning, and the fictitious play approach, using the data from the experiments.We arrive at three main conclusions. First, in all experiments subjects did "learn" how to play the game during the course of the sessions, adopting specific learning strategies. Second, only in one case (the Fouraker and Siegel's experiment (1963)), reinforcement behavioural rules seems to prevail among participants, whilst in both experiments carried out by C. Holt (1985), individual behaviour is better explained by a general model of individual choice which encompasses features of belief learning. Finally, we explore the ability of our models of learning to explain the equilibrium selection in individual markets within the three experiments. We analyse the data from six representative markets, in which behaviour converged towards Cournot or Collusive levels of output, (averaged over the last three periods of play), and try to assess which behavioural rule is more apt to explain the equilibrium selection in these experimental industries. As we will see, the belief model generally explains better both outcomes, that is, the process of convergence to the Nash-Cournot equilibrium and to the Joint Profit Maximising equilibrium, in all cases considered.

Keywords: Learning, Bounded rationality, Learning algorithms, Experimental oligopoly

JEL Classification: C91, D83