Adapting behaviors through a learning process

This paper concerns the concepts of conjectures and reactions of agents in a dynamic context, a framework where agents form conjectures on how the other agents (would) react to their actions and update this belief according to observations. They decide based on this conjecture without knowing profit functions of other agents. Agents somehow “learn” about the behavior (including the conjectures) of the other agents. We show that some type of “consistency” of these conjectures spontaneously appears. We discuss the Pareto-optimality and the local stability of the strategies obtained in the long run and analyze the example of Cournot’s oligopoly.