Probabilistic learning and emergent coordination in a non-cooperative game with heterogeneous agents: An exploration of minority game dynamics.

In this paper we present results of simulations in which we use a general probabilistic learning model to describe the behavior of heterogeneous agents in a non-cooperative game where it is rewarding to be in the minority group. The chosen probabilistic model belongs to a well-known class of learning models developed in evolutionary game theory and experimental economics, which have been widely applied to describe human behavior in experimental games. We test the aggregate properties of this population of agents (i.e., presence of emergent cooperation, asymptotic stability, speed of convergence to equilibrium) as a function of the degree of randomness in the agents' behavior. In this way we are able to identify what properties of the system are sensitive to the precise characteristics of the learning rule and what properties on the contrary can be considered as generic features of the game. Our results indicate that, when the degree of inertia of the learning rule increases, the market reaches a higher level of allocative and informational efficiency, although on a longer time scale.