On economic applications of the genetic algorithm: a model of the cobweb type

This paper explores the idea of using artificial adaptive agents in economic theory. In particular, we use Genetic Algorithms (GAs) to model the learning behavior of a population of adaptive and boundedly rational agents interacting in an economic system. We analyze the behavior of a GA in two versions of a model of the cobweb-type, one in which firms make only quantity choices, and the other one in which firms first decide to exit or to stay in the market, and subsequently decide how much to produce. We present simulations with different coding schemes and interpret the rather surprising differences between the results for different setups by employing the mathematical theory for GAs with state-dependent fitness functions. In particular, we explain the relationship between coding and convergence properties of GAs.     

Cycling in a stochastic learning algorithm for normal form games

In this paper we study a stochastic learning model for 2×2 normal form games that are played repeatedly. The main emphasis is put on the emergence of cycles. We assume that the players have neither information about the payoff matrix of their opponent nor about their own. At every round each player can only observe his or her action and the payoff he or she receives. We prove that the learning algorithm, which is modeled by an urn scheme proposed by Arthur (1993), leads with positive probability to a cycling of strategy profiles if the game has a mixed Nash equilibrium. In case there are strict Nash equilibria, the learning process converges a.s. to the set of Nash equilibria.     

Schumpeterian process competition, welfare and laissez-faire: An experiment in artificial economic evolution

A simulation model is used to construct a regime of artificial economic evolution, where Schumpeterian process competition prevails, in the presence of technological uncertainty and bounded rationality. The output decision of the firm is represented by a behavioural algorithm, which allows for the presence of collusive behaviour. The purpose of the experiments is to go some way towards addressing the twin issues of the nature of the relationship between market structure and industry performance in a dynamic setting, and the contention that the evolutionary metaphor implies a laissez-faire stance with respect to policy issues. Under the simplifying assumptions of the model, experiments suggest that industries which generate high average concentration over a given period can compare favourably with industries that generate low average concentration, if the time profiles of both welfare and concentration are analysed. Also, the experiments suggest that the industry will naturally evolve a structure best suited to exploit the technological environment, but despite this there is still a role for intervention into the competitive process.     

Heterogeneous beliefs and learning in forward looking economic models*

We study a class of forward looking economic models with heterogeneous agents in a bounded rationality setting. The agents employ the same recursive learning rule to update beliefs but are characterized by different memory parameters. The peculiarity of the learning mechanism is that the learning rate is not vanishing in the limit. Differently from what is obtained in the case of a vanishing learning step, i.e., the stability conditions in the heterogeneous agents case are those of the representative agent model, we show that heterogeneity matters for the expectational stability of a stationary perfect foresight equilibrium and that the stability parameter restrictions with heterogeneous agents are stronger than in the case of homogeneous agents.