A Genetic Algorithms Approach: Social Aggregation and Learning with Heterogeneous Agents

Summary

We analyze an economy in which increasing returns to scale incentivate social aggregation in a population of heterogeneous boundedly rational agents; however these incentives are limited by the presence of imperfect information on others’ actions. We show by simulations that the equilibrium coalitional structure strongly depends on agents’ initial beliefs and on the characteristics of the individual learning process that is modeled by means of genetic algorithms. The most efficient coalition structure is reached starting from a very limited set of initial beliefs. Furthermore we find that (a) the overall efficiency is an increasing function of agents’ computational abilities; (b) an increase in the speed of the learning process can have ambiguous effects; (c) imitation can play a role only when computational abilities are limited.

We thank two anonymous referees and all the participants to the workshops “New Tools for Qualitative Economic Dynamics” held at the CIMAT, Guanajuato - Mex (October 2002) and “VII WEHIA Workshop” held in Trieste (May 2002) for helpful comments. We acknowledge financial support from MURST PRIN 2000 ‘New Tools for qualitative analysis of economic dynamics’.