Simulations with Genetic Algorithms in Economic Systems

In chapter 5 I presented some simulations of genetic learning in evolutionary games. Evolutionary games, however, are always a stylized model of an economic system, since no explicit structure of the model is given. In many instances the special structure of a certain economic model allows considerably more precise insights into the behavior of individuals in such a situation than a more abstract game theoretic model. This holds also in particular true if we think of the adaptive behavior of boundedly rational agents. Basic dynamic properties of learning dynamics like convergence, dynamic equilibrium selection or complex behavior obviously depend on the framework the corresponding agents are living in. The purpose of this chapter is to study the dynamic properties of genetic learning in several standard economic models. Hereby we should not only try to derive these properties but also to understand their relation with the properties of the corresponding model. Altogether I will present three economic models. The first one is a very simple model of a market of price takers, where we will be able to see the power of the theoretical results from chapter 4 very impressively. The second model, an overlapping generations model with flat money, has a more complicated setup, and does neither fit into the one nor the more population models analyzed in section 4. Therefore we have no theoretical results for this model, but we will observe in simulations that GAs are also able to learn cyclical and stochastic equilibria. The third model describes the interaction of buyers and sellers in sealed bid double auction market. This model formally fits into the two-population framework of section 4.7 and we will be able to rely on analytical and numerical insights in our analysis.