CES-509 Market Microstructure: Can Dinosaurs Return? A Self-Organizing Map Approach under an Evolutionary Framework

This paper extends a previous model where we examined the markets' microstructure dynamics by using Genetic Programming as a trading rule inference engine, and Self Organizing Maps as a cluster- ing machine for those rules. However, an assumption we made in that model was that clusters, and thus trading strategy types, had to remain the same over time. This assumption could be considered unrealistic, but it was necessary for the purposes of our tests. For this reason, in this paper we extend this model by relaxing this assumption. Hence our framework does not lie on pre-speci?ed types, nor do these types remain the same throughout time. This allows us to investigate the dynamics of market behavior and more speci?cally whether successful strategies from the past can be successfully applied to the future. In the past, we investigated this phenomenon by using a simple ?tness test. Neverthe- less, a drawback of that approach was that because of its simplicity, it could only o?er limited understanding of the complex dynamics of mar- ket behavior. With the extended model we can thus have a more realistic view of the markets and hence draw safer conclusions about their behav- ior. Empirical results show that market behavior is non-stationary, and thus agents' strategies need to continuously co-evolve with the market, in order to remain eff?ective.