Selection, Genetic Drift, and Technical Trading

Abstract

A mutation operator modification suggested in chapter 7 resulted in drastically different model behavior. Instead of discovering profitable technical trading possibilities, most agents voluntarily gave up the use of their classifier system. On the one hand, this model behavior prompted me to analyze the implications for wealth accumulation by agents. On the other hand, the model behavior was so radically different from the original results that I kept designing tests and procedures that could bolster my argument of an inherent uselessness of the classifier system. In the course of these tests, the robustness of the zero-bit solution proved to be very robust. In fact, it still emerged when I did not expect it to under the given parameter combinations. Before presenting this simulation evidence in section 9.3, I will discuss a misconception about the proper way of detecting technical trading in the SFI-ASM. The main part of this chapter will then identify and discuss the zero-bit solution as a side effect of genetic drift. In population genetics and in evolutionary programming, genetic drift is known to affect the genetic variation in finite population sizes even in the absence of any selection pressure. By transferring the longstanding selectionist-neutralist debate from evolutionary biology to the field of genetic algorithms, I will separate the effects caused by genetic drift and those caused by selective forces in the two SFI-ASM versions.