The Santa Fe Artificial Stock Market Re-Examined - Suggested Corrections

Abstract

This paper rectifies a design problem in the Santa Fe Artificial Stock Market Model. Due to a faulty mutation operator, the resulting bit distribution in the classifier system was systematically upwardly biased, thus suggesting increased levels of technical trading for smaller GA-invocation intervals. The corrected version partly supports the Marimon-Sargent-Hypothesis that adaptive classifier agents in an artificial stock market will always discover the homogeneous rational expectation equilibrium. While agents always find the correct solution of non-bit usage, analyzing the time series data still suggests the existence of two different regimes depending on learning speed. Finally, classifier systems and neural networks as data mining techniques in artificial stock markets are discussed.