Minority Game Data Mining for Stock Market Predictions

The Minority Game (MG) is a simple model for understanding collective behavior of agents in an idealized situation for a finite resource. It has been regarded as an interesting complex dynamical disordered system from a statistical mechanics point of view. In previous work, we have investigated the problem of learning the agent behaviors in the minority game by assuming the existence of one “intelligent agent” who can learn from other agent behaviors. In this paper, we propose a framework, Minority Game Data Mining (MGDM), that assumes the collective data are generated from combining the behaviors of variant groups of agents following the minority game. We then apply this framework to real-world time-series data analysis by testing on a few stocks from the Chinese market and the US Dollar-RMB exchange rate. The experimental results suggest that the winning rate of the new model is statistically better than a random walk.