Exploring Market Behaviors with Evolutionary Mixed-Games Learning Model

* Final gross prices may vary according to local VAT.

Get Access

Abstract

The minority game (MG) is a simple model for understanding collective behavior of agents competing for a limited resource. In our previous work, we assumed that collective data can be generated from combination of behaviors of variant groups of agents and proposed the minority game data mining (MGDM) model. In this paper, to further explore collective behaviors, we propose a new behavior learning model called Evolutionary Mixed-games Learning (EMGL) model, based on evolutionary optimization of mixed-games, which assumes there are variant groups of agents playing majority games as well as the minority games. Genetic Algorithms then are used to optimize group parameters to approximate the decomposition of the original system and use them to predict the outcomes of the next round. In experimental studies, we apply the EMGL model to real-world time-series data analysis by testing on a few stocks from Chinese stock market and the USD-RMB exchange rate. The results suggest that the EMGL model can predict statistically better than the MGDM model for most of the cases and both models perform significantly better than a random guess.