The Combination of Investment Strategies Using the Replicator Equation.

Abstract

Recently, there has been a lot of interest in applying artificial intelligence to predicting future stock prices. One of the studied approaches makes use of multi-agent models to make those predictions. The current study is built upon this previous research that uses multi-agent prediction models. It answers the question whether a dynamic multi-agent model using the replicator equation to evolve a population of trading strategies over time can be used to combine stock index trend predictions. Of course, as a first step such a model is build. This model consists of several different trading strategies that all make up a certain part of the population. The replicator equation is used to evolve the proportions of those individual strategies in the population. Thereafter, the models ability to predict future trends of the S&P500 is tested. The model is not shown to be able to make good predictions. It is concluded that the replicator dynamic is too slow for this purpose. Thereafter, a second experiment is conducted with a revised prediction model. In this experiment largely the same model and method is used as in the first one. However, the revised model makes use of the ReDVaLeR algorithm to evolve the system, instead of the replicator equation. No good prediction performance is found from this revised model as well. Therefore, no evidence is found that it is useful to combine multiple investment strategies using a dynamic multi-agent model based on the (modified) replicator equation.