Neighborhood Evaluation in Acquiring Stock Trading Strategy Using Genetic Algorithms

Abstract

We propose a new method to evaluate individuals in genetic algorithms (GAs) for algorithmic trading in stock markets. In our previous work, we presented an effective method to acquire trading strategy in stock markets. However, it had a tendency of overfitting in genetic searches. Our new approach, namely neighborhood evaluation, involves evaluation for neighboring points of genetic individuals in fitness landscape as well as themselves. Empirical results for trading simulation in the first section of the Tokyo Stock Exchange for recent eleven years show the effectiveness of the neighborhood evaluation for reducing the overfitting tendency. We discuss suitable forms of neighborhoods on the performance of the genetic searches. We also propose a new method to reduce the computational cost of our method, because the neighborhood evaluation has a disadvantage on the cost.