Applying genetic algorithms with speciation for optimization of grid template pattern detection in financial markets

Abstract This paper presents a new computational finance approach. It combines a grid pattern recognition technique allied to an evolutionary computation optimization kernel based on Genetic Algorithms, creating a dynamic way to attribute a score to the signal that takes volatility into consideration and normalizing the pattern detection by fixing the grid size with the ultimate goal of reduce risk and increase profits. For pattern matching, a template based approach using a fixed size grid of weights is adopted to describe the desired trading patterns, taking not only the closing price into consideration, but also the variation of price in each considered time interval of the time series. The scores assigned to the grid of weights will be optimized by the Genetic Algorithm and, at the same time, the genetic diversity of possible solutions will be preserved using a speciation technique, giving time for individuals to be optimized within their own niche. The adoption of this approach has the goal of reducing the investment risk and check if it outperforms similar approaches. This system was tested against state-of-the-art solutions, namely the existing adaptable grid of weights and a non speciated approach, considering real data from the stock market. The developed approach using the grid of weights had 21.3% of average return over the testing period against 10.9% of the existing approach and the use of speciation improved some of the training results as genetic diversity was taken into consideration.