Predicting stock market time series using evolutionary artificial neural networks with hurst exponent input windows

Abstract

Predicting stock market time series is a challenging problem due to their random nature, non-stationarity and noise. In this study, we introduce an enhanced evolutionary artificial neural network (EANN) model to meet this challenge. Here, fractal analyses based on Hurst exponent calculations are used to characterize the time series and to identify appropriate input windows for the EANN. We investigate the efficacy of the model using closing price time series for a suite of stocks listed on the SPI index on the Australian Stock Exchange. The results show that Hurst exponent configured models out-perform basic EANN models in terms of average trading profit found using a simple trading strategy..