Forecasting of IBOVESPA returns using feedforward evolutionary artificial neural networks

Abstract

Facing the challenges of anticipating financial market uncertainties and movements, and the necessity of taking buy or sell decisions supported by rational methods, market traders found in statistics and econometrics methods, the base to support their decisions. In several scientific papers about forecasting financial time series, method selection keeps as central concern. This paper compares the performance of evolutionary feedforward artificial neural network (EANN) and an AR+GARCH model, for one step ahead forecasting of IBOVESPA returns. The EANN is trained by self-adapting differential evolution algorithm and AR+GARCH model is adjusted to be used as performance reference. The root mean square error (RMSE) and U-Theil inequality coefficient were used as performance metrics. Simulation results showed  EANN feedforward achieved better results, fit better and captured the nonlinear behavior of returns.