Effectiveness of artificial neural networks in solving financial time series

Abstract

This research aims to study and analyze which type of Artificial Neural Network (ANN) is more efficient and suitable in handling nonhomogenous variance for financial series. Apart from addressing the behavior and efficiency of ANN, the paper also aims to present an advanced methodology for ANN, as a replacement of GARCH and ARCH models in crisis management decision makers. The application part was applied to the Egyptian exchange market, to study the local currency exchange rate volatility (1/1/2009-4/6/2013) in order to develop a model describing those changes in the exchange rate. The research concludes that the best network type to represent such financial series is the Back Propagation. Moreover, comparing the result with general regression and probabilistic networks rendered the later two inefficient at handling such series.