Neural Network Prediction of Short-Term Dynamics of Futures on Deutsche Mark, Libor and S&P500

Abstract

The talk reports neural network modelling and its application to the prediction of short-term financial dynamics in three sectors of financial market: currency, monetary and capital. The methods of nonlinear dynamics, multifractal analysis and wavelets have been used for preprocessing of data in order to optimise the learning procedure and architecture of the neural network. The results presented here show that in all sectors of market mentioned above the useful prediction can be made for out-of-sample data. This is confirmed by statistical estimations of the prediction quality.