Application of Polynomial Neural Networks to Exchange Rate Forecasting

This research investigates the use of ridge polynomial neural network (RPNN) as non-linear prediction model to forecast the future trends of financial time series. The network was used for the prediction of one step ahead and five steps ahead of two exchange rate signals; the British Pound to Euro and the Japanese Yen to British Pound. In order to deal with a dynamic behavior which exists in time series signals, the functionality and architecture of the ordinary feedforward RPNN were extended to a novel recurrent neural network architecture called dynamic ridge polynomial neural network (DRPNN). Simulation results indicate that the proposed DRPNN offers significant advantages over feedforward RPNN and multilayer perceptron including such increment in profit return, reduction in network complexity, faster learning, and smaller prediction error.