Energy futures prices forecasting by novel DPFWR neural network and DS-CID evaluation

Abstract

In recent years, artificial neural networks have been employed a lot in forecasting financial price series. Crude oil and natural gas play the most important role in energy markets. Besides, crude oil price fluctuations are closely linked to financial markets. A novel hybrid neural network, DPFWR neural network, is put forward in this paper. The proposed DPFWR combines double parallel feedforward neural network and wavelet analysis theory with a random time effective function. We apply the DPFWR to forecast the energy futures price time series, including WTI crude oil, Brent crude oil, natural gas, RBOB gasoline, heating oil and Rotterdam coal. In order to compare the accuracy of forecasting results, several error criteria are applied to evaluate the forecasting errors of BP, DPF, LSTM, DPFWR and SARIMA models. A new method for error evaluation, called DS-CID, is developed to evaluate the forecasting errors in an attempt to observe the superiority of DPFWR neural network. Based on the empirical analysis, the forecast performance of DPFWR can be distinguished from other models by its great accuracy in this research.