A Dual Generalized Long Memory Modelling for Forecasting Electricity Spot Price: Neural Network and Wavelet Estimate

Abstract: In this paper, dual generalized long memory modelling has been proposed to predict the electricity spot price. First, we focus on modelling the conditional mean of the series so we adopt a generalized fractional k-factor Gegenbauer process ( k-factor GARMA). Secondly, the residual from the k-factor GARMA model has been used as a proxy for the conditional variance; these residuals were predicted using two different approaches. In the first approach, a local linear wavelet neural network model (LLWNN) has developed to predict the conditional variance using two different learning algorithms, so we estimate the hybrid k- factor GARMA-LLWNN based backpropagation (BP) algorithm and based particle swarm optimization (PSO) algorithm. In the second approach, the Gegenbauer generalized autoregressive conditional heteroscedasticity process (G-GARCH) has been adopted, and the parameters of the k-factor GARMAG- GARCH model have been estimated using the wavelet methodology based on the discrete wavelet packet transform (DWPT) approach. To illustrate the usefulness of our methodology, we carry out an empirical application using the hourly returns of electricity prices from the Nord Pool market. The empirical results have shown that the k-factor GARMA-G-GARCH model has the best prediction accuracy in terms of forecasting criteria, and find that this is more appropriate for forecasts.