Abstract :
Crude oil price volatility dynamics are governed by nonlinear and chaotic behaviour. This paper presents and compares the performance of four hybrid systems used to estimate and predict crude oil price volatility data. A GARCH family model is employed to estimate oil price volatility data and the Elman artificial neural network (ENN) system is used to model and predict the obtained data. Indeed, unlike previous studies found in the literature, recurrent artificial neural networks are chosen in this paper to model and predict future crude oil price volatility data estimated by GARCH family models since they are nonlinear systems capable of learning noisy and nonstationary data. In particular, four hybrid systems are tested and compared; including the GARCH-ENN, EGARCH-ENN, APARCH-ENN, and TARCH-ENN system. Using Brent crude oil price data, the obtained out-of-sample simulation results indicate that all hybrid systems provide very accurate forecasts of Brent future volatility. In addition, they show evidence of the superiority of the GARCH-ENN system over the EGARCH-ENN, TARCH-ENN, and APARCH-ENN systems. The presented four hybrid systems achieved very low forecasting errors. Thus, they could be effective in oil industry management and applications.
