Comparison of vector time series and ANN techniques for forecasting of WTI oil price

Abstract

Forecasting the changes of oil prices is of critical importance for authorities and plays a significant role in the dynamic global economy. This paper employs two prediction tools, including econometric and artificial neural network (ANN) models, for forecasting the price of WTI oil to conduct a comparative study. Forecasts from vector time series (vector autoregressive (VAR) and vector error correction (VEC) models) as econometric models are compared with those from ANN model based. For developing the models, 144 monthly data (2000/1-2011/12) comprising monthly oil price, production, reserves, fright rate, world GDP and inflation is applied. To obtain the best model for forecasting the oil price, various models comprising different combinations of training and testing dataset are tested. For achieving the aim, the most appropriate network structure and model is determined based on prediction accuracy and performance. The performance indexes for evaluating the VAR and ANN models contain of RMSE (Root Mean Square error), MAE (mean absolute error), and coefficient of determination (R2) criteria, indicate that ANN yields better results.