Comparison of SVR and GARCH Models in Forecasting Oil Price Volatility

Many researchers are interested in forecasting oil prices and understanding its behavior over time, given the importance of oil as a strategic factor in many industries and markets. Nowadays, non-classic approaches such as Artificial Intelligence and Machine Learning methods are used in modeling and forecasting complex systems. Recently, one of the Machine Learning techniques, entitled Suort Vector Machine, is extensively used in classification, regression and time series forecasting. This method, which is based on statistical learning theory, has attracted the interest of many researchers. This is due to some of its properties such as: simple geometric interpretation, unique general solution, ability to model non-linear behavior and minimization of generalization error instead of training error. In this paper we study and evaluate the ability of SVR in predicting oil price volatility in Iran. We then compare the performance of this model with the GARCH models. In this research, we have used monthly prices of Iranian crude oil for the period April 1981 to December 2011. The results indicate that the SVR method is superior to the GARCH models based on MSE, MAE, NMSE and TIC criteria.