Accuracy comparison of short-term oil price forecasting models

A forecasting model of the monthly crude oil price is investigated using the data between 1988 and 2009 from U.S. Energy Information Administration. First generalized auto-regressive conditional heteroskedasticity (GARCH) is applied to a state space model, a hybrid model (SS-GARCH) is proposed. Afterwards by computing a special likelihood function with two weak assumptions, model parameters are estimated by means of a faster algorithm. Based on the SS-GARCH model with the identified parameters, oil prices of next three months are forecasted by applying a Kalman filter. Through comparing the results between the SS-GARCH model and an econometric structure model, the SS-GARCH method is shown that it improves the forecasting accuracy by decreasing the index of mean absolute error (RMSE) from 7.09 to 2.99, and also decreasing the index of MAE from 3.83 to 1.69. The results indicate that the SS-GARCH model can play a useful role in forecasting short-term crude oil prices.