Forecasting the Real Prices of Crude Oil: A Dynamic Model Averaging Approach

Abstract

Forecasting oil prices has been of great interests for macroeconomists in the recent years. Our article contributes to this strand of the literature by using a dynamic model averaging (DMA) method to improve forecasting accuracy of real oil prices. The advantage of DMA is that the method combines models in a dynamic way using two forgetting factors to approximate the evolution of model parameters and model switching probabilities, respectively. Our empirical results show that DMA generates more accurate forecasts than the no-change forecasts at the relatively longer horizons. At a horizon of 12 months, the reduction of mean squared prediction error is as high as 30% and the accuracy of directional forecasts increases as high as 71%. It is also found that DMA performs better than Bayesian model averaging, the commonly-used mean combination of forecasts, and more sophisticated individual models such as a time-varying dimension model for the horizons of 3 and 12 months.