Forecasting Global Energy Commodity Prices: A Systematic Approach to Forecast Combination

Abstract

We investigate the accuracy of different forecasts of the global real price of oil and natural gas, on a monthly basis. In particular, we investigate whether combinations of forecasts including newer models are more accurate than those using only older state-of-the-art models from the literature.

Our new model combinations outperform previous state-of-the-art models when predicting both real oil and natural gas prices. For oil, the new combinations reduce MSPE ratios relative to the no-change forecast by an average of 4 percentage points compared to the old combinations. For natural gas, the reductions are on average 6 percentage points. Additionally, we find that using equal weights provide the highest accuracy for real oil price forecasts. On the other hand, for the real price of natural gas, the lowest MSPEs are achieved with performance-based weights derived from forecast errors from the past 24 months.

We also examine how using different indicators of global activity in our VAR and BVAR models impacts the combined forecasts. We find that the most accurate forecasts are obtained using the real commodity price factor (RCPF) for the real price of oil and world industrial production (WIP) for the real price of natural gas. Moreover, we find that using production-based VAR and BVAR models in our combination has the overall best performance for the real price of oil. In contrast, using consumption-based VAR and BVAR models is best for the combined forecasts of the real price of natural gas.