Forecasting oil prices: Can large BVARs help?

Large Bayesian vector autoregression (BVAR) is a successful tool for forecasting macroeconomic variables, but the benefits to predict crude oil prices are rarely discussed. In this paper, we test the ability of BVAR to predict the real price of crude oil using a large dataset with 108 variables, taking into account all potential error structures that could affect modeling and forecasting, and performing multivariate analysis of crude oil prices, filling in the gaps in the field. The results demonstrated that the large BVAR having an excellent out-of-sample forecast performance at long horizons. Small and medium sizes BVAR provide more accurate information for short forecast horizons. We also find that the advantages of utilizing a large dataset become more obvious when incorporating non-standard error terms.