Forecasting oil Prices: can large BVARs help?

Large Bayesian Vector Autoregressions (BVARs) have been a successful tool
in the forecasting literature and most of this work has focused on macroeconomic
variables. In this paper, we examine the ability of large BVARs to forecast the real
price of crude oil using a large dataset with over 100 variables. We find consistent
results that the large BVARs do not beat the BVARs with small and medium sizes
for short forecast horizons but offer better forecasts at long horizons. In line with
the forecasting macroeconomic literature, we also find that the forecast ability of the
large models further improves upon the competing standard BVARs once endowed
with flexible error structures.