Accuracy of the real crude oil price forecast for different specification of VAR models

The aim of the paper is to examine the forecasting ability of the real price of crude oil Brent. In order to forecast the crude oil price a large set of predictors including a variable describing a real and financial process of economy; prices of other commodity and supply of crude oil are collected. The analysis follows a recursive scheme and forecasts are generated for the period between January 2005 and October 2014. In the study all possible combinations of predictors are used for different specification of four-dimensional VAR models. Forecast accuracy of VAR models is compared with the na ve forecast. The results obtained indicate that, at short horizons, certain models generate more accurate forecasts than the benchmark models. The comparison of various specifications of the VAR models reveals that the most accurate forecasts are generated by the VAR(2) models.