Forecasting the Brent Oil Price: Addressing Time-Variation in Forecast Performance

This paper explores a range of different forecast methods for Brent oil prices and analyses their performance relative to oil futures and the random walk over the period 1995Q1 - 2015Q2, including periods of stable, upwardly trending and rapidly dropping oil prices. None of the individual methods considered outperforms either benchmark consistently over time or across forecast horizons. To address this instability, we propose a forecast combination for predicting quarterly real Brent oil prices. A four-model combination - consisting of futures, risk-adjusted futures, a Bayesian VAR and a DSGE model of the oil market - predicts oil prices more accurately compared to all methods evaluated up to 11 quarters ahead and generates forecasts whose performance is robust over time. The improvements in forecast accuracy and stability are noticeable in terms of both point forecasts – with MSPE gains of 23% relative to futures at the 11 quarter-ahead horizon and a directional accuracy of 70% – and density forecasts – with CRPS gains of 50% relative to futures and logarithmic score gains of 90%, both at the 7-quarter ahead horizon.