This paper documents the existence of a significant forecast error on crude oil futures, particularly evident since the mid-1990s, which is negative on average and displays a nontrivial cyclical component (risk premium). We show that the forecast error on oil futures could have been explained in part by means of real-time US business cycle indicators, such as the degree of utilized capacity in manufacturing. An out-of-the-sample prediction exercise reveals that futures which are adjusted to take into account this time-varying component produce significantly better forecasts than those of the unadjusted futures and randomwalk, particularly at horizons of more than 6 months.
Published in 2009 in: The B.E. Journal of Macroeconomics, v. 9, 1, Article 24