Predicting commodity returns: Time series vs. cross sectional prediction models

Abstract

Commodity cross-sectional models based on the commodity momentum, basis, and basis-momentum factors generate superior time-series and cross-sectional commodity return forecasts compared to the historical average and time-series forecasting models that use financial, macroeconomic, and commodity-specific variables as predictors. Timing and long-short strategies based on the commodity premium forecasts from cross-sectional models achieve significant utility gains compared to strategies based on the historical average or time series predictive models’ forecasts. Our evidence is robust across many commodities and different forecasting methodologies.