Predictive Dynamics in Commodity Prices

Using a sample of commodity spot price indexes over the period 1947-2010, we examine predictability of commodity returns at the monthly, quarterly, and annual horizons. We establish out-of-sample predictability by means of variables such as bond spreads, growth in money supply and industrial production. Predictability is strongest for raw industrials and metals indexes and weakest for foods and textiles. Some variables, such as the inflation rate, have little or no predictive power at the monthly horizon, but appear to have stronger predictive power over commodity spot prices at the quarterly and annual horizons. Our results suggest that predictability of commodity returns from macroeconomic variables such as inflation, industrial production and money supply is stronger during economic recessions than during expansions. This finding carries over to models for realized commodity volatility, where economic state variables add predictive power to a simple autoregression mostly during recessions.