Smoothing in USDA’s Commodity Forecasts

This study investigates the rationality of monthly revisions in annual forecasts of supply, demand, and price for U.S. corn, cotton, soybeans, and wheat, published in the World Agricultural Supply and Demand Estimates over 1984/85 through 2011/12. The findings indicate that USDA’s forecast revisions are not independent across months, and that forecasts are typically smoothed. Adjustment for smoothing in a subset of forecasts (2002/03 – 2011/12) showed weak results: marginal improvements in accuracy were limited to wheat production and cotton production and domestic use while deterioration in accuracy was observed in all other cases. Smoothing coefficients were highly unstable over time. Case studies for corn focused on correction for a structural break and the impact of forecast size and direction, but did not lead to improvements in accuracy. Case studies for October revisions of soybean production forecasts suggest that ten year rolling estimation and correcting for outliers using leverage may help improve accuracy in the adjusted forecasts.