What Can we Learn from our Mistakes? Evaluating the Benefits of Correcting Inefficiencies in USDA Cotton Forecasts.

This study investigated the magnitude of forecast improvements resulting from correction of inefficiencies in USDA cotton forecasts over 1999/00 to 2008/09 marketing years. The aspects of forecast performance included in this study were 1) bias and trends in bias, 2) correlation between forecast error and forecast level, 3) autocorrelation in forecast errors, 4) correlation in forecast revisions. Overall the results of this study demonstrated that some corrections of forecast inefficiencies, such as correction of correlation of error with forecast levels and correlation of error with previous year’s error resulted in consistent improvement of USDA cotton forecasts, while correction for correlation in forecast revisions did not benefit the forecasts. Correction for bias yielded mixed results likely because USDA has already been applying those corrections to some of the categories and thus our analysis resulted in over-correcting. The framework developed in this study can be used by USDA and other agencies to monitor and improve the performance of their forecasts.