The Confidence Intervals of Preharvest State Corn Yield Forecasts

This paper examines the accuracy of preharvest corn yield forecasts from crop-weather models for major U.S Corn Belt states. Cross section time series models using dummy variables to represent state crop reporting districts are estimated via ordinary least squares for Iowa, Illinois and Indiana. The models use normalized rainfall and temperature data and measures of crop maturity for the1972-1991 period. Four successive corn crop- weather models are estimated for each state using information available on July 1, August 1, September 1 and October 1, respectively. Crop reporting district level yield forecasts and forecast errors are derived via unconditional forecast error calculatlions. These forecasts and forecast errors are then aggregated together on a monthly basis throughout the growing season to form state yield forecasts and confidence intervals. During the 1992-1994 period, forcast accuracy was poor when weather conditions were abnormal compared to the average conditions during the period of model estimation. This is illustrated by inaccurate forecasts for 1992 and 1993 for Iowa caused by abnormally wet and cool conditions. Future crop yield forecasting efforts should focus on the use of resource capture models, which hold promise of producing more accurate forecasts than crop-weather models.