| In an attempt to reduce adverse selection, moral hazard, and administrative costs the Risk Management Agency (RMA) developed area yield crop insurance products. Area yield products require RMA to accurately predict county average yields two years in advance. Using data from 105 Kansas counties for four major crops, this study evaluates four yield trending models (linear time trend, precipitation adjusted time trend, Olympic and simple average) combined with three extrapolation techniques to identify the best out-of-sample estimation model based on the lowest out-of-sample error alone. The optimal number of years of historical data (N) to use in the models was also considered. Contrary to logical assumption, a lower value of N performed better for corn, grain sorghum, and soybeans. In general, the precipitation adjusted model out performed the linear time trend. The Olympic average model performed surprisingly well, indicating that a method of eliminating outliers may provide superior results. |
