Empirical Studies of Corn Yield Distribution Modeling

This research paper uses partially historical corn yields of each county over five states as training set for calibrating alternative yield models including single parametric distributions, nonparametric distributions, semiparametric distributions, mixing models, and other nontraditional models. The calibrated models are applied to predict the yield of the most recent eight years in historical yield data. The performance of alternative yield models is evaluated by the Bayesian information criterion (BIC) in the in-sample analysis and mean square error (MSE) in the out-of-sample analysis. From the in-sample and out-of-sample cases, this paper suggests that a single parametric distribution Weibull is more revealing and efficient to characterize corn yield distribution when training data set is small. The results are presented for the crop corn at the county levels.