Estimating Corn and Soybean Farm -Yield Distributions in Illinois

Abstract

Considerable disagreement exists about the most appropriate characterization of farm-level yield distributions. This dissertation investigates corn- and soybean-yield distributions at the farm level, expanding the existing literature by providing results based on unique farm-level yield data and by suggesting methodological improvements in estimating crop-yield distributions. Before estimating the distribution of yields for a given time period, a linear trend model is identified as an adequate representation of crop yield trend at the farm level. The R-test and the Jarque-Bera test are then used to test for normality, using critical values that are obtained for the exact sizes of the yield samples. The results show that normality is rejected by a large number of fields. Based on likelihood measures, and Kohnogorov-Smirnov and Anderson-Darling goodness-of-fit statistics, the logistic, Weibull and beta distributions consistently rank higher than the normal distribution. Goodness-of-fit results also suggest the danger of using a lognormal distribution to model farm-level yields. Finally, the results of this dissertation demonstrate that large differences in farmers' expected payments from Average Production History and Crop Revenue Coverage insurance policies can result solely from the parameterization chosen to represent yields. The results indicate that the typically unexamined premise of yield distribution may lead to incorrect conclusions in other important areas of insurance research, such as policy rating and quantitative assessment of expected losses from policies.