Evaluating Yield Models for Crop Insurance Rating

Generated crop insurance rates depend critically on the distributional assumptions of theunderlying crop yield loss model. Using farm level corn yield data from 1972-2008, we revisitthe problem of examining in-sample goodness-of-fit measures across a set of flexibleparametric, semi-parametric, and non-parametric distributions. Simulations are also conductedto investigate the out-of-sample efficiency properties of several competing distributions. Theresults indicate that more parameterized distributional forms fit the data better in-sample due tothe fact that they have more parameters, but are generally less efficient out-of-sample–and insome cases more biased–than more parsimonious forms which also fit the data adequately, suchas the Weibull. The results highlight the relative advantages of alternative distributions in termsof the bias-efficiency tradeoff in both in- and out-of-sample frameworks.