Rating Area-yield Crop Insurance Contracts Using Bayesian Model Averaging and Mixture Models

Abstract

Given increasing interest in area-yield crop insurance, many methods to estimate crop yield densities have been presented in the literature. Most of these methods are of parametric form, such as Normal, Beta, or Normal mixture. All previous research choose what they believe to be the single best method, which necessarily fails to take into account model uncertainty. This is problematic because any given model is only true with some level of probability less than one. Inference based on this single model methodology may lead to biased and inaccurate estimation results. An estimation method rooted in the Bayesian paradigm is proposed in this thesis. The proposed method, employing Bayesian Model Averaging and mixture models, takes into account model uncertainty and shows strong performance in simulations. In addition, this thesis considers a roughness penalty for the BIC in the Bayesian model averaging. The methodology is applied to rating area-yield crop insurance contracts for corn and soybean in Iowa, U.S.