This paper develops a regime-based framework for modeling county crop yields that explicitly accounts for rare catastrophic years and examines its implications for crop insurance rating. Yields are assumed to arise from two latent states of nature: a normal state, in which yields fluctuate around a deterministic trend, and a catastrophic state, in which yields experience substantial downward shocks. We apply the model to county-level corn and soybean yields for all 99 counties in Iowa and use the estimated yield distributions to derive crop insurance premiums. The model is estimated within a Bayesian hierarchical framework that incorporates spatial dependence across counties. The
empirical results show that catastrophic yield realizations occur with non-negligible probability and are associated with sizable losses. The proposed model generates predictive yield densities with negative skewness and bimodality in some cases, and provides a regime-based interpretation of these distributional features. The results show that the proposed model can identify relatively profitable insurance policies and, in several specifications, is more efficient than the benchmark approach.
