Three Essays on New Approaches for Agricultural Crop Insurance Premium Rate Policy

Crop insurance is one option producers have to mitigate against the many types of production risk they face -- for instance due to pests, drought or excessive rain. Insurance products come in many forms, both for production and price risk, but at the core of insurance premium rate policy is calculating expected yields. There are many types of data that can be used towards this purpose: yield data, past insurance experience, weather data, and program participation. This dissertation researches three different methods which may unite these data. The first method is a heuristic Bayesian approach; the second, using Factor Analysis and Bagging techniques; the third, a new application of overlap measures.;The first approach is an extension of the heuristic Bayesian approach developed in Borman et al. (2013). Previous research indicates a need to address the weighting of extreme loss events and adoption rates in the rate making process. Priors are estimated as a function of weather and participation variables, to reflect the nature of extreme events and additional information that can be gleaned from changing levels of adoption. Weighted priors are used to determine the distribution of the loss cost ratio (the ratio of indemnity payments to liability). Implied premium rates are simulated and compared to actual rates for individual yield products. An application to Iowa, Illinois and Indiana corn and soybean business from 1981 to 2010 reveals a redistribution of premium rates across counties that may be favorable.;The second method uses a factor analysis based approach to improve the forecasting of yields to address the well known short sample problem in yield distribution estimation. I consider contiguous county crop yields of the same or highly correlated crops as supplementary predictor series for the primary county. Again, the application considers Iowa, Illinois and Indiana corn and soybeans. Bootstrap aggregation and cross validation are used determine accuracy of the original estimates. Simulation methods are used with the estimated parameters to determine premium rates for an area based yield insurance contract. Area type products are often seen as favorable, due to their reduced moral hazard and adverse selection issues. Moreover, determining county level yields in regions with little or no data has become increasingly important for the U.S. Federal Crop Insurance Program with the establishment of shallow loss coverage products in the passing of the 2014 Agricultural Act. Whether yields are treated as established or unestablished, estimated premium rates are almost always higher that actual U.S. group plan rates. This suggests that previous rating systems may be biasing area based rates downward. It also implies that area based plans may need larger subsidies to attract producers if there are less costly individual products available.;Finally, this dissertation turns its attention to a common policy question - how much difference there is between choices for yield distributions. Although there is vast literature on the use of various choices for crop yields, little work has been done on the impacts of new choices. To reveal the difference in a meaningful way, overlap measures are presented for both parametrically and non-parametrically estimated distributions. Overlap measures give a scalar value between zero and one, where zero indicates no overlap and one perfect matching. Overlap measures are computed for Kansas irrigated and non-irrigated corn yields under three parametric distribution choices (Normal,Weibull and Beta) and one non-parametric (Gaussian kernel) distribution. The results indicate that irrigated and non-irrigated crop yield distributions tend to have high polarization for all distribution selections. This result echoes previous literature that irrigated and non-irrigated crop yields should not be pooled together for estimation. Comparisons of distributions within a particular practice reveals high similarity overall, but gives the statistician incomplete understanding of where the variation occurs. Nevertheless, overlap measures are able to indicate changes in commonality between distributions even when premium rates remain statistically similar. Thus, overlap measures are a good complement to other statistics in identifying changes from various distribution selections.