Spatial copula modeling of extreme crop insurance claims in Brazil

Abstract

We use robustly estimated spatial R-vine copula models to assess spatial dependencies among extreme crop insurance claims. A truthful predictive model for simultaneous extreme losses is derived based on the linear structure found between copula parameters and distances between groups. Findings are compared to those from classical estimation of pair-copulas. Univariate fits of the excess-losses are based on the Generalized Pareto distribution. The dependence implied by the spatial component is captured by the Gumbel copulas in Tree 1, whereas a few atypical points are handled by robust inference which reveals that the influence of joint multivariate extreme outliers can not be neglected. Our findings are useful for crop insurance firms as well as for local authorities trying to minimize the effects of the natural disasters.