Abstract:
The estimation of loss distribution is always a big issue for insurance companies. Several parametric or nonparametric methods are introduced to fit loss distributions. In this paper, we propose a method by combining both parametric and nonparametric methods to solve this problem. We first determine the threshold between large and small losses by observing the graph of mean excess function, then use the generalized Pareto distribution, the parametric method, to fit excess data, and use kernel density estimation, the nonparametric method, to fit the distribution below threshold. Finally, we use a data set about Chinese annual earthquake loss to compare this method with other existing methods.