Bayesian Premium Rating with Latent Structure (Q4455894)

This paper deals with a Bayesian approach to the non-life risk premium rating, based on hierarchical models with latent variables for both claim frequency and claim size. Inference is based on the joint posterior distribution and is performed by a Markov chain Monte Carlo. The authors assume that the number of claims follows a Poisson distribution and that, conditional on the number of claims, the claim size follows a Gamma distribution. A new class of hierarchical generalized linear models with latent variables is presented. Estimation is done using Markov chain Monte Carlo techniques to sample from the joint posterior distribution of all parameters including the latent variables. The latent variables are not only substitutes for unavailable but relevant covariate information. They also favour the sharing of risk among policyholders characterized by the same level of such latent variable. To enhance the mutual and social benefit of insurance the latent variables collect a part of the individual risk of each policyholder in a certain homogeneous group and distribute it over the same group. The authors demonstrate that the spatial interacting Markov random field latent variables are useful for the risk premium estimation.