Ratemaking by geographical area in the Boskov and Verrall model: a case study using Belgian car insurance data

This paper uses a method proposed by Boskov & Verrall (1994) for premium rating by postcode area. The aim is to analyze geographical variation in claim frequencies in order to estimate the local risk in each geographical area and to build homogeneous rating regions. The method accounts for spatial correlation in an hierarchical Bayesian framework and uses computer-intensive MCMC methods for statistical inference.