Modelowanie liczby szkód w ubezpieczeniach komunikacyjnych w przypadku występowania dużej liczby zer
Zero-inflated claim count modeling in automobile insurance. Case Study

Summary/Abstract: An important problem in the ratemaking process is to model the number of claims. Insurance portfolios are characterized by the fact that a significant number of policies have zero claims (called zero-inflated (ZI) effect). Thus, the classical Poisson regression does not give satisfactory results for the claim count with ZI effect. This paper presents a modified Poisson regression taking into account ZI effect. There is a ,,double” problem of the selection of variables into the model in this type of regression. One set of independent variables affects the number of accidents and the second set of variables (which may be the same) affects the occurrence of zero claims. In the paper we present a procedure for the selection of these two sets of independent variables in the claim count estimation. In the investigated case study, the automobile insurance dataset is taken from the literature [de Jong et al. 2008]. All calculations are based on a package {pscl} from R CRAN software.

• Details
• Contents