Zero-inflated claim count modeling and testing – a case study

In this paper the application of parametric count data models in claim counts modeling is investigated. Insurance portfolios have a very specific characteristic, i.e. for many policies there are no claims observed in the insurance history for a given period of time. As the zero-inflation and over-dispersion effects are a common situation in insurance portfolios, three models: zero-inflated Poisson (ZIP), zero-inflated negative binomial (ZINB) and zeroinflated generalized Poisson regression (ZIGP) are tested against the classic Poisson model. The 4-step procedure for modeling zero-inflation effect is proposed. This procedure is applied in the case study. For all calculations the R CRAN software was used.