6 Application
The pure premium method is widely used by actuaries for pricing insurance products. The pure premium is the average premium that must be collected to pay for losses only. It is calculated by dividing the loss by the number of exposures. In other way, the pure premium (PP) is calculated as a multiplication of the frequency
and the severity of claims. Therefore, the formula of PP is given by: PP = Frequency Severity To define the PP, an actuary needs to model separately the frequency and the severity. In the past, the one way analysis was commonly used by actuaries. A one way analysis does not take into account the effect of explanatory variables neither on frequency nor on severity. Two major problems had been encountered. First, one way analysis can be distorted by correlations between rating factors. Second, it ignores the interdependence or interactions between factors [2]. Recently, the one way analysis has been replaced by generalized linear models (GLMs). This class of models relates the response variable to the factors. They take into account correlations and interactions between factors. GLMs consist of a wide range of models. The use of each model depends essentially on the nature of the response variable. If this one is discrete, Poisson regression, negative binomial model and other discrete models can be used. If it is continuous, Gamma distribution and other continuous models can be fitted to the data. In insurance pricing, for example, the Gamma distribution and Poisson regression are widely used for modeling severity and frequency of a given risk respectively. In this paper, we will fit Poisson regression and ZIP regression to a private health insurance counts data. The database contains information about claims and policyholders. The response variable is the number of claims per year received from the insured. The covariate matrix contains socioeconomic variables describing the insured people (Table 1).