Compound model for two dependent kinds of claim (Q1892982)

We studied the case of a compound class of policies submitted to two kinds of claims whose yearly occurrence frequencies are a priori dependent random variables \(N_ 1\) and \(N_ 2\). This is the case, for example, with liability automobile insurance where the claims are damage only or bodily injury, or with a complementary health insurance concerning two kinds of medical acts (a consultation with a physician, radiological or biological analyses). The characteristics of the distribution of aggregate amounts of claims for the class are expressed in terms of those of the pair \((N_ 1, N_ 2)\). Whereas numerous bivariate discrete distributions are used in the statistic field, only a few of them, apart from the bivariate Poisson distribution, have been applied in the insurance field. In the present paper, apart from the standard bivariate Poisson distribution, we use the family of bivariate mixed-Poisson distributions with conditional independence (notably Poisson-Gamma than Poisson-inverse Gaussian distribution). The advantage of using this latter family is to put the parameter's estimation problem in the univariate mixed Poisson field. The theoretical results are checked with recent data concerning natural events insurance in the United States followed by that of liability automobile insurance in France.