Modelování četnosti pojistných škod v havarijním pojištění

Abstract

The thesis is focused on modeling the claim frequency for motor hull insurance. For modeling the frequency is used negative-binomial model in the thesis. The parameters are estimated using the method of maximum likelihood at significance level of 0.05. Most of the estimated coefficients were statistically significant. We have found that the negative binomial distribution is appropriate for modeling the claim frequency in vehicle insurance. We also analyzed the effect of categorical variables on the frequency of accidents. Continuous variables were categorized (converting into categorical factors) and the model was estimated only with categorical variables. Both estimated models were compared using predicted deviance residuals and smoothed Pearson residuals. After the comparison of these two models, we showed that the model with the categorical predictors is better fit model for modelling the frequency of claims in the vehicle insurance. The non-linearity between rating factors in model is showed by categorical variables. Models were estimated on the insurance portfolio of insurance company during the years 2005–2010.