4. CONCLUSIONS
An accurate insurance pricing system allows insurance companies to cover expected
losses, expenses and make adequate the provision for contingencies. The first step in auto
insurance pricing is the modeling of claim frequency, which represents an essential part for
obtaining a reasonable and equitable insurance premium.
In this paper, it was considered an analysis of the classical and mixed count data
models employed to estimate the frequency of claims made on vehicle insurance policies,
focusing on the factors used to explain the insured risk. After a distinct analysis of insured’s
age variable, we obtained five categories of age depending on different years intervals in
comparison with similar studies. This classification is used in the econometric modeling of
insurance premiums.
After testing the equidispersion assumptions of Poisson distribution, both statistics
presented in this paper reach the same conclusion, meaning the existence of overdispersion
within the studied insurance portfolio. Results of these tests showed that NB models correct
the overdispersion, providing a better fit to the data in comparison to the Poisson model.
Furthermore, the comparison of NB1 and NB2 models indicated that the last one is
preferred. By using the likelihood ratio in order to test the fit of the NB2 model, the results
suggest that this model is the most appropriate to deal with the problem of overdispersion
and to predict the claim frequency for the analyzed auto insurance portfolio.
While using Poisson and negative binomial models in the framework of GLMs, the
risk factors that appeared to explain significantly the frequency of claims was the age-group
and occupation of policyholders, the type, use and GPS device of vehicle, the bonus-malus
coefficient and duration of the insurance policy. Based on the obtained results, we observed
a decrease of claim frequency along with an increase of the insurance contracts duration,
and also an increase of the frequency of claims along with the increase of bonus-malus
coefficient. For these variables, there were obtained results which are similar with other
actuarial studies and also consistent with the reality of the studied phenomenon.
The results obtained for the three variables introduced as risk factors indicates that the
insured’s occupation and GPS device appears to be significant, while the value of vehicle
does not explain the frequency of claims. The modeling results could be considered as
interesting sugestions for the insurance companies while implementing their pricing policy.
Thus, the company could work with more age groups in order to evaluate the risk level of
each insured and implicitly to calculate the insurance premium. The insured’s occupation
represents another valid factor that could be considered by the company in order to group
the insurance portfolio in homogenous classes. Based on the GPS variable, the company
could implement some precautionary measures, suggesting the new insured to use a GPS
device. All this aspects aim at obtaining reasonable premium that corresponds to the risk
level of each insured, and therby respecting the principle of equity in insurance.
Our empirical study could be useful to the policy-makers by allowing a better control
on the insured risks and an accurate assessment of the insurance company liabilities leading
to solvency and profitability.