The Research Of Car Insurance Pricing Based On General Addictive Model

In 1980 China's insurance industry was resumed and,since then,auto insurance business has also been driven up,and in 1988 its proportion in the income of insurance industry has been to occupy the top and so far.While the loss ratios are not coordinated with the development of the premium income,it makes that the insurance company have not made a profit to date.Therefore,how to price the motor vehicle insurance products and how to make company profit,has became an issue concerned by the actuarial.The application of generalized linear models in car insurance pricing research has a long history,it's theory assumes that the influence of the independent variables on the response variable function is linear.But in the actual situation,a number of factors affecting the claims express not only a linear form.Simply using linear estimation may ignore the impact of some important variables.In order to better fit the data,it is necessary to optimize and extend model by introducing the non-parameter prediction term in general linear models,to get the general additive models,analyzing the impact of different factors on the response variable function from two aspects of the linear and nonlinear.Firstly,this paper analyzes the development present situation of the motor vehicle insurance industry in China.Then,using the punished least square method and the Bayesian method deduce respectively the parameters and non-parameters of general additive models.Finally,establishing the zero adjusted gamma models and general linear models for the car insurance loss data of our country in 2015,and analyzing the impacts of "vehicle,people and zone" on the claim numbers.The study shows that using the zero adjusted gamma distribution to fit loss data is more appropriate than some common distributions due to the excessive probability of claims,that is,the claims accumulate in zero-point.Establishing general additive model has more advantages than general linear model with the body value and owners age as the covariates of a nonlinear impact factor.