Research On The Application Of Double Generalized Linear Models In The Auto Insurance Ratemaking

In recent years, with the rapid development of Chinese economic and sustainable growth of automobile ownership, the vehicle insurance business whose market penetration and coverage has further expansion,it's market share and premium income is on the top in the property insurance. The vehicle insurance is the biggest business in non-life insurance market, which has huge impact on property insurance companies. However, due to the domestic insurance market construction time is shorter, not yet mature, so the vehicle insurance is still facing some difficulties in the development process. How to price the vehicle insurance products scientifically and fairly is one of the problems to be solved.Among the many methods of auto insurance pricing, Generalized Linear Models (GLM) is the most commonly used at this stage. As an extension of classical linear regression models, GLM has great progress in some aspects like theoretical framework, model structure and model diagnosis or so, which is used widely. But in the actual modeling process,GLM still has some limitations, for example, the assumption that dispersion parameter is a fixed constant limits the further promotion of GLM in auto insurance pricing. As an extension model of GLM, the Double Generalized Linear Models (DGLM) provides a new thought for solving the question above. DGLM retains the basic frame of GLM,relaxes the hypothesis of dispersion parameter in GLM, and allows to model the mean and dispersion parameter of dependent variable at the same time, that improves the adaptability of the model.Based on the existing research results, this paper further discussed the application of DGLM in the auto insurance ratemaking, the main work was as follows: Firstly, we sorted out the related theories and analyze the rationality of applying DGLM to auto insurance pricing;Secondly, in the empirical study, on a real dataset of automobile insurance loss, we used four variables, kilometers, bonus, makes and zone, as the rate factors, applied DGLM establish models of auto insurance ratemaking. We compared our model with GLM under assumptions of common distributions from three aspects: the significance of the parameters, the fitting effect of model and the relative rates. Results showed that the rate structure obtained by DGLM is more reasonable and practical for this dataset; Finally, according to the conclusion of empirical analysis and the actual situation of domestic ratemaking work, some suggestions on how to improve the scientificity of auto insurance pricing were putted forward: 1.Focus on data accumulation, establish and improve the claims experience database;2.Rationally develop the principles and objectives of auto insurance ratemaking; 3.Comprehensive use of a variety of pricing methods and innovate ratemaking mechanism.