2015

Research / 2015

Research

Prediction of Insurance Loss Based on Zero-Adjusted Random-Effect Regression Models

2019.06.06

Shengwang Meng, Zhengxiao Li

【Abstract】

In classification ratemaking of general insurance, the insurance company mainly focuses on predicting the aggregate claim losses of polices. The main method of predicting aggregate loss is establishing Tweedie regression model. However, Tweedie regression model may produce large deviations when predicting zero-claim numbers, as the zero-claim has a very high probability, far greater than the probability at zero in Tweedie distribution. Based on the assumption that aggregate insurance loss follows zero-adjusted distribution, zero-adjusted regression model can be established. If a categorical variable that contains too many levels is treated as random effect ,and also introduces quadratic function of continuous variables in the regression ,the accuracy of prediction can be further improved. Based on the empirical study of motor third-party liability insurance loss, several regression models with random or fixed effects are compared under different distributions, the empirical results shows that zero-adjusted random-effect regression models have superiority in predicting insurance loss.

【Keywords】

zero-adjusted distribution, Tweedie distribution, random effect, aggregate loss