基于分位回归的风险保费预测  

Prediction of Risk Premium Based on Quantile Regression

作　　者：杨亮[1,2] 孟生旺[1,2] 

机构地区：[1]中国人民大学应用统计科学研究中心,北京100872 [2]中国人民大学统计学院,北京100872

出　　处：《统计与信息论坛》2016年第9期83-88,共6页Journal of Statistics and Information

基　　金：国家自然科学基金项目<考虑风险相依的非寿险精算模型研究>(71171193);教育部重点研究基地重大项目<随机效应模型及其在非寿险风险管理中的应用>(12JJD790025)

摘　　要：风险保费预测是非寿险费率厘定的重要组成部分。在传统的分位回归厘定风险保费中,通常假设分位数水平是事先给定的,缺乏一定的客观性。为此,提出了一种应用分位回归厘定风险保费的新方法。基于破产概率确定保单组合的总风险保费,建立个体保单的分位回归模型,并与总风险保费建立等式关系,通过数值方法求解出分位数水平,实现对个体保单风险保费的预测。通过一组实际数据分析表明,该方法具有良好的预测效果。Prediction of risk premiums is an important part in the non-life insurance premium ratemaking. During the determining the risk premium in the traditional quantile regression, quantile level is generally given in advance, which is lack of objectivity. Therefore, we propose a new method for determining the risk premium by applying of quantile regression. Firstly, determining the total risk premiums based on the probability of bankruptcy; secondly, to establish quantile regression model in individual policies, and establish relationships with total risk premium; finally, to solve for quantile level by numerical methods, obtain the individual risk premiums. Basing on a set of actual data, the results demonstrates that the method has high forecasting accuracy.

关 键 词：保费原理 风险保费 分位回归 Tweedie回归 

分 类 号：F840[经济管理—保险] O212[理学—概率论与数理统计]