《保险研究》20190706-《提升算法对传统车险索赔频率建模模型的改进——基于我国五省交强险保单数据》（张连增、申晴）

［摘   要］广义线性模型作为非寿险定价的经典模型,在非寿险定价中得到了广泛的应用。近年来,以提升算法为代表的机器学习算法在保险领域取得了很好的效果,为保险产品定价提供了一种新的选择。本文将提升算法思想分别融入到回归树模型和广义线性模型(GLM)中去,用得到的新模型对我国车险索赔频率进行预测建模分析,并与传统的回归树模型和GLM进行比较。结果表明,加入提升算法后传统车险索赔频率建模模型的效果得到了很大的改善,并且在不存在过拟合的前提下,随着模型深度和迭代次数的增加,模型的效果也在不断优化。

［关键词］提升算法；回归树模型；广义线性模型；交强险；索赔频率

［基金项目］本文得到国家自然科学基金(No.71401041)、中国特色社会主义经济建设协同创新中心资助。

［作者简介］张连增,南开大学金融学院教授,博士生导师,研究方向：精算与风险管理、机器学习，E-mail:zhlz@nankai.edu.cn；申晴,南开大学金融学院博士研究生,研究方向：精算与风险管理、机器学习。

Improvement of the Traditional Auto Insurance Claims Frequency Model by Boosting Algorithm—Based on the Traffic Compulsory Insurance Data in Five Provinces of China

ZHANG Lian-zeng,SHEN Qing

Abstract：As a classic model of non-life insurance pricing,generalized linear model has been widely used in non-life insurance pricing. In recent years,the machine learning algorithm represented by Boosting algorithm has achieved good results in the insurance field,providing a new choice for insurance product pricing. In this paper,the Boosting algorithm was integrated into the regression tree model and the generalized linear model (GLM) respectively. The new model was used to forecast and analyze the claim frequency of China’s automobile insurance,and compared with the traditional regression tree model and GLM. The results show that the effect of the traditional model has been greatly improved by adding Boosting algorithm. Without over-fitting,as the increase in the depth of the model and the number of iterations,the effect of the model is also continuously improved.

Key words：boosting algorithm; regression tree model; generalized linear model; traffic compulsory third-party liability insurance; claim frequency