《保险研究》20201203-《长期护理保险的定价研究——基于XGboost算法及BP组合神经网络模型》（仇春涓、关惠琳、钱林义、王伟）

［摘   要］在我国人口老龄化日益严峻的背景下,老年人长期护理问题成为突出的社会问题,发展长期护理保险是解决这一问题的重要手段。本文分析中国中老年人口长期护理状态的影响因素,并据此研究长期护理保险的费率厘定。首先,利用中国健康与养老追踪调查(CHARLS)数据,对个体的长期护理需求等级进行划分；再根据2013年及2015年的追踪数据基于XGboost算法分析长期护理状态的影响因素；然后根据两年的追踪数据计算出两年期转移概率,并结合影响长期护理状态的因素按照基于Adaboost思想的BP组合神经网络模型计算得到分性别、分年龄的一年期转移概率；最后,根据一年期转移概率并用离散时间的多状态Markov模型进行长护险定价,得到一套分性别、分年龄、分初始状态的费率表。

［关键词］长期护理保险；长期护理状态；定价；机器学习

［基金项目］本文受国家社科基金重大项目“大数据背景下健康保险的精算统计模型与风险监管研究”(17ZDA091)资助。本文使用的数据来自中国健康与养老追踪调查(CHARLS),特此感谢。

［作者简介］仇春涓,华东师范大学经济与管理学部副教授；关惠琳,华东师范大学经济与管理学部硕士研究生；钱林义(通讯作者),华东师范大学经济与管理学部教授,博士生导师,上海市曙光学者；王伟,宁波大学数学与统计学院副教授。

A Research on the Pricing of Long-term Care Insurance—Based on XGboost algorithm and BP combined neural network

QIU Chun-juan,GUAN Hui-lin,QIAN Lin-yi,WANG Wei

Abstract：Under the background of population aging in China,the long-term care of the elderly has become a prominent social problem.The development of long-term care insurance is an important means to solve this problem.The paper studied the influencing factors of the long-term care status for the middle-aged and elderly population in China and the pricing of the long-term care insurance.First of all,this paper used CHARLS data to classify the level of individual long-term care needs.Secondly,based on the tracking data of 2013 and 2015,it analyzed the influencing factors of long-term care status based on Xgboost algorithm.Then,it calculated the two-year transfer probability based on the two-year tracking data,and the one-year transfer probability was calculated with BP combined neural network based on AdaBoost idea according to the factors that affected the long-term care status.Finally,using the discrete-time multi-state Markov model,the long-term care insurance pricing was carried out,and a set of rate table was obtained with gender,age and initial state.

Key words：long-term care insurance; long-term care state; pricing