基于两阶段分位数随机森林的非完全支付赔款预测  

Prediction of Not-fully-paid Claims Based on Two-stage Quantile Random Forest

作　　者：李政宵 刘钰冰 LI Zheng-xiao;LIU Yu-bing

机构地区：[1]对外经济贸易大学保险学院

出　　处：《保险研究》2024年第10期59-72,共14页Insurance Studies

基　　金：国家自然科学基金面上项目(72271056);对外经济贸易大学中央高校基本科研业务费专项资金(CXTD13-02)。

摘　　要：在健康保险精算研究中,准确度量保单非完全支付赔款能够帮助保险公司有效识别告知不实、保险欺诈、既往症和保险责任免除等问题,从而降低保险公司的财务负担。本文以短期健康险中的非完全支付赔款为研究对象,提出了“非完全支付概率-金额”两阶段框架,并运用随机森林分位数分类器和分位数回归森林对非完全支付概率和未付金额进行预测和风险因素分析,从点预测精度、区间预测精度和风险区分能力三个角度论证了两阶段分位数随机森林的优越性。本文对影响非完全支付概率和金额的风险因素的变量重要性排序进行分析表明,保单的疾病类型、医疗费用支出和保单生效至出险间隔时间是重要的风险因子。此外,两阶段分位数随机森林还能更加清晰地识别风险因子对非完全支付保险成本的非线性效应和交互效应,为保险公司的财务成本控制提供一定的理论参考。In actuarial research on health insurance,accurately measuring the not-fully-paid claims can help insurance companies effectively identify issues such as misrepresentation,insurance fraud,pre-existing conditions,and exemption from insurance liability,thereby reducing the financial burden on insurance companies.This paper focuses on the underpayed claims in short-term health insurance and proposes a two-stage framework of"underpayment probability-amount."The framework utilizes random forest quantile classifiers and quantile regression forests for predicting the probability and amount of underpayment.The superiority of the two-stage quantile random forest is demonstrated from the perspectives of point prediction accuracy,interval prediction accuracy,and risk discrimination ability.Analysis based on the importance ranking of risk factors affecting the probability and amount of underpayment reveals that the type of disease covered by the policy,medical expense expenditure,and the time interval between policy effective date and claim occurrence are important risk factors.Furthermore,the two-stage quantile random forest can more clearly identify the nonlinear and interactive effects of risk factors on the underpayment of insurance costs,providing theoretical reference for insurance companies to control financial costs.

关 键 词：两阶段分位数随机森林 非完全支付赔款 短期健康险 保险精算 

分 类 号：F842.6[经济管理—保险]