报告题目:Individual Bodily Injury Loss Assessment in Auto Insurance: A Group-Adaptive Weighted Quantiles Approach
报告时间:2026年11月23日 20:00-21:00
报告地点:腾讯会议828-369-907
报告摘要:In non-life insurance claims modeling, bodily injury (BI) claims materially affect auto insurers’ case reserving, claim triage, and tail-risk monitoring, yet their loss assessment is challenging due to heavy-tailed losses, claim-group heterogeneity, and unstructured medical information. We propose a quantile-based actuarial modeling framework, termed Group-Adaptive Weighted Quantiles (GAWQ), for individual BI loss assessment. The framework first estimates multiple conditional quantiles via gradient-boosted regression trees fitted with the pinball loss, and subsequently combines them through group-specific beta weight functions tailored to outpatient, inpatient, and disability claims. To capture clinical severity signals, medical diagnostic text is represented through domain-specific semantic embeddings and incorporated as additional covariates. Using 80,426 settled BI claims from a major Chinese property and casualty insurer during 2021–2025, GAWQ reduces out-of-sample mean squared error by 43.0% and mean absolute error by 27.4% relative to a text-augmented generalized linear model, while increasing the Gini index from 0.677 to 0.750. The estimated group-specific quantile weights provide transparent actuarial interpretations across claim groups, and ablation results show that medical-text features contain substantial predictive information.
报告人简介: 姚经,苏州大学杰出人才特聘教授,金融工程研究中心副主任,博士和硕士生导师,入选国家人才工程计划,江苏省特聘教授,重庆“巴渝学者”讲座教授等。主要研究方向为量化金融风险管理,随机优化和控制,统计机器学习方法及应用等。