报告题目：Individual Bodily Injury Loss Assessment in Auto Insurance: A Group-adaptive Quantile Combination Approach
报告时间：2026年10月28日10：30
报告人：姚经 教授
报告地点：阅江楼354
个人简介：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，入选国家人才计划，江苏省特聘教授，重庆“巴渝学者”讲座教授，苏州市优秀人才等。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。
报告摘要：Bodily injury (BI) claims are a major source of tail risk in auto insurance, but individual-level BI loss assessment has received limited attention in the actuarial literature. BI losses are difficult to predict because they are heavy-tailed, heterogeneous across claim types, and strongly influenced by medical information often recorded as unstructured diagnostic text. We propose a group-adaptive quantile combination framework for individual BI loss assessment. The method estimates multiple conditional quantiles of BI losses using gradient-boosted decision trees with the pinball loss and then combines these quantiles using entropy-regularized weights learned separately for outpatient, inpatient, and disability claims. We also incorporate medical diagnostic text through semantic embeddings from a domain-specific Chinese medical large language model. Using 80,426 individual BI claim records from a major Chinese property and casualty insurer, we find that the proposed framework substantially improves out-of-sample performance. Relative to a generalized linear model, it reduces mean squared error by 42.5% and mean absolute error by 28.7%, while increasing the Gini index from 0.677 to 0.747. The gains are especially pronounced for high-severity claims. The learned weights are actuarially interpretable: outpatient claims rely mainly on central quantiles, inpatient claims combine central and upper-tail information, and disability claims place dominant weight on the highest quantile. Ablation analysis shows that medical diagnostic text provides substantial incremental predictive value. The results demonstrate how quantile-based statistical learning, group-adaptive combination, and medical-text embeddings can support case reserving, claims triage, and tail-risk monitoring for BI claims.
报告题目：Optimal investment and robust valuation of climate-related derivatives under delay factors and model uncertainty
报告时间：2026年10月28日11：00
报告人：杨阳 讲师
报告地点：阅江楼354
个人简介：杨阳，苏州大学金融工程研究中心讲师。主要研究方向为保险精算、金融数学与风险管理，相关成果发表于 Insurance: Mathematics and Economics、Journal of Mathematical Analysis and Applications、Statistical Papers 等学术期刊。
报告摘要：Weather derivatives provide a means of managing climate-related financial exposures, but their valuation is complicated due to incomplete hedging opportunities and model uncertainty. In this paper, we investigate optimal investment and robust utility indifference pricing of temperature-linked claims when capital flows and terminal utility depend on past wealth performance. We model temperature by a mean-reverting Ornstein–Uhlenbeck process and allow for correlated shocks in the weather and financial dynamics. To account for different attitudes toward model uncertainty, separate drift distortions are assigned to the shared and stock-specific risk factors. Together with history-dependent capital flows, this formulation leads to a robust stochastic control problem with delay. Under exponential utility, we analyze the associated Hamilton–Jacobi–Bellman–Isaacs equation, establish a verification theorem, and derive analytical representations of the optimal investment strategy and indifference prices for both buyers and sellers. The resulting formulas separate speculative investment from claim-related hedging demand and show that stock-specific ambiguity aversion can affect weather-claim prices when temperature and stock-return shocks are correlated. In contrast, the effects of the capital-flow parameters and the weight on historical wealth are captured by a common effective discount factor. Numerical illustrations for a call option on terminal temperature, based on calibrated weather and stock-index dynamics, examine how correlation, ambiguity aversion, and delay parameters affect investment and valuation.
欢迎感兴趣老师参加！
管理工程学院
2026.10.26
