报 告 人： 李政宵（对外经济贸易大学）

报告时间： 2026年5月12 日18:30-19:30

报告地点：#腾讯会议：362-683-9628

报告摘要：

Insurance plays a critical role in protecting individuals and businesses from weather-related hazards, which increasingly pose significant financial risks. As such, it is essential for insurers to incorporate weather exposure into pricing decisions. In this paper, we propose a varying-coefficient approach for longitudinal insurance claims data that enables insurers to leverage high-dimensional weather information in both initial and renewal ratemaking. Our framework integrates deep learning components to tackle three key challenges in using weather data for pricing. First, we use an encoder-decoder architecture to extract salient features from spatiotemporal weather sequences, resolving their structural misalignment with traditional tabular insurance data. Second, we introduce an auxiliary loss function to handle the practical issue that future weather information is unavailable at the time of pricing. Third, we design a varying-coefficient structure that preserves the interpretability of traditional actuarial models while capturing the effects of weather and its interactions with standard rating variables. We demonstrate our approach using proprietary automobile insurance data from a major insurer in China. Through extensive empirical evaluation, we show that the proposed method improves pricing accuracy and highlights the role of dynamic weather patterns in shaping insurance risk.

 

报告人简介：

李政宵，副教授，博士生导师，鲁汶大学访问学者。 现任对外经济贸易大学保险学院副院长、统计与精算系主任， 对外经济贸易大学“惠园优秀青年学者”。

担任中国现场统计研究会风险管理与精算分会理事， 中国优选法统筹法与经济数学研究会量化金融与保险分会理事。主要研究领域包括非寿险精算、精算统计模型、 深度学习与精算方法、巨灾风险与环境风险度量。近年来在 Journal of Risk and Insurance、 North American Actuarial Journal、 Insurance: Mathematics and Economics、 ASTIN Bulletin: The Journal of the IAA、 《统计研究》《数理统计与管理》《系统工程理论与实践》《保险研究》 等国内外核心期刊发表多篇学术论文， 出版学术专著一部。主持国家自然科学基金项目两项， 对外经济贸易大学创新团队项目一项。