报告一

题目：New Ideas for Risk Measurers in Insurance and Finance

汇报人：  虞克明

会议时间：2024年7月22日(周一)  9：40-10：40

地点：综合楼644会议室

报告人简介：虞克明，英国伦敦布鲁内尔大学统计学与数据科学讲习教授(Chair Professor)、 数学学科研究影响中心主任；英国皇家统计学会会士、英国社科基金 (ESRC) 评审专家成员、英国自科基金 (EPSRC)评审专家成员 、欧洲科学基金(ESF) 评审专家成员。目前是《Journal of the American Statistical Association, A&CS》副主编，也担任过《Journal of the Royal Statistical Society-A》等多家国际SCI、SSCI期刊的副主编。目前他主要从事回归分析、 非参数统计、 机器学习、 贝叶斯推断、大数据及非常小的数据分析等方面的理论和方法研究，是贝叶斯分位数回归方法的开拓者，先后在《Journal of American Statistical Association》、《Journal of the Royal Statistical Society: Series B》、《Journal of Econometrics》、《Journal of Business & Economic Statistics》、《Bernoulli》等统计学顶级刊物上发表论文150多篇。

摘要： Due to the fact that tail behaviours are very important in risk measurement and have the most impact but are the most difficult to forecast. While  quantilebased  and expectile-based risk measurers such as VaR  (Value-at-Risk) and ES (Expected Shortfall) are widely used in insurance and finance,   this talk tries to introduce a new tail measurer, which potentially  includes  both quantile an expectile  as its special cases.At the same time, we introduce  √n-consistent  Linearextremile regression and its semi-supervised learning.

报告二

题目：Linear extremile regression and its  semi-supervised learning

汇报人：  虞克明

会议时间：2024年7月22日(周一)  10：50-11：50

地点：综合楼644会议室

摘要： The extremile (Daouia et al., 2019) is a novel and coherent measure of risk, determined by weighted expectations rather than tail probabilities. However, existing studies on extremile are difficult to generalize to linear models due to the challenge of obtaining a √n-consistent estimator of unknown parameters in the model. To address this issue, this article proposes a new definition of linear extremile regression and its estimation method. In many practical data analyses, linear regression models may not be accurate. Therefore, we have developed a semi-supervised framework for the proposed linear extremile regression using unlabeled data, which is often present in practical problems. Both simulations and real data analyses have been conducted to illustrate the finite sample performance of the proposed methods.