报告题目：Directional Residual-Based Recovery of Latent Functional Groups in Short Panel Models

报告人：虞克明

讲座时间：2026年11月8日（周日） 10:30

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 Royal Statistical Society-C》的编委，也担任过《Journal of the American Statistical Association, A&CS》、《Journal of the Royal Statistical Society-A》等多家国际SCI、SSCI期刊的编委。他连续入选2021、2022、2023及2025年斯坦福大学/爱思唯尔单年引用影响力排名全球前2%科学家名单。2024年，他被ScholarGPS评为高排名学者。2026年，他在Research.com评选的全球顶尖数学家排名中位列英国第196位、全球第3054位。虞教授是国际公认的贝叶斯分位数回归先驱，在统计方法论与数据科学领域具有深远影响。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。

报告摘要：

We propose a directional residual-based homogeneity pursuit procedure for varying-coefficient short panel models with individual fixed effects. The method identifies latent groups of individuals that share common coefficient functions without requiring preliminary estimation of individual-specific nonparametric trajectories. The key idea is to construct residual-score vectors from a pooled first-stage fit and exploit the associated directional information in the spline coefficient space to recover latent functional heterogeneity. This preserves directional information that is unavailable from scalar residual discrepancies and is designed for settings with a fixed time dimension. Given the estimated partition, we estimate group-specific coefficient functions through a combined adaptive SCAD procedure that distinguishes identically zero effects, nonzero constant effects, and genuinely varying effects. Under a fixed-T, large-N asymptotic framework, we establish exact recovery of the latent partition under directional separation conditions and derive oracle properties for the post-selection estimator. With an undersmoothed refitting step, we further obtain asymptotic normality for nonzero constant effects and studentized pointwise asymptotic normality for genuinely varying coefficient functions. Simulation studies demonstrate the advantages of retaining directional residual information for group recovery and functional estimation. An application to monthly observations from 21 United Kingdom weather stations identifies two latent station groups with distinct sunshine temperature relationships and heterogeneous roles of air frost and rainfall.