时 间:2025-06-06 (周五)10:30 - 12:00
报告人:刘旭上海财经大学报教授
地 点:普陀校区理科大楼1514
主持人:郁淼淼 华东师范大学助理教授
摘 要:
Heavy-tailed responses and strong dependence among covariates are common in economic and financial data, posing significant challenges for reliable inference, especially in high-dimensional settings. Quantile regression (QR) is a powerful tool for addressing such challenges, offering robust inference and providing insight across the entire conditional distribution of the response. In this paper, we consider the hypothesis testing problem for the significance of coefficients in high-dimensional QR in the presence of high-dimensional control factors. By incorporating a consistent estimator of the high-dimensional nuisance parameters and the projection score function of QR, we introduce a robust testing method that not only controls the Type I error well and significantly improves power performance but also remains valid when the parameters of interest are highly correlated with the control factors. Our theoretical framework avoids the estimation of the error density and the inversion of the covariance matrix and permits the estimation of the projection matrix from the same sample, thereby eliminating the need for additional data or sample splitting. The proposed test statistics are applicable to high-dimensional covariates of interest and control factors, while maintain robustness to heavy-tailed responses, and exhibit asymptotic normality under both the null and alternative hypotheses. Extensive numerical studies are conducted to evaluate the finite-sample performance of the proposed testing approach. We illustrate the proposed methods in financial econometrics by empirically analyzing the Shanghai Stock Exchange.
报告人简介:
刘旭博士是上海财经大学统计与管理学院常任教授。2011-2016年分别在美国西北大学和密歇根州立大学从事博士后研究。近年来主要研究兴趣为生成式学习、迁移学习、以及高维数据分析。在国际顶级统计期刊包括JASA,Biometrika,JoE,JMLR等发表30多篇论文。现担任International Journal of Organizational and Collective Intelligence (IJOCI) 和 Journal of Statistical Theory and Applications (JSTA)的副主编。主持两项国家自科面上项目、参与一项国家自科重点项目子课题。获得上海市第十六届哲学社会科学优秀成果二等奖。