Abstract: Multilevel Factor Analysis (MFA) is a powerful analytical tool for capturing hierarchical structures in economic and financial data, as it simultaneously incorporates global factors and group-specific local factors. However, existing MFA methodologies rely on restrictive high-order moment constraints, which are frequently violated by heavy-tailed data prevalent in real-world economic and financial applications. This incompatibility results in inconsistent estimators and constrains the practical applicability of MFA. To address this critical gap, this study proposes a novel Robust Multilevel Factor Analysis (RMFA) framework under the elliptical distribution assumption. RMFA integrates the robust factor estimation method under elliptical distributions with the robust canonical correlation analysis that requires no moment restrictions, forming a unified framework for factor number determination and factor estimation in MFA with heavy-tailed data. Additionally, a novel method is proposed to distinguish the global and local factor spaces within RMFA. Theoretical analysis demonstrates that the proposed RMFA estimators achieve the same convergence rate as traditional MFA estimators, while only imposing weak moment constraints. Monte Carlo simulations further confirm that the proposed method performs excellently under heavy-tailed data scenarios and maintains robustness even when model assumptions are not fully satisfied.

报告人简介：

  涂云东，北京大学博雅特聘教授，联合受聘于光华管理学院商务统计与经济计量系和北京大学统计科学中心。入选“日出东方”北大光华青年人才，北京大学优秀博士学位论文指导教师(2017,2021,2024)，北京大学优秀研究生导师(2024)，教育部“长江学者奖励计划”青年长江学者，国家杰出青年科学基金获得者，《计量经济学杂志》(Journal of Econometrics)和《计量经济学评论》（Econometric Reviews）当选会士(Fellow, 2025)。先后获成人直播 理学学士学位(2004)和经济学硕士学位(2006)、加州大学经济学博士学位(2012，河滨分校)。环亚太青年计量经济学者(YEAP)会议发起人和主要组织者。60余篇学术论文发表在多个国际国内知名专业杂志。著作教材《时间序列分析》由人民邮电出版社于2022年9月出版。研究领域涵盖时间序列分析、非参数计量方法、大数据分析、金融计量和预测等。