Stabilized Latent Group Structures Identification for Panel Data: A Regularization-Free Approach
Holder： Xingbai Xu（Xiamen University）
Time：2026-04-02 15:10-17:00
Location：Room 217, Guanghua Building 2
Abstract:
One of the major challenges in identifying latent group structures in panel data is to determine the number of groups. Most existing approaches rely on minimizing an information criterion. However, these methods are often sensitive to the choice of tuning parameters, rendering the selection process ad hoc and unstable in finite samples. This paper develops a fully data-driven selection procedure that circumvents the need for regularization by minimizing a clustering instability score. We propose a valid resampling scheme that partitions data along the cross-sectional dimension. We show that the proposed criterion achieves selection consistency under mild conditions, with the error probability decaying exponentially in the number of individuals. Unlike model-specific information criteria, our framework is versatile and can be seamlessly integrated with various detection algorithms (e.g., K-means, SBSA) across both linear and nonlinear models. Numerical experiments confirm its superior robustness and accuracy compared to conventional methods. We apply our algorithm to study the relationship between economic development and carbon dioxide emissions. The proposed methodology is implemented in the accompanying R package StableGroup.
About the Speaker:
许杏柏，厦门大学王亚南经济研究院、邹至庄经济研究院和经济学院长聘教授，博士生导师，厦门大学南强A类青年拔尖人才。2016年从俄亥俄州立大学获得经济学博士学位。主持多项国家自然科学基金项目。主要研究领域为空间计量经济学、网络计量经济学、稳健估计、贝叶斯、面板数据分析以及混合模型。多份研究成果发表在Journal of Econometrics, Econometric Theory, Journal of Business and Economic Statistics等学术期刊上。所教授课程《高级计量经济学》被评为首批国家级本科一流课程之一，所参与教学改革项目获得2022年国家级教学成果奖二等奖、福建省教学成果奖特等奖。
Your participation is warmly welcomed!
欢迎扫码关注北大统计科学中心公众号，了解更多讲座信息!
