Quantile Regression--Based Clustering for Panel Data

• Activity Number: 640 - Quantile Based Modeling for a Variety of Heteroscedastic Data
• Type: Invited
• Date/Time: Thursday, August 2, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Huixia Judy Wang* and Yingying Zhang and Zhongyi Zhu
• Companies: The George Washington University and Fudan University and Fudan University
• Keywords: Fixed effect; heterogeneity; panel data; quantile regression; subgroup identification

Abstract

In many applications such as economic and medical studies, it is important to identify subgroups of subjects who associate with covariates in different ways. In this paper, we propose a new quantile-regression-based clustering method for panel data. We develop an iterative algorithm using a similar idea of k-means clustering to identify subgroups at a single quantile level or at multiple quantiles jointly. Even in cases where the group membership is the same across quantile levels, the signal differentiating subgroups may vary with quantiles. It remains unclear which quantile is preferable or should we combine information across multiple quantiles. To answer this question, we propose a new stability measure to choose among multiple quantiles that gives the most stable clustering results. The consistency of the proposed parameter and group membership estimation is established. The finite sample performance of the proposed method is assessed through simulation and the analysis of an economy growth data.