报告题目：Quantile-regression-based clustering for panel data

报告地点：科学会堂710

报告时间：2018年5月3日（星期四）16: 30-17: 30

报告人简介：

朱仲义，男，复旦大学统计系教授，博士研究生导师；曾任中国概率统计学会第八、九届副理事长,国际著名杂志”Statistica Sinica”副主编，现任“应用概率统计”、“数理统计与管理”杂志编委，Elected Member of the ISI（国际数理统计学会），“中国科学：数学”杂志编委，中国现场统计研究会常务理事,中国统计教材编审委员会委员。主要研究邻域为保险精算、纵向数据(面板数据)模型、分位数回归模型等。主持完成国家自然科学基金四项，国家社会科学基金一项，作为子项目负责人完成国家自然科学基金重点项目一项。目前主持国家自然科学基金重大项目子项目一项，国家自然科学基金项目一项。近几年发表论文90多篇(其中包括在国际顶级刊物：J.R.Stat.Soc B, J.A.S.A., Ann. Statist.和Biometrika等SCI论文五十多篇)。作为第一完成人获得教育部自然科学二等奖一次。

报告摘要：

In many applications it is important to identify subgroups of units with heterogeneous parameters. We propose a new quantile-regression-based method for panel data to identify subgroups and estimate group-specific parameters. In practice the signal differentiating subgroups may vary across quantiles though the group membership may be common. It remains unclear which quantile is preferable or should one combine information across quantiles to perform clustering. To answer this question, we consider a stability measure to choose among single quantiles and the composite quantile. We establish the asymptotic properties of the proposed estimators, and assess their performance through simulation and the analysis of an economic growth data.