主讲人：
苏良军（清华大学经济与管理学院讲席教授）
主持老师：
（北大经院）王熙
参与老师：
（北大经院）王一鸣、刘蕴霆
（北大国发院）沈艳、黄卓、孙振庭、张俊妮
（北大新结构经济学研究院）胡博
时间：
2021年4月23日（周五）10：00-11:30
地点：
永利集团3044am官方入口107会议室
主讲人简介：
苏良军， 2004年获得加州大学San Diego分校经济学博士学位。2004-2008年在北京大学光华管理学院商务统计与计量经济系担任助理教授与副教授，2008-2020年在新加坡管理大学永利集团3044am官方入口先后担任副教授、教授与李光前讲席教授。2020年7月加盟清华大学经济管理学院，为经济系讲席教授。苏良军教授长期从事理论计量经济领域的研究工作,主要集中在非参数计量经济学、面板数据分析、大数据与机器学习等方向。目前已经在Econometrica、 Econometric Theory、 IEEE Transactions on Information Theory、 Journal of Machine Learning Research、Journal of Applied Econometrics、 Journal of Econometrics、 Journal of the American Statistical Association、 Journal of Business & Economic Statistics、 Quantitative Economics 等国际一流经济学、统计学与信息学杂志发表论文七十余篇，并编辑出版了两本书。研究结果已被多部世界权威或知名面板数据与非参数计量经济学教科书引用，包括Li 和Racine (2007, Nonparametric Econometrics)、Hsiao (2014, Panel Data Analysis, 3rd edition)，Pesaran (2015, Time Series and Panel Data Econometrics)、Henderson 和 Parmeter (2015, Applied Nonparametric Econometrics)、Racine (2019, An Introduction to the Advanced Theory and Practice of Nonparametric Econometrics)等。
摘要：
This paper studies uniform inference in a linear panel data model when the slope coefficients may exhibit heterogeneity over both the individual and time dimensions and they can be correlated with the regressors. For this model, we can not eliminate either the individual or time heterogeneity in the slope coefficients through the usual demeaning or differencing procedure, and conventional fixed effects estimators are generally inconsistent. We propose a generalized fixed effects (GFE) estimation procedure to estimate the model under suitable identification restrictions. To establish the asymptotic properties of the GFE estimators, we invert a number of large dimensional square matrices by approximating them with quasi-Kronecker structured matrices. We establish the asymptotic normality of our GFE estimators and show that their convergence rates depend on the unknown degree of parameter heterogeneity. To make a uniform inference on the common slope component, we propose a novel triple-bootstrap procedure and a hybrid procedure to estimate the asymptotic variance which are valid uniformly over a broad space of parameter heterogeneity.Simulations show the superb performance of our estimators and inference procedures in various scenarios. We apply our method to study the relationship between savings and investments in a cross-country study. We find significant parameter heterogeneity along both the individual and time dimensions and provide some new insight to explain the celebrated Feldstein-Horioka puzzle.
