講座主題：Quantile Regression with Group-level Treatments

主講嘉賓：陳松年，浙江大學教授、世界計量經濟學會院士

講座時間：2023年5月26日（周五） 10:00-12:00

講座地點：bevictor伟德官网沙河校區學院樓11号樓308

嘉賓簡介：陳松年，著名經濟學家、世界計量經濟學會院士、浙江大學青山商學高等研究院引進的首位浙大青山講席教授。陳教授1986年畢業于複旦大學數學系，1994年獲普林斯頓大學經濟學博士學位。研究領域為理論與應用微觀計量學。他曾任新加坡國立大學經濟學系講席教授、香港科技大學經濟系講席教授。陳教授在計量經濟學領域享有盛譽，特别是在微觀計量經濟學領域具有突出成績，在截斷删失回歸、分位數回歸、樣本選擇模型等領域的研究享譽國際。他已在Econometrica, Review of Economic Studies, Journal of Econometrics等國際學術期刊發表論文四十餘篇，并在計量經濟學頂刊Journal of Econometrics長期擔任副主編，是該期刊的榮譽會員。

内容摘要：To study the distributional effects of group level treatments, Angrist and Lang (2004) applied quantile regression with group level regressors, and Chetverikov et al. (2016) proposed a grouped instrumental variables quantile regression estimator, a quantile extension of the Hausman and Taylor’s (1981) instrumental variables estimator for panel data. However, the analyses of distributional effects of group level treatments in Angrist and Lang (2004) and Chetverikov et al. (2016) are incomplete and their models are quite restrictive, and they only allow for heterogenous distributional effects of group-level treatments that corresponds to individual-level unobserved characteristics, but not group-level unobserved characteristics. In other words, Angrist and Lang (2004) and Chetverikov et al. (2016) allow for within group hetergeneous distributional treatment effects, but not between group heterogeneous distributional treatment effects. In this article, we provide a comprehensive analysis by proposing a quantile regression model that allows for heterogenous distributional effects of group level treatments associated with both individual level and group level unobserved characteristics, corresponding to within-group and between-group distributional effects. We propose two step quantile regression and instrumental variables quantile regression estimators, depending on whether the group level treatments are correlated with the group level unobserved characteristics. Large sample properties are presented and simulation results indicate our estimators perform well in finite samples.