系统GMM估计方法的比较研究  

Comparative Study of System GMM Estimation Method

作　　者：尹康 洪丽[2,3] Yin Kang;Hong Li(School of Economics and Trade,Hubei University of Economics,Wuhan 430205,China;School of Political Science and Public Administration,Wuhan University,Wuhan 430072,China;Centre for Social Security Studies,Wuhan University,Wuhan 430072,China)

机构地区：[1]湖北经济学院经济与贸易学院,武汉430205 [2]武汉大学政治与公共管理学院,武汉430072 [3]武汉大学社会保障研究中心,武汉430072

出　　处：《统计与决策》2020年第21期38-42,共5页Statistics & Decision

基　　金：教育部人文社会科学研究青年基金项目(16YJC790027);武汉大学自主科研项目(人文社会科学)(413000037);湖北经济学院校级教学研究项目(2018003)。

摘　　要：文章从系统GMM估计原理出发,分析了R和STATA软件在实现动态面板模型中系统GMM估计程明,基于经典文献编写的估计程序,从估计偏误和均方误差标准看,要优于R软件和STATA软件的估计结果。在权重矩阵的选择方面,R和STATA软件所推荐的协方差结构(H3)估计表现较差,而最简单的单位矩阵结构(H1)估计表现最佳。当个体效应的方差相对较大时,H1的优势更为明显。This paper starts from the principle of system GMM estimation and analyzes design differences between R and STATA software in the implementing of the system GMM estimation procedure in the dynamic panel model. At the same time, the paper makes a comparative analysis on the three weight matrix structures commonly used in system GMM estimation. The numerical simulation results show that the estimation program based on classical literature is superior to the estimation result of R software and STATA software in terms of estimation bias and mean-square error standard, that in the selection of the weight matrix,the covariance structure(H3) estimation recommended by R and STATA software performs poorly, while the simplest unit matrix structure(H1) estimation performs best, and that when the variance of individual effects is relatively larger, the advantage of H1 is more significant.

关 键 词：动态面板数据 系统GMM 权重矩阵 工具变量 

分 类 号：F064.1[经济管理—政治经济学]