统计研究 ›› 2018, Vol. 35 ›› Issue (3): 112-128.doi: 10.19343/j.cnki.11-1302/c.2018.03.010

• • 上一篇    

截面相关面板数据的部分异质结构的识别

徐凤 黎实   

  • 出版日期:2018-03-25 发布日期:2018-03-25

Identifying Heterogeneity in the Partial Structure of Panel Data with Cross-Sectional Dependence

Xu Feng & Li Shi   

  • Online:2018-03-25 Published:2018-03-25

摘要: 在大维面板数据中,截面之间很可能呈现出部分异质的特征,即参数在截面间具有组群效应,同组参数相同而不同组参数相异。如果忽略部分异质性而采用完全异质或同质的方法,可能导致估计的不一致性以及统计推断无效性。鉴于已有的部分异质性的研究要么限定截面独立,要么局限于强因子情形,本文尝试在Reese和Westerlund(2015)[1]提出的允许强因子或非强因子存在的较一般的框架下探讨面板数据部分异质结构的识别问题。采用Pesaran(2006)[2] CCE (Common Correlated Effects)方法处理不同强弱的共同因子,并借鉴Su et al.,(2016)[3]的C-Lasso (Classifier- Least Absolute Shrinkage and Selection Operator)方法,对CCE变化后的方程构造带有加法-乘法惩罚项的惩罚最小二乘,优化后以同步地实现分组和参数的估计。理论分析表明,在强因子或半强因子情形中,本文所提方法在分组方面具有渐近一致性,即所有个体被正确分组的概率随着 而趋于1。同时,参数的Lasso估计和事后Lasso估计均具有渐近正态性。另外分析结果也表明,因子的强弱不会影响分组的一致性但会影响以上两种估计量的渐近正态性,因子越强,两种估计量收敛得越快。模拟结果则表明有限样本下,本文所提的方法在分组、参数估计和分组数确定方面均具有良好的表现。具体的,在强因子和不同的半强因子情形中,随着N,T的增加,分组和分组数正确率很快地上升到100%,而两种参数估计的均方根误差和偏差则明显地降低。最后,利用本文所提的方法,研究了人力资本对经济增长影响的部分异质性。

关键词: 面板数据 部分异质 截面相关

Abstract: Partial heterogeneity often exists in large panels and are characterized by group structure in which the parameters are constant within group and vary between groups. Neglecting this group structure may lead to inconsistent estimates and invalid inferences. Up till now, Researchers have not fully investigated partial heterogeneity in panel data models, for example partial heterogeneity in panel data with strong or non-strong factors. In view of this, we try to study the identification of unknown partially heterogeneous structure under a quite general framework with strong and semi-strong factors(Reese and Westerlund, 2015). The method is proved to achieve uniformly consistent classification regardless of strong or semi-strong factors. Whereas the Lasso estimators and post-Lasso estimators converge fast to zero-mean normal distribution with the rise in the strength of factors. Simulation results indicate that the proposed method performs well under finite samples. Specifically, with the rise of N and T, the accuracy for classification and group numbers increases to 100% quickly, and the root-mean-square error and bias for the two types of estimators decrease significantly. Finally, an empirical analysis of Human Capital-Growth nexus is presented.

Key words: Panel Data, Partial Heterogeneity, Cross-sectional Dependence