动态广义最小二乘法在面板协整中的应用研究

The Dynamic GLS Estimation of a Cointegrated Regression in Panel Data

在进行非平稳面板数据的协整分析时,使用动态最小二乘法(DOLS)可以有效消除内生性问题,从而得到具有渐进正态分布的统计量。但在小样本条件下,由于可使用解释变量差分项的阶数有限,导致模型中均衡误差项的序列相关,使得DOLS统计量出现严重的检验水平畸变。为此,本文将单一时间序列的动态广义最小二乘法(DGLS)应用于非平稳的同质面板数据模型。在序贯极限分布的条件下,DGLS统计量仍具有正态的条件极限分布。而仿真实验表明,对于非平稳的同质面板数据模型,即使在均衡误差项存在高序列相关的条件下,DGLS统计量仍具有较好的小样本性质。

In the cointegration model of non-stationary panel data,the dynamic ordinary least square（DOLS） estimator can eliminate the endogeneity that caused by the long-run correlation between the equilibrium error and the first difference of the regressors. And under the sequential limit theory,the estimator of the DOLS is asymptotic normality. But in small samples,the leads and lags of the first differences of the independent variables are finite,which causes the serial correlation in the equilibrium errors,and this results in the great size distortion to the DOLS estimator. In this paper,we use panel dynamic generalized least squares （ DGLS） to estimate the cointegrating vectors in non-stationary homogeneous panel data,and the estimator of the DGLS still has a Gaussian sequential limit distribution. In a series Monte-Carlo experiments,we find that the estimator of the DGLS performs better in the sizes than that of the panel DOLS and FMOLS in small samples when there is serial correlation in the equilibrium errors.