面板数据下的EG两步法

针对文献中将时间序列模型的EG两步法直接应用到面板数据模型这一现象，推导了面板数据中使用EG两步法的极限分布，进而指出在面板数据中直接使用时间序列EG两步法存在的问题，并通过MonteCarlo模拟展示了这一错误应用所存在的过度拒绝零假设的后果.同时，模拟结果还表明，当前实证研究中常用的两种基于EG两步法的面板协整检验，Kao协整检验和Pedroni协整检验均有其适用范围，对于不同的样本容量组合，两者有限样本表现不尽相同，因此，实证研究应根据样本数据的实际情况选用合适的检验统计量.此外，本文的实证结果表明，弱PPP定理在OECD国家并不成立.

Engle-Granger two-step method in panel data

Plenty of empirical studies apply Engle-Granger (EG) two-step method which is designed for time series models directly to panel data models. However, it is valuable to study the validity of this extension and we find that it may result in severe consequences in some situations through Monte Carlo simulations. For example, this misuse may lead to over rejections in hypothesis testing. Meanwhile, we compare two cointegration tests suggested by Kao and Pedroni, respectively, and find that their finite sample performances vary according to sample size in the data generating process (DGP). Therefore, empirical applications may be considered on case by case basis. Besides, our empirical study indicates that the weak PPP theorem does not hold in OECD countries.