The Comparison Of Property Of Panel Unit Root Tests With Small Samples And The Research Of Panel Cointegration

The theory of panel unit root test is more and more popular because of its complete advantage. In China , however , the research for panel data econometrics is still introductory.Therefore,intensifyingapplied research has significant value to both theory and reality to impulse science of economic theory and practice. Panel data with moderate or small sample is often used in empirical research,especially small-sample. So this article focus on panel data with small sample in order to provide some kind of theoretic reference for empirical research. Subsequently, two main mathods of panel cointegration are introduced. Finally, empirical research is presented as an practical example of this article.First, we make panel data-generating process when we do some reseach of properties of panel unit root test. In each experiment, then we consider the cases of N =10, 15, 25 and T = 10, 15,20,25,30,35,40,45, 50. A total of 10000 simulations are used in computing the empirical size and power of the tests. The innovation of this article is relaxing assumption of cross-uncorrelations.The followings are our conclusions:(1) In general,the LLC test has the best power, and whereas IPS test has the lowest power.But we should notice that LLC test has the highest size that means when panel unit root exists,the possibility of its rejecting unit root is the highest.So we suggest in the case of less individual and time period we'd better choose Fisher test.(2) We generate the error terms according to an AR(1)model thus using ADF(1) is appropriate. From the results we can see the danger of under-selecting the order of ADF regression which is also pointed out by IPS. The IPS test is greatest influenced when the lags of ADF test is insufficient. When N is fixed, the power of both Fisher test and LLC test increases with period of time T extends, while the power of IPS test decreases with period of time T extends.