Theory Study And Application On Panel Unit Root Test With Cross-sectional Dependency And Heavy-tailed Properties

Since the1970s, study on unit root test has became one of the research hot spots in modern econometrics and has big significance to the development of modern econometrics, and the application of which involved in various fields of economics. Generally, The process of research on unit root test is from time series unit root test to the cross-section independency (first generation) panel data unit root test, and then to the cross-section dependency (second generation) panel unit root test. Among the studies of second generation panel unit root test, Bootstrap method is an effective approach to solve the problem of cross-section dependency.For panel data structure which allows various forms of cross-sectional dependence including (but not exclusive to) the popular common factor framework, Block Bootstrap method is a good choice, which is bringing to the researchers’ much attention recently.Notice that both Bootstrap panel unit test and Block Bootstrap panel unit test required the error term is independent along the time index, However the conditions are not necessarily satisfied when in practical application, such as those with heavy-tailed nature of economic and financial data is difficult to satisfy the conditions. There is no related theories on how to conduct unit root test when panel data both are cross-section dependency and has heavy-tailed effects, such as GARCH effect Existing research shows that:when the error term is conditional heteroscedasticity, Wild Bootstrap method has a better finite sample properties than those Bootstrap methods which are under the assumption of independent identically distributed in time series unit root testFor this reason, this article will expand the Wild Bootstrap method from the time series unit root test to the panel unit root test and build a basic theory system of Wild Bootstrap panel unit root test, used to solve the problem of panel unit root test with cross-section dependency and heavy-tailed effects, and we compare the finite sample properties of Wild Bootstrap panel unit test and Block Bootstrap panel unit test with a wide range of dependencies through Monte Carlo simulation. Monte Carlo simulation results show that when there is heavy tail sex, the block bootstrap test-statistic has large size distortion and poor power when T is large or T and N are comparative, wild bootstrap method has smaller size distortions and better power than block bootstrap method for moderate and large sample.The wild bootstrap panel unit root test procedure is then applied to the Chinese securities market weak efficiency hypothesis, the empirical results implies that LLC test and IPS test are all reject the null hypothesis of unit root, On the contrary, Wild Bootstrap test and Block Bootstrap test are all accept the null hypothesis of unit root, and Wild Bootstrap test results are easier to accept the null hypothesis than Block Bootstrap.so we can make a conclusion that the Chinese securities market is general weak efficiency.