Panel Data Analysis With Hierarchical Structure And Application

Firstly, in the preface part, the paper elaborate the development process and mainresearch result of panel data analysis, including basic theories and the latest researchresult about unit root test and cointegration.Secondly ,in foundation described basic theories model of the panel data analysis,the article built up the general block linear model expression form of panel data analysismodel. Deduced parameter estimation and hypothesis test statistic and its probabilitydistribution. The relationship between GDP and Export of the PRC's province, cities,and autonomous regions is analysed in panel data set : GDP and Export annul datafrom 1992 to 2004 in the PRC's province, cities, and autonomous regions. Empiricalanalysis show that the fixed two panel model is better than other model, but exist theproblem of residual errors sequence correlation .There are two issues. First, the standard errors computed under the assumptionthat the error term is independent identical distribution will be biased. Second, theassumption of independence is unlikely to satisfied. In the panel Data Analysis modelwith hierarchical Structure, hierarchical effects, nested effects, time effects are seted.Then the dissertation deduced parameter estimation and hypothesis test statistic andits probability distribution and analyze the hierarchical panel data set : Eastern China,Central China and Western China are the top level, and the PRC's province, cities,and autonomous regions are bottom level. Empirical analysis show that hierarchicalStructure panel data Analysis model is the better one.Finally, the dissertation study unit root test and cointegration of panel data set anddiscuss nonstationary of GDP and Export annul data from 1992 to 2004 in the PRC'sprovince, cities, and autonomous regions. Empirical analysis show that the panel datahas a unit root ,so it is nonststionary. The conclusion is valuable in studing relationbetween exports and economic growth. Taking account of nonstationary, dynamic paneldata model is estimated, and revise the problem of residual errors sequence correlation.