Theory And Applications Of Dynamic Panel Data Model Estimates And Its Endogenous Structural Breaks Test

The existing literatures show that panel data model has been an important branch of modern econometrics ever since proposed in 1960s,of which the dynamic panel data model is the forefront and research focus of this branch.The so-called Dynamic Panel Data Model, is the panel data model which reflects the dynamic hysteresis effect by introducing lagged response variables to the static model. The specificity of this model lies in that the lagged item of dependent variable is related to the individual effect of random error components, resulting in the estimator endogenous. Further, how to test the endogenous structural breaks of dynamic panel data model is undoubtedly the most difficult problems in this direction. This article focuses on the dynamic panel data model estimate and the endogenous structural breaks testing, and its research contents and conclusions can be summarized as following:Firstly,this paper formulates thoroughly and systematically the estimator bias, the main estimate method and assumptions of GMM, as well as the weak instrumental variable and so on. On this basis, it reveals the statistical nature and application condition, the source of bias or weak effect of the major estimators of GMM,and how to correct. The major findings are that the various estimate methods of dynamic panel data model applies different data generation process; the estimator bias resulting from the system GMM is the smallest and most effective in most cases, but if the variance ratio of individual effects and idiosyncratic shocks is very small,especially when very large,the bias degree of estimator may be greater than the estimate results from differenced GMM; the bias degree of estimator from the differenced GMM increases significantly with the auto-regression coefficients close to 1, but its sensitivity to the changes of variance ratio is weaker;the estimator of GMM of the level equation shows larger sensitivity to the changes of variance ratio, the bias degree of estimator is smaller than the estimator of differenced GMM when the auto-regression coefficient is larger; The bias correction method based on the fixed effect bias estimator, especially the approach proposed by Hansen (2001) has the smallest degree of bias when there are no other predetermined variables in model.otherwise,the nature of limited samples if relatively poor.Secondly, the basic feature of China's provincial-level panel data is that the largest useful number of cross-sectional is 31 .but in existing literatures, the simulation study of dynamic panel data model estimation bias under the framework of the GMM designs much greater cross-section numbers than the number of China's provinces,so that Chinese scholars have not a better measure of estimate bias when analysing China's provincial-level dynamic panel data. Considering all the factors effecting the estimated bias,the author designs and realizes the corresponding simulation study, and finds the bias degree of autoregression coefficient estimator with three representative dynamic panel data model on the condition that the cross-section number is 30. Based on simulation results, when doing dynamic model analysis in China's provincial-level panel data ,this paper recommends as following: the appropriate values of time dimensionis 10 to 15; for smaller auto-regression coefficient and larger variance ratio of individual effects and idiosyncratic shocks, applies differenced GMM;the variance ratio of individual effects and idiosyncratic shocks close to 1, using the system GMM;when the variance ratio of individual effects and idiosyncratic shocks is large,if the auto- regression coefficient is close to 0.5,the forward orthogonal deviation GMM applies;if close to 1, the system GMM applies;Unless the auto-regressive coefficients and the variance ratio of individual effects and idiosyncratic shocks are all very small, all three estimation methods do not use the potential instrumental variables at all.and it is recommended to select instrumental variables with lag periods less than 4.Thirdly, this paper formulates thoroughly and systematically the structure breaks testing and estimation of dynamic panel data model, if there is indeed structural breaks, ignoring it,the estimator of GMM is not only biased but also non-consistent through the change of the moment function before and after, as well as breaks point. this paper Focuses on the endogenous structural breaks testing's theoretical framework, identification methods and the structure of test statistic of Wachter (2004)'s dynamic panel data model; And do unknown structural breaks test to the model of inflation inertia in Wachter (2004)'s endogenous structural breaks test method of dynamic panel data model. in accordance with the test results, this paper estimates the model parameters in the system GMM and compares to the estimator without considering structural breaks.Based on the above, this paper's innovation and meaning lie in: firstly,from the influence of data generation process to estimator bias and validity, this article reveals the statistical nature and application condition of major dynamic panel data model estimator,and this is different from the existing literature;secondly,the cross-section number in the simulation study designed and realized is the least in the existing literature on dynamic panel data model's estimator bias simulation study.And the result shows how to study the economic problems in China's provincial-level panel data;thirdly,this paper firstly introduces the endogenous structural breaks test method of dynamic panel data model method which allowing the slope coefficient and the individual effects of unknown time to be structural breaks ,and the later to be of the endogenous;forthly,do unknown structural breaks test to the model of inflation inertia and inhabitants consumption function in Wachter(2004) test method firstly. The result is that there are all significant structural changes in the two models. So it is much of significance to do structural breaks test to prevent the wrong model-setup in the study of China's economic problems with econometrics analysis.