Asymptotical Inference for a Simple Dynamic Panel Data Model when Both N and T Are Large

• Main Authors: Li, Chu-Chien, 李竹健
• Other Authors: Wang, Chien-Ho
• Format: Others
• Language: zh-TW
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/48657423842257760654

碩士 === 國立臺北大學 === 經濟學系 === 93 === My thesis investigates the asymptotics of dynamic panel data when cross section and time series dimensions tend to infinity. In traditional dynamic panel data analysis, we only consider asymptotics under N closed to infinity and T fixed. GMM methods always use to handle dynamic panel data model. However the panel data collection has a lot of programs recently. Economists will face some panel data sets with N closed to infinity and T closed to infinity in empirical analysis. In thesis, we inspect the traditional estimation methods for dynamic panel data under N closed to infinity and T closed to infinity within traditional ways to handle dynamic panel data. We set T/N ratio for well defined convergence path. The ratio of T/N we set in our thesis are 0, 1/2, 1, and infinity. There are totally five estimation methods considered in this paper: within-group estimation, instrumental variable estimation, generalized method of moments estimation, two stage least square estimation and maximum likelihood estimation. The results of simulation show the estimator using the lag two period observation as the instrumental variable is the best. When the parameter is 0.9, the two stage least square estimator is better than generalized method of moments estimation. If 0<T/N 1/2, researcher can use two stage least square estimation with T being smaller and generalized method of moments estimation with T being larger. When T/N=1, the best way is instrumental variable estimation. Finally, the within-group estimator may be a better choice if 1<T/N<infinity. Because its asymptotical bias will be converge with T increasing.