Font Size: a A A

Panel Unit Root And Panel Cointegration: Testing Methods And Applications To Chinese Economy

Posted on:2008-04-02Degree:DoctorType:Dissertation
Country:ChinaCandidate:J S YangFull Text:PDF
GTID:1119360272466912Subject:Quantitative Economics
Abstract/Summary:PDF Full Text Request
The panel unit root and panel cointegration analysis can not only improve the finite sample size and test power, but also provides more information about the individuals and that between each other. It is well known that cross-sectional dependency must be considered in most panel data, but the first-generation panel unit root tests are based on the cross-sectional independency. The second-generation panel unit root tests are either based on the special specification for the cross-sectional dependency, or characterized by serious size distortion when the dependency is strong. Panel cointegration test can be achieved based on the residuals or on the panel vector error correction model (PVECM). The panel cointegration test based on residuals is not valid for cases with more than one cointegration vectors, and does not allow for the interaction of short-run dynamics and cointegration relationship among individuals. Groen and Kleibergen (2003) proposed the panel cointegration test based on PVECM, but it also suffer from the assumption that there is no cointegration between individuals. Based on the comprehensive analysis of the current literature on panel unit root tests and panel cointegration tests, some existing tests are modified or developed, and some new test statistics are providedIn view of the panel unit root test, one purpose of this paper is to construct the new test for panel data with general cross-sectional dependency by modifying or developing the existing tests. The combing p-value test by Maddala and Wu (1999) and Choi(2001)employs the nonlinear transformation from the t-statistics to its p-value, and then the cross-sectional dependency is weakened significantly. Therefore, the combing p-value test based on ADF statistic can be extended to panel data with weak cross-sectional dependency, and the simulation result argues for its wonderful performances. Chang (2002) suggests removing the cross-sectional dependency and ensuring the asymptotic standard normality of her test statistic by using the nonlinear instrumental variables. However, we found that her SN test can only be used for panel with very weak cross-sectional dependency and would become biased when determinant component is included. To expand its application, the SN test has been modified based on our simulation results. For the lower power of the existing test in panel data with cross-sectional dependency, finally, we combined the SUR type feasible GLS and Nonlinear Instrumental variable (NIV) estimator to construct the generalized NIV (GNIV) test statistic available for panel data with none, weak, moderate or even strong cross-sectional dependency, where the FGLS is for removing the cross-sectional dependency and the NIV is for bringing about the consistent estimates. the simulation result offered the evidence that the GNIV test is superior to the SN test by Chang (2002), the CIPS test by Pesaran(2003)and the LLC test by Levin, Lin and Chu (2002).In view of panel cointegration, for the limitations of panel cointegration tests based on residuals, this paper focus most attention on the PVECM based. Firstly, Groen and Kleibergen (2003) have not proposed the critical values for their test statistics with asymptotic distribution as the functional of Brown motion, which would hamper its application. This paper provided those critical values under varied setting based on a set of simulation to enable the test in applications. Secondly, since the interaction among individuals is popular, we pooled the variables (not pooled data) in different cross-sectional units to include the cointegration relationship among individuals, and then constructed the panel likelihood ratio (LR) test statistics based on Johansen (1988,1991,1995). However, the number of parameters to be estimated for such a test would increase with a rate N, the number of the cross-sectional units, so the critical values obtained from the asymptotic distribution cannot be used in finite sample. To solve such a problem, we propose a LR test produce based on the bootstrap simulation, and then investigate the performance of the test based on bootstrap procedure by a small-scale simulation design. Finally, as the examples of empirical study employing panel cointegration test based on residuals, panel cointegration test based on PVECM, or panel unit root test only, respectively, the energy consumption of Chinese industries, Renminbi real equilibrium exchange rate, and weak effectiveness of Chinese securities market are investigated based on panel data. To show how the different test techniques are carried out, the example of energy consumption is based on LM panel cointegration test by McCoskey and Kao (1998),our GNIV test and the IPS panel unit root test by Im, Pesaran and Shin (2003). While, Renminbi real equilibrium exchange rate is discussed based on panel cointegration test by Groen and Kleibergen (2003) and our modified SN panel unit root test. To test the weak effectiveness of Chinese securities market, the combing p-value test and our GNIV test are employed. The analysis result implies the existence of the long-run equilibrium relationship between energy consumption and growth in industries, and between Renminbi real exchange rates and the fundamental economic factors; evidence for weak effectiveness of Chinese securities market is also observed.
Keywords/Search Tags:Panel Unit Root, Panel Cointegration, Modification and Development, Generalized Nonlinear Instrumental Variables Test, critical value, Unrestricted Panel LR Test, Empirical Study
PDF Full Text Request
Related items