Robust inference in panel data model / Nurul Sima Mohamad Shariff

Mohamad Shariff, Nurul Sima (2012) Robust inference in panel data model / Nurul Sima Mohamad Shariff. PhD thesis, University of Malaya.

PDF (Title page, abstract, content) - Submitted Version
Download (287Kb) | Preview
    PDF (chapter 1 - 7) - Submitted Version
    Download (3213Kb) | Preview
      PDF (references) - Submitted Version
      Download (158Kb) | Preview
        PDF (appendix) - Submitted Version
        Download (164Kb) | Preview


          Panel data is a group of many individual units observed for a specific time period. In general, researchers may tend to pool the units together where each observation is treated as independently among the others. Such restriction is invalid because most of the economic data are cross correlated between cross sectional units which may arise from a common influence which affects all units. This is known as cross sectional dependence (CD). The presence of outliers may result in rejection of null hypothesis, that is, in support of cross sectional independence. To overcome such problem, alternative methods which are insensitive to the presence of outliers are needed. To address the problems of incorrect test statistics and parameter estimates in the presence of CD and outliers, this study will focus on several parts. Firstly, robust versions of CD tests are proposed to investigate the presence of CD and outliers in both the pure static and dynamic models. The asymptotic behaviours and simulation study of power for the finite sample behaviour based on Monte Carlo simulation study are considered. We find evidence that, in the presence of mild (low) CD and outliers in panels, our tests outperform the commonly used CD tests that are the LM and CD tests. Secondly, we propose a robust version of Common Correlated Mean Group (CMG), namely RCMG, for estimating parameters in pure static model. Some properties and statistical inference for the parameter are also considered. To better understand the finite sample behavior of these approaches, we run a Monte Carlo simulation study. Our proposed estimator yields unbiased estimates with small MSE in the presence of outliers occur in X and Y directions. The hypothesis test for the robust estimator indicates that RCMG estimator has reasonable size and power with and without the presence of outliers and outperform CMG estimator in contaminated panel. In addition to the measure of bias and MSE, its accuracy is also measured by the length of confidence interval for RCMG estimator to supports these findings. Thirdly, we further explore the unit root tests for the dynamic framework. The currently available tests such as ADF and CIPS are very much affected by the presence of outliers which subsequently result in wrong decision making by favoring to the null hypothesis of a unit root. An alternative of CIPS denoted by RCIPS is introduced based on the RCMG procedure. The performance and robustness of the RCIPS is discussed and comparisons are made to ADF ad CIPS. Our simulation results show that while the CIPS performs well for large T , the RCIPS tends to provide a good size and power even for smaller N and T , as well as with and without the presence of outliers. Finally, we revisit two real datasets that are related to panel data; 1) the gasoline data for the pure static case; and 2) the PPP panel of ASIAN and CEEC countries for the dynamic model. Here, we employ the methods discussed above and reanalyze the data accordingly.

          Item Type: Thesis (PhD)
          Additional Information: Thesis submitted in fulfillment of the requirement for the degree of Doctor of Philosophy
          Uncontrolled Keywords: Applied Statistics; Panel data; Common Correlated Mean Group
          Subjects: Q Science > QA Mathematics
          Divisions: Faculty of Science
          Depositing User: Ms Rabiahtul Adauwiyah
          Date Deposited: 09 Apr 2013 15:32
          Last Modified: 30 Aug 2013 12:35

          Actions (For repository staff only : Login required)

          View Item