Estimation and inference with nonstationary panel data
This PhD thesis applies the time-series concepts of unit-roots and cointegration to nonstationary panel data. The first three chapters set the scene for what follows and together are the first methodological core of the thesis, on nonstationary panel data estimation and testing. In chapter 1 we consider the established panel unit root tests of Levin, Lin and Chu (2002) and Im, Pesaran and Shin (2003) and also Pesaran (2005) for cross-sectional dependence, with a panel of 20 OECD inflation rates. In chapter 2 we consider the established panel cointegration tests of Kao (1999), Pedroni (1999) and Larsson, Lyhagen and Lothgren (200 1) with a panel of 25 OECD exchange rates to test for long run PPP, again including cross-sectional dependence. In chapter 3 a more original contribution is given. We conduct an extensive empirical study of the long run determinants of consumption expenditure for a panel of 20 OECD countries. A panel data cointegrating regression is estimated using the panel DOLS and FMOLS estimators of Kao and Chiang (2000) and Pedroni (2000,2001). Using Bai and Kao (2005) we again consider cross-sectional dependence. The second methodological core is the statistical inference of nonstationary panel data, in the last two chapters. In chapter 4 is another original contribution using the bootstrap with nonstationary panel data. New bootstrap algorithms are presented for the panel DOLS estimators mentioned above and also the group-mean estimator of Pesaran and Smith (1995). In our last original contribution, in chapter 5, we consider the asymptotic properties of nonstationary panel data estimators. The asymptotic normality and asymptotic consistency of our panel FMOLS, DOLS and OLS estimators are proved for the simple case of the panel cointegrating regression with a constant intercept and trend. The new sequential limit asymptotic theory of Phillips and Moon (1999) is highlighted.