670 Panel Methods to Test for Unit Roots and Cointegration
Table 13.2 Results of first-generation panel unit root and
stationarity testsLLC Breitung IPSt MW HadriEuro-area −1.31 −1.37 −1.34 24.65 12.79
CEEC –8.87 −0.58 –4.17 98.69 14.52
Industrial −1.04 –2.49 –1.73 66.10 22.68
Worldwide –9.64 6.48 –2.60 207.30 51.70Notes: Bold entries indicate rejection of the unit root null hypothesis
at the 5% critical level and italic entries indicate rejection at the 10%
level.
LLC: Levin, Lin and Chu test in section 13.2.1.1.
Breitung: Breitung modification of LLC test in section 13.2.1.1.
IPSt:t-test proposed by Im, Pesaran and Shin in section 13.2.1.2.
MW: Maddala and Wu’sp-value test in section 13.2.1.3.
Hadri: Test with stationarity as null in section 13.2.1.4.economies with the sample ranging from January 1993 to June 2004. The start
date is chosen to exclude the high inflation period of the early 1990s. Two other
datasets containing a larger number of countries are also considered. One of these
is an industrial countries dataset consisting of 29 countries, including the countries
of the euro-area dataset, for which the sample period coincides with the sample
period for the euro-area. The other, labeled Worldwide, contains 57 non-euro-area
countries for which monthly data are available back to January 1981.
The detailed discussion in Wagner (2008a) shows clearly that all four panel
datasets exhibit cross-sectional dependence, investigated by computing the long-
run covariance matrix ofqt =
[
q1,t,...,qN,t]′
(respectively sub-vectorsthereof), by inspecting the cross-correlation functions and by cointegration anal-
ysis. In particular there appears to be evidence for the presence of common
non-stationary components, in line with the discussion above.
The results obtained when applying a battery of first-generation panel unit root
tests are reported in Table 13.2.
Depending upon the panel considered, two or three of the four reported panel
unit root tests lead to a rejection of the unit root null hypothesis. For the euro-
area dataset the three rejections only occur at the 10% level. The quite substantial
number of rejections, and hence the strong evidence “in favor” of PPP, is a typical
finding in this literature and is in line with the results discussed in Lyhagen (2000).
When the null is taken to be stationarity of theqi,tseries, however, the reported
results for the Hadri test show rejections of the null hypothesis of stationarity for
all the panels. This is consistent with the evidence concerning non-stationarity in
the RER panels, as documented in detail in Wagner (2008a), and runs counter to
the evidence from the unit root tests. However, as illustrated by Hlouskova and
Wagner (2006), the poor performance of the Hadri test, with rejections occurring
far too often whenever the data exhibit sizeable serial correlation, severely limits
the usefulness of this test, even if the panels were cross-sectionally independent.