Anindya Banerjee and Martin Wagner 639Note, furthermore, that for many empirical applications it may be necessary to
also allow for structural change, which is most often modeled in the determin-
istic component of the processes. In this chapter we put particular emphasis on
structural change and present some evidence that allowing for structural change
may alter conclusions drastically. This fact is to a certain extent well-established in
the time series unit root literature, starting with Perron (1989), but has found less
attention in the non-stationary panel literature.
In section 13.2 we start with a general formulation of testing for a unit root
in a panel data setting. This is first specialized to the case of panels with cross-
sectionally independent members, but is then extended to a discussion of a more
general framework, considering both cross-sectional dependence and structural
change. Several sub-sections discuss the results of simulation and empirical studies
to evaluate the properties of the tests devised for unit roots. As mentioned above,
we illustrate the methods by an empirical study of purchasing power parity and
the environmental Kuznets curve. Section 13.3 offers an analogous discussion of
testing and estimation for cointegration, and as an additional empirical study, con-
tains an analysis of exchange rate pass-through in the euro-area. In this application
issues such as structural stability of the import pass-through equations (in the face
of policy changes in the euro-area) are studied with reference to the core problem
of testing for cointegration. Section 13.4 concludes.
Three appendices follow the main text. In Appendix A we collect some details on
the datasets employed in this chapter; Appendix B contains a brief discussion on
cross-sectional dependence; and Appendix C mentions a few aspects with respect
to limit theory in non-stationary panels.
It is helpful, perhaps, before starting on the main account, to discuss here briefly
how our study differs from some of the excellent studies which are already avail-
able; see, for example, Breitung and Pesaran (2008) or the special issue of the
Journal of Applied Econometrics(2007) devoted to the topic of heterogeneity and
cross-sectional dependence in panel data models, or more particularly the chapter
by Choi (2006b) in the first volume of this handbook. Some of the material which
appears in those sources finds some repetition here, since many of the themes
which we deal with cannot be presented without context or introduction. Never-
theless, we believe our chapter contributes to the literature in three distinct and
important ways. First, it presents a unified and general formulation of testing for
unit roots and cointegration in panel data. Second, in doing so, almost uniquely
this chapter pays considerable attention to the role of structural change in panel
unit root and cointegration analysis. Structural change has been found to be a sub-
stantial issue in the time series context and continues to be so in the panel setting.
Finally, by presenting a range of evidence on the performance of the methods,
based on both simulation and empirical evidence using a variety of data sources,
this chapter attempts to demonstrate unit root and cointegration analysis in panels
in action. The material collected in Appendices B and C will also be of some interest
by adding to the discussion both on modeling and accounting for dependence and
on some key concepts of limit theory for integrated panels.