Anindya Banerjee and Martin Wagner 709heterogeneity across countries, only very limited evidence for an EKC arises, with
the main observation being that for several countries very implausible parameter
estimates occur.
The scarcity of evidence for an EKC for sulphur dioxide in the panel at hand is in
line with the observation that the common factors driving GDP and SO 2 appear not
to be cointegrated. This long-run disconnect between these two variables poten-
tially shows the limitation of reduced form EKC modeling and it may be necessary
to resort to more structural modeling approaches to shed further light on the EKC
hypothesis, from either a time series or a panel perspective.
13.4 Conclusions
We have sought in this chapter to provide an up-to-date analysis of the methods
involved for estimation and inference in non-stationary panels. Our overriding
objective has been not only to provide information on the tools but also to interpret
the literature and to highlight the considerable challenges that remain. Starting
with the motivating example and concluding with the appendices, this chapter has
sought to emphasize the difficulties involved in formulating hypotheses within a
panel framework, estimating them and conducting inference coherently.
Features of the data that the methods need to incorporate include (in addition to
non-stationarity) cross-sectional dependence and structural instability. For exam-
ple, we have argued that assuming cross-sectional independence of the units leads
to relatively simple sequential central limit theory that provides the asymptotic
normality of many of the test statistics. As soon as this assumption is relaxed,
matters become much more complicated and in ways which are still not fully
understood in the literature.
Modeling dependence via factors is a popular device but is only one of the
many ways of formulating the problem. Account must be taken of short-run versus
long-run dependence, which must be dealt with appropriately. The link between
cointegration and factor models in panels needs to be adequately explored (see
Appendix B).
More generally, the asymptotic theory must be put on a surer footing to deal
not only with many of the joint limiting arguments that arise in the consideration
of any form of dependence in the panel but also to deal with cases where there is
potential structural instability in the data. In this chapter we have demonstrated
some preliminary (and fairly crude) ways of modeling structural instability but a
closer look is clearly warranted. The extension of systems methods to panels – to
allow for multiple cointegrating vectors – in the possible presence of cross-sectional
dependence and structural breaks is an important task but one of considerable
complexity. Dealing with these problems should constitute fruitful areas of research
in the years ahead.
13.5 Appendix A: Datasets employed
Data for exchange rate pass-through example
Sources: Eurostat, COMEXT.