Published Online: 2013-06-01
Published in Print: 2014-02-01
The definition of convergence, in this literature, hinges on the time series concept of stationarity. In this literature, stationarity is assumed to imply convergence irrespective of whether the underlying specification used in testing contains an intercept or a linear trend. This issue has attracted attention in the context of absolute or conditional convergence in the growth literature where the issue of convergence has been extensively analyzed, see Dufrénot, Mignon, and Naccache (2009) and Stengos and Yazgan (2013) for a discussion of the different definitions of convergence.
This can be explained by the following example (see De Grauwe 1996, 97). Let us suppose an improvement in the terms of trade of the home country following a shift in world preferences in favor of the products of that country. As a result, the country will experience an improvement in its current account position causing a need for real appreciation to re-equilibrate the current account. In this case, the nominal exchange rates and domestic prices may even move in opposite directions, whereas in a PPP framework they move in the same proportion.
The source and the description of the data used and the countries covered in the study are given in Appendix B.
Although these estimators are consistent for
asymptotically normally distributed for
they are also known to exhibit nonstandard behavior when
For instance, they have a non-normal limit distribution for
and they converge to unity in probability and are inconsistent for
d>1 (see Shimotsu and Phillips 2005, 2006)
Hence, FEWL estimators cannot be used under the null hypothesis of test 3 below. Nevertheless we still used them for this case also for completeness as they yielded similar results to the others.
The results remain qualitatively same across different choices of υ such as υ=T0.50, 0.55, 0.60, 0.7
It becomes clear from the table that the quantiles of the reported distributions converge to those of the standard normal as T increases, but slowly and show significant differences across estimators. The graphics and some summary statistics of these distributions are available upon request.
These countries are: Brazil, China, Hong Kong, Costa Rica, Hungary, Iran, Ireland, Saudia Arabia.
Although, the underlying individual tests are not cross-sectionally independent, under the null, the fraction of rejections is expected to converge to α, as N and T→∞, where α is the size of the underlying test.
To conserve space we only report the results of the FELW, FELWd, 2FELW and the 2FELWd estimators as the other two estimators give very similar results. We also do not report the results for the 10% significance level for the same reason. These results are available upon request. As mentioned above, FELWd and 2FELWd apply (linear) prior detrending the data. Therefore we also control for linear trends that may be present in the data via these estimators.
We use T=500 critical values.
The reported results were obtained by using 2FELW estimator of d. However we obtained qualitatively similar results with other 3 estimators (FELW,FELWd, 2FELWd) of d. The 780 pairs are obtained from 40 countries by excluding Colombia, Costa Rica, Egypt, Pakistan, Sri Lanka, Trinidad, and Uruguay from 47 countries which can be found in Appendix B. These countries are excluded because of the unavailability of trade data.
This result may due to the problems in constructing the trade data as explained in Appendix B.
Following the suggestion of Charles Engel we also included the real exchange rate volatility among our regressors which did not turn out to be significant. These results are available upon request
TRADE becomes significant in Asia and Europe with the unexpected sign. One should be cautious about this result since this may also due the data problems as mentioned in Appendix B.
Comments (0)