Limitation of panel cointegration: application to PPP in the EU.(RESEARCH NOTE)

While panel-based tests of cointegration have clear benefits in terms of size and power, empirical tests indicate that the panel outcome may be driven by the time series properties of a small number of panel members. Using the Larsson et al. test [Larsson et al., Econometrics Journal, 2001, 4: 109-142] to test for both cointegration and PPP amongst the nominal (euro) exchange rate, domestic prices, and foreign (euro area) prices for a panel of 12 former Central and Eastern European countries, one cointegrating vector is identified (consistent with Johansen tests of individual countries). Subsequent tests for PPP, however, strongly reject the hypothesis with a chi-squared test statistic of 55.43 (against a 5% critical value of 24.99). This is a surprising result, since individual Johansen-based tests of PPP for the countries do not reject the proposition in eight out of the 12 cases.

The results suggest a limitation in the panel cointegration test, akin to the Taylor and Sarno [Journal of International Economics, 1998, 46: 281-312] concern that panel unit root test outcomes can be driven by one or more individual series. This was because the tests were based on testing the null of non-stationarity across all members of the panel. Rejection of the null therefore implies that at least one of the series is stationary. Those authors attempted to resolve this problem by developing a new type of test that tests the hypothesis that at least one series within a panel contains a unit root. Rejection of this hypothesis implies stationarity across all panel members. This type of test suffers from severe power problems, however, and this concern over panel unit root results continues to remain.

My results show that while the Larsson et al. test seems to be robust to the identification of the cointegration rank (compared to the Johansen test of individual countries), tests of restrictions on the identified vector (in this case to test for PPP by normalizing on the exchange rate and imposing symmetry conditions on the domestic and foreign price series) seem to be driven by a relatively small number of panel members. Thus, in interpreting this panel cointegration test (which is based on the null of no cointegration), one must exercise caution. Going forward, it would be of interest to explore alternative panel cointegration tests and to examine whether a Taylor and Sarno type solution using an alternative null would provide more consistent results.

J. Beime

Department of Economics and Finance, Brunei University, Cleveland Road, Uxbridge, Middlesex,

UB8 3PH, UK

e-mail: John.Beirne@brunel.ac.uk

Published online: 25 January 2009

Print Get the Mag Weekly Updates