Recent panel studies of purchasing power parity have reported strong e
vidence of mean-reversion in real exchange rates. However, these studi
es fail to control for cross-sectional dependence in the data. This fa
ilure has dramatic consequences, raising the significance level of tes
ts with a nominal size of 5 percent to as much as 50 percent. It is sh
own in this paper that, controlling for cross-sectional dependence, no
evidence against the random walk null can be found in panels of up to
64 real exchange rates. This finding cannot be attributed to low powe
r, as there is ample power in panels of this size to reject the unit-r
oot null. (C) 1998 Elsevier Science B.V.