Abstract :
This paper considers various tests of the unit root hypothesis in panels where the
cross-section dependence is due to common dynamic factors+ Three situations are
studied+ First, the common factors and idiosyncratic components may both be nonstationary+
In this case test statistics based on generalized least squares ~GLS!
possess a standard normal limiting distribution, whereas test statistics based on
ordinary least squares ~OLS! are invalid+ Second, if the common component is
I ~1! and the idiosyncratic component is stationary ~the case of cross-unit cointegration!,
then both the OLS and the GLS statistics fail+ Finally, if the idiosyncratic
components are I ~1! but the common factors are stationary, then the
OLS-based test statistics are severely biased, whereas the GLS-based test statistics
are asymptotically valid in this situation+ A Monte Carlo study is conducted
to verify the asymptotic results+
