On the asymptotic tt-test for large nonstationary panel models

The asymptotic t  -test for the long-run average in a heterogeneous nonstationary panel model is derived. The asymptotics of the Least Squares Dummy Variable (LSDV) and of the Pooled-OLS (POLS) estimators for the slope parameter are studied under various circumstances (serial correlation, strong cross-sectional dependence in the errors and in the regressors and mixed stationary/nonstationary errors) and a modified estimator of the asymptotic variance is derived. The asymptotic variance is computed up to a simple transformation of the residual and no nuisance parameters need to be estimated. The resulting tt-statistics are shown to have a standard normal limiting distribution. Asymptotic tests based on the standardized version of the tt-statistic are shown to have good power properties, and the correct size, even for nn as small as 25.