This paper considers unit root regressions in data having simultaneously extensive cross-section and time-series variation. The standard least-squares estimators in such data structures turn out to have an asymptotic distribution that is neither Op(T-1) Dickey-Fuller, nor Op(N-?) normal and...