On the limiting behaviour of augmented seasonal unit root tests

In a recent paper, Taylor (2003) has shown that the seasonal unit root tests of Dickey et al. (1984) [DHF] have non-degenerate limiting distributions for series which admit unit roots at any of the zero or seasonal frequencies. In this note we go a stage further and show that the standard practice of augmenting the DHF regression with lagged dependent variables alters the limiting distributions of the DHF statistics in the above scenario. Associated Monte Carlo evidence shows that this may either increase or decrease (possibly even below the nominal level) the rejection frequencies of the tests, relative to the unaugmented case.