As it is clear from [Patterson, K. D. Unit root tests in time series, volume 1, 13.7]
in the absence of seasonal unit roots, seasonality is just a particular case
of autocorrelation, which is routinely dealt with including sufficient lags
in ADF-type tests or sufficiently large window in PP-type tests.
So, with the first generations tests we have two possibilities:
include seasonal dummies, or set minimum lag length to ($pd-1)
With tests having no include seasonal dummies option the second
one is the only valid way to proceed.
So, my proposition is to include option:
set minimum lag length as ($pd-1) for zero frequency (i.e. usual,
non-seasonal) unit root tests
Show replies by date