Panel data Perron and Dickey-Fuller unit tests give OPPOSING results for half of the variables in my data set.

I believe that this has to do with lag selection process.

After I saw the opposing results with:

xtunitroot fisher ldelay, dfuller lags(10)

I used:

xtunitroot fisher ldelay, dfuller lags(aic 10)

in order for the lag order to be selected automatically according to Akaike information criteria,

but it doesn't work.


I don't know: (a) how to manually select the optimal lag and (b) lets say the Perron and ADF test results are still opposing, which one should I choose?