EViews.com

Hello

There is an option of automatic selection of lag length in Dickey-Fuller GLS test (depending on info criterion chosen). It seems that there is an inconsistency in results obtained.
Please find attached a workfile with a series of 26 elements (some of them are missing values). If I perform DFGLS test with automatic selection of lag length (with Akaike Info Criterion and max lags = 4) on this series I get the optimal lag equal to 3 with AIC =0.636060.
However, the value of AIC equals to 0.619124 for lag length=1, which is less than AIC for lag length=3 and is in fact the smallest value of AIC for lags = 0,1,2,3,4. This means that the automatic selection should have given us the optimal lag = 1, not 3.

Is this really a mistake in EViews calculations or maybe you use a different version of AIC for choosing an optimal lag length?

Before I look at the attachment, a question: were you comparing the information criteria using the same samples?

I didn't change the sample at all. 1-26.

Which means you were using different samples, since each time you change the lag length, a different number of observations is chopped off from the start of the sample.

So, for correct comparison I should estimate test on sample 6-26 for all lags, am I right?

Yes.