[Gretl-devel] ADF-GLS test question

Hello devels, I recently got an email from Peter Sephton (he of the critical values for the KPSS test) raising a question about our implementation of the ADF-GLS test. I was thinking of going ahead and trying to implement Peter's suggestion but then it occurred to me it would be good to get some more input -- it's been a while since I've thought about this test. Here's Peter's message: <quote> I have a question about the ADF-GLS code. I’ve been programming the maic search in Matlab and I’m not getting the same answers as GRETL and I think I know why. I think GLS detrending should be done on the entire sample, BEFORE adjustment for lags in the ADF-type test, if you look at the ERS or the Ng Perroon 2001 papers. Then the sample should be changed to remove lags+1 observations from the beginning of the sample so that the search for the “optimal” lag length is done using the model selection criterion over a common sample period, or one can simply set the lag at a fixed value, but this should be done over an unchanging observation space. It appears GRETL doesn’t do the search for the optimal lag length over a common period. This might be solved by explicitly removing the lags+1 obs at the start of the sample, but this then removes those observations from the GLS detrending. This is probably a pretty picky point but I thought I should raise it. </quote> Allin