Re: [Gretl-users] About the automatic ADF lag length selection

On Wed, 18 Feb 2009, Olle Olsson wrote: > I can't find any reference in the manual about this so I'm > asking the question here: is anyone aware of literature on how > Gretl's automatic lag length selection for the ADF test perform > compared to lag selection by the AIC or BIC criteria? I'm not aware of any literature on that, but it wouldn't be difficult to write a gretl script to make the comparison. That's "left as an exercise" ;-) Hint: set up an artificial Data Generating Process with known lag length. Run gretl's automatic lag-selection procedure and also a scripted lag-selection procedure using $aic, $bic or $hqc. Wrap this in a loop with a few thousand interations and calculate appropriate statistics -- for example, at the simplest, the relative frequencies of cases where the selected lag order is "too short", "just right" and "too long", relative to the known DGP, for gretl's automatic version and the alternatives. I suspect $aic will select an excessive lag order more often than the other criteria; other than that, I'm not sure what you'll find. Bonus: if you can show convincingly that an alternative method is better than what gretl does currently, we'll change the method. Allin Cottrell _______________________________________________