Ardl aic selection: no lag for dependent variable in the short run

[Fabiana De Cristofaro]

Hi all,

reading all the usual general forms of an ARDL reparametrised into an ECM there is always the first difference of Y in the short run.

I run:

ardl y x1 x2 x3 x4, maxlag(2) ec1 aic

and I got ARDL(1,0,1,1) process. Sometimes even using maxlag(4).

In this way I cannot have Y différence in the short run. I was wondering if it is ammissibile or maybe I need to force the command. I have annual data and 50 obs.

Thank you for your attention

Fabiana

[Sebastian Kripfganz]

In principle, there is nothing wrong with a model in which there are no short-run terms of Y. That's just what the data tells you. You could of course force the command to include such a short-run effect by manually imposing a higher lag order, e.g.

ardl y x1 x2 x3 x4, maxlag(2) ec1 aic lags(2 . . .)

You will probably find that the short-run coefficient associated with the first difference of Y is statistically insignificant and therefore can be omitted, which is what the automatic lag selection does.