The results from Pesaran and Shin indicated that the ARDL approach to testing cointegration can involve variables of I(0), I(1), or a combination of them.
However, if I am not concerned about cointegration and just want to calculate the long-run multiplier, can I just use an ARDL with I(1) variables? Or do I still need to do differencing?
Hi: I don't want to go into all the details ( I'm tired and it's late and the references explain it better than I could anyway ) but you can't keep it in I(1) form because you'll get the well known spurious regression problem when you estimate. So, you need to estimate it using the error correction form and then, take the estimated parameters and backout the long term multiplier. Page 29 of this shows what I mean by the error correction form.
https://web.sgh.waw.pl/~atoroj/econometric_methods/lecture_6_ecm.pdf
ADDENDUM A:
I went and looked more and found the paper that I was looking for (see below ). Note The title will make you think you're not interested ( cointegration is in the title ) but, if you remember that cointegration is not required for the ecm formulation, then it's still useful. There's one more paper that I want to link to that I haven't found yet. If I find it, I'll add it as a link.
https://www.econstor.eu/bitstream/10419/28020/1/507401956.PDF
ADDENDUM B: Here's the one that was avoiding me. Title is right up your alley but I still recommend reading what's at all three links.
