ARDL procedure appears to take care of multicollinearity(Pesaran et al 2001 page 299, although one should be mindful of dummies, see Pesaran et al 2001 page 307, footnote 17,
yes, I agree that ARDL is used to avoid serial correlation and endogeneity but I do believe it can suffer from those diseases when not adequate cared for.
It is an issue and that is why a researcher cannot only really accept any criterion (e.g., SBC) derived model without checking and it is the checking that ultimately leads to attempting more lag augmentations.
We might have to agree to disagree on some aspects of ARDL.
If a model is specified adequately, whether a priori or a posteriori based on a variety of diagnostic checks, it should be useful for interpretation, analysis and forecasting.
We are not disagreeing in one sense. I accept everything you said above. We differ because you earlier you were so sure but you can‘t be. Here, you covered it cleverly by posteriori and a priori. All that I am saying is that simply running an ARDL and obtaining the results thru a selection criteria does not suggest that all is well. I really don’t believe that passing a variety of diagnostics tests assures one that model is excellent. Just acceptable
ARDL procedure appears to take care of multicollinearity(Pesaran et al 2001 page 299, although one should be mindful of dummies, see Pesaran et al 2001 page 307, footnote 17,