Automatic lag selection

[vapniks]

It would be nice if dynardl could do automatic lag selection, e.g. using a stepwise method, or if there was another routine to help with this process.

[hannesdatta]

Hi all,
what's your current thinking about this? Is this still on the wishlist?

Considering formula (9) in Philips (2018, Am. Journal Political Science)...:

...we could iteratively try out up to p and q lags here, and return as an optimal model the one with highest BIC and NO autocorrelation.

I have this code ready (but still a bit messy), wrote it for my ongoing project (https://research.tilburguniversity.edu/en/publications/universality-or-differences-in-marketing-elasticities-in-emerging) (no mentioning of dynardl here yet, I'm working on the revision).

If there is interest, I could try to incorporate that one.

It would be good to think this through in terms of infrastructure.

Maybe passing an estimated dynardl object to a new method, which then automatically expands the lag structure, where needed?

[andyphilips]

Hi @hannesdatta! Soren and I are rushing to get another project done at the moment, but we plan on conferring in the next few weeks to talk about various possible dynamac updates...we will definitely talk about your ideas given here and get back to you.

Thanks,
--AP--