2-Step Optimal GMM

[Davidvandijcke]

Hi Gabriele

Insofar as I can tell, the GMM estimators in most of the code just use a first guess for the optimal weighting matrix, is that correct?

If you want, I can add some code to compute the 2-step GMM estimator with the asymptotically optimal weighting matrix?

Best David

[GabrieleRovigatti]

Dear David,

thanks for the comment.

I remember having coded a version with the optimal weighting matrix - in one of my experiments long ago - and, if I am not mistaken, downsides offset the advantages (again, I do not recall precisely, but sometimes the optimal W matrix could not be computed). However, if that is not true, I would welcome the addition. Thanks you very much!!

[Davidvandijcke]

Dear Gabriele

Okay, thanks for the info! Did you use simulated data for your experiments?

If you still have the code for it around somewhere, I'd be happy to test if I can get it to work consistently. It does seem to work on the data I'm using right now.

On another note, I've been coding up the estimator by Doraszelski and Jaumandreu (2019) using your prodest framework ( https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3496603). Could maybe be a good addition in the future?

All best David

On Sun, 15 Mar 2020 at 13:58, Gabriele Rovigatti [email protected] wrote:

Dear David,

thanks for the comment.

I remember having coded a version with the optimal weighting matrix - in one of my experiments long ago - and, if I am not mistaken, downsides offset the advantages (again, I do not recall precisely, but sometimes the optimal W matrix could not be computed). However, if that is not true, I would welcome the addition. Thanks you very much!!

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/GabrieleRovigatti/prodest/issues/17#issuecomment-599213972, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEA5XBST7Y5L7X2N36SABF3RHTNHHANCNFSM4LJ4TJFA .

[GabrieleRovigatti]

Dear David,

I will try to dig into my old code and let you know whether I'll be able to find the version I was mentioning. I confirm that I used simulated data.

On a different page, I would not only be interested in adding the estimator by Doraszelski and Jaumandreu to prodest, but I would be personally interested in testing it on a paper I am working on at the moment - on Italian micro-data.

Let me know whether your timing on that.

Thank you very much,

Gabriele