Questions on irfs and confidence bands

[fuyangzhao]

Hi Professor!

After running the Bayesian estimation, I got the figure in the attachment, in which the confidence bands look strange and are not smooth at all. Is that normal?

Another question is that the irfs (solid lines) do not always lie in the middle of the bands. This is even more strange, since after estimation with Metropolis iterations, the parameters and the variance matrix are set to the posterior mean. Irfs are computed at M_.params and M_.Sigma_e, meaning that they must lie exactly in the middle of the confidence bands, aren't they?

Thanks in advance!
Fuyang

[jpfeifer]

That looks really weird. How did you generate those figures?

[fuyangzhao]

That looks really weird. How did you generate those figures?

The irfs are extracted from oo_.irfs, and the confidence bands are extracted from oo_.PosteriorIRF.dsge.HPDinf/sup.

Lately I found that irfs stored in oo_.PosteriorIRF.dsge.mean lie in the middle of the confidence bands. But they are slightly different from those stored in oo_.irfs. Is that normal?

And...what can I do about the confidence bands? Can I make them smoother?

Thanks!

[jpfeifer]

Yes, that is normal. You are looking at different concepts. The

Code: Select all

PosteriorIRF.dsge.mean

are the mean IRFs, averaged over the posterior parameter draws. The stoch_simul IRFs in

Code: Select all

oo_.irfs

are IRFs at the mean parameter draw. This conceptual distinction explains why the latter does not lie in the middle of the HPDIs.

To get smoother results, it often helps to use more draws (both mh_replic and sub_draws).

[fuyangzhao]

Thank you Professor! I will try more!