Hi Tom,
In the users' guide, you mention the invalidity of MC procedures to produce IRFs as if it is a OLS full-VAR.
I wonder it's possible using bootstrap to generate error bands, isn't it? I do SVAR for a two-blocks model.
By doing so, I do not need GIBB sampling as you suggest in the UG 7.9 section.
So, within bootstrapping loop, I can do CVMODEL and IMPULSE to get IRFs, then run MCPROCESSIRF and MCGRAPHIRF.
Is it OK?
Many thanks,
Boostrap Near-VAR
Re: Boostrap Near-VAR
Hi Tom,
For Near-VAR, you tell us using SUR, so Kilian-Bootstrap correction is correct, isn't it?
Because if I just estimate A-B model, I do not need re-estimate VAR within the loop with Kilian BS.
That is I call VCV to re-compute %sigma from residuals generated by KilianBootDraw. However, since Near-VAR is estimated by SUR but Kilian correct OLS bias. I am not sure whether Kilian BS is correct or not as there are two different estimators (SUR vs OLS).
Should I only use standard bootstrap instead the Kilian BS? Thus, I'll do SUR within the loop. Is it right?
Many thanks,
For Near-VAR, you tell us using SUR, so Kilian-Bootstrap correction is correct, isn't it?
Because if I just estimate A-B model, I do not need re-estimate VAR within the loop with Kilian BS.
That is I call VCV to re-compute %sigma from residuals generated by KilianBootDraw. However, since Near-VAR is estimated by SUR but Kilian correct OLS bias. I am not sure whether Kilian BS is correct or not as there are two different estimators (SUR vs OLS).
Should I only use standard bootstrap instead the Kilian BS? Thus, I'll do SUR within the loop. Is it right?
Many thanks,
Re: Boostrap Near-VAR
Kilian's correction would still apply---SUR will have the same type of small-sample bias as OLS.