SVAR bootstrap

For technical questions regarding estimation of single equations, systems, VARs, Factor analysis and State Space Models in EViews. General econometric questions and advice should go in the Econometric Discussions forum.

Moderators: EViews Gareth, EViews Moderator

SilvioBerlusconi
Posts: 8
Joined: Wed Nov 07, 2018 4:14 am

SVAR bootstrap

Postby SilvioBerlusconi » Tue Apr 13, 2021 2:19 am

Dear All,
when using Eviews12 there are few issues that I do not understand

1. It appears internal error 500. it seems that bootstrap method can be applied only to Cholesky factorization and not to different identification strategies using for instance the structural decomposition method. If so, can this issue be also fixed and extended to all feasible identification and decomposition methods?
2. The "coefficient" of estimated IRFs sometimes is out of the computed bands when using bootstrap (please see the attached Photo). This is strange as estimated coefficients should always be within the estimated bands. The
Thank you in advance for your help
Best
S
Attachments
irfs.jpg
irfs.jpg (26.29 KiB) Viewed 15162 times

EViews Mirza
Posts: 80
Joined: Sat Apr 22, 2017 8:23 pm

Re: SVAR bootstrap

Postby EViews Mirza » Tue Apr 13, 2021 9:45 am

Hi Silvio.

Thanks for bringing this to our attention. Would you please share your workfile with the VAR equation where this unintended behaviour occurs? That will help tremendously in diagnosing the issue.

SilvioBerlusconi
Posts: 8
Joined: Wed Nov 07, 2018 4:14 am

Re: SVAR bootstrap

Postby SilvioBerlusconi » Wed Apr 14, 2021 3:53 am

Please see the attached Workfile with 3 variables: INV, TR, and Y. This is just an example.

1. In VAR01, you can find the first model identified NOT through a Cholesky factorization (indeed, there is a -8 in Matrix A and this is commonly used when employing SVAR modelling. Please open VAR01 and then click on proc and on estimate structural factorization to see that). When you compute IRFs with any bootstrap methods, you cannot compute IRFs and bands as an “internal Error 500” appears (for the estimation method, you have to choose in the ‘Impulse definition section’ the Structural Decomposition method as the identification is NOT a Cholesky). However, only when using a Cholesky factorisation and then a “Cholesky -dof adjested” method in the “Impulse definition section” (e.g., in VAR02 identification – see matrix A and B), bands can be computed with the bootstrap method.

2. If you estimate the VAR02 by imposing a Cholesky factorization (a “Cholesky -dof adjested” method in the Impulse definition), when you estimate IRFs’ bands with 68% confidence interval (for instance - by using the Bootstrap Standard Percentile, Hall’s percentile and also others), the ‘coefficients’ (Black line) are in some cases outside the computed bands (red bands). This should not occur.

3. Last point. Should be possible to allow the interface the choose the confidence interval levels (e.g. 95%/90%/68% etc etc) even when using the Monte Carlo and the Analytic Methods?

Thank you very much for your help.
Best,
S.
Attachments
dataset.wf1
(51.83 KiB) Downloaded 203 times

SilvioBerlusconi
Posts: 8
Joined: Wed Nov 07, 2018 4:14 am

Re: SVAR bootstrap

Postby SilvioBerlusconi » Wed Apr 21, 2021 3:37 am

Dear All,
I am using the "Scaled IRFs" add-in to create bands at 90% level when using SVAR modelling.
I have found something strange since, when computing scaled IRFs, the coefficient (blue line) and the estimated bands (red line) coincide on impact, namely at time 1 (Please see the attached figure). This should not occur. Do you have any explanation? There could be a bug in the add-in?
Thank you in advance for your reply and for your amazing work.
Best,
S.
Attachments
aaa.pdf
(56.35 KiB) Downloaded 240 times

EViews Mirza
Posts: 80
Joined: Sat Apr 22, 2017 8:23 pm

Re: SVAR bootstrap

Postby EViews Mirza » Tue Oct 05, 2021 8:49 am

Hi Silvio!

I believe both of the issues you raised are now fixed. The Cholesky re-ordering issue has been fixed for a while now. The internal error 500 on custom impulse matrices fix should go out in the next patch.

SilvioBerlusconi
Posts: 8
Joined: Wed Nov 07, 2018 4:14 am

Re: SVAR bootstrap

Postby SilvioBerlusconi » Fri Dec 17, 2021 7:38 am

Dear all,
Thank you for your help.
The internal error 500 is still occurring when using bootstrap methods together with a non-Cholesky identification.
Do you have any idea for that?
Best
SB

EViews Mirza
Posts: 80
Joined: Sat Apr 22, 2017 8:23 pm

Re: SVAR bootstrap

Postby EViews Mirza » Fri Dec 17, 2021 11:02 am

We haven't pushed the fix out yet. I'll keep you posted when it goes out. However, it has been addressed.


Return to “Estimation”

Who is online

Users browsing this forum: Baidu [Spider] and 29 guests