Page 2 of 2

Re: Large Bayesian VAR

Posted: Tue Nov 21, 2017 9:17 am
by anranal
I keep getting the following error Dot,diagonal and solve require a vector argument" what does this refer to?

Re: Large Bayesian VAR

Posted: Tue Nov 21, 2017 4:07 pm
by dakila
Did you create random walk prior vector before the estimation?
Could you provide the data file and description of the model?

Re: Large Bayesian VAR

Posted: Tue Nov 21, 2017 6:46 pm
by anranal
I have create random walk prior vector irw following LBVAR.pdf and fill it in the dialog box.
The data contains 39 variables from 2005m7 to 2017m8. Further, are there some rules in setting training sample and sample size?
Much appreciation for your reply!

Re: Large Bayesian VAR

Posted: Wed Nov 22, 2017 2:55 pm
by dakila
What is your Eviews version? Can you run a example of the lbvar add-in? Sample size should not include the missing observations. Training sample should be at least 30-40.

Re: Large Bayesian VAR

Posted: Thu Nov 23, 2017 9:01 am
by anranal
dakila wrote:What is your Eviews version? Can you run a example of the lbvar add-in? Sample size should not include the missing observations. Training sample should be at least 30-40.


I used Eviews8.0 to estimate the model. When I update Eviews to the latest version,everything goes well.
Much appreciation!

Re: Large Bayesian VAR

Posted: Wed Aug 08, 2018 12:37 am
by awatt43
Hi there

I am trying to estimate a LBVAR in using the add in but I have a couple of questions on the estimation procedure.

1. Is the model estimated following the hierarchical approach of Giannone et al (2012) in their paper prior selection for vector autoregressions?
2. Are the impulse responses formulated in the same manner as the Banbura et al paper which is referenced in the add in doc? I.e. can we specify slow moving versus fast moving variables for the computation of impulse responses?
3. In the add in how is the covariance prior specified i.e. the prior for the third block of dummies?

Hope that you can help with these questions.

Thanks,
Abigail

Re: Large Bayesian VAR

Posted: Wed Aug 08, 2018 1:24 am
by dakila
1. Is the model estimated following the hierarchical approach of Giannone et al (2012) in their paper prior selection for vector autoregressions?

No.
2. Are the impulse responses formulated in the same manner as the Banbura et al paper which is referenced in the add in doc? I.e. can we specify slow moving versus fast moving variables for the computation of impulse responses?

Yes. the identification is recursive (cholesky). But you cannot specify slow moving versus fast moving variables. It is related with FAVAR model. If you use FAVAR add-in you can specify that variables.
3. In the add in how is the covariance prior specified i.e. the prior for the third block of dummies?

Yes.

Re: Large Bayesian VAR

Posted: Wed Aug 08, 2018 2:14 am
by awatt43
Thank you for your quick response dakila. How is the cholesky ordering specified? Is this taken from the ordering of the variables in the model specification?

Re: Large Bayesian VAR

Posted: Wed Aug 08, 2018 2:34 am
by dakila
Yep

Re: Large Bayesian VAR

Posted: Wed Aug 08, 2018 6:24 am
by awatt43
Thank you. Is it possible to view the final coefficient estimates and the impulse responses in an extractable format rather than a frozen graph object?
Thanks,
Abigail

Re: Large Bayesian VAR

Posted: Wed Aug 08, 2018 4:20 pm
by dakila
The lbvar add-in is updated. It now includes the generalized IRF and option to save IRF.

Re: Large Bayesian VAR

Posted: Thu Aug 16, 2018 5:36 am
by awatt43
Thanks Dakila - could you explain to me the difference between the generalized and cholesky impulse response functions? Am i right in thinking that generalized allow you to estimate the impact of the shocks independent of the impact of the shock on other variables within the system, hence you do not need to specify an ordering?

Your help is much appreciated!

Re: Large Bayesian VAR

Posted: Thu Aug 16, 2018 7:19 am
by dakila
Unlike the traditional impulse response analysis, GIRF does not require orthogonalization of shocks and is invariant to the ordering of the variables in the VAR. In other words, the definition of GIRF is different from Cholesky IRF.