SVAR restrictions

[nicote]

Hi,

I am writing my thesis and I am experiencing little trouble with the SVAR model and cannot find out how and where to impose restriction in my model in EVIEWS.

Basically, I have done the VAR with two lags and I only need to add restrictions for doing the structural VAR that I am looking for. The original SVAR was done in this paper : http://www.sciencedirect.com/science/ar ... 8910000197.

I feel like I am only missing a simple step. Attached are some screenshots of EVIEWS concerning my problem.

If anyone could help, I would be very very very thankful !!

Thank you for your consideration.

Best regards,

Nicolas.

( PS: Working on EVIEWS version 7 )

[trubador]

That would be the A matrix (mata) with an identity B matrix. However, since the S matrix is underidentified they further impose two long run restrictions. In that case, you'll need the svarpatterns add-in as EViews does not allow for both short and long run restrictions.

[nicote]

Thank you very much for your reply trubador. I really appreciate.

I have included the add-in after running my VAR and tried many combination but without success.
There is something that I do not get. Might be the steps

-How could I get the S matrix?

With the add-in I can only add one pattern matrix for short-run and one for the long-run. Maybe I do something wrong when adding the two pattern matrix. Could you explicite which matrix I should use for what ?

Again, thank you !

[trubador]

Short run matrix is the S matrix. However, since it is underidentified they impose two additional restrictions to control the long run impact of monetary policy shock on real exchange rate and real output rate. This is done by creating another 6x6 matrix (say longmat) with (2,6) and (5,6) elements are zero and the rest are filled with NAs.

[nicote]

Alright.

It is working as far as I know (the add-in does not give any confirmation but I had no warning message this time so I guess it is fine).

However, it provides any significant change neither in the results nor in the impulse response functions.

I thank you again very much trubador for your precious help!!!

Best regards.

Nicolas

[trubador]

SVAR is actually a different (more general) factorization method of the residuals than a simple cholesky. So it does not (and should not) change the original estimation. You can generate impulse responses, as I already mentioned here: svarpatterns

The add-in allows you to impose both short-run and long-run restrictions to obtain non-recursive orthogonalization of the error terms (as opposed to recursive Cholesky decomposition) for impulse response analysis that would make more sense from a macroeconomic/structural point of view. In order to use the add-in, you should first estimate a regular VAR model. After that, you can either supply the name of your model or the covariance matrix. The output will be a factor matrix, which can further be used in generating impulse responses (i.e. as a user-specified impulse definition). In short, this add-in aims to extend the current functionality of EViews' Structural VAR estimation toolbox.

[terrya]

If I can horn in here, is there any way we can get the residuals from the estimation using SVARPatterns? I can't see anything that allows this.

Also, I think that Bjornland probably used RATS for this work.

[trubador]

If I can horn in here, is there any way we can get the residuals from the estimation using SVARPatterns?

I think you are referring to structural residuals. svarpatterns add-in will produce a factor matrix, which can be used as an input to hdecomp add-in to generate the structural residuals.

[terrya]

Thanks.

[Helia]

Hi,
Thanks for the add-in. It is very helpful.
However, how can i get the impulse response funtions graphs? It would be just perfect if i can have them

[tiffkatie]

Thanks for your guidelines on using both short and long run restrictions on Svar. I have a question related to the sign of the shock.
Suppose I have a SVAR matrix (5,5) and the factormat matrix (5,5) after using the add-in. Looking at the elements of factormat matrix, suppose factormat(4,5)=-0.15, factormat(5,5)=0.27. Am I correct if interpret that the shock of 0.27 on the 5th element have an impact of -0.15 on the the 4th variables?

if the SVar is not symmetric and I need to measure the impact of a negative shock of 5th variable on the 4th variables, how could I do in Eviews.

Thanks

[dakila]

Yes, your interpretation is correct.
You can multiply factormat by -1. However the result will not change since svar is symmetric.

[tiffkatie]

Thanks Dakila.