Panel SVAR

[shvercer]

Hi everyone,

I would like to know whether it is possible to estimate Panel VARs in EViews, and then impose identification restrictions, such as Cholesky or other short-term restrictions (eg like Blanchard and Perotti 2002 on US SVAR)? In addition, do you know whether it is possible to then compute impulse responses and variance decomposition for the panel SVAR?

I am in the start of planning and working on my thesis, and at this point it would be very helpful for me to know whether such a feature exists. It appears that RATS does have this feature, but I have no experience of that program at all. On the other hand, I have used EViews for a while, and I am currently using EViews 6.

Thanks a lot for your help.

Rilind Kabashi

[EViews Gareth]

Could you define what you mean by a panel VAR?

[shvercer]

Hi,

Thanks for your reply. I was trying to replicate some papers which use the following:

The Panel VAR in structural form is

A0Zit = A(L)Zit-1 + CXit + Eit;

Zit is a vector of endogenous variables in log levels. Xit is a vector with the country-specific intercepts (ci), country-specific linear trends (tit) and year dummies (dt). Matrix A0 captures the contemporaneous relations between the endogenous variables. Matrix A(L), is the matrix polynomial in the lag operator L. Matrix C contains the coefficients of the country fixed effects, the country-specific linear trends and the time fixed effects. The vector Eit contains the orthogonal structural shocks to each equation of the VAR.

Premultiplying by A{-1} gives the model in reduced-form

Zit = B(L)Zit-1 + DXit + Uit

In order to recover the structural shocks in Eit from the reduced form, restrictions are imposed to the A0 matrix. Most authors using panel VARs here use Cholesky, thus restricting the upper triangular to zeros. However, in country studies (ie standard time series VAR) some authors also impose other short-run restrictions by restricting various elements of the A0 matrix, and not only setting the upper triangular to zeros.

So I was wondering whether it is possible to do all this in EViews, either with Cholesky or other short-run restrictions. In addition, is it possible to compute impulse responses and standard errors. I am currently using EViews 6, but I think I will also have access to EViews 7.

Once more, thank you very much for your help.

Best regards,

Rilind

[EViews Gareth]

Well, that doesn't make it much clearer...

EViews has no Panel-specific VAR estimators built in. It does not do routines such as those suggested by Binder and Pesaran 2005 (which, as far as my limited research shows, appears to be one of the current favourites).

However you can estimate a normal VAR inside EViews. It will work in the same way as a standard VAR on the stacked data, with the exception that lags will not cross across cross-sections. It offers no options for fixed effects, although, of course, you can always create those with dummy variables. I can make no recommendation as to whether such a VAR is appropriate for your model.

[einstein999]

Hi,

I have additional question to this.

Will such "panel var" on stacked data have correct IRFs estimated?

Eviews estimates IRFs and calculates variance decompositions as well as structural factorization BUT.... Are the calculations correct?

Thanks in advance