Comparing SVAR results in RATS with Eviews

[alvarezcc]

Hi! I estimated a SVAR in both Eviews and RATS but the IRF's that I get in both programs are different in direction even if the magnitud is similar (one seems to mirror the other but with the oposite direction). For what reason could I be getting these results?

On the other hand, I estimated an A-B model for the structural innovations. Comparing the estimations performed in both programs I get similar coeficients estimates for the parameters of the A matrix, but the parameters in the B matrix are strikingly different. What could be the source of these difference? (I know that there could exist differences because of the method used in the estimation: BFGS in RATS and analytic derivatives in Eviews)

Last, I would like to know if it is possible to calculate impulse responses for shocks in the exogenous variables and how to calculate 1% shocks instead of one standar deviation shock.

[TomDoan]

If FF' factors the matrix sigma, then so does F with any collection of sign flips on its columns. For most factorizations, we use the sign convention that F has positive values on its diagonal, which makes sense when we associate shock i with variable i. However, with some structural VAR's, there is no obvious association between shocks and variables---for instance in the Blanchard-Quah model, both the supply and demand shocks are interpreted as (positive) shocks to GDP. But the signs produced by estimating the model might not match with that. If you're using CVMODEL, you can get the sign pattern you want by putting the normalizing 1's at the locations you want; they don't have to be on the diagonal. If you're using a constructive factorization like BQ, you can flip the column signs if they come out wrong. For more, see section 7.5.3 in the Version 8 User's Guide.

[trangan_20142011]

Dear all,
When I run SVAR procedure, I receive the notice that “Hessian of Structural VAR likelihood is singular at starting values. Reset starting values or respecify restrictions to ensure that the model is (locally) identified”.
What should I do next?
Please give me your advice (my email: thanhnha_nh@yahoo.com). Thank you for your attention.
Regards,

[TomDoan]

Dear all,
When I run SVAR procedure, I receive the notice that “Hessian of Structural VAR likelihood is singular at starting values. Reset starting values or respecify restrictions to ensure that the model is (locally) identified”.
What should I do next?
Please give me your advice (my email: thanhnha_nh@yahoo.com). Thank you for your attention.
Regards,

You'd have to show me your program. That would usually indicate a rather serious problem with your setup.

[trangan_20142011]

Sorry, I forgot to mention it. I used Eviews. Please give me some advice. Thanks a lot.
Best regards,

[TomDoan]

Shouldn't you ask the people at EViews about that?

[trangan_20142011]

I did, but there is no reply. What should I do?
Regards,

[TomDoan]

Don't use EViews??

If you attach your program, I can take a look at it. However, again, that usually would indicate a rather serious problem with the model such as a singular covariance matrix or identical rows or columns, depending upon whether you are using an A or B form of the model.