Output Problem: User Specified Impulse Responses in SVAR

[csm11]

Hi

I am trying to generate user specified impulse responses in a 3*3 SVAR. I want to generate 1% of GDP shocks as opposed to standard devaition shocks. I ran my SVAR as usual, then created a mtarix with the size of the shocks. Since i want to shock each variable by 1%, and i want to see the response of the whole sysytem, my matrix looks like this;
1 1 1
1 1 1
1 1 1
I enter the name of that matrix under user specified, since I'm using the drop dowm menu commands. My problem is that there is no variation in the impulse responses. The output i get in the second and 3rd columns is an exact replica of the 1st column, which leaves me to believe I only generated a response to the first shcok, not the second and third. Attached is a copy of one of the IRFs.

How do I get EViews to compute responses to all the shocks so that I get accurate results for responses to the second and third shock in columns 2 and 3? I am using EViews 9.

Thank you.

[EViews Glenn]

Can you post the workfile with the estimated SVAR, etc.?

[csm11]

I have attached a sample. The matrices a and b are the short run restrictions. "SVAR model" is the model before imposing restrictions a and b. The matrix "user_specified" are the one unit shocks I'm trying to generate and the graph "user_specified_irf" are the problematic impulse responses I am getting which just replicate.

[EViews Glenn]

Your responses should be the same by row since your user-specified response basically defines responses of a given variable X to impulses that sum across all of the corresponding impulse-responses in the unit shock case with equal weights. Another way to think about this is that one can think about decomposing the response calculation as first computing the responses in the unit shock (identity) case, and then applying the initial impulse weighting to those responses. In your case, the matrix of ones simply sums the values.

You can verify this by taking the results for the Residual - one unit shock case and summing the results in the table for each period. You'll see that they equal your common value.

[csm11]

Thank you. I understand where the values are coming from now. However, what I'm actually trying to generate is the response of variable Y 1% of GDP shock to variable X. My variables GOV1 and CA_DEF are percentages of GDP, and output is a growth rate, so I assume a 1 unit shock to these variables is a 1% of GDP shock.

In this case, how do isolate the response of a variable X to a 1 unit shock in one variable only, not the summed up impulses? Would that be the same as using the Residual-1 Unit shock?

Also, does the user specified function sum up the responses because of the way my user specified matrix is laid out, or is that generally how user specified responses are calculated? I ask this because when I change the user specified matrix to read
0 0 0
0 0 0
0 1 0

I still get the same values for the single variable that I shock, as I would if I had shocked all variables. The same happens if u use the vector
1
1
1

and ask it to shock the second variable only.

[EViews Glenn]

Just include one shock when you do the impulse-response. By default, we list all responses and all shocks.
Or use the one-unit shock which is equivalent to the identity matrix

[caromartinez]

Hello,
I have a similar question.
I have a estimated SVAR and I generate impulse response to a shock on the first variable of my model.
However, I want to generate a specific size for the shock using the structure of my SVAR. How can I do that?

[dakila]

Use the sirf add-in.