Impulse response to Cholesky 1% shock

[raphbrun]

Hi,

Eviews provides an easy way to get impulse responsed to Cholesky one standard deviation innovations. I am interested to get the impulse response to a Cholesky 1% shock, instead of a one standard deviation shock. This way I could interpret the outcome as, let's say, a 1% shock in my X variable, gives a yy% shock in my Y variable.

How can I do that?

kind regards,
Raphael

[trubador]

If your model is in logarithmic form, then one unit shock will do what you want.

For customized impulse responses, you should use "User Specified" feature. Create a matrix/vector that contains the impulses and type the name of that matrix/vector in the edit box. For example, if your model has p endogenous variables, then you should create a (p x p) impulse matrix or (p x 1) impulse vector. Here, you can fill the matrix/vector with values of your interest. You should then manually compute the resulting responses in terms of percentages.

Please refer to users' manual for more details...

[raphbrun]

Thanks for your reply. I am still a bit confused on how to get the Cholesky 1% shock. I present a simple example for sake of illustration:

I have a VAR(1) with two variables in log-differences:

x_t = c + alpha_1 x_t-1 + beta_1 y_t-1 + eps_1,t
y_t = c + alpha_2 x_t-1 + beta_2 y_t-1 + eps_2,t

Using monthly WTI oil price and European headline inflation I can estimate my model. The variance-covariance matrix of the residuals is:
204.74 1.90
1.90 0.07

And the Cholesky factorization of this variance-covariance matrix is:
14.31 0.00
0.13 0.23

You suggest using the "user specified" feature and apply a one unit shock. What matrix would you generate to obtain a 1 unit shock instead of the Cholesky factorization above?

You then say "You should then manually compute the resulting responses in terms of percentages". You mean relating the unit shock to the units of the variables in the model? My variables are log differences, i.e. growth rates, then a one unit shock is a one percent shock. Is that right? Then no need to make any further manual computation if that is correct...

Thanks again for your help,
Raphael

[trubador]

Create a user defined shock matrix:

Code: Select all

matrix(2,1) myimpulse
myimpulse.fill(by=c) 0,1

And then type the name (i.e. myimpulse) in the edit box.

As far as I understand, you are trying to estimate the impact of an oil price shock on the inflation. If that is your case, then I strongly recommend you to use Structural VAR analysis, which I believe serves the purpose best.

[mayxanh]

Hi trubador,

I am building a VAR model of the form LOG (X) LOG(Y) LOG(Z)

Then if I get the impulse function with one unit shock, the graph titled "Response of LOG(X) to LOG(Y)" should be interpreted as: response of X to 1% shock of Y?


Thank you very much.


T

[trubador]

Yes, that is correct.

[mayxanh]

Thank you trubador

T

[Kostandin]

Hi gentlemen
I dont know if I'm writing in the right place, but could you please help me with a matter. I've constructed a 7 variable SVAR model, and I wanted to know if I could perform stress tests with impulses to several variables contemporaneously, for example consider a case, when inflation has risen by 2 s.d. and at the same time net export declined for 1 s.d.? How do you think can I perform it with Eviews?
Thanks in advance

[arthur]

Hi trubador,

I am building a VAR model of the form LOG (X) LOG(Y) LOG(Z)

Then if I get the impulse function with one unit shock, the graph titled "Response of LOG(X) to LOG(Y)" should be interpreted as: response of X to 1% shock of Y?


Thank you very much.


T

Hello

Let me ask a question, guys. if we are saying "...response of X to 1% shock of Y", how we acn interpret this? say Y was 8% and becomes 9% (8%+1%), or Y was 8% and becomes 8.08 (8%+8*001)? which variant is correct?

[arthur]

guys, the question above is very urgent issue and any reply will be valuable for me.
write at least your opinions, pls

[mjfl]

Create a user defined shock matrix:

Code: Select all

matrix(2,1) myimpulse
myimpulse.fill(by=c) 0,1

And then type the name (i.e. myimpulse) in the edit box.

As far as I understand, you are trying to estimate the impact of an oil price shock on the inflation. If that is your case, then I strongly recommend you to use Structural VAR analysis, which I believe serves the purpose best.

ok got it!

[brylle28]

The identification of monetary policy shocks in structural VARs is often achieved by assuming delayed impacts on inflation and output like Cholesky economy.

[Tibovwh]

Hi,

I am a bit confused. I am interested in interpreting the responses of 3 variables (all in logs) after innovations in 2 other variables (one in log and one in percent (i.e; 1% is 0,01))

now i rescale my impulse responses by dividing everything by the one SD innovation in the variable in logs. Now i can interpret everything as elasticities? i.e dy / y / dx / x

for the impulse respones following an innovation in the variable in percent, i rescale everything by dividing by the one SD innovation in the variable in percent. However, it's a bit unclear how to interpret. the innovation is now 1. So can I now say " an X% increase / decrease in variables following a 1 percentage point increase in R* (i.e. 0,01 -> 0,02 ; i.e. 100 basis points increase)

I know it's probably not the best place for a question like this but I hope somebody could help out

thx in advance,

[zarfarzane]

[quote="trubador"]If your model is in logarithmic form, then one unit shock will do what you want.

Trubadr, Thank you for the response. Can you please introduce a reference that confirm this : when variables are in logarithm form, one unit shock means 1% shock to the variable.