creating negative shocks in a VAR model

[nazif]

Dear Tom
Is it possible to create a negative shock in simple VAR model using impulse instruction. In my paper I am planning to compare the effects of negative and positive credit shocks. Could please provide me a simple example?
Thanks

[TomDoan]

If you have a factor of the covariance matrix, you can get another factor by flipping the signs on any combination of columns. For instance, if you have a four variable VAR and want the second shock to have the opposite sign from what would be created by the factor process (call the standard factor F), then do

Code: Select all

compute flipper=||1.0,-1.0,1.0,1.0||
compute f=f*%diag(flipper)

and use the new F matrix in IMPULSE.

[nazif]

Dear Tom,
Thanks. I have further questions about shocks. How can I change the procedure if i want to obtain negative one standard deviation shocks instead of one unit? Can apply the same thing for also MCGraphIRF procedure. If not, would you show me how to introduce negative shocks using this procedure?
Regards

[TomDoan]

Don't try to change MCGRAPHIRF. Instead, alter (a copy of) MCVARDoDraws to use the factor that you want:

http://www.estima.com/forum/viewtopic.php?f=7&t=333

[nazif]

Dear Tom,
I have further questions about negative shocks. if i change the flipper=||1.0,1.0,-2.0,1.0,1.0||does that give me negative 2 unit shocks to the third variable?
Besides, my estimates shows that flipping the row of the factor in this way just produce the same shocks but with opposite sign. Therefore it does not account for asymmetry. So how can i estimate the asymmetric effects of monetary policy if i have a following simple model?

Thanks for any help
Nazif

Code: Select all

system(model=creditmodel1)
variables LIP LCPI LM1 INTBRATE LCRE
lags 1 to p
det constant lpetpri{1 to 4} ffr{1 to 4} indus{1 to 4}
end(system)
estimate(noprint)
*
dec frml[rect] bfrml
nonlin cr1 rf1 rf2 rf5 uf1 uf2 uf3 tf1 tf2 tf4
frml  bfrml = ||1.0,0.0,0.0,0.0,0.0|$
                cr1,1.0,0.0,0.0,0.0|$
        			 rf1,rf2,1.0,0.0,rf5|$
					 uf1,uf2,uf3,1.0,0.0|$
					 tf1,tf2,0.0,tf4,1.0||
compute cr1=rf1=rf2=rf5=uf1=uf2=uf3=tf1=tf2=tf4=0.0
cvmodel(b=bfrml,method=bfgs,pmethod=genetic,factor=bfactor) %sigma
compute flipper=||1.0,1.0,-1.0,1.0,1.0||
compute f=bfactor*%diag(flipper)
impulse(model=creditmodel1,factor=f,labels=||"IP","Prices", "M1", "INBRATE", "LOANS"||,window="Impulse Responses(High)", steps=nsteps)

[TomDoan]

Dear Tom,
I have further questions about negative shocks. if i change the flipper=||1.0,1.0,-2.0,1.0,1.0||does that give me negative 2 unit shocks to the third variable?

That's correct.

Besides, my estimates shows that flipping the row of the factor in this way just produce the same shocks but with opposite sign. Therefore it does not account for asymmetry. So how can i estimate the asymmetric effects of monetary policy if i have a following simple model?

Thanks for any help
Nazif

You can't. A VAR is linear in the shocks. You need some form of non-linearity, like a threshold effect, in order to get asymmetrical responses.

[nazif]

Dear Tom

Thanks for your help. I finally obtained my threshold VAR estimates. After estimating the threshold level of inflation by runing Tsay's (1998) rats codes, I code the following VAR model for two different subsamples, the results of impulse responses for 1 unit shock are already attached.

As expected, I obtained quite different responses in the high inflation regime compared to that of low inflation. I would like to ask you how to scale low inflation impulse responses with respect to high inflation. Does it make sense to equate them in terms of size? I find a paper on Threshold VAR applying this kind of normalization. The title of the paper is "Monetary policy shocks, systematic monetary policy, and inflation regimes. Results from threshold vector autoregressions" you can reach the working paper of Mandler by the following link.

http://www.iwh-halle.de/d/start/News/wo ... andler.pdf

I would really appreciate if you clarify this point.

Best Regards,

Nazif

[nazif]

Dear Tom,
I just got response from the author. For the impulse responses in the high inflation regime the standard deviations of the structural shock in the low inflation regime are used, so he equated the mean response in the first period . Does this make sense? or would it be more useful to scale the responses of each variable in
terms of standard deviations as in the Enders RATS manual on page 87?
Thanks
Nazif

[TomDoan]

Dear Tom,
I just got response from the author. For the impulse responses in the high inflation regime the standard deviations of the structural shock in the low inflation regime are used, so he equated the mean response in the first period . Does this make sense? or would it be more useful to scale the responses of each variable in
terms of standard deviations as in the Enders RATS manual on page 87?
Thanks
Nazif

Equating the mean response emphasizes the difference in the shape of the responses rather than the sizes. That certainly might be useful in some circumstances.