Constrained VAR in EViews?

[yanivd]

Is there a way to run constrained VAR in EViews?
For example, I wish to run a model with 4 variables, say x, y , z and w. But I don't want w to depend on itself only on the other variables...
Or, as another example, I don't want x to depend on z but on the rest.

Is this handled through SVAR? Does this problem have a name? Like constrained VAR or something similar?

Thanks!

[Seeking_Knowledge]

Hello

An important part of VAR analysis is to let the 'data decide what is x and y '
Quick ---> Estimate VAR ------> Endogenous variable..... and put x or w in exognous after C
in that way at least you have defined some variables as predictors...

Ln [ w ] = c β1x β2y β2z

Ln [ x ] = c β1x β2y

not sure if I understood your problem..... but good luck

[EViews Glenn]

http://forums.eviews.com/viewtopic.php? ... stem#p8745

[yanivd]

Thanks for the prompt reply!

Let me explain some more:
Let's say that I have x, y and z. Let's assume that x is salary and y and z are work output. I know that at every time t, x(t) is simply a*y(t-1)+b*z(t-1), in other words, the employer rewards the employee according to her output. So, I know for a fact that x(t) does not depend on x(t-1) and so I would like to exclude it. What I am afraid could happen (if I do include x(t-1) in x(t)'s equation) is that it will be co-linear and will "absorb" the variance so that I will not see the dependence of x(t) on z(t-1) and y(t-1) and so cannot estimate a and b.
This is exactly what I see in the data, i.e., even though I know x(t)=a*y(t-1)+b*z(t-1) is the right equation, a and b only sometimes are significant and I am afraid it is because I include x(t-1) in the estimation.
Does this make any sense?

P.S.
Making an equation system seems like a nice solution, but losing the IRFs is critical. Do you think there's a way to recreate IRFs if I use the eqtn system?

[yanivd]

Just to stress the point - I KNOW that x(t) does not depend explicitly on x(t-1): assume that a computer calculates x(t) according to the formula I gave above (which is actually the case for my data - x(t) for the data is calculated by an automatic algorithm...)

[EViews Glenn]

What you are asking for is reasonable, though not typical of VAR analysis. Having said that, unfortunately, when estimating as a system you will lose the IRF functionality. You'd have to do the calculations yourself.

[yanivd]

I understand, thanks!

[yanivd]

Does anyone know if I can do this in the context of an SVAR model? (i.e., create some sort of restrictions that will produce such a model...)