Is it possible to construct VAR so that one x affects y & Z?

[Eviews90a]

Sorry for my ignorance and non-technicality here.

I know the idea of VAR is (of C-decomposition) A affects B with lags, B affects C with lags, and so on.

BUt is it possible to construct or write a code so that A affects B and C at the same time, which simulutenously affect D?

I think it is a matter of variance and covariance matrix, but I don't know if it possible and how to imterpret it in Eview8.

Please help, anybody?

[BobJ]

The idea behind a VAR is that the lags of the endogenous variables affect each other (lags of x, y and z affect each other, dpending on the identifying assumptions you use). You can certainly do what the VAR that you're describing. It would not be a VAR per se but a near VAR (where some coefficients are restricted to be zero). The only advantage this has over simple linear regressions (if you run 3 of them) is that you can estimate it using Seemingly Unrelated Regressions (SUR) which allows for correlation in disturbance across time and equations. I suppose you can do it in eviews by estimating the VAR and then imposing parameter restrictions (zero restrictions). Haven't tried it though.

[Eviews90a]

Thank you, BobJ (and I am sorry I did not reply for long).

[EViews Glenn]

The EViews VAR doesn't allow for parameter restrictions. The easiest way is just to make a system object from the VAR estimated without restrictions, and then to impose the restrictions yourself in the system.

[trubador]

The gist of VAR approach is not the parameter estimation per se. It is the impulse response analysis that really matters. So yes, what you need to do is proper factorization of the variance-covariance matrix of residuals. If you want to carry out different factorizations other than the default (i.e. Cholesky), please do a (re)search for "Structural VAR estimation"...