VAR(1)

[kristin]

Hi Everybody,

I want to replicate a study including VAR-models.

It is said that the following VAR(1) was estimated

Δy(t)=µ+Πy(t-1)+ε(t)

I learned that in a VAR the variables on the left hand side and on the right hand side have to be the same, so that you must have Δy(t-1) on right hand side when you have Δy(t) on the left hand side (what is exactly that what appears when I estimate the model in EViews.

Many thanks in advance for any help.

[EViews Gareth]

If there is a question here, it isn't immediately clear what it is...

[kristin]

Ok. I'm sorry. I try to make myself clear.

I have an ouput table that I want to replicate.

On top of the tabe it is said that the following bivariate VAR(1) is estimated:

Δy(t)=µ+Πy(t-1)+ε(t).

The output containts one equation for Δf(t) and one for Δs(t):

Δf(t)=0.009-1.771*f(t-1)+1.744*s(t-1)+ε(ft)

Δs(t)=0.009-1.676*f(t-1)+1.649*s(t-1)+ε(st).

I tried to estimate it in EViews with "Estimate VAR" or "Estimate VEC" and entered Δf(t) and Δs(t) as endogenous variables. But I never get an output like the one mentioned above as the right hand side of a VAR contains the lagged endogenous variables Δf(t-1) and Δs(t-1) and not f(t-1) or s(t-1).

So my question is how I can estimate the equations above. If I can just estimate them separately with "ls" or if I would miss important dynamics doing so.
Or do I have to estimate a VEC without including lags?!

[EViews Gareth]

Not sure I follow that - assuming you're using Δf(t) to stand for differencing, if you enter a difference term as one of the left hand side variables, EViews will report difference terms on the right hand side too.

[kristin]

That's the problem. I need differences on the left side, but not on the right side.

[EViews Gareth]

Ah I understand now. You won't be able to do that in a VAR object. You'll have to use a System instead.

[kristin]

It confused me that the guy in my study called it VAR.

Will I miss important dynamics if I estimate a System instead of a VAR?

[EViews Gareth]

The coefficient estimates will be the same. You'll lose some post-estimation diagnostics (notably impulse responses).

[Steffen]

I have a question that might fit into this context.

I want to estimate the VECM:

Δs(t+1)=α(s)*[f(t)-β(s)s(t)-µ]+ε(s,t+1)

Δf(t+1)=α(f)*[f(t)-β(s)s(t)-µ]+ε(f,t+1)

Can I estimate this with "Estimate VAR" or do I also have to choose a system?

If I can estimate it with "Estimate VAR" (choosing VEC), how can I estimate the above equation with the paranthesis?