EViews.com

Hi everybody

I'm using Eviews 7.2 and a stacked panel dataset.

How does Eviews estimate the coefficients of the var/vecm model in a panel setting? Is the var/vecm estimated for each cross-section-entity separately and afterwards the coefficients are averaged? I don't need to know the exact formulas, only the concept.

Thanks a lot for advice!

Cheers, Harry

The VAR/VEC estimation in EViews is not panel style estimation. It performs exactly the same calculations that would be performed if the data was just regular stacked data, apart from the fact that EViews know where lags end (i.e. the lags will not go across cross-sections).

Hi Gareth

Thanks a lot for the quick answer!

I still haven't figured out how Eviews can display one coefficient per variable (and time lag) in the var/vecm. How can Eviews consider stacked data as a time series of one country (e.g.)?

I'm not sure I understand your question.

time: t = 1,...,T
cross-section: i = 1,...,N

A time series first order var would look like this:
y(t) = a1 + b11y(t-1) + b12x(t-1) + c1(t)
x(t) = a2 + b21y(t-1) + b22x(t-1) + c2(t)
The coefficients a1, b11 etc. would be estimated by using OLS for the y(t)-equation and the x(t)-equation.

A panel first order var could look like this:

y(1,t) = a11 + b111y(1,t-1) + b112x(1,t-1) + c11(t)
y(2,t) = a21 + b211y(2,t-1) + b212x(2,t-1) + c21(t)
...
y(N,t) = aN1 + bN11y(N,t-1) + bN12x(t-1) + cN1(t)

x(1,t) = a12 + b121y(1,t-1) + b122x(1,t-1) + c12(t)
x(2,t) = a22 + b221y(2,t-1) + b222x(2,t-1) + c22(t)
...
x(N,t) = aN2 + bN21y(n,t-1) + bN22x(n,t-1) + cN2(t)

If each equation would be estimated by using OLS you would get different coefficients for each cross-section.

I guess Eviews doesn't consider different coefficients for each cross-section, so we have:

y(1,t) = a1 + b11y(1,t-1) + b12x(1,t-1) + c1(t)
y(2,t) = a1 + b11y(2,t-1) + b12x(2,t-1) + c1(t)
...
y(N,t) = a1 + b11y(N,t-1) + b12x(N,t-1) + c1(t)

x(1,t) = a2 + b21y(1,t-1) + b22x(1,t-1) + c2(t)
...
x(N,t) = a2 + b21y(N,t-1) + b22x(N,t-1) + c2(t)

Does Evies estimate these coefficients by pooled OLS (minimizing the sum of squares for all t an i)?

Sorry for the confusing display, I don't know how to incorporate mathematical formulas into my post.

It does least squares on the stacked data, which is equivalent to your 2nd specification.

Thanks for the clarification. This is pretty helpful for me.
However, one final thing remains: Is it really true that Eviews estimates a single constant per equation? In other words, there are no fixed-effects estimated. Right?

Correct.

Hi everybody,

is it possible to use EViews to estimate a pooled panel VAR with fixed effects? How would I include these? At the moment my data is stacked so I don't see how to include dummies.

I would highly appreciate your help!

Thanks,
Bobi

As already mentioned, EViews doesn't have specific panel VAR estimation built in. If you want fixed effects, and are happy using standard OLS on the stacked data, you'll have to create dummy variables yourself.

Hello, I am also wanting to include fixed effects in a VAR model but am coming up with similar technical problems when attempting to include dummies in the VAR specification - eg"Near singular matrix"

I have developed an approach to this issue, and would appreciate feedback on its robustness, or any flaws etc.

You can do VAR analysis without using the VAR tool in EViews, rather by simply entering in the equation one dependent variable at a time. So to test a VAR relationship between X Y and Z using 3 lags you can enter the equations
ls x c x(-1) x(-2) x(-3) y(-1) y(-2) y(-3) z(-1) z(-2) z(-3)
and so on for the other terms y and z.

Thus if you have panel data, by estimating a VAR in this way you can opt to include fixed effects as part of the estimation process.

You can also opt to include period effects. I have not heard of period effects being incorporated into a VAR analysis but I can't see why they shouldn't be, when you have panel data.

With this approach you lose access to the other analytical features built in to the EViews VAR module, but at least you can do fixed effects!

Comments appreciated.