> Tang writes
> Anybody has comments? I am a dummy at this stuff.
> Thanks
> . webuse grunfeld
> . mvreg invest mvalue kstock = L(1/2).invest L(1/2).mvalue L
(1/2).kstock
>
> Smells like a panel; VAR to me... remember that VAR is just regression
on the
> same set of lagged regressors for each of the dependent variables. This
> implmentation imposes the constraints that the coefficients do not
differ
> across panels, but isn't that what a panel VAR implies?
> A VAR Is merely the regression of y1,y2,..yn on P lags of y1,y2,..,yn.
> Since each regression has the same regressors the estimation is >
efficiently performed by OLS on each equation. mvreg takes account of the
> fact that (X'X)^{-1} need only be computed once in this case. If
> y1,...yn are organized as a panel, then the VAR says that y1(i) is > >
regressed on y1(i),y2(i),..yn(i) and so on for each i. That is what the >
command above is doing: OLS of invest, mvalue, kstock on two lags of >
each of those variables. That is a VAR on panel data as far as I >
understand it.
> Kit
Yes. But I think an important point not mentioned so far is that usually
the researcher would like to introduce fixed effects in the panel VAR. My
intuition is that in this case simple OLS will not do the job. Similar
problems should arise as in the estimation of a dynamic panel data model
with fixed effects, i.e. some kind of Nickel-bias.
I think a Stata implementation of a panel VAR is related to some
modification of xtabond or xtabond2. Tang needs to extend xtabond to work
on a system of equations.
Hope this general comment is helpful,
Falko.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/