st: Re: system estimation with dynamic panel

[Kit Baum <baum@bc.edu>]

< >

In an earlier message I wrote

Estimation of each of these equations with Arellano-Bond would be  
perfectly appropriate. There are no cross-equation constraints. It is  
a common misconception that if you can write down a system of  
equations (in this case a recursive system) you need a systems  
estimator. Systems estimators potentially gain efficiency; they are  
not needed for consistency. "Limited information" estimation, that  
is, estimating each equation separately with Arellano-Bond is just  
fine here. In the absence of cross-equation constraints, you never  
'need' to use a systems estimator.


My advice still holds re the example below. Arellano-Bond and similar  
dynamic panel estimators are often applied to models in which one or  
more of the RHS variables are considered jointly determined with the  
LHS variable. Estimation of each equation by A-B or Blundell-Bond  
should yield consistent estimates if tests show that the  
specification is adequate and the lagged instruments are suitably  
uncorrelated with the error.

And no, at this point I do not know other Stata solutions for a  
systems estimator of dynamic panels. Perhaps if you could cite a  
specific estimator that you are interested in, it would be easier to  
ascertain whether it could readily be implemented using existing  
Stata commands.


Kit Baum, Boston College Economics and DIW Berlin
http://ideas.repec.org/e/pba1.html
An Introduction to Modern Econometrics Using Stata:
http://www.stata-press.com/books/imeus.html


On Aug 5, 2008, at 02:33 , Hewan Belay wrote:


> If I may pick up a question below by Ozlem in May, that was only  
> very partially answered by Kit/Nicola: Nicola's/Kit's response was  
> an econometric one rather than a stata one. To follow up on this  
> response:
> What if the system, however, looks like this instead:
>  y(it)=b0*y(it-1) + b1*X(it) + b2*Z(it)
>  Z(it)=a0*Z(it-1) + a1*W(it) + a2*y(it)
> In other words, what if the model *does* call for a systems  
> approach: Is there a way in stata?
>
> I have grappled with this once a couple years ago, and there was no  
> response on the list which led me to believe that stata doesn't  
> have a set of commands for systems estimations of panel data  
> equations -- however, I got lots and lots of follow ups by other  
> listers over the years to my private mail who had seen my question  
> on the list, asking if I ever discovered an approach since then.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/