st: Re: xtabond2 - variables in level

[Kit Baum <baum@bc.edu>]

This reflects a confusion in the notion of a model in first differences.

If we have y = b0 + b1 X + e
Then D.y = b1 D.X + D.e
and b1 is still the estimate of dy/dX. If you are using any form of  
Arellano-Bond/Blundell-Bond estimation, you are working with the  
difference transformation (or the orthogonal difference  
transformation), so that the model is fit to the differences of the  
variables specified. Thus

y_it = b0 + b1 y_i,t-1 + b2 k)_t + b3 l_t + b4 rents_t-1  properly  
becomes

D.y_it = b1 D.y_i,t-1 + b2 D.k + b3 D.l + b4 D.rents_t-1

and b4 is still the estimate of dy/dL.rents. The only way in which it  
would make sense to include L.rents in the differenced equation is if  
it were a flow, and the original model contained a stock variable.  
But I do not imagine that rents paid in a particular period can be  
considered as a stock variable.


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 Apr 9, 2008, at 02:33 , Tomas wrote:

> However, I want to find out whether it is possible to estimate
> that function using xtabond2. In Stata, I would write
>
> xi: xtabond2 y yL1 k l rentsL1 i.year, gmm(yL1) iv(i.year k l rentsL1)
> noleveleq small
>
> where all the variable are in logs, and *L1 means one period lag. If I
> am not wrong, if I do so the coefficient I get for rentsL1 will be on
> its first difference (i.e. rentsL1 - rentsL2). Is there a way to have
> the coefficient for level of rents, keeping all the other variables in
> first difference?
*
*   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/