Re:RE: Re: st: Residuals in Panel Data regression

[Christopher Baum <kit.baum@bc.edu>]

<>
On Apr 25, 2012, at 2:33 AM, Eric wrote:

> FE treats the unobserved time invariant component as random and, therefor, requires strict exogeneity, which means that the lagged dependent variable is ruled out as a regressor. 
> OLS with dummies (constants) only requires contemporaneous exogeneity and can include the lagged dependent variable as a regressor.

If you do

webuse grunfeld,clear
reg invest L.invest mvalue i.company
xtreg invest L.invest mvalue,fe

you will find, unsurprisingly, that the two sets of point and interval slope estimates are identical. We know that the FE results are subject to Nickell bias,
so they should not be relied upon, as the within transformation performed by xtreg,fe induces correlation between the transformed LDV and error.
But the OLS results are, by construction, identical, as the LSDV model is just another way of producing the within transformation.
In what sense do you consider the OLS results more acceptable, in terms of bias and consistency in the N dimension, than those from xtreg, fe?

Kit

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


*
*   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/