Re: Re: st: RE: FGLS vs. OLS

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

<>
On Jul 17, 2013, at 2:33 AM, Jonathan wrote:

> This makes sense. But my question is--if sureg uses FGLS, and if FGLS
> yields coefficient estimates that differ from those of OLS, then why
> do the sureg and regress commands (which should be using FGLS and OLS,
> respectively) yield the exact same regression coefficients? Is sureg
> perhaps not really using FGLS?

sureg does use FGLS, but the name of the command refers to seemingly unrelated regressionS, plural.
You do not have regressionS, but rather a single regression. The FGLS aspect of SUR relates to taking
advantage of the cross-equation correlations of OLS residuals. As you have only one residual series, there are
no such correlations. If you do something like

sureg (y1 x1) (y2 x2)

you will see that FGLS does not give you the same point estimates unless (a) the residuals are numerically orthogonal or
(b) x1 = x2, as mentioned elsewhere. Otherwise estimating these equations as a system will matter.

Cheers
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
                                                                                                   | http://www.crup.com.cn/Item/111779.aspx	


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/