RE: st: Testing for Autocorrelation after Xtreg?

["Wright, Brad" <bradwright@unc.edu>]

I think I may be able to do the Breusch-Godfrey test by hand as follows:

Following xtreg, fe:
1. predict residuals, e
2. reg residuals l.residuals $x1 (where $x1 = original right hand side variables)
3. di chiprob(e(df_m), e(N)*e(r2))

Anyone see a reason why this wouldn't work?
________________________________________
From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] on behalf of Wright, Brad [bradwright@unc.edu]
Sent: Tuesday, January 18, 2011 5:23 PM
To: statalist@hsphsun2.harvard.edu
Subject: RE: st: Testing for Autocorrelation after Xtreg?

Fabio,

Are you suggesting running a Hausman test to compare two FE models, one that controls for autocorrelation via xtregar and one that doesn't? If so, I don't see how this would work, as autocorrelation doesn't bias coefficient estimates, only standard errors. So both models should yield the same coefficients, and we would fail to reject the null--but that doesn't really tell us anything about efficiency as far as I can tell. If I'm off the mark, please enlighten me.

-Brad
________________________________________
From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] on behalf of Fabio Zona [fabio.zona@unibocconi.it]
Sent: Tuesday, January 18, 2011 4:57 PM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: Testing for Autocorrelation after Xtreg?

just one more information: but test for autocorrelation precedes or follows the hausman test? shouldn't you run a housman test with the xtregar option, in case of autocorrelation?




----- Messaggio originale -----
Da: "Brad Wright" <bradwright@unc.edu>
A: statalist@hsphsun2.harvard.edu
Inviato: Martedì, 18 gennaio 2011 22:48:57 GMT +01:00 Amsterdam/Berlino/Berna/Roma/Stoccolma/Vienna
Oggetto: st: Testing for Autocorrelation after Xtreg?

I need to do a specification test for autocorrelation following a fixed effects regression using xtreg. Unfortunately, bgodfrey, dwatson, durbinalt, and the like don't seem to work in that context. Any suggestions for how I might go about doing this?

-Brad
*
*   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/

*
*   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/

*
*   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/

*
*   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/