Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: RE: testing heteroksedasticity and autocorrelation fixed effect model


From   "Wooldridge, Jeffrey" <[email protected]>
To   <[email protected]>
Subject   st: RE: RE: testing heteroksedasticity and autocorrelation fixed effect model
Date   Wed, 2 Mar 2011 08:58:26 -0500

My comments about xtscc were incorrect. From your description I thought this was a program that estimates models of heteroskedasticity and/or serial correlation. (I guess an example of that is xtarreg.) But xtscc also computes robust standard errors for standard panel data estimators (pooled OLS, FE). In your case, clustering seems to be sufficient -- unless you want to allow spatial correlation in the cross section, in which case xtscc would be appropriate.

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Wooldridge, Jeffrey
Sent: Wednesday, March 02, 2011 6:29 AM
To: [email protected]
Subject: st: RE: testing heteroksedasticity and autocorrelation fixed effect model

The beauty of "clustering" is that it produces asymptotically valid inference whether or not heteroskedasticity/autocorrelation are problems. To be safe, you should have a reasonably large cross section relative to the number of time periods. And it seems you have that. Of course, the usual FE standard errors probably are better behaved in small samples if you do not have the above problems, but it is now acceptable to use clustering in your kind of setting.

With T around 14 the xtscc may produce reliable results if your models of hetero./autocorrelation are correct. Ideally one could compute robust standard errors after using these GLS-type procedures to guard against having the variances and covariances misspecified -- just as in the GEE literature. But I don't think Stata currently allows that. The idea is that you might get more efficientt estimators by assuming some form of variances/covariances even though the form might be wrong. Currently Stata allows this for RE estimation, recognizing the RE structure might be too restrictive and so allowing cluster-robust standard errors.

Unfortunately, comparing the cluster standard errors from FE to the nonrobust standard errors from xtscc is not fair to FE. Both sets should be fully robust.

In Chapter 10 of my MIT Press book I do show how to test for autocorrelation in the errors after having done FE. It can be made robust to arbitrary heteroskedasticity.

JMW

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Jan Lid
Sent: Wednesday, March 02, 2011 5:11 AM
To: [email protected]
Subject: st: testing heteroksedasticity and autocorrelation fixed effect model

Hi everybody,

I was wondering if it is a necessary to test for heteroskedasticity and autocorrelation in either a fixed or random effect model? Or can i just ust he cluster (csid) option that will correct in case there is heteroskedasticity and autorcorrelation and won't correct for it in case there isn't. I can´t see to find a good way of testing for thm in a fixed effect model.  

Thx!



----------------------------------------
> From: [email protected]
> To: [email protected]
> Subject: RE: st: fixed effect correcting auto correlation and heteroskedasticity
> Date: Tue, 1 Mar 2011 18:59:35 +0100
>
>
> Total N is ci 1200. Divided in ci 150 groups So large N. T is ci 12 periods if T stands for time (not entirely sure...). My paneldata is unbalanced.
>
> Is the t and N-large enough to use stscc or better to use cluster option?
>
> Thx for the help!
>
> ----------------------------------------
> > Subject: RE: st: fixed effect correcting auto correlation and heteroskedasticity
> > Date: Tue, 1 Mar 2011 12:25:43 -0500
> > From: [email protected]
> > To: [email protected]
> >
> > What are the dimensions of your N and T (roughly?) Many of the routines
> > that claim to correct for serial correlation and/or heteroskedasticity
> > are only guaranteed to work (in the sense of eliminating the problems)
> > when T is fairly large.
> >
> > If N is large and T is not very large, the "cluster" option after FE --
> > or, for that matter, RE -- is attractive. So
> >
> > xtreg y x1 ... xk, fe cluster(csid)
> >
> > where csid is the cross section identifier. The resulting standard
> > errors are completely robust to any kind of serial correlation and/or
> > heteroskedasticity. The other approaches assume parametric forms and,
> > like I said, typically rely on large T approximations.
> >
> > Jeff
> >
> > Jeffrey M. Wooldridge
> > University Distinguished Professor
> > Department of Economics
> > Michigan State University
> > 110 Marshall-Adams Hall
> > East Lansing, MI 48824-1038
> > Phone: 517-353-5972
> > Fax: 517-432-1068
> > http://www.msu.edu/~ec/faculty/wooldridge/wooldridge.html
> >
> > -----Original Message-----
> > From: [email protected]
> > [mailto:[email protected]] On Behalf Of Jan Lid
> > Sent: Tuesday, March 01, 2011 12:20 PM
> > To: [email protected]
> > Subject: RE: st: fixed effect correcting auto correlation and
> > heteroskedasticity
> >
> > Thx for the reply. The xtscc command works fine! I also tried it with a
> > random effect model since I do have some regressions that prefer re. but
> > that doesn't worl.
> >
> > the command xtscc dependent independents, re however does not work.
> > Unfortunately i still have the same problem that i can correct for
> > heteroskedasticity or autocorrelation. Is there a command that corrects
> > for both in a random effect model?
> > or is xtregar enough? or xtreg, re robust?
> >
> > How does it work in this situation exactly.
> >
> > ----------------------------------------
> > > Date: Tue, 1 Mar 2011 08:52:31 -0800
> > > From: [email protected]
> > > Subject: Re: st: fixed effect correcting auto correlation and
> > heteroskedasticity
> > > To: [email protected]
> > >
> > > Hi,
> > >
> > > You can consider the user written -xtscc- which corrects for both
> > > autocorrelation and heteroskedasticity. It also permits unbalanced
> > panel and
> > > allows for fixed effects. For further info, see:
> > > http://www.stata-journal.com/article.html?article=st0128
> > >
> > > ---------------------------
> > > Syed Abul Basher
> > > Qatar Central Bank
> > > P.O. Box 1234
> > > Doha, Qatar.
> > > ---------------------------
> > >
> > >
> > >
> > >
> > > ----- Original Message ----
> > > From: Jan Lid
> > > To: [email protected]
> > > Sent: Tue, March 1, 2011 7:44:44 PM
> > > Subject: st: fixed effect correcting auto correlation and
> > heteroskedasticity
> > >
> > >
> > > Dear statalisters,
> > >
> > > I have a question about correcting for autocorrelation and
> > heteroskedasticity in
> > > panel data. I have read many posts but are still very confused. First
> > of all my
> > > hausman test say i have to use fixed effect model so i will use that
> > one
> > >
> > > I can correct my paneldata for autocorrelation using xtregar in stead
> > of xtreg.
> > > This does not correct for heteroskedasticity however.
> > > i can use xtreg ,fe robust. This corrects for heteroskedasticity but
> > not for
> > > autocorrelation.
> > >
> > > Maybe the fixed effect model does already correct for either one or
> > both? Am I
> > > missing the obvious... i can't find a good explanation Is it possible
> > to use a
> > > newey west option somehow e.g?
> > >
> > > I also had a question about unbalanced data set. MY dataset is
> > unbalanced. Do i
> > > need to correct for this?
> > >
> > > Help is very much appreciated.
> > > *
> > > * 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/
> *
> * 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/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index