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]

Re: st: Serial Correlation


From   Gordon Hughes <G.A.Hughes@ed.ac.uk>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Serial Correlation
Date   Sat, 05 Feb 2011 12:04:51 +0000

This is too small a panel for any kind of sophisticated analysis with small N and T. --xtgls-- will almost certainly underestimate the standard errors. Probably the best option is to use --xtpcse-- with the --corr(ar1)-- option. This allows for heteroscedastic errors across countries and AR1 errors over time within each panel. With a panel of these dimensions you will have to use random effects because FE uses up too many degrees of freedom.

But, equally, important - expand the size of your panel if at all possible.

Gordon Hughes
g.a.hughes@ed.ac.uk


Date: Fri, 4 Feb 2011 15:14:33 -0300 (ART)
From: =?utf-8?Q?Gabriel_Nicol=C3=A1s_Michelena?= <hmg@mrecic.gov.ar>
Subject: Re: st: Serial Correlation

Under Heterocedasticity the OLS estimators are still consistent, but t-stastic are no the right ones because you are using the wrong VCE matrix. You solve this problem just adding robust option in the regression.

If you suspect that are individidual effects in you model, then you should use Fixed or Random Effects for a better estimation. In my opinion you should pick one or another model depending of the characteristic of the model that you are working on. In this case, previous works on the matter always help. Again, if you suspect of serial correlation, the easy solution is employ the robust VCE matrix again.

Regards!

- ----- Mensaje original -----
De: "Robert Mills" <R.Mills@sms.ed.ac.uk>
Para: statalist@hsphsun2.harvard.edu
Enviados: Viernes, 4 de Febrero 2011 12:46:39
Asunto: st: Serial Correlation

Hi all,

I'm performing a panel data regression across six countries and ten
years in Stata.

I'm a little confused as to which methodology I should use, so far I
have:

Run my regression in OLS, then used the Breush-Pagan Lagrange
Multiplier Test, which rejected the null hypothesis that the variance
of errors is zero (homoskedastic), thus OLS is inconsistent so I need
to use Random or Fixed Effects

I've used a Hausman Test in which determined Random effects to be
inconsistent, so I'm going to use Fixed Effects.

So my errors are heteroskedastic, and I need to correct for this - do
I simply use robust standard errors in Stata? Or should I use the
Huber-White Standard Errors? Or are these the same thing?

I've read that using Huber-White Standard Errors requires no serial
correlation in error terms. To check for this, I need to perform a
Durbin-Watson Test, and if I find serial correlation, use
Prais-Winsten (GLS) to correct this.

However, can you use GLS for fixed effects? And if so, how do you do
this in Stata?

Or, should I use Newey West Standard Errors, which correct for both
heteroskedasticity and for serial correlation (AR 1). This would seem
like the best option, but I'm not sure if you can use NW SE's for
fixed effects? If so, how is this done in stata?

Thanks in advance for any help you may have!

Cheers,

Robert Mills




At 07:33 05/02/2011, you wrote:
Date: Fri, 4 Feb 2011 15:14:33 -0300 (ART)
From: =?utf-8?Q?Gabriel_Nicol=C3=A1s_Michelena?= <hmg@mrecic.gov.ar>
Subject: Re: st: Serial Correlation

Under Heterocedasticity the OLS estimators are still consistent, but t-stastic are no the right ones because you are using the wrong VCE matrix. You solve this problem just adding robust option in the regression.

If you suspect that are individidual effects in you model, then you should use Fixed or Random Effects for a better estimation. In my opinion you should pick one or another model depending of the characteristic of the model that you are working on. In this case, previous works on the matter always help. Again, if you suspect of serial correlation, the easy solution is employ the robust VCE matrix again.

Regards!





- ----- Mensaje original -----
De: "Robert Mills" <R.Mills@sms.ed.ac.uk>
Para: statalist@hsphsun2.harvard.edu
Enviados: Viernes, 4 de Febrero 2011 12:46:39
Asunto: st: Serial Correlation

Hi all,

I'm performing a panel data regression across six countries and ten
years in Stata.

I'm a little confused as to which methodology I should use, so far I
have:

Run my regression in OLS, then used the Breush-Pagan Lagrange
Multiplier Test, which rejected the null hypothesis that the variance
of errors is zero (homoskedastic), thus OLS is inconsistent so I need
to use Random or Fixed Effects

I've used a Hausman Test in which determined Random effects to be
inconsistent, so I'm going to use Fixed Effects.

So my errors are heteroskedastic, and I need to correct for this - do
I simply use robust standard errors in Stata? Or should I use the
Huber-White Standard Errors? Or are these the same thing?

I've read that using Huber-White Standard Errors requires no serial
correlation in error terms. To check for this, I need to perform a
Durbin-Watson Test, and if I find serial correlation, use
Prais-Winsten (GLS) to correct this.

However, can you use GLS for fixed effects? And if so, how do you do
this in Stata?

Or, should I use Newey West Standard Errors, which correct for both
heteroskedasticity and for serial correlation (AR 1). This would seem
like the best option, but I'm not sure if you can use NW SE's for
fixed effects? If so, how is this done in stata?

Thanks in advance for any help you may have!

Cheers,

Robert Mills

*
*   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