How could I understand the command "xtgls" in the Stata

[wanhaiyou]

Hi,I have a question on "xtgls". I know "xtgls" is fit the panel data model using GLS method.
My question is that whether this model is fixed effect model or random effect model? or this
method is not make a distinction between fe model and re model?

I have a panel data and choose fe and re using hausman test, the result indicates that the re model
is good. Howerer, the model residual exist serial correlation and heteroskedasticity. Can I use xtgls to
fit this model?

Thanks very much for any suggestions.

Best regards,
wanhaiyou

[Carlo Lazzaro]

Wanhaiyou:
let's make a step backwards.
-xtgls- was conceived for dealing with small N, large T panel datasets. Hence, if your dataset is a large N, small T one, you should stick with -xtreg, re-.
If you suspect serial correlation and heteroskedasticity, just add -vce(robust)- option.

[wanhaiyou]

Thanks very much, Carlo. I want to know that xtreg,re with vec(robust) can handle with both serial correlation and heteroskedasticity?

I know that we can use xtpcse or xtscc to handle with both serial correlation and heteroskedasticity, but I am not sure these two routines
are fixed or random effect model.

Thanks for your help!

Best regards,
wanhaiyou

[wanhaiyou]

Wanhaiyou:
let's make a step backwards.
-xtgls- was conceived for dealing with small N, large T panel datasets. Hence, if your dataset is a large N, small T one, you should stick with -xtreg, re-.
If you suspect serial correlation and heteroskedasticity, just add -vce(robust)- option.

Hi Carlo
Another important question is that whether this model (fitted by xtgls or xtpcse)is fixed effect model or random effect model?
I can not understand.

Thanks
Best regards,
wanhaiyou

[Carlo Lazzaro]

wanhaiyou:
they basically estimate -fe- models, as you may read from this enligthening article :http://www.stata-journal.com/sjpdf.h...iclenum=st0084

[wanhaiyou]

wanhaiyou:
they basically estimate -fe- models, as you may read from this enligthening article :http://www.stata-journal.com/sjpdf.h...iclenum=st0084

Thanks very much for your help,Carlo. I will read the literature seriously.

Best regards,
wanhaiyou

[Grace Gondwe]

Wanhaiyou:
let's make a step backwards.
-xtgls- was conceived for dealing with small N, large T panel datasets. Hence, if your dataset is a large N, small T one, you should stick with -xtreg, re-.
If you suspect serial correlation and heteroskedasticity, just add -vce(robust)- option.

Dear Carlo,

In addition to the article provided in this discussion, are there any other articles for referencing particularly on the T>N dataset requirement for xtgls,

Regards,
Grace

[Carlo Lazzaro]

Grace.
welcome to this forum.
From the reference section of -xtgls- entry in Stata .pdf manual, you can find, among Others, the following textbook on econometrics:

Davidson, R., and J. G. MacKinnon. 1993. Estimation and Inference in Econometrics. New York: Oxford University Press.

Greene, W. H. 2012. Econometric Analysis. 7th ed. Upper Saddle River, NJ: Prentice Hall.

Greene, W. H. 2018. Econometric Analysis. 8th ed. New York: Pearson.

[Grace Gondwe]

Dear Carlo, thanks a lot for the information

[Carlo Lazzaro]

Grace:
see also https://www.stata.com/bookstore/micr...metrics-stata/, pages 274-276.

[Manh Hoang Ba]

wanhaiyou:
they basically estimate -fe- models, as you may read from this enligthening article :http://www.stata-journal.com/sjpdf.h...iclenum=st0084

Dear Carlo,
I feel that your answer may be misleading.
According to the STATA manual.pdf (and also in the article you linked), the xtgls command is based on pooled ols, and to obtain an estimate of fe, one needs to add dummy variables that identify panels.
Therefore, to obtain FE estimates for the coefficients of the model with the regression equation:
y_it = x_it*b + u_i + e_it
The STATA command used is:
xtgls y x i.id
Where id is the panel variable.

[Carlo Lazzaro]

Manh:
it may be.
But see 4 Estimation and testing of fixed-effect panel-data systems in https://www.stata-journal.com/sjpdf....iclenum=st0084