R-Square/Goodness of Fit - FGLS panel data model

[Bimal Krishna]

Hi

I am using FGLS model due to auto-correlation, heteroskedasticity and cross sectional dependence in my panel data. Can anyone please advise how do i interprete/find the goodness of fit in the absence of R-squared value.

Bimal Krishna

[Carlo Lazzaro]

Bimal:
welcome to this forum.
Your query does not give enough details (eg, is your panel N>T or the other way round? What's the -xt- command you used?) to get any helpful repies from interested listers.
Please read (and act on) the FAQ recommendations on posting topics. Thanks.

[Bimal Krishna]

Thanks a lot Carlo for your response, apologies for my ignorance.

The details requested are as follows:

1. Panel: N=4, T=12; T>N.
2. My dependent variables are Return on Assets, Return on Equity and Net Interest Margin with 10 independent variables
3.I am using xtgls, in Stata 16.0 MP, after post estimation diagnostics done using xtreg reveals problems of auto-correlation, heteroskedasticity and cross sectional dependence and robust option in stata is unable to calculate F-statistics.

The post estimation diagnostic results are as follows:





Pooled OLS estimation is same as RE estimation for all the three models.

Bimal Krishna

[Tom Scott]

Am I to understand that you have 4 (people, companies, cities, etc.)? If so, why use a regression that is meant to produce population estimates? Maybe try something descriptive?

[Bimal Krishna]

Hi Tom

The sample of four is out of a population of 6 banks.

Regards
Bimal

[Carlo Lazzaro]

Bimal:
as Tom wisely highlighted, with (4*12)=48 observations (at best, that is without missing values that would cause the omission of the related observations), you are pushing your data above the limit.
Besides, you seem to have more that one regressand (and probably more than one regression to run) with 10 predictors, which are not susteinable by a sample composed of 48 observations.
As an aside, if you had more observations you'd be correct in preferring -xtgls- to -xtreg-, due to T>N.

[Bimal Krishna]

Hi Carlo
Many thanks for your advise. I will try to increase N and T and see how things work out, however, please do advise on the goodness of fit for the xtgls model.

Bimal

[Carlo Lazzaro]

Bimal:
see -help xtgls postestimation-.