Determining if FGLS is a better estimator than OLS

[Hob]

Hi all,

I am writing some c++ code to automatically do OLS regression, however, it appears using the Breusch-Pagan and White test that the data is heteroskedastic. Based on this I then weighted the residuals via FGLS outlined in Introductory Econometrics: A Modern Approach (Jeffrey Wooldridge)

The question I have is, once you have obtained a revised regression using FGLS, how do you establish that the errors are homoscedastic and that the FGLS is a better fit? (I believe \(R^{2}\) is not valid under GLS?)

Many thanks,

Hob

 

[Yike Lu]

There should be a test for heteroskedasticity? And the sum squared errors should still be valid.