Talk:Generalized least squares

I would like to see some links on the page to the most common litterature as well as any litterature of historical significance (no pun intended). —Preceding unsigned comment added by Thedreamshaper (talk • contribs) 07:13, 24 October 2010 (UTC)

New data, X0
This topic is well done, but should include using the gls slopes to estimate with new data (which is a function of the error covariance, not just X0 and b). —Preceding unsigned comment added by Davezes (talk • contribs) 02:33, 8 May 2011 (UTC)

Feasible generalized least squares
Is there a reference book or article or website for "Feasible generalized least squares" ?? Rjensen (talk) 02:03, 4 January 2012 (UTC)


 * A guess is that the non-citation to "White" is something to do with Heteroscedasticity-consistent standard errors. Otherwise, a web search does yield a range of possible sources, but they seem to yield something somewhat different to the present content of the section. Melcombe (talk) 15:01, 20 January 2012 (UTC)

Yes, now it has been corrected, also a link to Newey-West page is introduced — Preceding unsigned comment added by 163.117.79.154 (talk) 11:03, 4 June 2014 (UTC)

General remark
This article focusses too much of GLS estimation of the linear regression model. GLS however is a general method of estimation for a larger class of models and not limited to this case. I think the article should start with a general discussion and present the linear regression model estimation as a spcial (improtant) application. — Preceding unsigned comment added by 84.80.48.163 (talk) 13:48, 20 January 2012 (UTC)


 * If you have a citation, or preferably more than one, for people using the term "Generalized least squares" for something other than what is included here, then you should state it (them). Preferably it should be a widespread usage, not just someone/some-group misusing established terminology. Melcombe (talk) 14:49, 20 January 2012 (UTC)

The method can be applied in a nonlinear regression context, but the term GLS usually refers to the linear case, so that the current focus is probably the most appropriate. — Preceding unsigned comment added by 163.117.79.154 (talk) 11:04, 4 June 2014 (UTC)

T
in the FGLS section, last equation, what is T? 104.129.196.79 (talk) 22:47, 7 October 2015 (UTC)

I second that discussion. T is undefined. I suspect this equation was hijacked from an econometrics book that was using SUR or SUR with time series and T indicates time. However T is not appropriate in the context of this page, which only focuses on the general approach. Kgierach (talk) 21:38, 19 March 2016 (UTC)

Dr. Reed's comment on this article
Dr. Reed has reviewed this Wikipedia page, and provided us with the following comments to improve its quality:

"This article should mention that the estimator for the variance-covariance matrix associated with the FGLS estimates is obtained by replacing the population parameters of the error variance covariance matrix with their estimates. This makes use of the "analogy principle" (Manski, Charles F. Analog Estimation Methods in Econometrics. Chapman and Hall, 1988.).  However, the variability of the FGLS estimator is not adjusted to account for this fact, which can cause bias in the estimation of the standard errors.  For an example of where this bias can be severe, see  Beck, N. and Katz, J.N.  What do do (and not to do) with Time-Series Cross-Section Data, The American Politican Science Review, Vol. 89, No. 3 (1995): pp. 634-647."

We hope Wikipedians on this talk page can take advantage of these comments and improve the quality of the article accordingly.

We believe Dr. Reed has expertise on the topic of this article, since he has published relevant scholarly research:


 * Reference : W. Robert Reed & Haichun Ye, 2007. "A Monte Carlo Evaluation of Some Common Panel Data Estimators when Serial Correlation and Cross-sectional Dependence are Both Present," Working Papers in Economics 07/01, University of Canterbury, Department of Economics and Finance.

ExpertIdeasBot (talk) 16:07, 12 July 2016 (UTC)

How to minimize the quadratic form
To my understanding this makes use of https://math.stackexchange.com/questions/2606391/minimization-of-a-convex-quadratic-form this could be expaned here Biggerj1 (talk) 18:55, 18 August 2021 (UTC)

Needs software section
Needs software section. I am sure NLME implements. 68.134.243.51 (talk) 21:08, 24 August 2022 (UTC)

Whereas GLS is more efficient than OLS under heteroscedasticity (also spelled heteroskedasticity) or autocorrelation,
Which one is it?

More efficient under heteroskedasticity or autocorrelation? 68.134.243.51 (talk) 21:42, 24 August 2022 (UTC)

Sigma Omega inconsistent notation in HAC, Huber-White pages
abuse of notation! 68.134.243.51 (talk) 03:13, 15 September 2022 (UTC)

Lacks a autocorrelation and heteroscedasticity section in FGLS
Perhaps FGLS should be its separate page. 68.134.243.51 (talk) 13:39, 15 September 2022 (UTC)

Lacks section discussing Bayesian perspective
I believe that a section that discusses the derivation of GLS from Bayesian origins would add a nice perspective. I also believe that this could be a good place to include a brief discussion about what is done when correlations/covariances between errors are not explicitly known. I may come back and write this myself in a bit.

Best, Magenta.lily (talk) 03:38, 28 September 2023 (UTC)

I have added this section. Any feedback is appreciated!

Magenta.lily (talk) 03:29, 1 October 2023 (UTC)