Inconsistency with Weighted Statistics for Panel model

[PROFI_USER]

Could you please help me.

What is the reason for Mean dependent variable to have different value for the same models.

We can estimate simple panel model (Cross-section = None) with GLS Weights = Cross-Section weights as ordinary linear regression (with weights) by rearranging factors and dependent variables.

It was verified that we can get the same values for Coefficient, and for some Weighted statistics.
Specifically, R2, adjR2, F-stat, DWstat are the same, as well as residuals series. Any way some characteristics like Mean dependent variable, SD dependent var, and SSR are different.
I was not able to find reason for it in help system.

thanks!

[PROFI_USER]

I've attached the picture with models estimates to explain my previous post

[EViews Glenn]

I'm guessing that the scaling you used for the variance weights in your "by-hand" calculation doesn't match the scaling used by the internal panel computation (which uses the "EViews" scaling -- scaling is discussed in the docs for weighted least squares). The following example should allow you to construct weighted estimates that match the built-in computation:

Code: Select all

wfcreate a 2001 2010 10 series y = nrnd series x = nrnd pagestruct crossid @date(dateid) ' least squares to get the residuals equation eq0.ls y c x eq0.makeresid resid01 ' compute by-cross section variances (no df correction) series var = @sumsby(resid01*resid01,crossid)/@obsby(resid01,crossid) ' compute variance scale scalar scale = @sumsq(resid01)/@obs(resid01) ' compute inverse scaled standard deviations series invstd = @sqrt(scale / var) ' built-in panel cross-section weighting equation eq1.ls(wgt=cxdiag) y c x ' restructure workfile as non-panel pagestruct(freq=u, start=1, end=100) ' estimate matching weighted with inverse scaled standard deviations equation eq2.ls(wgt=invstd, wscale=none) y c x

[PROFI_USER]

Thank you for your answer. I understand meaning.