Very high R - squared in Panel EGLS (Cross-section SUR)

[Jacek78]

Dear All,
I estimate one-way FE models with corrections by the non-spherical character of the error terms - using the EViews8 package, settings for GLS Weights were introduced by marking the cross-section SUR option. I estimated 5-6 models with the same dependent variable but with different sets of independent ones (3-5 right hand variables for a model) and every time I gain Adjusted R-squared higher than 0.99 (e.g. 0.994985). It is displayed in "Weighted Statistics" section of results.

Dependent Variable: log(LTX)
Method: Panel EGLS (Cross-section SUR)
Sample: 2004Q1 2013Q4
Periods included: 40
Cross-sections included: 8
Total panel (balanced) observations: 320
Linear estimation after one-step weighting matrix

Could somebody tell me what can be wrong with my estimations? Because it is rather strange to have different models with the same (very high) rsq....

[Jacek78]

Ok, maybe differently formulated question...

Can anybody tell me what kind of R-squared statistic Eviews provides in "panel EGLS (Cross-section SUR)" estimation?
- Is it standard "goodness-of-fit statistic" - less useful as a diagnostic tool for GLS regressions (because not bounded between zero and one)?
Or
- it provides different way of defining a goodness-of-fit statistic in the GLS case. Like "Pseudo R2" proposed for instance by Baxter and Cragg (1970)?

EViews 8 Users Guide I and II doesn't provide (any) information on it :-(

I would be very thankful for your kindly effort to reply...

Kind regards
J

[EViews Glenn]

Based on the comparison of the sums of squares for the estimated model against those from the weighted estimation of the dependent variable on the weighted constant alone, using the EGLS weights.