I am interested in this as well. Besides getting more significant variables and higher Durbin-Watson value, my R^2 jumps also.
Here are the results when I do a normal Panel Least Squares:
Dependent Variable: MLEVW-MLEVW(-1)
Method: Panel Least Squares
Date: 11/15/12 Time: 13:13
Sample (adjusted): 1993 2011
Periods included: 19
Cross-sections included: 10348
Total panel (unbalanced) observations: 66241
White cross-section standard errors & covariance (d.f. corrected)
Variable Coefficient Std. Error t-Statistic Prob.
KZCONSTRAINED(-1)*MDEV2 0.009581 0.004592 2.086482 0.0369
MDEV2*MDEV2 0.016816 0.016334 1.029520 0.3032
MTOB(-1)*MDEV2 -5.24E-06 6.27E-06 -0.835260 0.4036
LIQUIDITY(-1)*MDEV2 1.02E-05 1.04E-05 0.982654 0.3258
TANGIBILITY(-1)*MDEV2 0.008468 0.015940 0.531217 0.5953
EBITTA(-1)*MDEV2 1.19E-05 1.24E-05 0.961401 0.3364
LNASSETSDEFLATED(-1)*MDEV2 -0.002317 0.001981 -1.169856 0.2421
R-squared 0.001391 Mean dependent var 0.008193
Adjusted R-squared 0.001300 S.D. dependent var 0.109958
S.E. of regression 0.109887 Akaike info criterion -1.578629
Sum squared resid 799.7806 Schwarz criterion -1.577667
Log likelihood 52291.98 Hannan-Quinn criter. -1.578332
Durbin-Watson stat 2.121004
As you see, all the independent variables are insignificant and R^2 is very low. When I regressed with only one independent variable, I got a negative R^2 (?). When I apply cross-section weights as GLS weights I get the following results:
Dependent Variable: MLEVW-MLEVW(-1)
Method: Panel EGLS (Cross-section weights)
Date: 11/15/12 Time: 13:16
Sample (adjusted): 1993 2011
Periods included: 19
Cross-sections included: 10348
Total panel (unbalanced) observations: 66241
Linear estimation after one-step weighting matrix
White cross-section standard errors & covariance (d.f. corrected)
Variable Coefficient Std. Error t-Statistic Prob.
KZCONSTRAINED(-1)*MDEV2 0.009662 9.49E-05 101.8232 0.0000
MDEV2*MDEV2 0.016657 0.001710 9.744021 0.0000
MTOB(-1)*MDEV2 -5.48E-06 2.08E-06 -2.638492 0.0083
LIQUIDITY(-1)*MDEV2 1.04E-05 1.11E-06 9.295758 0.0000
TANGIBILITY(-1)*MDEV2 0.008213 0.000978 8.397623 0.0000
EBITTA(-1)*MDEV2 1.18E-05 1.32E-06 8.963587 0.0000
LNASSETSDEFLATED(-1)*MDEV2 -0.002346 0.000191 -12.26996 0.0000
Weighted Statistics
R-squared 0.900697 Mean dependent var 0.008739
Adjusted R-squared 0.900688 S.D. dependent var 0.438289
S.E. of regression 0.109843 Sum squared resid 799.1513
Durbin-Watson stat 2.059037
Unweighted Statistics
R-squared 0.001381 Mean dependent var 0.008193
Sum squared resid 799.7880 Durbin-Watson stat 2.121280
Now R^2 is very high, independent variables are significant and Durbin-Watson is higher. What does this imply?