这个是OLS回归的white检验结果:
Heteroskedasticity Test: White
F-statistic 3.781195 Prob. F(2,356) 0.0237
Obs*R-squared 7.467489 Prob. Chi-Square(2) 0.0239
Scaled explained SS 14.84751 Prob. Chi-Square(2) 0.0006
Test Equation:
Dependent Variable: RESID^2
Method: Least Squares
Date: 11/27/15 Time: 17:50
Sample: 1980M02 2009M12
Included observations: 359
Variable Coefficient Std. Error t-Statistic Prob.
C 45.35918 7.631269 5.943859 0.0000
FED^2 0.296081 0.122108 2.424749 0.0158
FED -5.694752 2.090076 -2.724662 0.0068
R-squared 0.020801 Mean dependent var 26.04900
Adjusted R-squared 0.015300 S.D. dependent var 52.30914
S.E. of regression 51.90744 Akaike info criterion 10.74512
Sum squared resid 959200.2 Schwarz criterion 10.77757
Log likelihood -1925.750 Hannan-Quinn criter. 10.75803
F-statistic 3.781195 Durbin-Watson stat 1.621757
Prob(F-statistic) 0.023716
这个是GLS回归的white检验结果:Heteroskedasticity Test: White
F-statistic 3.857664 Prob. F(2,356) 0.0220
Obs*R-squared 7.615303 Prob. Chi-Square(2) 0.0222
Scaled explained SS 15.22953 Prob. Chi-Square(2) 0.0005
Test Equation:
Dependent Variable: WGT_RESID^2
Method: Least Squares
Date: 11/27/15 Time: 17:51
Sample: 1980M02 2009M12
Included observations: 359
White heteroskedasticity-consistent standard errors & covariance
Collinear test regressors dropped from specification
Variable Coefficient Std. Error t-Statistic Prob.
C 45.80540 12.13417 3.774909 0.0002
FED^2*WGT^2 0.297242 0.142353 2.088067 0.0375
FED*WGT^2 -5.779052 2.812417 -2.054835 0.0406
R-squared 0.021213 Mean dependent var 26.06828
Adjusted R-squared 0.015714 S.D. dependent var 52.49996
S.E. of regression 52.08584 Akaike info criterion 10.75198
Sum squared resid 965804.9 Schwarz criterion 10.78444
Log likelihood -1926.981 Hannan-Quinn criter. 10.76489
F-statistic 3.857664 Durbin-Watson stat 1.622561
Prob(F-statistic) 0.022006
这个是按照网上以1/abs(residual)为权重得到的white检验结果:
Heteroskedasticity Test: White
F-statistic 678.9638 Prob. F(3,355) 0.0000
Obs*R-squared 305.7179 Prob. Chi-Square(3) 0.0000
Scaled explained SS 2678.685 Prob. Chi-Square(3) 0.0000
Test Equation:
Dependent Variable: WGT_RESID^2
Method: Least Squares
Date: 11/27/15 Time: 17:52
Sample: 1980M02 2009M12
Included observations: 359
White heteroskedasticity-consistent standard errors & covariance
Variable Coefficient Std. Error t-Statistic Prob.
C -72.54260 14.18626 -5.113581 0.0000
FED^2*WGT^2 0.655126 0.653456 1.002556 0.3168
FED*WGT^2 -21.38808 11.89167 -1.798576 0.0729
WGT^2 273.7781 30.98947 8.834552 0.0000
R-squared 0.851582 Mean dependent var 115.1192
Adjusted R-squared 0.850328 S.D. dependent var 485.2832
S.E. of regression 187.7440 Akaike info criterion 13.31912
Sum squared resid 12512970 Schwarz criterion 13.36238
Log likelihood -2386.781 Hannan-Quinn criter. 13.33632
F-statistic 678.9638 Durbin-Watson stat 2.078626
Prob(F-statistic) 0.000000
网上有人说比较n*R^2和卡方,有人说看p-value。现在头好疼,通宵搞这个,已经彻底晕了。GLS回归感觉异方差性基本没变是什么情况啊。wls又算出来个p=0,同时得到了好大的n*R^2。。。求各路大神拯救