⑵在方程窗口上点击View\\Residual Test\\White Heteroskedastcity,即可以得到检验结果。下图是怀特检验中no cross terms的结果。
White Heteroskedasticity Test: F-statistic
0.713421 Probability 4.917313 Probability
Std. Error
1.16E+08 58.13777 0.000177 242.9534 0.002606 2129086. 9756.465
t-Statistic
0.199172 -1.716907 1.130907 1.683536 -1.418741 -0.175701 0.175414
Coefficient
23149016 -99.81716 0.000201 409.0208 -0.003697 -374082.6 1711.416
0.645005 0.554461
Prob.
0.8450 0.1080 0.2771 0.1144 0.1779 0.8630 0.8633
Obs*R-squared
Test Equation:
Dependent Variable: RESID^2 Method: Least Squares Date: 12/15/12 Time: 05:19 Sample: 1991 2011 Included observations: 21
Variable C X1 X1^2 X2 X2^2 X3 X3^2
R-squared
0.234158 Mean dependent var -0.094060 S.D. dependent var 1651657. Akaike info criterion 3.82E+13 Schwarz criterion -326.2034 F-statistic 2.893273 Prob(F-statistic)
945767.9 1579063. 31.73366 32.08183 0.713421 0.645005
Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat
从上图中的怀特检验中我们可以发现P值很大，那么也就意味着该模型不存在异方差。
1.2 Goldfeld-Quant检验
A. 将样本按解释变量排序(SORT X1)并分成两部分(分别有1到10共10个样本合19到28共10个样本)
B. 利用样本1建立回归模型1，其残差平方和为RSS1=895534。 C. 利用样本2建立回归模型2，其残差平方和为RSS2=8271287。 D. 计算F统计量：F?RSS2/RSS1＝9.2362。
取??0.05时，查F分布表得F0.05(8?4,8?4)?6.39,F?9.24?F0.05?6.39，所以存在异方差性。 1.3 park检验
A. 建立回归模型。
B. 生成新变量序列：GENR LNE2=log(RESID^2)
GENR LNX1=log(X1) GENR LNX2=log(X2) GENR LNX3=log(X3)
C. 建立新残差序列对解释变量的回归模型：LS LNE2 C LNX1 LNX2 LNX3，回归结果如
????Dependent Variable: LNE2 Method: Least Squares Date: 12/21/12 Time: 19:18 Sample: 13 21
Included observations: 9
Variable C LNX1 LNX2 LNX3
R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood F-statistic Prob(F-statistic)
Coefficient -140.7097 -11.31236 9.469729 41.31118
Std. Error 143.2992 28.26318 22.47073 44.37096
t-Statistic -0.981929 -0.400251 0.421425 0.931041
Prob. 0.3712 0.7055 0.6909 0.3946 12.22358 1.689082 4.370270 4.457925 4.181110 1.959886
0.249551 Mean dependent var -0.200719 S.D. dependent var 1.850850 Akaike info criterion 17.12823 Schwarz criterion -15.66621 Hannan-Quinn criter. 0.554226 Durbin-Watson stat 0.667410
D. 从上图所示的回归结果中可以看出，LNX1 LNX2 LNX3的系数估计值不为0但不能通过显著性检验，即随机误差项的方差与解释变量不是存在较强的相关关系，即认为不存在异方差性。
1.4Gleiser检验(Gleiser检验与Park检验原理相同)
A. 建立回归模型。
B. 生成新变量序列：GENR E=ABS(RESID)
C. 分别建立新残差序列(E)对各解释变量的回归模型：LS E C X1; LS E C X2;LS E C X3回归结果如下所示：
????Dependent Variable: E Method: Least Squares Date: 12/21/12 Time: 19:23 Sample: 13 21
Included observations: 9
Variable C X1
R-squared Adjusted R-squared S.E. of regression Sum squared resid
Coefficient 1.450207 -9.60E-07
Std. Error 0.727079 2.45E-06
t-Statistic 1.994566 -0.392606
Prob. 0.0863 0.7063 1.184311 0.750408 2.568480 2.612308
0.021546 Mean dependent var -0.118234 S.D. dependent var 0.793530 Akaike info criterion 4.407833 Schwarz criterion
Log likelihood F-statistic Prob(F-statistic)
????Dependent Variable: E Method: Least Squares
-9.558160 Hannan-Quinn criter. 0.154140 Durbin-Watson stat 0.706291
2.473900 2.739094
Coefficient 1.409742 -3.94E-06
Std. Error 0.605681 9.52E-06
t-Statistic 2.327530 -0.413617
Prob. 0.0528 0.6915
1.184311 0.750408 2.566115 2.609943 2.471535 2.750817
Date: 12/21/12 Time: 19:25 Sample: 13 21
Included observations: 9
Variable C X2
R-squared
0.023857 Mean dependent var -0.115592 S.D. dependent var 0.792593 Akaike info criterion 4.397421 Schwarz criterion -9.547518 Hannan-Quinn criter. 0.171079 Durbin-Watson stat 0.691533
Adjusted R-squared S.E. of regression Sum squared resid Log likelihood F-statistic Prob(F-statistic)
????Dependent Variable: E Method: Least Squares
Coefficient 15.64669 -0.141322
Std. Error 10.99407 0.107405
t-Statistic 1.423194 -1.315783
Prob. 0.1977 0.2297
1.184311 0.750408 2.369259 2.413086 2.274679 2.690820
Date: 12/21/12 Time: 19:24 Sample: 13 21
Included observations: 9
Variable C X3
R-squared
0.198285 Mean dependent var 0.083755 S.D. dependent var 0.718296 Akaike info criterion 3.611639 Schwarz criterion -8.661664 Hannan-Quinn criter. 1.731285 Durbin-Watson stat 0.229699
Adjusted R-squared S.E. of regression Sum squared resid Log likelihood F-statistic Prob(F-statistic)
由上述各回归结果可知，各回归模型中解释变量的系数估计值显著不为0且均能通过显著性检验。所以认为存在异方差性。
D. 由F值或R2确定异方差类型
Gleiser检验中可以通过F值或R2值确定异方差的具体形式。 (3) 调整异方差性 ① 确定权数变量
根据Gleiser检验，可以取以下三种形式作为权数变量：
W2?1~W?1e~2 XiW3?1ei4i生成权数变量：GENR W1=1/X1^0.5 GENR W2=1/X2^0.5 GENR W3=1/X3^0.5
GENR W4=1/ABS(RESID） GENR W5=1/ RESID ^2
② 利用加权最小二乘法估计模型 在Eviews命令窗口中依次键入命令：
LS(W=Wi) Y C X1 X2 X3
????Dependent Variable: Y Method: Least Squares Date: 12/21/12 Time: 19:39 Sample: 13 21
Included observations: 9 Weighting series: W1
Variable C X1 X2 X3
R-squared
Coefficient -35745.78 0.127272 0.291195 308.9713
Std. Error 18509.48 0.035439 0.136844 195.5354
t-Statistic -1.931215 3.591297 2.127928 1.580130
Prob. 0.1113 0.0157 0.0866 0.1749
43490.05 12813.75
Weighted Statistics
0.999075 Mean dependent var 0.998520 S.D. dependent var
Adjusted R-squared