计量经济学消除异方差实验报告.doc

《计量经济学消除异方差实验报告.doc》由会员分享，可在线阅读，更多相关《计量经济学消除异方差实验报告.doc（5页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、 计量经济学消除异方差实验报告姓名：赵国伟 班级：市场营销三班 学号：1.建立模型：Y=kX+cDependent Variable: YMethod: Least SquaresDate: 11/13/12 Time: 19:26Sample: 1 20Included observations: 20VariableCoefficientStd. Errort-StatisticProb.C272.3635159.67731.0.1053X0.0.32.386900.0000R-squared0.Mean dependent var5199.515Adjusted R-squared0.S

2、.D. dependent var1625.275S.E. of regression216.8900Akaike info criterion13.69130Sum squared resid.0Schwarz criterion13.79087Log likelihood-134.9130Hannan-Quinn criter.13.71073F-statistic1048.912Durbin-Watson stat1.Prob(F-statistic)0.由软件普通最小二乘法统计结果如下：Y=0.X+272.3635R2=0. F=1048.912 RSS=.0 2.下面对该模型进行异方

3、差性检验 已知如果存在异方差性时由X引起的。 首先采用怀特检验结果如下：Heteroskedasticity Test: WhiteF-statistic23.81765Prob. F(1,18)0.0001Obs*R-squared11.39120Prob. Chi-Square(1)0.0007Scaled explained SS5.Prob. Chi-Square(1)0.0252Test Equation:Dependent Variable: RESID2Method: Least SquaresDate: 11/13/12 Time: 19:39Sample: 1 20Inclu

4、ded observations: 20VariableCoefficientStd. Errort-StatisticProb.C-5602.79711961.18-0.0.6451X20.0.4.0.0001R-squared0.Mean dependent var42337.15Adjusted R-squared0.S.D. dependent var45279.67S.E. of regression30521.10Akaike info criterion23.58486Sum squared resid1.68E+10Schwarz criterion23.68444Log li

5、kelihood-233.8486Hannan-Quinn criter.23.60430F-statistic23.81765Durbin-Watson stat1.Prob(F-statistic)0.由表知怀特统计量nR2=11.3912，该值大于5%显著性水平下，自由度为1的2分布的相应临界值3.84，因此拒接同方差的原假设。3.再采用G-Q检验，将原始数据按X排成升序，去掉中间6个数据，得到两个容量为7的子样本。对1-7回归得到结果如下Dependent Variable: YMethod: Least SquaresDate: 11/13/12 Time: 20:17Sample:

6、 1 7Included observations: 7VariableCoefficientStd. Errort-StatisticProb.C1619.5771936.4140.0.4411X0.0.1.0.2748R-squared0.Mean dependent var3992.683Adjusted R-squared0.S.D. dependent var163.6466S.E. of regression157.1866Akaike info criterion13.18770Sum squared resid.2Schwarz criterion13.17225Log lik

7、elihood-44.15695Hannan-Quinn criter.12.99669F-statistic1.Durbin-Watson stat2.Prob(F-statistic)0.RSS1=.2对14-20回归得到结果如下Dependent Variable: YMethod: Least SquaresDate: 11/13/12 Time: 20:28Sample: 14 20Included observations: 7VariableCoefficientStd. Errort-StatisticProb.C-126.2083623.2100-0.0.8475X0.0.1

8、1.623230.0001R-squared0.Mean dependent var6980.719Adjusted R-squared0.S.D. dependent var1541.068S.E. of regression318.9180Akaike info criterion14.60270Sum squared resid.6Schwarz criterion14.58725Log likelihood-49.10946Hannan-Quinn criter.14.41169F-statistic135.0995Durbin-Watson stat1.Prob(F-statisti

9、c)0.RSS2=.6 计算F统计量：F=RSS2/RSS1=4.1165在5%和10%的显著水平下，自由度为（5,5）的F分布的临界值分别为10.97 与5.05，因此都拒接同方差。4.现在我们已经确认原模型存在异方差，而要修正异方差必须进行帕克检验得出f(x),下面进行帕克检验：Dependent Variable: LNE2Method: Least SquaresDate: 11/13/12 Time: 21:14Sample: 1 20Included observations: 20VariableCoefficientStd. Errort-StatisticProb.C-20.

10、4965610.51447-1.0.0670LNX3.1.2.0.0098R-squared0.Mean dependent var9.Adjusted R-squared0.S.D. dependent var1.S.E. of regression1.Akaike info criterion3.Sum squared resid41.43840Schwarz criterion3.Log likelihood-35.66353Hannan-Quinn criter.3.F-statistic8.Durbin-Watson stat2.Prob(F-statistic)0.3.显著不为0

11、得:lnE2=20+3.lnx 所以E2=x3.e-20f(x)=x3.5.得到了f(x)我们就可以用以w=1/x3.作为适当的权对原模型进行加权最小二乘法来消除异方差，由软件得 Dependent Variable: YMethod: Least SquaresDate: 11/13/12 Time: 21:38Sample: 1 20Included observations: 20Weighting series: W1VariableCoefficientStd. Errort-StatisticProb.C487.0051317.07351.0.1419X0.0.11.584300.

12、0000Weighted StatisticsR-squared0.Mean dependent var4346.773Adjusted R-squared0.S.D. dependent var2244.303S.E. of regression163.8523Akaike info criterion13.13045Sum squared resid.3Schwarz criterion13.23002Log likelihood-129.3045Hannan-Quinn criter.13.14988F-statistic134.1960Durbin-Watson stat1.Prob(

13、F-statistic)0.Unweighted StatisticsR-squared0.Mean dependent var5199.515Adjusted R-squared0.S.D. dependent var1625.275S.E. of regression237.1468Sum squared resid.Durbin-Watson stat1.所以新模型为:Y=0.X+487.0051R2=0. F=134.1960 可以看出X参数的t统计量值有了显著地改进。6.最后在检验新模型是否还存在异方差性 直接用怀特检验有Heteroskedasticity Test: WhiteF

14、-statistic3.Prob. F(2,17)0.0708Obs*R-squared5.Prob. Chi-Square(2)0.0688Scaled explained SS3.Prob. Chi-Square(2)0.2105Test Equation:Dependent Variable: WGT_RESID2Method: Least SquaresDate: 11/13/12 Time: 22:14Sample: 1 20Included observations: 20VariableCoefficientStd. Errort-StatisticProb.C-5026.707

15、13748.96-0.0.7192WGT2-48602.3239158.15-1.0.2314X2*WGT20.0.1.0.1495R-squared0.Mean dependent var24162.81Adjusted R-squared0.S.D. dependent var29719.53S.E. of regression26886.86Akaike info criterion23.37414Sum squared resid1.23E+10Schwarz criterion23.52350Log likelihood-230.7414Hannan-Quinn criter.23.40330F-statistic3.Durbin-Watson stat2.Prob(F-statistic)0.由表知怀特统计量nR2=5. ，该值小于5%显著性水平下，自由度为2的2分布的相应临界值5.99，因此接受同方差的原假设。所以最终的修正结果为：:Y=0.X+487.0051R2=0. F=134.1960