实验五异方差模型地检验和处理学生实验报告材料.docx

1、实验五异方差模型地检验和处理学生实验报告材料 实 验 报 告课程名称： 计量经济学 实验项目： 实验五 异方差模型的 检验和处理 实验类型：综合性 设计性 验证性 专业班别： 姓 名： 学 号： 实验课室： 指导教师： 石立 实验日期： 2014.5.30 广东商学院华商学院教务处 制 一、实验项目训练方案小组合作：是 否 小组成员：无实验目的：掌握异方差模型的检验和处理方法实验场地及仪器、设备和材料实验室：普通配置的计算机，Eviews软件及常用办公软件。实验训练内容（包括实验原理和操作步骤）：【实验原理】异方差的检验：图形检验法、Goldfeld-Quanadt检验法、White检验法、G

2、lejser检验法；异方差的处理：模型变换法、加权最小二乘法(WLS)。【实验步骤】本实验考虑三个模型：【1】广东省财政支出CZ对财政收入CS的回归模型；（数据见附表1：附表1-广东省数据）【2】广东省固定资产折旧ZJ对国内生产总值GDPS和时间T的二元回归模型；（数据见附表1：附表1-广东省数据）【3】广东省各市城镇居民消费支出Y对人均收入X的回归模型。（数据见附表2：附表2-广东省2005年数据）（一）异方差的检验1.图形检验法分别用相关分析图和残差散点图检验模型【1】、模型【2】和模型【3】是否存在异方差。注：相关分析图是作因变量对自变量的散点图（亦可作模型残差对自变量的散点图）；残差散

3、点图是作残差的平方对自变量的散点图。模型【2】中作图取自变量为GDPS来作图。模型【1】相关分析图 残差散点图模型【2】相关分析图 残差散点图模型【3】相关分析图 残差散点图【思考】相关分析图和残差散点图的不同点是什么？*在模型【2】中，自变量有两个，有无其他处理方法？尝试做出来。（请对得到的图表进行处理，以上在一页内）2.Goldfeld-Quanadt检验法用Goldfeld-Quanadt检验法检验模型【3】是否存在异方差。注：Goldfeld-Quanadt检验法的步骤为：排序：删除观察值中间的约1/4的，并将剩下的数据分为两个部分。构造F统计量：分别对上述两个部分的观察值求回归模型，

4、由此得到的两个部分的残差平方为和。为较大的残差平方和，为较小的残差平方和。算统计量。判断：给定显著性水平，查F分布表得临界值。如果，则认为模型中的随机误差存在异方差。（详见课本135页）将实验中重要的结果摘录下来，附在本页。将样本进行排序后，区间定义为17，然后用OLS方法求得如下结果：将样本进行排序后，区间定义为1218，然后用OLS方法求得如下结果：有上图可知，=17472943，=1757380F=/=17472943/1757380=2.90586在 =0.05下，上式中分子、分母的自由度均为5，查F分布表得临界值F0.05（5,5）=5.05，因为F=2.90586 F0.05（5,

5、5）=5.05，所以接受原假设，说明模型不存在异方差。（请对得到的图表进行处理，以上在一页内）3.White检验法分别用White检验法检验模型【1】、模型【2】和模型【3】是否存在异方差。Eviews操作：先做模型，选view/Residual Tests/White Heteroskedasticity (no cross terms/cross terms)。摘录主要结果附在本页内。模型【1】Heteroskedasticity Test: WhiteF-statistic4.940866Prob. F(2,25)0.0156Obs*R-squared7.932189Prob. Chi-

6、Square(2)0.0189Scaled explained SS14.57723Prob. Chi-Square(2)0.0007Test Equation:Dependent Variable: RESID2Method: Least SquaresDate: 05/30/14 Time: 11:48Sample: 1978 2005Included observations: 28CoefficientStd. Errort-StatisticProb.C-879.85131125.376-0.7818290.4417CS12.937204.6513282.7813980.0101CS

7、2-0.0066200.002964-2.2335610.0347R-squared0.283292Mean dependent var1940.891Adjusted R-squared0.225956S.D. dependent var4080.739S.E. of regression3590.225Akaike info criterion19.31077Sum squared resid3.22E+08Schwarz criterion19.45351Log likelihood-267.3508Hannan-Quinn criter.19.35441F-statistic4.940

8、866Durbin-Watson stat2.144291Prob(F-statistic)0.015552模型【2】Heteroskedasticity Test: WhiteF-statistic1.993171Prob. F(5,22)0.1195Obs*R-squared8.729438Prob. Chi-Square(5)0.1204Scaled explained SS14.67857Prob. Chi-Square(5)0.0118Test Equation:Dependent Variable: RESID2Method: Least SquaresDate: 05/30/14

9、 Time: 11:50Sample: 1978 2005Included observations: 28CoefficientStd. Errort-StatisticProb.C1837.8986243.7010.2943600.7712GDPS-3.3950935.407361-0.6278650.5366GDPS2-9.08E-050.000185-0.4895370.6293GDPS*T0.1603000.3151760.5086040.6161T-491.56141982.891-0.2479010.8065T249.08543152.98750.3208460.7514R-sq

10、uared0.311766Mean dependent var3461.910Adjusted R-squared0.155349S.D. dependent var7240.935S.E. of regression6654.775Akaike info criterion20.63147Sum squared resid9.74E+08Schwarz criterion20.91694Log likelihood-282.8405Hannan-Quinn criter.20.71874F-statistic1.993171Durbin-Watson stat1.971537Prob(F-s

11、tatistic)0.119510模型【3】Heteroskedasticity Test: WhiteF-statistic7.670826Prob. F(2,15)0.0051Obs*R-squared9.101341Prob. Chi-Square(2)0.0106Scaled explained SS14.09286Prob. Chi-Square(2)0.0009Test Equation:Dependent Variable: RESID2Method: Least SquaresDate: 06/22/14 Time: 10:35Sample: 1 18Included obse

12、rvations: 18CoefficientStd. Errort-StatisticProb.C1865425.2810916.0.6636360.5170X-354.7917388.1454-0.9140690.3751X20.0188100.0116861.6095970.1283R-squared0.505630Mean dependent var1232693.Adjusted R-squared0.439714S.D. dependent var2511199.S.E. of regression1879689.Akaike info criterion31.88212Sum s

13、quared resid5.30E+13Schwarz criterion32.03052Log likelihood-283.9391Hannan-Quinn criter.31.90258F-statistic7.670826Durbin-Watson stat2.010913Prob(F-statistic)0.005074用Glejser检验法检验模型【1】是否存在异方差。分别用残差的绝对值对自变量的一次项、二次项，开根号项和倒数项作回归。检验异方差是否存在，并选定异方差的最优形式。对CS回归，结果为Dependent Variable: ABS(RESID)Method: Least

14、 SquaresDate: 06/22/14 Time: 11:01Sample: 1978 2005Included observations: 28CoefficientStd. Errort-StatisticProb.CS0.0138490.0054882.5236820.0181C6.9227313.6912251.8754560.0720R-squared0.196762Mean dependent var13.14854Adjusted R-squared0.165868S.D. dependent var15.90841S.E. of regression14.52928Aka

15、ike info criterion8.258958Sum squared resid5488.600Schwarz criterion8.354116Log likelihood-113.6254Hannan-Quinn criter.8.288049F-statistic6.368969Durbin-Watson stat1.172635Prob(F-statistic)0.018061常数项不显著，去掉常数项再进行回归，得结果为：Dependent Variable: ABS(RESID)Method: Least SquaresDate: 06/22/14 Time: 11:03Sam

16、ple: 1978 2005Included observations: 28CoefficientStd. Errort-StatisticProb.CS0.0171380.0022737.5386050.0000R-squared0.350502Mean dependent var9.940665Adjusted R-squared0.350502S.D. dependent var10.04016S.E. of regression8.091511Akaike info criterion7.054569Sum squared resid1767.759Schwarz criterion

17、7.102148Log likelihood-97.76396Hannan-Quinn criter.7.069114Durbin-Watson stat1.874280对CS2回归，得结果为：Dependent Variable: ABS(RESID)Method: Least SquaresDate: 06/22/14 Time: 11:05Sample: 1978 2005Included observations: 28CoefficientStd. Errort-StatisticProb.CS21.94E-061.10E-061.7560300.0909C4.2562630.998

18、9184.2608710.0002R-squared0.106027Mean dependent var5.132011Adjusted R-squared0.071643S.D. dependent var4.753345S.E. of regression4.579909Akaike info criterion5.949984Sum squared resid545.3647Schwarz criterion6.045142Log likelihood-81.29978Hannan-Quinn criter.5.979075F-statistic3.083641Durbin-Watson

19、 stat2.365101Prob(F-statistic)0.090859对回归，得结果为：Dependent Variable: ABS(RESID)Method: Least SquaresDate: 06/22/14 Time: 11:08Sample: 1978 2005Included observations: 28CoefficientStd. Errort-StatisticProb.CS(1/2)0.0627530.0161993.8737860.0006R-squared0.022805Mean dependent var1.302296Adjusted R-square

20、d0.022805S.D. dependent var1.838557S.E. of regression1.817472Akaike info criterion4.067832Sum squared resid89.18656Schwarz criterion4.115410Log likelihood-55.94964Hannan-Quinn criter.4.082377Durbin-Watson stat2.323523对1/cs回归，得结果为：Dependent Variable: ABS(RESID)Method: Least SquaresDate: 06/22/14 Time

21、: 11:09Sample: 1978 2005Included observations: 28CoefficientStd. Errort-StatisticProb.1/CS-46.3186129.48793-1.5707650.1283C1.3420280.3887583.4520900.0019R-squared0.086671Mean dependent var0.910752Adjusted R-squared0.051543S.D. dependent var1.495389S.E. of regression1.456341Akaike info criterion3.658

22、480Sum squared resid55.14412Schwarz criterion3.753637Log likelihood-49.21872Hannan-Quinn criter.3.687570F-statistic2.467302Durbin-Watson stat2.137709Prob(F-statistic)0.128329从四个回归的结果看，第二个不显著，其他三个显著，比较这三个回归，还是选择第三个，方程为ABS(RESID)= 0.062753*CS(1/2)即异方差的形式为（0.062753*CS(1/2)2（二）异方差的处理1.模型【1】中CZ对CS回归异方差的处

23、理已知CZ对CS回归异方差的形式为：，选取权数，使用加权最小二乘法处理异方差。并检验处理异方差之后模型是否仍存在异方差，若仍然存在异方差，请继续处理异方差。把W=1/作为权数进行最小二乘法Dependent Variable: CZMethod: Least SquaresDate: 06/22/14 Time: 11:18Sample: 1978 2005Included observations: 28Weighting series: 1/(CS(1/2)CoefficientStd. Errort-StatisticProb.CS1.2756770.01940665.736280.00

24、00C-21.243654.264097-4.9819800.0000Weighted StatisticsR-squared0.994019Mean dependent var254.4606Adjusted R-squared0.993789S.D. dependent var189.1988S.E. of regression22.86683Akaike info criterion9.166001Sum squared resid13595.19Schwarz criterion9.261159Log likelihood-126.3240Hannan-Quinn criter.9.195092F-statistic4321.259Durbin-Watson stat1.550317Prob(F-statistic)0.000000Unweighted