EViews计量经济学实验报告-异方差的诊断及修正.docx
JK（一）图形法1、在"Workfile”页面:选中x,y序列，点击鼠标右键,点击OpenasGroup-Yes2,在“Group”页面：点击VieWGraphScatterSimpleScatter,得到X,Y的散点图（图3所示）：3,在“Workfile”页面：点击Generate,输入"e2=residZ-OK4、选中X,e2序列，点击鼠标右键,Open-asGroupYes5,在“Group”页面：点击VieWGraphScatterSimpleScatter,得到到e2的散点图（图4所示）：6、推断由图3可以看出，被说明变量:Y随着说明变量X的增大而渐渐分散，离散程度越来越大；同样，由图4可以看出，残差平方，对说明变量X的散点图主要分布在图形中的下三角部分，大致看出残差平方e,W随X,的变动呈增大趋势。因此,模型很可能存在异方差。但是否的确存在异方差还应当通过更近一步的检验。JK（二）White检验1,在“Equation”页面：点击View-ResidualTestsWhite检验（nocross）,（本例为一元函数，没有交叉乘积项）得到检验结果，如图5：White检验结果WhiteHeteroskedasticityTest:R-SqUared6dependentvar0S.E.of4709.74Akaikeinfo19.8536regression4criterion15.55E+0Schwarz19,9963Sumsquaredresid8criterion5-274.953.607211.oglikelihood06F-statistic8Durbin-Watson1.47990Prob(F-stat0.04203stat8istic)6AdjustedO.16186S.D.5144.472、因为本例为一元函数，没有交叉乘积项，则协助函数为l2=a0+a,x,+a2+vl从上表可以看出，nK=6.270612，有While检验知，在=0,05下,查/分布表,得临界值力2qm(2)=5.99147。比较计算的炉统计量与临界值,因为nR?=6.270612>z2o,m(2)=5.99147,所以拒绝原假设，不拒绝备择假设，这表明模型存在异方差。(四)异方差的修正在运用加权最小二乘法估计过程中，分别选用了权数,为=1X,Mil=1/Xl»<yvl)XloR在“Workfile”页面:点击“Generate"，输入wl=lxw-OK；同样的输入''w2=lx"2''"w3=lsqr(x)”:R-squared67dependentvar9S.E.of32.0711Akaikeinfo9.84254regression7criterion126742.5Schwarz9.93769Sumsquaredresid6criterion9-135.79177.3511.oglikelihood56F-stalistic5Durbin-Watson1.46514Prob(F-stat0.00000slat8istic)0UnweightedStatistics0.85309Mean213.465R-SqUared5dependentvar0Adjusted0.84744S.D.146.489R-squared5dependentvar5S.E.of57.2163Sumsquared85116.4regression2resid0Durbin-Watson1.26146stat93、在“Equation”页面：点击“EstimateEquation",输入"ycx点击“weighted"，输入w2"，出现如图7：用权数%的结果DependentVariable:YMethod:1.eastSquaresDate:10/22/10Time:00:16Sample:128Includedobservations:28Weightingseries:H'2VariableCoefficientSld.Errort-StatistProb.icCX6.4967030.1068923.4865260.0109911.8633749.7252600.07370.0000WeightedStatisticsR-SqUarCdAdjustedR-SqUared0.92271Mean5dependentvar0.91974S.D.3dependentvar67.9212975.51929S.E.of21.3943kaikeinfo9.03288regression9criterion49.1280411900.7SchwarzSumsquaredresid2criterion1-124.4694.58061.ogIikelihood04F-Statistic8Durbin-Watson1.90567Prob(F-stat0.00000stat0istic)0UnweightedStatistics0.85418Mean213.465R-sqared2dependentvar0Adjusted0.84857S.D.146.489R-SqUared3dependentvar5S.E.of57.0013Sumsquared84486.8regression4resid8Durbin-Watson1.24221stat24、在"Equation”页面：点击*EstimateEqUaIion",输入"yc点击“weighted"，输入“*3”，出现如图8：DependentVariable:YMethod:1.eastSquaresDate:10/22/10Time:00:17Sample:128Includedobservations:28Weightingseries:W3CoefficStd.t-StatistProb.VariableientErroric8.64034C111.187330.7723330.44690.10615X30.00774613.704730.0000WeightedStatistics0.61155Mean165.842R-sqared2dependentvar0Adjusted0.59661S.D.67.1304R-sqared2dependentvar4S.E.of42.6364Akaikeinfo10.4120regression6criterion5Sumsquaredresid6criterion0-143.76187.8191.oglikelihood86F-statistic7Durbin-Walson1.27542Prob(l7-stal0.00000stat9istic)0UnweightedStatistics0.85445Mean213.465R-SqUared3dependentvar0Adjusted0.84885S.D.146.489R-squared5dependentvar5S.E.of56.9512Sumsquared84329.4regression1resid4Durbin-Watson1.23354stat547264.5Schwarz10.5072经估计检验，发觉用权数小,%的结果，其可决系数反而减小I只有用权数%的效果*好，可决系数增大.用权数的的结果DependentVariable:YMethod:1.eastSquaresDate:10/22/10Time:00:16Sample:128Includedobservations:28Weightingseries:W2VariableCoefficientStd.ErrortStatistProb.ic6.49670C33.4865261.8633740.07370.10689X20.0109919.7252600.0000WeightedStatistics0.92271Mean67.9212R-squarcd5dependentvar9Adjusted0.91974S.D.75.5192R-SqUarCd3dependentvar9S.E.of21.3943kaikeinfo9.03288regression9criterion4Sumsquaredresid11900.7Schwarz9.12804-124.4694.58061.oglikelihood04F-stalistic8Durbin-Watson1.90567Prob(F-stat0.00000slat0istic)0UnweightedStatistics0.85418Mean213.465R-SqUared2dependentvar0Adjusted0.84857S.D.146.489R-sqared3dependentvar5S.E.of57.0013Sumsquared84486.8regression4resid8Durbin-Watson1.24221stat22criterion1用权数内的估计结果为：Y-6.496703+0.106892X,(1.863374)<9.725260)R2=O.922715DW=I.905670F=94.58068括号中的数据为t统计量值。由上可以看出，运用加权最小二乘法消退了异方差后，参数用的t检验显著，可决系数提高了不少，F检验也显著，并说明销售收入每熠长1元，销售利洞平均增长0.106892元。四、实践结果报告：1、用图示法初步推断是否存在异方差：被说明变量Y随着说明变量X的增大而渐渐分散，离散程度越来越大；同样的，残差平方/对说明变量X的散点图主要分布在图形中的下三角部分，大致看出残差平方/随X,的变动呈增大趋势。因此，模型很可能存在异方差。但是否的确存在异方差还应当通过更近步的检验。再用White检验异方差：因为nK=6.270612>z2o,w(2)-5.99147,所以拒绝原假设，不拒绝备择假设，这表明模型存在异方差。2、用加权最小二乘法修正异方差：发觉用权数外的效果最好，则估计结果为：Y1=6.496703+0.106892X,(1.863374)(9.725260)R:=0.922715DW=1.905670F=94.58068括号中的数据为t统计量值。由上可以看出，R'=0922715,拟合程度较好。在给定=0.0时，t=9.725260>ftwj(26)=2.056,拒绝原假设，说明销售收入对销售利润有显著性影响。F=94.58068>105(l,26)=4.23,表明方程整体显著。运用加权最小二乘法后，参数区的t检验显著，可决系数提高了不少，F检验也显著，并说明销售收入每增长1元，销售利涧平均增长0.106892兀。3、再用Ihite检验修正后的模型是否还存在异方差:While检验结果WhiteHeteroskedasticityTest:3.144590.06050F-statistic7Probability95.628050.059960bs*R-sqared8Probability3TestEquation:DependentVariable:STDRESID2Method:1.easlSquaresDate:10/22/10Time:00:17Sample:128Includedobservations:28CoefficStd.Errort-StatistProb.icVariableientC1927.346675.22462.8543780.0085-1.4566130.734838-1.9822230.0585O.00024父250.0001541.5863420.12520.20100Mean425.025R-SqUared2dependentvar8Adjusted0.13708S.D.1198.21R-sqared2dependentvar0S.E.of1113.05Akaikeinfo16.9685regression7criterion73097241Schwarz17.1113Sumsquaredresid4criterion0-234.553.144591.oglikelihood99F-statistic7Durbin-Watson2.55950Prob(F-stat0.06050stat6istic)9由上看出，nR2=5.628O58,由White检验知,在a=0,05下，查/分布表,得临界值:Z1OW(2)=5.99147»比较计算的/统计量与临界值，因为nR2=5.628058</。(2)=5.99147.所以接受原假设，这说明修正后的模型不存在异方差。老师评阅看法；
