计量经济学试验-Eviews.doc

计量经济学异方差模型实验一、实验目的与要求掌握异方差模型的检验及处理方法，熟悉图形法检验和掌握G-Q检验、Park检验或Gleiser检验，理解同方差性变换和掌握加权最小二乘法（WLS）。能运用计量经济学软件包Eviews对异方差模型进行检验及处理。二、实验内容及步骤1.选择异方差模型实际经济问题从本实验指导书提供的参考选题中或从其它途径选择合适的实际经济问题。2.回归模型的理论形式设定针对所选的实际经济问题，依据有关的经济理论设定恰当的回归模型的理论形式。3.模型异方差检验应用计量经济学软件包Eviews对已设定的回归模型进行初步估计并进行异方差检验。4.模型异方差处理应用计量经济学软件包Eviews对异方差模型进行处理。三、实验题目、实验具体步骤及实验结论已知我国29个省、直辖市、自治区1994年城镇居民人均生活费支出Y和可支配收入X的数据资料如下：序号YX序号YX12940354716160919632232227691720482450318982334182087268841560195719377746325158518932023032895619772314212404307271596195322203424218166019602318762313935304297242186265310231127742517142102112856362626168020031218462248271870212713234128392818772171141577191929183524231519472515试根据以上数据建立并估计恰当的城镇居民人均生活费支出模型。实验步骤及内容如下：1.回归模型的理论形式设定 以城镇居民人均生活费支出(Y)作为被解释变量，可支配收入(X)作为解释变量。因此回归模型的理论形式设定为 Yi=a0+a1Xi+mi i=1,2,292.用OLS法初步估计参数并进行异方差检验(1)用OLS法初步估计参数，结果如下：=58.31791+0.795570XiDependent Variable: YMethod: Least SquaresDate: 05/31/12 Time: 21:10Sample: 1 29Included observations: 29VariableCoefficientStd. Errort-StatisticProb. X0.7955700.01837343.301930.0000C58.3179149.049351.1889640.2448R-squared0.985805 Mean dependent var2111.931Adjusted R-squared0.985279 S.D. dependent var555.5470S.E. of regression67.40436 Akaike info criterion11.32577Sum squared resid122670.4 Schwarz criterion11.42006Log likelihood-162.2236 F-statistic1875.057Durbin-Watson stat1.893970 Prob(F-statistic)0.000000生成残差绝对值序列E=ABS(resid)序号E序号E159.79522526231611.0218598171260.7487024141740.5355394466317.178337996118109.8101258455.24843968481933.6017513426520.66804172612058.4931203638677.73306244482198.3090142659716.06615959672249.6070700859842.3648502492322.4713675332953.11770872792417.03482497131045.77085230372516.60609288151187.05480647932628.1553393011120.75931610017627119.5046565671324.05880087072891.4995755974148.016778847129150.98406995815112.176511986生成残差绝对值平方E2=E2序号E2序号E213575.4689641716121.48139382823690.40484498171643.129958223295.0952963091812058.263728343052.39008761191129.077693295427.167948791203421.445129966042.42899705219664.662285947258.121484185222460.8614025181794.7805366223504.9623588192821.4909805124290.185261802102094.9709206125275.762320789117578.5393311526792.723131157120.5765609399862714281.362941313578.82589933828837226874308332922796.18938111512583.5698414(2)异方差检验1)G-Q检验对观测值按X值从小到大排序去掉中间7个约为(1/4)n个样本观测值，将剩下的22个样本观测值分成容量相等的两个子样本，一个子样本的X值较小，另一个子样本的X值较大，样本容量n1=n2=11。分别对两个子样求Y对X的回归方程，并计算残差平方和。定义样本区间1至11得到回归方程为=-287.1872+0.974751X 。残差平方和RSS1=21688.89Dependent Variable: YMethod: Least SquaresDate: 06/07/12 Time: 20:05Sample: 1 11Included observations: 11VariableCoefficientStd. Errort-StatisticProb. X0.9747510.1339267.2782960.0000C-287.1872271.8586-1.0563840.3183R-squared0.854777 Mean dependent var1688.545Adjusted R-squared0.838641 S.D. dependent var122.2083S.E. of regression49.09050 Akaike info criterion10.78817Sum squared resid21688.89 Schwarz criterion10.86052Log likelihood-57.33496 F-statistic52.97359Durbin-Watson stat2.306656 Prob(F-statistic)0.000047定义样本区间19至29得到回归方程为=-27.68345+0.820337X 。残差平方和RSS2=43702.65Dependent Variable: YMethod: Least SquaresDate: 06/07/12 Time: 20:19Sample: 19 29Included observations: 11VariableCoefficientStd. Errort-StatisticProb. X0.8203370.03216925.500950.0000C-27.68345106.7596-0.2593060.8012R-squared0.986349 Mean dependent var2641.545Adjusted R-squared0.984832 S.D. dependent var565.8140S.E. of regression69.68393 Akaike info criterion11.48878Sum squared resid43702.65 Schwarz criterion11.56113Log likelihood-61.18830 F-statistic650.2986Durbin-Watson stat2.610584 Prob(F-statistic)0.000000计算F检验值：F=RSS2/RSS1=2.015在5%的显著性水平下，第一和第二自由度为11-1-1的F0.05分布临界值为3.14,因为F检验值2.015小于F临界值3.14，接受原假设，即随机误差项不存在递增异方差。2)格里瑟检验残差绝对值对解释变量X平方进行回归，在方程设置窗口直接以X2为解释变量。模型估计为=45.19286+1.05E-06Xi2Dependent Variable: EMethod: Least SquaresDate: 06/05/12 Time: 21:35Sample: 1 29Included observations: 29VariableCoefficientStd. Errort-StatisticProb. X21.05E-061.71E-060.6150830.5436C45.1928614.217563.1786650.0037R-squared0.013818 Mean dependent var52.70317Adjusted R-squared-0.022707 S.D. dependent var38.78479S.E. of regression39.22265 Akaike info criterion10.24286Sum squared resid41537.25 Schwarz criterion10.33715Log likelihood-146.5214 F-statistic0.378327Durbin-Watson stat1.727432 Prob(F-statistic)0.543650残差绝对值对解释变量X进行回归。模型估计为=28.65130+0.009318XiDependent Variable: EMethod: Least SquaresDate: 06/05/12 Time: 21:31Sample: 1 29Included observations: 29VariableCoefficientStd. Errort-StatisticProb. X0.0093180.0106150.8777620.3878C28.6513028.339631.0109980.3210R-squared0.027744 Mean dependent var52.70317Adjusted R-squared-0.008265 S.D. dependent var38.78479S.E. of regression38.94474 Akaike info criterion10.22864Sum squared resid40950.71 Schwarz criterion10.32293Log likelihood-146.3152 F-statistic0.770466Durbin-Watson stat1.698005 Prob(F-statistic)0.387817残差绝对值对解释变量X平方根进行回归，在方程设置窗口直接以平方根SQR(X)为解释变量。模型估计为=-6.023802+1.164759Dependent Variable: EMethod: Least SquaresDate: 06/05/12 Time: 21:37Sample: 1 29Included observations: 29VariableCoefficientStd. Errort-StatisticProb. SQR(X)1.1647591.1505831.0123210.3204C-6.02380258.45719-0.1030460.9187R-squared0.036567 Mean dependent var52.70317Adjusted R-squared0.000885 S.D. dependent var38.78479S.E. of regression38.76763 Akaike info criterion10.21952Sum squared resid40579.08 Schwarz criterion10.31382Log likelihood-146.1830 F-statistic1.024793Durbin-Watson stat1.676836 Prob(F-statistic)0.320371残差绝对值对解释变量X倒数进行回归，在方程设置窗口直接以1/X为解释变量。模型估计为=99.71475-114882.9Dependent Variable: EMethod: Least SquaresDate: 06/05/12 Time: 21:38Sample: 1 29Included observations: 29VariableCoefficientStd. Errort-StatisticProb. 1/X-114882.982555.43-1.3915850.1754C99.7147534.517642.8888060.0075R-squared0.066923 Mean dependent var52.70317Adjusted R-squared0.032364 S.D. dependent var38.78479S.E. of regression38.15201 Akaike info criterion10.18751Sum squared resid39300.54 Schwarz criterion10.28180Log likelihood-145.7188 F-statistic1.936508Durbin-Watson stat1.597750 Prob(F-statistic)0.175408经过比较以上各形式回归的决定系数以及常数项是否显著可知，从图2中可确定异方差的形式为3.模型异方差处理以1/Xt为权数，采用加权最小二乘法估计模型如下：=4182.281+-51151894XiDependent Variable: YMethod: Least SquaresDate: 06/05/12 Time: 21:25Sam