序列相关性的判定与补救

《序列相关性的判定与补救》由会员分享，可在线阅读，更多相关《序列相关性的判定与补救（6页珍藏版）》请在装配图网上搜索。

1、实验名称：序列相关性的判定与补救（实验序号：B142011O5）5.1 实验目的掌握自相关问题出现的来源、后果、检验及修正的原理，以及相关的EViews软件 操作方法。5.2 实验内容以实验五已克服异方差的中国的消费函数模型（见输出结果图 5.4）为例，练习检 查和克服模型的自相关的操作方法。由图 5.4 得到的回归式为：Lnyt=-0.0486+0.9561 Lnxt.（6.1）（-0.05）（68t.7）R2=0.997,DW=0.55 Ft =47215.3 实验步骤5.3.1 检验模型是否存在自相关1）观察残差图，如图 6.1，可初步判断残差项存在一定程度的正自相关。2）用 DW 检验

2、判断是否存在自相关由 EViews 输出结果（图 5.4）知 DW=0.55,若给定 a=0.05,查附表,d =1.08, d =1.36。LU因为DW=0.551.26,依据判别规则，认为误差项u存在严重的正自相关。（3）用LM检验判断是否存在自相关在估计窗 口选择 View/Residual Tests/Serial Correalation LM Test （见图 6.2）。图 6.2点击后会自动弹出一个设定滞后期(Lag Specification)对话框。输入1,点击0K 键，得到LM检验结果，见图6.3。Breusch-Godfrey Serial Correlation LM

3、Test:F-statistic12.34154Probability0.004278Obs*R-squared7.605234Probability0.005820Test Equation:Dependent Variable: RESIDMethod: Least SquaresDate: 10/02/07 Time: 23:37Presample missing value lagged residuals set to zero.VariableCoefficientStd. Errort-StatisticProb.C0.0817150.1145170.7135630.4891LN

4、X-0.0077750.010407-0.7470830.4694RESID(-1)0.8823450.2511623.5130530.0043R-squared0.507016Mean dependent var-1.78E-15Adjusted R-squared0.424852S.D. dependent var0.032289S.E. ofegression0.024487Akaike info critEfion-4.404468Sum squared resid0.007196Schwarz criterion-4.262858Log likElihood36.03351F-sta

5、tistic6.170771Durbin-Watson stat0.929987Prob(F-statistic)0.014355图 6.3根据p-值判断拒绝原假设，所以BG (LM)检验结果也说明(6.1)式存在自相关。( 4)用回归检验法判断自相关6.4。 将估计结果(6.1)式得到的残差定义为ut，首先做一阶自回归，得到估计结果见图 对该估计式采用LM检验法检验其自相关性，如图6.5。可以判断出仍然存在自 相关。 用残差的二阶自回归形式重新建立模型，见图6.6。 再次用LM检验法判断其自相关性，如图6.70从图6.7可以看出，此时p-值已经 达到0.3，落在接受域，即认为误差项不存在自相

6、关。对图 6.6的输出结果进行整理，可以得到残差的二阶回归式为u t =1.3436 u t _1 -0.8175 u 一 2 +v(6.2)(5.18)(-3.03)R2=0.71,s.e.=0.02,TR2=i.iDependent Variable: UMethod: Least SquaresDate: 10/02/07 Time: 23:45Sample (adjusted): 1991 2004Included observations: 14 after adjustmentsVariableCoefficient Std. Error t-Statistic Prob.U(-1

7、)0.8185960.2408513.3987600.0048R-squared0.470493Mean dependent var-0.000137Adjusted R-squared0.470493S.D. dependent var0.033503S.E. of regression0.024379Akaike info criterion-4.521418Sum squared resid0.007727Schwarz criterion-4.475771Log likelihood32.64992Durbin-Watson stat0.841759图6.4Breusch-Godfre

8、y Serial Correlation LM TestF-statistic10.01115Probability0.008159Obs*R-squared6.140424Probability0.013213Test Equation:Dependent Variable: RESIDMethod: Least SquaresDate: 10/02/07 Time: 23:45Presample missing value lagged residuals set to zeroVariableCoefficientStd. Error t-StatisticProb.u(-n-0.472

9、6110.237049-1.9870210.0702RESID(-1)0.9970060.3153583.1640400.0002R-squared0.430602Mean dependent var-0.003993Adjusted R-squared0.391018S.D. dependent var0.024025S.E. of regression0.010736Akaike info criterion-4.985203Sum squared resid0.004212Schwarz criterion-4.893909Log likelihood36.89642Durbin-Wat

10、son stat2.307851图6.5Dependent Variable: UMethod: Least SquaresDate: 10/02/07 Time: 23:46Sample (adjusted): 1992 2004Included obsen/ations: 13 after adjustmentsVariableCoefficientStd. Errort-StatisticProb.u(-n1.3436080.2592895.1018970.0003U(-2)-0.0175490.269255-3.0363340.0113R-squared0.711789Mean dep

11、endent var-0.000472Adjusted R-squared0.605588S.D. dependent var0.034047S.E. of regression0.019539Akaike info criterion-4.892131Sum squared resid0.004200Schwarz criterion-4.805216Log likelihood33.79885Durbin-Watson stat2.312116图 6.6Breusch-Godfrey Serial Correlation LM TestF-statistic0.935986Probabil

12、ity0.356128Obs*R-squared1.057317Probability0.303828Test Equation:Dependent Variable: RESIDMethod: Le a st SquaresDate: 10/02/07 Time: 23:46sample missing value lagged residuals SEt to zero.VariableCoeffici EntStd. Errort-StatisticProb.U(-1)0.2678450.37983107051690.4968U(-2)-0.2063300.344103-0.599619

13、0.5621RESID(-1)-0.4425370.457419-0.9674640.3561R-squarEd0.081332Mean dipEndent var-0.001223Adjusted R-squared-0.102401S.D. dependent var0.018664S.E. ofegression0.019596Akaike info crit门-4.827759Sum squared resid0.003840Schwarz criterion-4.697386Log likelihood34.38043Durbin-Watson stat2.087198图 6.75.

14、3.2 克服自相关Dependent Variable: GDLNYMethod: Least SquaresDate: 10/02/07 Time: 23:56Sample adjusted): 1992 2004Included observations: 13 after adjustmentsVariableUoe fficientStd. Errort-StatisticProb.C-0.0351410.121147-0.2900670.7772GDLNX0.9581890.02302241.620920.0000R-squared0.993690Mean dependent var

15、5.002073Adjusted R-squared0.993117S.D. dependent var0.234076S.E. of regression0.019487Akaike info criterion-4.897510Sum squared resid0.004177Schwarz criterion-4.810595Log likelihood33.83382F-statistic1732.301Durbin-Watson stat2.330588Prob(F-statistic)0.000000图6.8Breusch-Godfrey Serial Correlation LM

16、 Test:F-statistic0.479087Probability0.504590Obs*R-squared0.594339Probability0.440746Test Equation:Dependent Variable: RESIDMethod: Least SquaresDate: 10/03/07 Time: 00:02sample missing value 1日弱刨 residuals set to zero.VariableUoEffiuiEntStd. Errort-StatisticProb.C-0.0107190.125084-0.0856970.9334GDLN

17、X0.0020900.0237800.0879040.9317RESID(-1)-0.2227860.321870-0.6921610.5046R-squared0.045718Mean dependent var4.78E-16Adjusted R-squared-0.145138S.D. dependent var0.018657S.E. of regression0.019965Akaike info criterion-4790460Sum squared resid0.003986Schwarz criterion-4.660087Log likelihood34.13799F-st

18、atistic0.239543Durbin-Watson stat1.905244Prob(F-statistic)0791376图 6.9用广义最小二乘法估计回归参数。根据(6.2)式残差项的回归系数，对变量Lny和Lnx 作二阶广义差分GDLny=Lny-1.3436 Lnyt 1+0.8175 Lnyt 2GDLnx =Ln x - 1 . 343 6 Ln x +0.8175 Ln xttt-1t-2以GDLny ,GDLnx (t=2,3,15为样本再次回归，得EViews输出结果如图6.8。 此时 LM 检验结果见图 6.9。可以判断已经很好的克服了自相关。整理广义最小二 乘回归结果为GDLnyt=-0.035+0.9582 GDLnx(6.3)t(-0.29)(41.62)R2=0.99,s.e.=0.02,DW=2.33 TR2=0.59因为得，卩0(1-1.3436+0.8175)=-0.035卩=-0.0739所以0，原模型的广义最小二乘估计是Lny =-0.0739+0.9582Lnx (6.4)(6.3) 式残差图见图6.10。