多重共线性计量经济学.docx
《多重共线性计量经济学.docx》由会员分享,可在线阅读,更多相关《多重共线性计量经济学.docx(22页珍藏版)》请在冰点文库上搜索。
![多重共线性计量经济学.docx](https://file1.bingdoc.com/fileroot1/2023-6/19/9b260756-f6d5-44e7-8423-08e6394de59e/9b260756-f6d5-44e7-8423-08e6394de59e1.gif)
多重共线性计量经济学
唐山学院
实训报告
系别:
_________________________
实习名称:
_________________________
班级:
_________________________
姓名学号:
_________________________
指导教师:
_________________________
2017年10月
多重共线性实训报告
一、模型设定及其估计
经分析,影响中国粮食生产的主要因素,与农业化肥施用量X1、粮食播种面积X2、成灾面积X3、农业机械总动力X4以及农业劳动力X5的相关投入资料有关。
各影响变量中X1、X2、X4、X5与中国粮食产量之间呈现正相关,X3与之呈负相关。
为此设定了如下形式的计量经济模型:
Yt=β0+β1X1t+β2X2t+β3X3t+β4X4t+β5X5t+μt
式中,Yt为中国粮食产量(万吨);X1为农业化肥施用量(万公斤);X2为粮食播种面积(千公顷);X3为成灾面积(公顷);X4为农业机械总动力(万千瓦);X5为农业劳动力(万人)。
各解释变量前的回归系数预期都大于零。
为估计模型参数,收集1983-2007年阶段的中国粮食产量与各相关投入资料的统计数据。
(1)直接观测法
利用Eviews软件,生成Y、X1、X2、X3、X4、X5等数据,采用OLS方法估计模型参数,得到的回归结果如图1所示。
DependentVariable:
Y
Method:
LeastSquares
Date:
10/16/17Time:
08:
26
Sample:
19832007
Includedobservations:
25
Variable
Coefficient
Std.Error
t-Statistic
Prob.
C
-26689.44
7504.813
-3.556310
0.0021
X1
5.995098
0.609501
9.836080
0.0000
X2
0.536693
0.057836
9.279634
0.0000
X3
-0.135917
0.029708
-4.575032
0.0002
X4
-0.090845
0.042038
-2.161033
0.0437
X5
-0.007532
0.070478
-0.106866
0.9160
R-squared
0.980843
Meandependentvar
44945.64
AdjustedR-squared
0.975802
S.D.dependentvar
4150.729
S.E.ofregression
645.6787
Akaikeinfocriterion
15.98404
Sumsquaredresid
7921119.
Schwarzcriterion
16.27657
Loglikelihood
-193.8006
Hannan-Quinncriter.
16.06518
F-statistic
194.5613
Durbin-Watsonstat
1.715878
Prob(F-statistic)
0.000000
图1OLS回归结果
该模型R2=0.9808,
2=0.9758,可决系数很高,F检验值为194.56,明显显著。
但是当α=0.05时,tα/2(n-k)=t0.025(25-6)=2.09,不仅X5的系数不显著,而且X4、X5的符号与预期的相反,这表明可能存在严重的多重共线性。
(2)简单相关系数法
为了验证各解释变量之间的多重共线性,计算各解释变量的相关系数,如表1所示:
表1相关系数矩阵
变量
X1
X2
X3
X4
X5
X1
1.000000
-0.616566
0.400794
0.952746
0.314885
X2
-0.616566
1.000000
-0.238146
-0.741538
-0.060970
X3
0.400794
-0.238146
1.000000
0.310301
0.409300
X4
0.952746
-0.741538
0.310301
1.000000
0.128834
X5
0.314885
-0.060970
0.409300
0.128834
1.000000
由相关系数矩阵可以看出,各解释变量相互之间的相关系数较高,证实确实存在一定的多重共线性。
(3)方差扩大因子法
为了进一步了解多重共线性的性质,我们做辅助回归,即将每个X变量分别作为被解释变量都对其余的X变量进行回归。
回归结果如下图所示:
DependentVariable:
X1
Method:
LeastSquares
Date:
01/05/14Time:
08:
44
Sample:
19832007
Includedobservations:
25
Variable
Coefficient
Std.Error
t-Statistic
Prob.
C
-7769.179
2136.013
-3.637234
0.0016
X2
0.051610
0.017805
2.898564
0.0089
X3
0.010375
0.010649
0.974228
0.3416
X4
0.066416
0.004159
15.97035
0.0000
X5
0.072349
0.020170
3.587012
0.0018
R-squared
0.962459
Meandependentvar
3384.080
AdjustedR-squared
0.954951
S.D.dependentvar
1116.053
S.E.ofregression
236.8794
Akaikeinfocriterion
13.94984
Sumsquaredresid
1122237.
Schwarzcriterion
14.19361
Loglikelihood
-169.3729
Hannan-Quinncriter.
14.01745
F-statistic
128.1883
Durbin-Watsonstat
0.425652
Prob(F-statistic)
0.000000
图2X1为被解释变量的回归分析
DependentVariable:
X2
Method:
LeastSquares
Date:
01/05/14Time:
08:
48
Sample:
19832007
Includedobservations:
25
Variable
Coefficient
Std.Error
t-Statistic
Prob.
C
125103.6
7704.106
16.23856
0.0000
X1
5.731790
1.977458
2.898564
0.0089
X3
-0.074579
0.113643
-0.656255
0.5191
X4
-0.499729
0.118021
-4.234244
0.0004
X5
-0.365359
0.259950
-1.405495
0.1752
R-squared
0.684295
Meandependentvar
109339.6
AdjustedR-squared
0.621154
S.D.dependentvar
4055.785
S.E.ofregression
2496.356
Akaikeinfocriterion
18.65991
Sumsquaredresid
1.25E+08
Schwarzcriterion
18.90368
Loglikelihood
-228.2489
Hannan-Quinncriter.
18.72752
F-statistic
10.83755
Durbin-Watsonstat
0.708192
Prob(F-statistic)
0.000077
图3X2为被解释变量的回归分析
DependentVariable:
X3
Method:
LeastSquares
Date:
01/05/14Time:
08:
52
Sample:
19832007
Includedobservations:
25
Variable
Coefficient
Std.Error
t-Statistic
Prob.
C
36202.18
55903.41
0.647585
0.5246
X1
4.366882
4.482403
0.974228
0.3416
X2
-0.282648
0.430698
-0.656255
0.5191
X4
-0.226530
0.312324
-0.725306
0.4767
X5
0.394643
0.523076
0.754467
0.4594
R-squared
0.270588
Meandependentvar
24412.44
AdjustedR-squared
0.124706
S.D.dependentvar
5194.503
S.E.ofregression
4859.830
Akaikeinfocriterion
19.99225
Sumsquaredresid
4.72E+08
Schwarzcriterion
20.23603
Loglikelihood
-244.9031
Hannan-Quinncriter.
20.05986
F-statistic
1.854837
Durbin-Watsonstat
1.911428
Prob(F-statistic)
0.157987
图4X3为被解释变量的回归分析
DependentVariable:
X4
Method:
LeastSquares
Date:
01/05/14Time:
08:
52
Sample:
19832007
Includedobservations:
25
Variable
Coefficient
Std.Error
t-Statistic
Prob.
C
134775.9
26179.12
5.148223
0.0000
X1
13.96183
0.874235
15.97035
0.0000
X2
-0.945905
0.223394
-4.234244
0.0004
X3
-0.113138
0.155987
-0.725306
0.4767
X5
-0.993832
0.301919
-3.291717
0.0036
R-squared
0.968715
Meandependentvar
41334.80
AdjustedR-squared
0.962458
S.D.dependentvar
17725.77
S.E.ofregression
3434.494
Akaikeinfocriterion
19.29798
Sumsquaredresid
2.36E+08
Schwarzcriterion
19.54176
Loglikelihood
-236.2248
Hannan-Quinncriter.
19.36560
F-statistic
154.8218
Durbin-Watsonstat
0.485178
Prob(F-statistic)
0.000000
图5X4为被解释变量的回归分析
DependentVariable:
X5
Method:
LeastSquares
Date:
01/05/14Time:
08:
53
Sample:
19832007
Includedobservations:
25
Variable
Coefficient
Std.Error
t-Statistic
Prob.
C
56210.02
20222.95
2.779516
0.0116
X1
5.410972
1.508490
3.587012
0.0018
X2
-0.246038
0.175054
-1.405495
0.1752
X3
0.070123
0.092943
0.754467
0.4594
X4
-0.353576
0.107414
-3.291717
0.0036
R-squared
0.495363
Meandependentvar
34716.40
AdjustedR-squared
0.394436
S.D.dependentvar
2632.495
S.E.ofregression
2048.554
Akaikeinfocriterion
18.26451
Sumsquaredresid
83931507
Schwarzcriterion
18.50829
Loglikelihood
-223.3064
Hannan-Quinncriter.
18.33213
F-statistic
4.908121
Durbin-Watsonstat
0.624956
Prob(F-statistic)
0.006376
图6X5为被解释变量的回归分析
表2辅助回归的R2值
被解释变量
可决系数R2的值
方差扩大因子VIFj
X1
0.9625
26.67
X2
0.6843
3.1676
X3
0.2706
1.3710
X4
0.9687
31.9489
X5
0.4954
1.9818
经验表明,方差扩大因子VIFj≥10时,通常说明该解释变量与其余解释变量之间有严重的多重共线性,这里X1、X4的方差扩大因子远大于10,表明存在严重多重共线性问题。
二、对多重共线性的处理
将各变量进行对数变换,再对以下模型进行估计。
lnYt=β0+β1lnX1t+β2lnX2t+β3lnX3t+β4lnX4t+β5lnX5t+εt
利用Eviews软件,对Yt、X1、X2、X3、X4、X5分别取对数,分别生成lnY、lnX1、lnX2、ln3、lnX4、ln5的数据,采用OLS方法估计模型参数,得到回归结果如下图所示。
一元回归结果:
DependentVariable:
LNY
Method:
LeastSquares
Date:
11/25/15Time:
16:
50
Sample:
19832007
Includedobservations:
25
Variable
Coefficient
Std.Error
t-Statistic
Prob.
C
8.902008
0.206034
43.20657
0.0000
LNX1
0.224005
0.025515
8.779293
0.0000
R-squared
0.770175
Meandependentvar
10.70905
AdjustedR-squared
0.760182
S.D.dependentvar
0.093396
S.E.ofregression
0.045737
Akaikeinfocriterion
-3.255189
Sumsquaredresid
0.048114
Schwarzcriterion
-3.157679
Loglikelihood
42.68986
Hannan-Quinncriter.
-3.228144
F-statistic
77.07599
Durbin-Watsonstat
0.939435
Prob(F-statistic)
0.000000
图7
DependentVariable:
LNY
Method:
LeastSquares
Date:
11/25/15Time:
16:
50
Sample:
19832007
Includedobservations:
25
Variable
Coefficient
Std.Error
t-Statistic
Prob.
C
15.15748
5.912971
2.563429
0.0174
LNX2
-0.383434
0.509669
-0.752321
0.4595
R-squared
0.024017
Meandependentvar
10.70905
AdjustedR-squared
-0.018417
S.D.dependentvar
0.093396
S.E.ofregression
0.094252
Akaikeinfocriterion
-1.809063
Sumsquaredresid
0.204321
Schwarzcriterion
-1.711553
Loglikelihood
24.61329
Hannan-Quinncriter.
-1.782018
F-statistic
0.565986
Durbin-Watsonstat
0.335219
Prob(F-statistic)
0.459489
图8
DependentVariable:
LNY
Method:
LeastSquares
Date:
11/25/15Time:
16:
51
Sample:
19832007
Includedobservations:
25
Variable
Coefficient
Std.Error
t-Statistic
Prob.
C
9.619385
0.859809
11.18782
0.0000
LNX3
0.108101
0.085278
1.267632
0.2176
R-squared
0.065302
Meandependentvar
10.70905
AdjustedR-squared
0.024663
S.D.dependentvar
0.093396
S.E.ofregression
0.092237
Akaikeinfocriterion
-1.852285
Sumsquaredresid
0.195678
Schwarzcriterion
-1.754775
Loglikelihood
25.15357
Hannan-Quinncriter.
-1.825240
F-statistic
1.606891
Durbin-Watsonstat
0.597870
Prob(F-statistic)
0.217613
图9
DependentVariable:
LNY
Method:
LeastSquares
Date:
11/25/15Time:
16:
51
Sample:
19832007
Includedobservations:
25
Variable
Coefficient
Std.Error
t-Statistic
Prob.
C
8.949090
0.298255
30.00479
0.0000
LNX4
0.16