admin 管理员组

文章数量: 887018


2023年12月23日发(作者:制作网页时不能使用图案作为网页背景)

指导老师 xxx

1

回归模型分析

工商07-4班 xxx

070614409

我国改革开放以来固定资产投资与GDP关系分析

【摘要】本文旨在对我国改革开放以来固定资产投资与GDP关系进行计量分析。首先我们对已有的部分关于固定资产投资的观点和评论进行了评述;然后再收集的数据的基础上利用EViews软件进行了计量分析,从数据本身出发验证了两者的因果关系,并寻求设定合理的经济关系模型;接着运用软件对设定的模型进行了参数估计,检验及修正;最后我们利用所

2

得的结果进行了经济预测以评估所得结果的价值并对结果本身提出了政策意见。

一 问题的提出

我国自改革开放以来已保持了国民经济20多年的快速增长,GDP年均增长率在10%以上,如此高的增长速度不经要引起人们对其增长动力或原因的兴趣。今年来关于投资,消费和出口“三驾马车”拉动经济增长的理论较为突出。尤其是进入90年代后直到90年代末到新世纪最近几年,不论是学术界还是公众媒体都对固定资产投资的高增长表现出不同程度的担忧,因而才引出关于经济软着陆和怎样减少固定资产投资的讨论。那么,究竟固定资产投资同GDP之间的关系如何?新世纪的前后几年是不是存在固定资产投资过热拉动经济过热的情况?本文通过回归分析给出答案。

二 数据收集

为进行计量分析,我们寻求改革开放至今的GDP和固定资产的可比数据,数据来源为《中国统计年鉴》及中国国家统计局网站()的数据资料,两项数据样本数都为27,满足一元回归的要求。

1978-2004年GDP及固定资产投资年度数据

obs

1978

1979

1980

1981

1982

1983

1984

GDP

3624.100

4038.200

4517.800

4862.400

5294.700

7171.000

7171.000

FAI

780.2000

846.2000

910.9000

961.0000

1230.400

1430.100

1832.900

3

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

8964.400

10202.20

11962.50

14928.30

16909.20

18547.90

21617.80

26638.10

34634.40

46759.40

58478.10

67884.60

74462.60

78345.20

82067.50

89468.10

97314.80

104790.6

117251.9

136515.0

2543.200

3120.600

3791.700

4753.800

4410.400

4517.000

5594.500

8080.100

13072.30

17042.94

20019.26

22974.03

24941.10

28406.17

29854.71

32917.73

37213.49

43499.91

55566.61

70072.71

三 数据分析

由于相关数据为时间序列,很可能为非平稳序列,直接回归可能造成伪回归。因此对

4

两时间序列进行平稳性检验,方法为ADF检验。EViews5默认情况下检验结果如下:

GDP的ADF检验

Null Hypothesis: GDP has a unit root

Exogenous: Constant

Lag Length: 2 (Automatic based on SIC, MAXLAG=6)

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic 2.588925 1.0000

Test critical values: 1% level -3.737853

5% level -2.991878

10% level -2.635542

*MacKinnon (1996) one-sided p-values.

Augmented Dickey-Fuller Test Equation

Dependent Variable: D(GDP)

Method: Least Squares

Date: 05/28/05 Time: 16:45

Sample (adjusted): 1981 2004

Included observations: 24 after adjustments

Variable CoefficieStd. Error t-Statistic Prob.

5

nt

GDP(-1) 0.045772 0.017680 2.588925 0.0175

D(GDP(-1)) 1.327562 0.217150 6.113574 0.0000

-0.73283D(GDP(-2)) 1 0.231507 -3.165485 0.0049

C 399.8333 664.7932 0.601440 0.5543

R-squared 0.851066 Mean dependent var 5499.883

Adjusted R-squared 0.828726 S.D. dependent var 4860.139

S.E. of regression 2011.383 Akaike info criterion 18.20204

8091321Sum squared resid 9 Schwarz criterion 18.39839

-214.424Log likelihood 5 F-statistic 38.09586

Durbin-Watson stat 2.019994 Prob(F-statistic) 0.000000

FAI的ADF检验

Null Hypothesis: FAI has a unit root

Exogenous: Constant

Lag Length: 6 (Automatic based on SIC, MAXLAG=6)

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic 4.261202 1.0000

6

Test critical values: 1% level -3.808546

5% level -3.020686

10% level -2.650413

*MacKinnon (1996) one-sided p-values.

Augmented Dickey-Fuller Test Equation

Dependent Variable: D(FAI)

Method: Least Squares

Date: 05/28/05 Time: 16:48

Sample (adjusted): 1985 2004

Included observations: 20 after adjustments

CoefficieVariable nt Std. Error t-Statistic Prob.

FAI(-1) 0.259573 0.060915 4.261202 0.0011

D(FAI(-1)) 0.762570 0.231930 3.287925 0.0065

-0.27295D(FAI(-2)) 7 0.300382 -0.908699 0.3814

-0.74513D(FAI(-3)) 3 0.292414 -2.548210 0.0255

D(FAI(-4)) -0.608990.290971 -2.092966 0.0583

7

3

D(FAI(-5)) 0.739293

-1.25799D(FAI(-6))

C

5

189.1530

0.331724 -3.792298

376.0138 0.503048

0.0026

0.6240

0.338712 2.182655 0.0497

R-squared 0.956233 Mean dependent var 3411.991

3814.703

16.95094

Adjusted R-squared 0.930703 S.D. dependent var

S.E. of regression 1004.198 Akaike info criterion

1210097Sum squared resid 4 Schwarz criterion

-161.509Log likelihood 4 F-statistic

17.34923

37.45431

0.000000 Durbin-Watson stat 2.303034 Prob(F-statistic)

由上述结果可以看到两序列的ADF统计量均大于5%水平下的临界值,因而不能拒绝原假设,序列为非平稳序列。

由于两序列均为非平稳序列,因而需要进行两序列协整的检验,否则其回归将是没有意义的。

协整检验第一步,对两序列运用OLS法进行简单一元回归,得到回归参数估计和残差序列。

回归结果:

Dependent Variable: GDP

Method: Least Squares

8

Date: 05/30/05 Time: 02:57

Sample: 1978 2004

Included observations: 27

CoefficieVariable nt Std. Error t-Statistic Prob.

C 7569.484 2054.444 3.684445 0.0011

FAI 2.157312 0.083696 25.77556 0.0000

R-squared 0.963736 Mean dependent var 42756.36

Adjusted R-squared 0.962285 S.D. dependent var 41079.01

S.E. of regression 7977.694 Akaike info criterion 20.87787

Sum squared resid 1.59E+09 Schwarz criterion 20.97386

-279.851Log likelihood 3 F-statistic 664.3797

Durbin-Watson stat 0.288516 Prob(F-statistic) 0.000000

残差序列

Last updated: 05/30/05 - 02:57

1978 -5628.519

1979 -5356.801

9

1980 -5016.780

1981 -4780.261

1982 -4929.141

1983 -3483.656

1984 -4352.621

1985 -4091.560

1986 -4099.393

1987 -3786.865

1988 -2896.615

1989 -174.8939

1990 1233.837

1991 1979.233

1992 1637.317

1993 -1136.117

1994 2422.972

1995 7720.820

1996 10752.96

1997 13087.37

1998 9494.736

1999 10092.08

2000 10884.79

2001 9464.196

10

2002 3378.225

2003 -10192.12

2004 -22223.20

协整检验第二步,运用ADF法检验残差序列平稳性从而检验两序列是否存在协整。

残差序列ADF检验

Null Hypothesis: ET has a unit root

Exogenous: Constant

Lag Length: 5 (Automatic based on SIC, MAXLAG=6)

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic -3.730854 0.0113

Test critical values: 1% level -3.788030

5% level -3.012363

10% level -2.646119

*MacKinnon (1996) one-sided p-values.

Augmented Dickey-Fuller Test Equation

Dependent Variable: D(ET)

Method: Least Squares

Date: 05/30/05 Time: 03:25

Sample (adjusted): 1984 2004

11

Included observations: 21 after adjustments

CoefficieVariable nt Std. Error t-Statistic Prob.

-0.37046ET(-1) 8 0.099298 -3.730854 0.0022

D(ET(-1)) 0.962063 0.162816 5.908896 0.0000

D(ET(-2)) 0.461881 0.288497 1.600993 0.1317

D(ET(-3)) 0.245461 0.277022 0.886071 0.3905

D(ET(-4)) 0.307336 0.272255 1.128853 0.2779

D(ET(-5)) 1.203676 0.285665 4.213592 0.0009

-1293.08C 8 582.6998 -2.219132 0.0435

-892.358R-squared 0.846063 Mean dependent var 4

Adjusted R-squared 0.780090 S.D. dependent var 4690.288

S.E. of regression 2199.491 Akaike info criterion 18.49104

6772866Sum squared resid 3 Schwarz criterion 18.83922

-187.155Log likelihood 9 F-statistic 12.82436

Durbin-Watson stat 1.764056 Prob(F-statistic) 0.000055

12

由结果显示残差序列的ADF统计量小于5%水平下的临界值,因而不能拒绝原假设残差序列是平稳的,因而就有两序列间存在协整。也证实了两序列间存在长期稳定关系。

由于两序列被证实存在长期稳定关系,进一步检验GDP同固定资产投资间因果关系及程度。采用检验方法为Granger检验。

调整滞后长度为2-5,得到如下结果。

Pairwise Granger Causality Tests

Date: 05/30/05 Time: 03:33

Sample: 1978 2004

Lags: 2

Null Hypothesis: Obs F-Statistic Probability

GDP does not Granger Cause FAI 25 1.51006 0.24503

FAI does not Granger Cause GDP 12.8015 0.00026

Pairwise Granger Causality Tests

Date: 05/30/05 Time: 03:34

Sample: 1978 2004

Lags: 3

Null Hypothesis: Obs F-Statistic Probability

GDP does not Granger Cause FAI 24 0.60966 0.61786

13

FAI does not Granger Cause GDP 6.40575 0.00422

Pairwise Granger Causality Tests

Date: 05/30/05 Time: 03:34

Sample: 1978 2004

Lags: 4

Null Hypothesis: Obs F-Statistic Probability

GDP does not Granger Cause FAI 23 0.43905 0.77841

FAI does not Granger Cause GDP 4.74551 0.01249

Pairwise Granger Causality Tests

Date: 05/30/05 Time: 03:34

Sample: 1978 2004

Lags: 5

Null Hypothesis: Obs F-Statistic Probability

GDP does not Granger Cause FAI 22 3.05627 0.05695

FAI does not Granger Cause GDP 3.69720 0.03297

对上述结果总结如下:

14

滞后长度m=n

2

Granger因果性

GDP->FAI

FAI->GDP

F值 P值 结论

拒绝

不拒绝

拒绝

不拒绝

拒绝

不拒绝

不拒绝

不拒绝

1.51006 0.24503

12.8015

0.60966

6.40575

0.43905

4.74551

3.05627

3.6972

0.00026

0.61786

0.00422

0.77841

0.01249

0.05695

0.03297

3 GDP->FAI

FAI->GDP

4 GDP->FAI

FAI->GDP

5 GDP->FAI

FAI->GDP

可见GDP与固定资产投资存在明显的因果关系,受制于序列的不平稳才使得结论看上去仍受滞后长度的影响。。

四 模型设定,参数估计与检验

由数据分析可知,GDP与固定资产投资不但存在长期稳定关系更存在因果关系。因此可设定初步模型为:

GDP=C+β1* FAI+u

应用OLS法进行参数估计。得到如下结果:

Dependent Variable: GDP

Method: Least Squares

Date: 05/31/05 Time: 14:13

Sample: 1978 2004

Included observations: 27

Variable CoefficieStd. Error t-Statistic Prob.

15

nt

C

FAI

7569.484

2.157312

2054.444

0.083696

3.684445

25.77556

0.0011

0.0000

R-squared 0.963736 Mean dependent var 42756.36

41079.01

20.87787

20.97386

Adjusted R-squared 0.962285 S.D. dependent var

S.E. of regression

Sum squared resid

7977.694 Akaike info criterion

1.59E+09 Schwarz criterion

-279.851Log likelihood 3 F-statistic 664.3797

0.000000 Durbin-Watson stat 0.288516 Prob(F-statistic)

a经济意义检验:由经济理论以及此前的因果检验可知固定资产投资与GDP存在长期稳定的正线性关系,模型估计与此相符。

b统计推断检验:可决系数为0.963736,模型拟合情况较理想。T统计量为25.77556而显著水平0.05下临界值为2.060因此T统计量显著。说明参数估计是显著的,固定资产投资对GDP有显著影响。F统计量为664.3797,0.05显著水平下临界值为3.33,因此F统计量也是显著的。说明模型设定也是显著的。

c计量经济检验

1 多重共线性检验。由于是一元回归不存在多重共线性问题,无须检验。

2 异方差检验。

16

2.00E+071.60E+071.20E+072E8.00E+064.00E+060.00E+FAI

ARCH检验,设定滞后期为3得到如下结果

ARCH Test:

F-statistic 10.08751 Probability 0.000294

Obs*R-squared 14.45014 Probability 0.002352

Test Equation:

Dependent Variable: RESID^2

Method: Least Squares

Date: 05/31/05 Time: 14:53

Sample (adjusted): 1981 2004

Included observations: 24 after adjustments

CoefficieVariable nt Std. Error t-Statistic Prob.

17

C

RESID^2(-1)

9059274.

1.820983

-2.13

0.421753

0.444520

4.317653

0.6614

0.0003

RESID^2(-2)

RESID^2(-3)

8

1.488624

0.567807 -3.775697

0.491305 3.029939

0.0012

0.0066

6273110R-squared 0.602089 Mean dependent var 7

1.04E+08 Adjusted R-squared 0.542403 S.D. dependent var

7048675S.E. of regression

Sum squared resid

1 Akaike info criterion

9.94E+16 Schwarz criterion

-465.569Log likelihood 1 F-statistic

39.13076

39.32710

10.08751

0.000294 Durbin-Watson stat 1.787900 Prob(F-statistic)

比较obj*R2=14.45014>显著程度0.05,自由度P=3时的λ临界值7.81473。因此决绝原假设,判断模型误差项存在异方差。

3 自相关检验。

18

5000RESID10000-1000-2000-3000-4040006000RESID(-1)

由此前回归结果可知D-W统计量为0.288516。给定显著水平0.05,查D-W表n=27,k=1得下限临界值为1.316,上限临界值为1.469。而0.288516<下限1.316因此模型误差项存在一阶自相关。

五 模型修正

(一)异方差修正

WLS估计法。生成权数w=1/fai的估计结果为

Dependent Variable: GDP

Method: Least Squares

Date: 05/31/05 Time: 15:16

Sample: 1978 2004

Included observations: 27

Weighting series: 1/FAI

CoefficieVariable nt Std. Error t-Statistic Prob.

19

C 1984.233 211.1094 9.399078 0.0000

FAI 2.757995 0.109874 25.10151 0.0000

Weighted Statistics

R-squared 0.779430 Mean dependent var 10214.52

Adjusted R-squared 0.770607 S.D. dependent var 2721.756

S.E. of regression 1303.584 Akaike info criterion 17.25481

4248329Sum squared resid 9 Schwarz criterion 17.35080

-230.939Log likelihood 9 F-statistic 630.0859

Durbin-Watson stat 0.677084 Prob(F-statistic) 0.000000

Unweighted Statistics

R-squared 0.878099 Mean dependent var 42756.36

Adjusted R-squared 0.873223 S.D. dependent var 41079.01

S.E. of regression 14626.47 Sum squared resid 5.35E+09

Durbin-Watson stat 0.201485

换用对数变换法将gdp和fai替换成Lgdp和Lfai。的如下结论

Dependent Variable: LGDP

Method: Least Squares

20

Date: 05/31/05 Time: 15:22

Sample: 1978 2004

Included observations: 27

CoefficieVariable nt Std. Error t-Statistic Prob.

C

LFAI

2.713661

0.829008

0.115957

0.012904

23.40222

64.24241

0.0000

0.0000

R-squared 0.993979 Mean dependent var 10.06867

1.208213

-1.78592Adjusted R-squared 0.993738 S.D. dependent var

S.E. of regression 0.095609 Akaike info criterion 1

-1.68993Sum squared resid

Log likelihood

0.228525 Schwarz criterion

26.10993 F-statistic

3

4127.088

0.000000 Durbin-Watson stat 0.876328 Prob(F-statistic)

比较两种方法可知gdp与固定资产投资在对数线性回归下拟合最好!此时的ARCH检验结果为

ARCH Test:

F-statistic

Obs*R-squared

0.685945 Probability

2.239024 Probability

0.571103

0.524303

21

Test Equation:

Dependent Variable: RESID^2

Method: Least Squares

Date: 05/31/05 Time: 15:24

Sample (adjusted): 1981 2004

Included observations: 24 after adjustments

CoefficieVariable nt Std. Error t-Statistic Prob.

C 0.006041 0.003046 1.983333 0.0612

RESID^2(-1) 0.173273 0.226838 0.763864 0.4539

-0.01532RESID^2(-2) 0 0.226278 -0.067705 0.9467

RESID^2(-3) 0.232025 0.225640 1.028298 0.3161

R-squared 0.093293 Mean dependent var 0.009341

-0.04271Adjusted R-squared 3 S.D. dependent var 0.007696

-6.70339S.E. of regression 0.007859 Akaike info criterion 3

-6.50705Sum squared resid 0.001235 Schwarz criterion 1

22

Log likelihood 84.44072 F-statistic 0.685945

0.571103 Durbin-Watson stat 1.892617 Prob(F-statistic)

其obj*R2=2.239024<临界值7.81473。异方差修正!

此时模型修正为:

LGDP=C+β1* LFAI+u

二 自相关修正

广义差分。此前结论有DW=0.876328,因此计算出β估计量为0.561836,从而分别得到GDP和FAI的差分序列,再进行OLS参数估计得到:

Dependent Variable: DLGDP

Method: Least Squares

Date: 06/07/05 Time: 02:13

Sample (adjusted): 1979 2004

Included observations: 26 after adjustments

CoefficienVariable t Std. Error t-Statistic Prob.

C 1.275681 0.104470 12.21095 0.0000

DLFAI 0.807090 0.025677 31.43270 0.0000

R-squared 0.976285 Mean dependent var 4.521712

Adjusted R-squared 0.975297 S.D. dependent var 0.512440

S.E. of regression 0.080542 Akaike info criterion -2.126285

Sum squared resid 0.155686 Schwarz criterion -2.029508

Log likelihood 29.64170 F-statistic 988.0147

Durbin-Watson stat 1.663055 Prob(F-statistic) 0.000000

D-W=1.663055,此时的不能拒绝区域为(1.464,2.531)因此D-W落在不能拒绝的区域,修正了自相关。

再使用迭代法可得到:

Dependent Variable: LGDP

Method: Least Squares

23

Date: 05/31/05 Time: 15:38

Sample (adjusted): 1979 2004

Included observations: 26 after adjustments

Convergence achieved after 15 iterations

CoefficieVariable nt Std. Error t-Statistic Prob.

C 3.059726 0.387301 7.900121 0.0000

LFAI 0.791156 0.041009 19.29245 0.0000

AR(1) 0.670792 0.177606 3.776861 0.0010

R-squared 0.995495 Mean dependent var 10.14072

Adjusted R-squared 0.995103 S.D. dependent var 1.171495

-2.05653S.E. of regression 0.081979 Akaike info criterion 9

-1.91137Sum squared resid 0.154573 Schwarz criterion 4

Log likelihood 29.73501 F-statistic 2541.114

Durbin-Watson stat 1.815953 Prob(F-statistic) 0.000000

Inverted AR Roots .67

DW=1.815953Y也落在不能拒绝的区域,修正了自相关。

比较两种方法取得的结果,可知,使用迭代法更为准确。

24

六 时期考虑

为了验证是否在近10年内的固定资产投资是否过热,我们运用上述方法和结果对1994-2004期间的数据进行了分析,得到如下结果:

Dependent Variable: LGDP

Method: Least Squares

Date: 05/31/05 Time: 16:33

Sample: 1994 2004

Included observations: 11

Convergence achieved after 6 iterations

CoefficienVariable t Std. Error t-Statistic Prob.

C 5.891059 0.547243 10.76497 0.0000

LFAI 0.531348 0.050359 10.55127 0.0000

AR(1) 0.627602 0.058372 10.75170 0.0000

R-squared 0.996148 Mean dependent var 11.32730

Adjusted R-squared 0.995185 S.D. dependent var 0.309808

S.E. of regression 0.021497 Akaike info criterion -4.614782

Sum squared resid 0.003697 Schwarz criterion -4.506265

Log likelihood 28.38130 F-statistic 1034.457

Durbin-Watson stat 2.062084 Prob(F-statistic) 0.000000

Inverted AR Roots .63

可见模型拟合良好,但固定资产投资对GDP的系数明显下降,证明了其拉动GDP增长的效率是下降了。而在GDP增速和经济结构均无重大变化的情况下,不难得出固定资产投资必然增长过快即过热的结论!

七 模型结论:

1978-2004修正后的模型I为

LGDP = 3.059725644 + 0.7911561329*LFAI

(0.387301) (0.041009)

t= (7.900121) (19.29245)

经济意义:固定资产投资每增长1%将拉动GDP增长0.7911561329%,可见固定资产投资作为经济增长“三驾马车”的称谓名至实归。

25

1994-2004的模型II为

LGDP = 5.891059452 + 0.53134764*LFAI

(0.0547243) (0.050349)

t= (10.55127) (10.75170)

经济意义:固定资产投资每增加1%将拉动GDP增长0.53134764%,可见固定资产投资的拉动效率下降了,造成固定资产投资过热的原因之一。

八 模型预测应用效果

最新数据有2005年第一季度全国固定资产投资额为10583亿元,GDP为31318.98亿元。四月份固定资产投资额为3441.67亿元。

为保证准确性使用模型I进行预测。将一季度数据代入模型内有:

LGDP预测值=3.059725644 + 0.7911561329*ln10583

=10.39137286555

进而得GDP预测值=e(10.39137286555)=32577.36亿元

实际GDP值为31318.98,误差为(32577.36-31318.98)/31318.98*100%=4.01795%实证表明预测效果良好!

继续预测4月份的GDP:

LGDP预测值=3.059725644 + 0.7911561329*ln3441.67

=9.5

得到:GDP预测值=e(9.5)=13439.77亿元

26


本文标签: 投资 检验 序列