我國改革開放以來固定資產投資與GDP關系分析

1、我國改革開放以來固定資產投資與GDP關系分析【摘要】本文旨在對我國改革開放以來固定資產投資與GDP關系進行計量分析。首先我們對已有的部分關于固定資產投資的觀點和評論進行了評述；然后再收集的數(shù)據(jù)的基礎上利用EViews軟件進行了計量分析，從數(shù)據(jù)本身出發(fā)驗證了兩者的因果關系，并尋求設定合理的經濟關系模型；接著運用軟件對設定的模型進行了參數(shù)估計，檢驗及修正；最后我們利用所得的結果進行了經濟預測以評估所得結果的價值并對結果本身提出了政策意見。一問題的提出我國自改革開放以來已保持了國民經濟20多年的快速增長，GDP年均增長率在10%以上，如此高的增長速度不經要引起人們對其增長動力或原因的興趣。今年來關于

2、投資，消費和出口“三駕馬車”拉動經濟增長的理論較為突出。尤其是進入90年代后直到90年代末到新世紀最近幾年，不論是學術界還是公眾媒體都對固定資產投資的高增長表現(xiàn)出不同程度的擔憂，因而才引出關于經濟軟著陸和怎樣減少固定資產投資的討論。那么，究竟固定資產投資同GDP之間的關系如何？新世紀的前后幾年是不是存在固定資產投資過熱拉動經濟過熱的情況？本文試圖運用計量經濟學的方法尋求答案。二數(shù)據(jù)收集為進行計量分析，我們尋求改革開放至今的GDP和固定資產的可比數(shù)據(jù)，數(shù)據(jù)來源為中國統(tǒng)計年鑒及中國國家統(tǒng)計局網(wǎng)站(http:)的數(shù)據(jù)資料，兩項數(shù)據(jù)樣本數(shù)都為27,滿足一元回歸的要求。1978-2004年GDP及固定資

3、產投資年度數(shù)據(jù)obsGDPFAI1978197919801981198219831984198519861987198819891990199119921993199419951996199719983624.1004038.2004517.8004862.4005294.7007171.0007171.0008964.40010202.2011962.5014928.3016909.2018547.9021617.8026638.1034634.4046759.4058478.1067884.6074462.6078345.20780.2000846.2000910.9000961.0000

4、1230.4001430.1001832.9002543.2003120.6003791.7004753.8004410.4004517.0005594.5008080.10013072.3017042.9420019.2622974.0324941.10284065029854.71200089468.1032917.73200197314.8037213.492002104790.643499.912003117251.955566.612004136515.070072.71數(shù)據(jù)分析由于相關數(shù)據(jù)為時間序列，很可能為非平穩(wěn)序列，直接回歸可能造成偽回歸。因此對兩時間

5、序列進行平穩(wěn)性檢驗，方法為ADF檢驗。EViews5默認情況下檢驗結果如下：GDP的ADF檢驗NullHypothesis:GDPhasaunitrootExogenous:ConstantLagLength:2(AutomaticbasedonSIC,MAXLAG=6)t-StatisticProb.MacKinnon(1996)one-sidedp-values.AugmentedDickey-Fullerteststatistic2.5889251.0000Testcriticalvalues:1%level-3.7378535%level-2.99187810%level-2.6355

6、42AugmentedDickey-FullerTestEquationDependentVariable:D(GDP)Method:LeastSquaresDate:05/28/05Time:16:45Sample(adjusted):19812004Includedobservations:24afteradjustmentsCoefficieProb.VariablentStd.Errort-StatisticGDP(-1)0.0457720.0176802.5889250.0175D(GDP(-1)1.327562-0.732830.2171506.1135740.0000D(GDP(

7、-2)10.231507-3.1654850.0049C399.8333664.79320.6014400.5543R-squared0.851066Meandependentvar5499.883AdjustedR-squared0.828726S.D.dependentvar4860.139S.E.ofregression2011.3838091321Akaikeinfocriterion18.20204Sumsquaredresid9-214.424Schwarzcriterion18.39839Loglikelihood5F-statistic38.09586Durbin-Watson

8、stat2.019994Prob(F-statistic)0.000000FAI的ADF檢驗NullHypothesis:FAIhasaunitrootExogenous:ConstantLagLength:6(AutomaticbasedonSIC,MAXLAG=6)t-StatisticProb.*AugmentedDickey-Fullerteststatistic4.2612021.0000Testcriticalvalues:1%level-3.8085465%level-3.02068610%level-2.650413*MacKinnon(1996)one-sidedp-valu

9、es.AugmentedDickey-FullerTestEquationDependentVariable:D(FAI)Method:LeastSquaresDate:05/28/05Time:16:48Sample(adjusted):19852004Includedobservations:20afteradjustmentsVariable2E.FAI(-1)0.2595730.0609154.2612020.0011D(FAI(-1)0.7625700.2319303.2879250.0065-0.27295D(FAI(-2)70.300382-0.9086990.3814-0.74

10、513D(FAI(-3)30.292414-2.5482100.0255-0.60899D(FAI(-4)30.290971-2.0929660.0583D(FAI(-5)0.7392930.3387122.1826550.0497-1.25799D(FAI(-6)50.331724-3.7922980.0026C189.1530376.01380.5030480.6240R-squared0.956233Meandependentvar3411.991AdjustedR-squared0.930703S.D.dependentvar3814.703S.E.ofregression1004.1

11、98Akaikeinfocriterion16.950941210097Sumsquaredresid4Schwarzcriterion17.34923-161.509Loglikelihood4F-statistic37.45431Durbin-Watsonstat2.303034Prob(F-statistic)0.000000由上述結果可以看到兩序列的ADF統(tǒng)計量均大于5%水平下的臨界值，因而不能拒絕原假設,序列為非平穩(wěn)序列由于兩序列均為非平穩(wěn)序列，因而需要進行兩序列協(xié)整的檢驗，否則其回歸將是沒有意義的。協(xié)整檢驗第一步，對兩序列運用OLS法進行簡單一元回歸，得到回歸參數(shù)估計和殘差序列。回

12、歸結果：DependentVariable:GDPMethod:LeastSquaresDate:05/30/05Time:02:57Sample:19782004Includedobservations:27CoefficieVariablentStd.Errort-StatisticProb.C7569.4842054.4443.6844450.0011FAI2.1573120.08369625.775560.0000R-squared0.963736Meandependentvar42756.36AdjustedR-squared0.962285S.D.dependentvar41079

13、.01S.E.ofregression7977.694Akaikeinfocriterion20.87787Sumsquaredresid1.59E+09Schwarzcriterion20.97386-279.851Loglikelihood3F-statistic664.3797Durbin-Watsonstat0.288516Prob(F-statistic)0.000000殘差序列Lastupdated:05/30/05-02:571978-5628.5191979-5356.8011980-5016.7801981-4780.2611982-4929.1411983-3483.656

14、1984-4352.6211985-4091.5601986-4099.3931987-3786.8651988-2896.6151989-174.893919901233.83719911979.23319921637.3171993-1136.11719942422.97219957720.820199610752.96199713087.3719989494.736199910092.08200010884.7920019464.19620023378.2252003-10192.122004-22223.20協(xié)整檢驗第二步，運用ADF法檢驗殘差序列平穩(wěn)性從而檢驗兩序列是否存在協(xié)整。殘差

15、序列ADF檢驗NullHypothesis:EThasaunitrootExogenous:ConstantLagLength:5(AutomaticbasedonSIC,MAXLAG=6)t-StatisticProb.MacKinnon(1996)one-sidedp-values.AugmentedDickey-Fullerteststatistic-3.7308540.0113Testcriticalvalues:1%level-3.7880305%level-3.01236310%level-2.646119AugmentedDickey-FullerTestEquationDepe

16、ndentVariable:D(ET)Method:LeastSquaresDate:05/30/05Time:03:25Sample(adjusted):19842004Includedobservations:21afteradjustmentsCoefficieStd.Errort-StatisticProb.VariablentET(-1)-0.3704680.099298-3.7308540.0022D(ET(-1)0.9620630.1628165.9088960.0000D(ET(-2)0.4618810.2884971.6009930.1317D(ET(-3)0.2454610

17、.2770220.8860710.3905D(ET(-4)0.3073360.2722551.1288530.2779D(ET(-5)1.2036760.2856654.2135920.0009C-1293.088582.6998-2.2191320.0435R-squared0.846063MeandependentvarAdjustedR-squared0.780090S.D.dependentvar4690.288S.E.ofregression2199.491Akaikeinfocriterion18.491046772866Sumsquaredresid3Schwarzcriteri

18、on18.83922-187.155Loglikelihood9F-statistic12.82436Durbin-Watsonstat1.764056Prob(F-statistic)0.000055由結果顯示殘差序列的ADF統(tǒng)計量小于5%水平下的臨界值，因而不能拒絕原假設殘差序列是平穩(wěn)的，因而就有兩序列間存在協(xié)整。也證實了兩序列間存在長期穩(wěn)定關系。由于兩序列被證實存在長期穩(wěn)定關系，進一步檢驗GDP同固定資產投資間因果關系及程度。采用檢驗方法為Granger檢驗。調整滯后長度為2-5,得到如下結果。PairwiseGrangerCausalityTestsDate:05/30/05Time:

19、03:33Sample:19782004Lags:2NullHypothesis:ObsF-StatisticProbabilityGDPdoesnotGrangerCauseFAI251.510060.24503FAIdoesnotGrangerCauseGDP12.80150.00026PairwiseGrangerCausalityTestsDate:05/30/05Time:03:34Sample:19782004Lags:3NullHypothesis:ObsF-StatisticProbabilityGDPdoesnotGrangerCauseFAI240.609660.61786

20、FAIdoesnotGrangerCauseGDP6.405750.00422PairwiseGrangerCausalityTestsDate:05/30/05Time:03:34Sample:19782004Lags:4NullHypothesis:ObsF-StatisticProbabilityGDPdoesnotGrangerCauseFAI230.439050.77841FAIdoesnotGrangerCauseGDP4.745510.01249PairwiseGrangerCausalityTestsDate:05/30/05Time:03:34Sample:19782004L

21、ags:5NullHypothesis:ObsF-StatisticProbabilityGDPdoesnotGrangerCauseFAI223.056270.056953.697200.03297FAIdoesnotGrangerCauseGDP對上述結果總結如下:滯后長度m=nGranger因果性F值P值結論2GDP->FAI1.510060.24503拒絕FAI->GDP12.80150.00026不拒絕3GDP->FAI0.609660.61786拒絕FAI->GDP6.405750.00422不拒絕4GDP->FAI0.439050.77841拒絕FA

22、I->GDP4.745510.01249不拒絕5GDP->FAI3.056270.05695不拒絕FAI->GDP3.69720.03297不拒絕可見GDP與固定資產投資存在明顯的因果關系，受制于序列的不平穩(wěn)才使得結論看上去仍受滯后長度的影響。四模型設定，參數(shù)估計與檢驗由數(shù)據(jù)分析可知，GDP與固定資產投資不但存在長期穩(wěn)定關系更存在因果關系。因此可設定初步模型為：GDP=C+1*FAI+u應用OLS法進行參數(shù)估計。得到如下結果:DependentVariable:GDPMethod:LeastSquaresDate:05/31/05Time:14:13Sample:197820

23、04Includedobservations:27CoefficieVariablentStd.Errort-StatisticProb.C7569.4842054.4443.6844450.0011FAI2.1573120.08369625.775560.0000R-squared0.963736Meandependentvar42756.36AdjustedR-squared0.962285S.D.dependentvar41079.01S.E.ofregression7977.694Akaikeinfocriterion20.87787Sumsquaredresid1.59E+09Sch

24、warzcriterion20.97386-279.851Loglikelihood3F-statistic664.3797Durbin-Watsonstat0.288516Prob(F-statistic)0.000000a經濟意義檢驗：由經濟理論以及此前的因果檢驗可知固定資產投資與GDP存在長期穩(wěn)定的正線性關系，模型估計與此相符。b統(tǒng)計推斷檢驗：可決系數(shù)為0.963736，模型擬合情況較理想。T統(tǒng)計量為25.77556而顯著水平0.05下臨界值為2.060因此T統(tǒng)計量顯著。說明參數(shù)估計是顯著的，固定資產投資對GDP有顯著影響。F統(tǒng)計量為664.3797,0.05顯著水平下臨界值為3.33,

25、因此F統(tǒng)計量也是顯著的。說明模型設定也是顯著的。c計量經濟檢驗1多重共線性檢驗。由于是一元回歸不存在多重共線性問題，無須檢驗。2異方差檢驗。2.00E+071.60E+071.20E+072E8.00E+064.00E+060.00E+00020000400006000080000FAIARCH檢驗，設定?t后期為3得到如下結果ARCHTest:F-statistic10.08751Probability0.000294Obs*R-squared14.45014Probability0.002352TestEquation:DependentVariable:RESIDA2Method:Lea

26、stSquaresDate:05/31/05Time:14:53Sample(adjusted):19812004Includedobservations:24afteradjustmentsCoefficieVariablentStd.Errort-StatisticProb.C9059274.203799130.4445200.6614RESID2-1)1.8209830.4217534.3176530.0003-2.14386RESID2-2)80.567807-3.7756970.0012RESID2-3)1.4886240.4913053.0299390.00666273110R-s

27、quared0.602089Meandependentvar7AdjustedR-squared0.542403S.D.dependentvar1.04E+087048675S.E.ofregression1Akaikeinfocriterion39.13076Sumsquaredresid9.94E+16Schwarzcriterion39.32710-465.569Loglikelihood1F-statistic10.08751Durbin-Watsonstat1.787900Prob(F-statistic)0.000294比較obj*R2=14.45014>顯著程度0.05,自

28、由度P=3時的入臨界值7.81473。因此決絕原假設，判斷模型誤差項存在異方差。3自相關檢驗。50004000-3000_«PSER2000-.1000-、，Q-*0*.*-1000-2000-*-3000一J,一J11-4000-20000200040006000RESID(-1)由此前回歸結果可知D-W統(tǒng)計量為0.288516。給定顯著水平0.05,查D-W表n=27,k=1得下限臨界值為1.316,上限臨界值為1.469。而0.288516<下PM1.316因此模型誤差項存在一階自相關。五模型修正（一）異方差修正WLS估計法。生成權數(shù)w=1/fai的估計結果為Depend

29、entVariable:GDPMethod:LeastSquaresDate:05/31/05Time:15:16Sample:19782004Includedobservations:27Weightingseries:1/FAICoefficieVariablentStd.Errort-StatisticProb.C1984.233211.10949.3990780.0000FAI2.7579950.10987425.101510.0000WeightedStatisticsR-squared0.779430Meandependentvar10214.52AdjustedR-squared

30、0.770607S.D.dependentvar2721.756S.E.ofregression1303.584Akaikeinfocriterion17.254814248329Sumsquaredresid9Schwarzcriterion17.35080-230.939Loglikelihood9F-statistic630.0859Durbin-Watsonstat0.677084Prob(F-statistic)0.000000UnweightedStatisticsR-squared0.878099Meandependentvar42756.36AdjustedR-squared0

31、.873223S.D.dependentvar41079.01S.E.ofregression14626.47Sumsquaredresid5.35E+09Durbin-Watsonstat0.201485換用對數(shù)變換法將gdp和fai替換成Lgdp和Lfai。的如下結論DependentVariable:LGDPMethod:LeastSquaresDate:05/31/05Time:15:22Sample:19782004Includedobservations:27CoefficieStd.Errort-StatisticProb.VariablentC2.7136610.1159572

32、3.402220.0000LFAI0.8290080.01290464.242410.0000R-squared0.993979Meandependentvar10.06867AdjustedR-squared0.993738S.D.dependentvar1.208213-1.78592S.E.ofregression0.095609Akaikeinfocriterion1-1.68993Sumsquaredresid0.228525Schwarzcriterion3Loglikelihood26.10993F-statistic4127.088Durbin-Watsonstat0.8763

33、28Prob(F-statistic)0.000000比較兩種方法可知gdp與固定資產投資在對數(shù)線性回歸下擬合最好！此時的ARCH檢驗結果ARCHTest:F-statistic0.685945Probability0.571103Obs*R-squared2.239024Probability0.524303TestEquation:DependentVariable:RESIDA2Method:LeastSquaresDate:05/31/05Time:15:24Sample(adjusted):19812004Includedobservations:24afteradjustments

34、CoefficieVariablentStd.Errort-StatisticProb.C0.0060410.0030461.9833330.0612RESIDA2(-1)0.1732730.2268380.7638640.4539-0.01532RESIDA2(-2)00.226278-0.0677050.9467RESIDA2(-3)0.2320250.2256401.0282980.3161R-squared0.093293Meandependentvar0.009341AdjustedR-squared-0.04271S.D.dependentvar0.007696-6.70339S.

35、E.ofregression0.007859Akaikeinfocriterion3-6.50705Sumsquaredresid0.001235Schwarzcriterion1Loglikelihood84.44072F-statistic0.685945Durbin-Watsonstat1.892617Prob(F-statistic)0.571103其obj*R2=2.239024<臨界值7.81473。異方差修正！此時模型修正為：LGDP=C+1*LFAI+u二自相關修正,從而分別得廣義差分。此前結論有DW=0.876328,因此計算出明古計量為0.561836到GDP和FAI

36、的差分序列，再進行OLS參數(shù)估計得到：DependentVariable:DLGDPMethod:LeastSquaresDate:06/07/05Time:02:13Sample(adjusted):19792004Includedobservations:26afteradjustmentsVariableCoefficientStd.Errort-StatisticProb.C1.2756810.10447012.210950.0000DLFAI0.8070900.02567731.432700.0000R-squared0.976285Meandependentvar4.521712A

37、djustedR-squared0.975297S.D.dependentvar0.512440S.E.ofregression0.080542Akaikeinfocriterion-2.126285Sumsquaredresid0.155686Schwarzcriterion-2.029508Loglikelihood29.64170F-statistic988.0147Durbin-Watsonstat1.663055Prob(F-statistic)0.000000D-W=1.663055,此時的不能拒絕區(qū)域為(1.464,2.531)因此D-W客在不能拒絕的區(qū)域,修正了自相關。再使用迭

38、代法可得到：DependentVariable:LGDPMethod:LeastSquaresDate:05/31/05Time:15:38Sample(adjusted):19792004Includedobservations:26afteradjustmentsConvergenceachievedafter15iterationsCoefficieVariablentStd.Errort-StatisticProb.C3.0597260.3873017.9001210.0000LFAI0.7911560.04100919.292450.0000AR(1)0.6707920.177606

39、3.7768610.0010R-squared0.995495Meandependentvar10.14072AdjustedR-squared0.995103S.D.dependentvar1.171495-2.05653S.E.ofregression0.081979Akaikeinfocriterion9-1.91137Sumsquaredresid0.154573Schwarzcriterion4Loglikelihood29.73501F-statistic2541.114Durbin-Watsonstat1.815953Prob(F-statistic)0.000000InvertedARRoots.67DW=1.815953Y也落在不能拒絕的區(qū)域，修正了自相關。比較兩種方法取得的結果，可知，使用迭代法更為準確。六時期考慮為了驗證是否在近10年內的固定資產投資是否過熱，我們運用上述方法和結果對1994-2004期間的數(shù)據(jù)進行了分析，得到如下結果:DependentVariable:LGDPMethod:LeastSquaresDate:05/31/05Ti