《2021-2022年收藏的精品资料计量经济模型.doc》由会员分享,可在线阅读,更多相关《2021-2022年收藏的精品资料计量经济模型.doc(20页珍藏版)》请在淘文阁 - 分享文档赚钱的网站上搜索。
1、关于我国影响税收增长因素的实证分析【摘要】:税收是我国财政收入的重要组成部分,对维持社会稳定和促进经济增长有很大的作用。影响税收收入的因素来自于很多方面,从国内生产总值,财政支出和物价这三个方面进行研究,得出税收与三者的关系,为现行政策提供参考。【关键词】:国内生产总值 财政支出 零售商品物价水平 税收 计量模型 检验一、问题的提出 改革开放以来,中国经济高速增长,1978-2008年的31年间,国内生产总值从3645.2亿元增长到314045亿元,一跃成为世界第二大经济体。随着经济体制改革的深化和经济的快速增长,中国的财政收支状况也发生了很大的变化,中央和地方的税收收入1978年为519.2
2、8亿元,到2008年已增长到54223.79亿元,31年间平均每年增长16.76%。税收作为财政收入的重要组成部分,在国民经济发展中扮演着不可或缺的角色。为了研究影响中国税收增长的主要原因,分析中央和地方税收收入的增长规律,以及预测中国税收未来的增长趋势,我们需要建立计量经济模型进行实证分析。而且从进入21世纪以来,我国的经济发展面临着巨大的挑战与机遇,在新的经济背景下,基于知识和信息的产业发展迅速,全球一体化日渐深入,中国已是WTO的一员。新形势的经济发展是经济稳定和协调增长的结果,由于税收具有敛财与调控的重要功能,因而它在现实的经济发展中至始至终都发挥着非常重要的作用,所以研究影响我国税收
3、收入的主要原因具非常重要的作用。 二、模型设定(一)为了具体分析各要素对提高我国税收收入的影响大小,选择能反映我们税收变动情况的“各项税收收入”为被解释变量(用Y表示),选择能影响税收收入的“国内生产总值(用X1表示)”、“财政支出(用X2表示)”和“ 商品零售价格指数(用X3表示)”为解释变量。计量经济学模型的设定lnY= 0+ 1 lnX1+ 2 lnX2 + 3 X3 + ui(二)确定参数估计值范围由经济常识知,因为国内生产总值(X1)、财政支出(X2)和商品零售价格指数(X3)的增加均会带动税收收入的增加,所以国内生产总值(X1)、财政支出(X2)和商品零售价格指数(X3)与税收收入
4、应为正相关的关系,所以可估计011 ,021, 031。表1为由中国统计年鉴得到的1990-2009年的有关数据。表1 税收收入模型的时间序列表年份税收收入(Y)国内生产总值(X1)财政支出(X2)商品零售价格指数(X3)(单位:%)(单位:亿元) (单位:亿元)(单位:亿元)19902821.86 18667.80 3083.59102.119912990.17 21781.50 3386.62102.919923296.91 26923.48 3742.2105.419934255.30 35333.92 4642.3113.219945126.88 48197.86 5792.62121
5、.719956038.04 60793.73 6823.72114.819966909.82 71176.59 7937.55106.119978234.04 78973.03 9233.56100.819989262.80 84402.28 10798.1897.4199910682.58 89677.05 13187.6797200012581.51 99214.55 15886.598.5200115301.38 109655.17 18902.5899.2200217636.45 120332.69 22053.1598.7200320017.31 135822.76 24649.95
6、99.9200424165.68 159878.34 28486.89102.8200528778.54 184937.37 33930.28100.8200634804.35 216314.43 40422.73101200745621.97 265810.31 49781.35103.8200854223.79 314045.43 62592.66105.9200959521.59 340506.87 76299.9398.8资料来源:中国统计年鉴2009;三、参数估计模型为:lnY= 0+ 1 lnX1+ 2 lnX2 + 3 X3 + uiY=税收收入 (亿元)X1=国内生产总值 (亿
7、元)X2=财政支出 (亿元)X3=零售商品物价指数 (%)用Eviews估计结果为:表2Dependent Variable: LOG(Y)Method: Least SquaresDate: 06/12/11 Time: 10:53Sample: 1990 2009Included observations: 20VariableCoefficientStd. Errort-StatisticProb.C-0.3475030.284342-1.2221280.2394LOG(X1)-0.0052180.073387-0.0711010.9442LOG(X2)0.9878780.0644271
8、5.333210.0000X30.0035430.0017482.0271930.0596R-squared0.998444Mean dependent var9.405915Adjusted R-squared0.998152S.D. dependent var0.972567S.E.of regression0.041805Akaike info criterion-3.334744Sumsquared resid0.027963Schwarz criterion-3.135598Log likelihood37.34744F-statistic3422.460Durbin-Watson
9、stat0.986881Prob(F-statistic)0.000000根据表中数据,模型设计的结果为: (-1.222128) (-0.071101) (15.33321) (2.027193) R2=0.998444 R()2=0.998152 DW=0.986881 F=3422.460 n=20四、检验及修正(一)经济意义检验经济意义检验主要检验模型参数估计量在经济意义上的合理性。所估计的参数= -0.005218,=0.987878, =0.003543,且0, 01 , 01 ,不符合变量参数中确定的参数范围,、符合变量参数中确定的参数范围。模型估计结果说明,在假定其他变量不变的
10、情况下,当年国内生产总值每增长1%,平均来说税收收入会减少0.005218%;在假定其他变量不变的情况下,当年财政支出每增长1%,平均来说税收收入会增加0.987878%;在假定其他变量不变的情况下,当年商品零售价格指数上涨1%,平均来说税收收入会增加0.003543%。这里与理论分析和经验判断相一致,符合中国现实的国情具有经济意义应保留,符号为负不符合经济检验不具有经济意义,应剔除。(二)统计意义检验1、拟合优度检验(R2检验)拟合优度检验,顾名思义,是检验模型对样本观测值的拟合程度。从回归估计的结果看模型拟合较好:可绝系数 R2=0.998444 R()2=0.998152 ,这说明所建模
11、型整体上与样本观测值拟合的很好说明“解释变量”国内生产总值 财政支出 商品零售价格指数 对“被解释变量” 税收收入的绝大部分差异作了解释。2、 F检验假设:=0,=0,=0 :(j=1,2,3)不全为零给定显著性水平=0.05,在F分布表中查出自由度为F(k=3,n-k-1=16)的临界值(3,16)3.24,由表2中得到F3422.460(3,16)3.24,应拒绝原假设:=0,=0,=0 ,接受:(j=1,2,3)不全为零说明回归方程显著,即表明模型的线性关系在95%的置信水平下成立,即列入模型的解释变量“解释变量”国内生产总值 财政支出 商品零售价格指数 联合起来确实对“被解释变量”税收
12、收入有显著影响。3、t检验分别针对:=0,=0,=0 ,给定显著性水平0.05,查t分布表的自由度为n-k-116的临界值2.120。由表2中的数据可得,与、对应的t统计量分别为(-1.222128)(-0.071101)(15.33321) (2.027193)其绝对值不全大于2.120,这说明在显著水平0.05下,只有能拒绝:=0,=0,=0 ,也就是说,当在其他解释变量不变的情况下,各个解释变量“国内生产总值(X1)”、“财政支出(X2)”和“ 商品零售价格指数(X3)”分别对被解释变量“各项税收收入(Y)”不全都有显著影响,这可能是由于多重共线性或自相关性的影响。(三)计量经济意义检验
13、1、多重共线性检验让lnY分别对lnX1、lnX2、X3做回归。(1)将lnY与lnX1做回归得到结果如表3: 表 3Dependent Variable: LOG(Y)Method: Least SquaresDate: 06/12/11 Time: 19:53Sample: 1990 2009Included observations: 20VariableCoefficientStd. Errort-StatisticProb.C-3.4252980.499347-6.8595560.0000LOG(X1)1.1235750.04361025.764170.0000R-squared0.
14、973599Mean dependent var9.405915AdjustedR-squared0.972132S.D. dependent var0.972567S.E.of regression0.162357Akaike info criterion-0.703402Sum squared resid0.474475Schwarz criterion-0.603829Log likelihood9.034023F-statistic663.7926Durbin-Watson stat0.204663Prob(F-statistic)0.000000R2=0.973599 D.W.=0.
15、204663(2)将lnY与lnX2做回归得到结果如表4: 表 4Dependent Variable: LOG(Y)Method: Least SquaresDate: 06/12/11 Time: 20:05Sample: 1990 2009Included observations: 20VariableCoefficientStd. Errort-StatisticProb.C0.0889020.0987940.8998760.3801LOG(X2)0.9743730.01027994.793760.0000R-squared0.998001Mean dependent var9.40
16、5915Adjusted R-squared0.997890S.D. dependent var0.972567S.E. of regression0.044677Akaike info criterion-3.284085Sum squared resid0.035928Schwarz criterion-3.184512Log likelihood34.84085F-statistic8985.857Durbin-Watson stat0.835853Prob(F-statistic)0.000000 (0.899876) (94.79376)R2=0.998001 D.W.=0.8358
17、53(3)将lnY与X3做回归得到结果如表5:表 5Dependent Variable: LOG(Y)Method: Least SquaresDate: 06/12/11 Time: 20:07Sample: 1990 2009Included observations: 20VariableCoefficientStd. Errort-StatisticProb.C15.423253.4441614.4780850.0003X3-0.0581160.033204-1.7502610.0971R-squared0.145438Mean dependent var9.405915Adjust
18、ed R-squared0.097962S.D. dependent var0.972567S.E. of regression0.923702Akaike info criterion2.773786Sum squared resid15.35807Schwarz criterion2.873359Log likelihood-25.73786F-statistic3.063413Durbin-Watson stat0.129252Prob(F-statistic)0.097100 (4.478085) (-1.750261)R2=0.145438 D.W.=0.129252计算各解释变量的
19、相关系数,选择lnX1、lnX2 、X3的数据,得到相关系数矩阵如表6:表6 相关系数表LOG(X1)LOG(X2)X3LOG(X1)1.0000000000000000.986621320268606-0.344133098922941LOG(X2)0.9866213202686061.000000000000000-0.401039177546979X3-0.344133098922941-0.4010391775469791.000000000000000可见财政支出对税收收入的影响最大,与经验相符合,因此选(2)得出的回归形式为初始的回归模型。逐步回归将lnY与lnX1、lnX2做回归
20、得到下表7: 表7Dependent Variable: LOG(Y)Method: Least SquaresDate: 06/12/11 Time: 20:11Sample: 1990 2009Included observations: 20VariableCoefficientStd. Errort-StatisticProb.C-0.0651200.269604-0.2415390.8120LOG(X1)0.0461240.0749130.6157070.5462LOG(X2)0.9353950.06416614.577760.0000R-squared0.998044Mean de
21、pendent var9.405915Adjusted R-squared0.997814S.D. dependent var0.972567S.E. of regression0.045468Akaike info criterion-3.206140Sum squared resid0.035145Schwarz criterion-3.056780Log likelihood35.06140F-statistic4338.136Durbin-Watson stat0.807678Prob(F-statistic)0.000000将lnY与lnX1、X3做回归得到下表8: 表8Depend
22、ent Variable: LOG(Y)Method: Least SquaresDate: 06/12/11 Time: 22:57Sample: 1990 2009Included observations: 20VariableCoefficientStd. Errort-StatisticProb.C-0.3594330.222734-1.6137320.1250LOG(X2)0.9833580.01018896.525010.0000X30.0035000.0015922.1990730.0420R-squared0.998444Mean dependent var9.405915A
23、djusted R-squared0.998260S.D. dependent var0.972567S.E.of regression0.040563Akaike info criterion-3.434428Sum squared resid0.027971Schwarz criterion-3.285068Log likelihood37.34428F-statistic5452.820Durbin-Watson stat0.981206Prob(F-statistic)0.000000将其他解释变量分别倒入上述初始回归模型,寻找最佳回归方程表9Clnx2Lnx1x3D.W.Y=f(x2
24、)0.0889020.974370.9980010.835853t值0.89987694.79376Y=f(x2,x1)-0.0651200.9353950.0461240.9978900.807678t值-0.24153914.577760.615707Y=f(x2,x3)-0.3594330.9833580.0035000.9984440.981206t值-1.6137296.525012.199073讨论:第一步,在初始模型中引入X1,模型修正的拟合优度反而略有下降,同时X1的参数未能通过t检验,参数符号与经济意义相符。第二步,去掉X1,引入X3,拟合优度提高,且参数符号与经济意义相符,
25、变量也通过了t检验,D.W.检验也表明不存在1阶序列相关性。因此最终的税收收入函数应以Y=f(x2,x3)为最优,拟合结果如下: (-1.613732) (96.52501) (2.199073)R2=0.998444(2)异方差检验怀特检验 利用怀特检验法检验模型是否存在异方差。 残差相关图 表10表11White Heteroskedasticity Test:F-statistic0.664468Probability0.626329Obs*R-squared3.010411Probability0.556085Test Equation:Dependent Variable: RESI
26、D2Method: Least SquaresDate: 06/12/11 Time: 23:33Sample: 1990 2009Included observations: 20VariableCoefficientStd. Errort-StatisticProb.C0.0991970.1469140.6752070.5098LOG(X2)-0.0066090.010283-0.6426730.5301(LOG(X2)20.0003770.0005340.7055880.4913X3-0.0012920.002206-0.5856880.5668X325.98217941.0151289
27、0.5893020.5644R-squared0.150521Mean dependent var0.001399Adjusted R-squared-0.076007S.D. dependent var0.001794S.E. of regression0.001861Akaike info criterion-9.523470Sum squared resid5.1929968Schwarz criterion-9.274537Log likelihood100.2347F-statistic0.664468Durbin-Watson stat2.373028Prob(F-statisti
28、c)0.626329表12White Heteroskedasticity Test:F-statistic0.637251Probability0.675110Obs*R-squared3.707910Probability0.592187Test Equation:Dependent Variable: RESID2Method: Least SquaresDate: 06/12/11 Time: 23:54Sample: 1990 2009Included observations: 20VariableCoefficientStd. Errort-StatisticProb.C0.29
29、28470.2911051.0059850.3315LOG(X2)-0.0212070.021545-0.9843000.3417(LOG(X2)20.0004400.0005470.8036740.4350(LOG(X2)*X30.0001300.0001680.7741900.4517X3-0.0037140.003845-0.9658390.3505X321.1778081.2724890.9255940.3703R-squared0.185396Mean dependent var0.001399Adjusted R-squared-0.105535S.D. dependent var
30、0.001794S.E. of regression0.001886Akaike info criterion-9.465391Sum squared resid4.979800Schwarz criterion-9.166671Log likelihood100.6539F-statistic0.637251Durbin-Watson stat2.603946Prob(F-statistic)0.675110记为对原始模型进行普通最小二乘回归的道德残差平方项,将其与X2 、X3及其平方项与交叉项做辅助回归,得 (1.005985) (-0.984300) (0.803674) (0.7741
31、90) (-0.965839) (0.925594)R2=0.185396怀特统计量nR2=20*0.185396=3.70792,该值小于5%显著性水平下,自由度为5的分布的相应临界值=11.07,因此,不拒绝同方差的原假设。去掉交叉项后的辅助回归结果为 (0.675207) (-0.642673) (0.705588) (-0.585688) (0.589302)R2=0.150521怀特统计量nR2=20*0.150521=3.01042,该值小于5%显著性水平下,自由度为5的分布的相应临界值=11.07,因此,不拒绝同方差的原假设。拉格朗日乘数检验表13Breusch-Godfrey
32、Serial Correlation LM Test:F-statistic4.184117Probability0.057610Obs*R-squared4.145950Probability0.041734Test Equation:Dependent Variable: RESIDMethod: Least SquaresDate: 06/13/11 Time: 01:00Presample missing value lagged residuals set to zero.VariableCoefficientStd. Errort-StatisticProb.C0.0597630.
33、2064890.2894260.7760LOG(X2)-0.0032020.009480-0.3377790.7399X3-0.0002930.001468-0.1993470.8445RESID(-1)0.4829150.2360852.0455110.0576R-squared0.207297Mean dependent var5.83E-16Adjusted R-squared0.058666S.D. dependent var0.038369S.E. of regression0.037226Akaike info criterion-3.566736Sum squared resid
34、0.022173Schwarz criterion-3.367589Log likelihood39.66736F-statistic1.394706Durbin-Watson stat1.432424Prob(F-statistic)0.280658含1阶滞后残差项的辅助回归为 (0.289426) (-0.337779) (-0.199347) (2.045511)R2=0.207297于是,LM=19*0.207297=3.938643,该值大于显著性水平自由度为1的分布的相应临界值=3.84,由此判断原模型存在1阶序列相关性。表14Breusch-Godfrey Serial Corr
35、elation LM Test:F-statistic3.468355Probability0.057796Obs*R-squared6.324294Probability0.042335Test Equation:Dependent Variable: RESIDMethod: Least SquaresDate: 06/13/11 Time: 01:14Presample missing value lagged residuals set to zero.VariableCoefficientStd. Errort-StatisticProb.C0.0413580.1984270.208
36、4320.8377LOG(X2)-0.0026650.009100-0.2928150.7737X3-0.0001660.001410-0.1174620.9081RESID(-1)0.6958210.2650572.6251780.0191RESID(-2)-0.4135020.267512-1.5457310.1430R-squared0.316215Mean dependent var5.83E-16Adjusted R-squared0.133872S.D. dependent var0.038369S.E. of regression0.035708Akaike info crite
37、rion-3.614540Sum squared resid0.019126Schwarz criterion-3.365606Log likelihood41.14540F-statistic1.734177Durbin-Watson stat1.943582Prob(F-statistic)0.194866含2阶滞后残差项的辅助回归为 (0.208432) (-0.292815) (-0.117462) (2.625178) (-1.545731)R2=0.316215于是,LM=18*0.316215=5.69187,该值小于显著性水平自由度为2的分布的相应临界值=5.99,仍说明原模型不存在序列相关性,表明并不存在2阶序列想关性结合1阶滞后残差项的辅助回归情况,可判断存在1阶序列相关性。表15Breusch-Godfrey Serial Correlation LM Test:F-statistic2.347569Probability0.116783Obs*R-squared6.693728Probability0.082328Test Equation:Dependent Variable: RESIDMethod: Least SquaresDate: 06/13/11
限制150内