1、198610275.182204.912090.73198712058.622262.18107.32140.36198815042.822491.21118.52390.47198916992.322823.78117.82727.4199018667.823083.59102.12821.86199121781.53386.62102.92990.17199226923.483742.2105.43296.91199335333.924642.3113.24255.3199448197.865792.62121.75126.88199560793.736823.72114.86038.04
2、199671176.597937.55106.16909.82199778973.039233.56100.88234.04199884402.2810798.1897.49262.8199989677.0513187.679710682.58200099214.5515886.598.512581.512001109655.218902.5899.215301.382002120332.722053.1598.717636.452003135822.824649.9599.905920017.312004159878.328486.89102.8062257182005183867.9339
3、30.28100.777430866200621087140422.73101.028237636表1. 1980-2006年我国税收收入相关因素统计表(三)模型的建立与构造在EVIEWS软件中输入数据,观察Y与三个解释变量X1、X2、X3之间的散点图,如图1、图2、图3所示:图1图2图3由以上散点图发现存在较强的线性关系,故此选择建立线性模型。建立模型:、利用EVIEWS软件对数据进行普通最小二乘回归,得到如图4结果:Dependent Variable: YMethod: Least SquaresDate: 12/16/12 Time: 12:50Sample: 1980 2006Inc
4、luded observations: 27VariableCoefficientStd. Errort-StatisticProb.C-6357.3062589.143-2.4553710.0221X1-0.0111910.014037-0.7972610.4335X20.9670820.07682112.588750.0000X357.1184124.003452.3795920.0260R-squared0.994954Mean dependent var8681.087Adjusted R-squared0.994296S.D. dependent var9909.343S.E. of
5、 regression748.4057Akaike info criterion16.20972Sum squared resid12882553Schwarz criterion16.40170Log likelihood-214.8312F-statistic1511.718Durbin-Watson stat0.691548Prob(F-statistic)0.000000图4Y = -6357.306 - 0.011191*X1 + 0.967082*X2 + 57.11841*X3 (2589.143) (0.014037) (0.076821) (24.00345)t =(-2.4
6、55371) (-0.797261) (12.58875) (2.379592)=0.994954 =0.994296 F=1511.718(四)模型检验1.经济意义检验我国税收收入与财政支出及商品零售物价指数呈正相关关系,当国内其他因素不变时,财政支出每增加1单位,我国税收收入增加0.967082单位;当其他因素不变时,商品零售物价指数每增加1单位,我国税收收入增加57.11841单位,两者与税收收入呈正相关符合现实经济意义,但模型中国内生产总值与税收收入呈负相关,不符合现实经济意义。2.统计检验由=0.994954 ,=0.994296与1十分接近,说明模型拟合优度很好。F统计量等于151
7、1.718大于5%显著性水平下F(3,23)的临界值3.03,表明模型整体的显著性较高。除X1外,X2与X3的t检验值均大于5%显著性水平下自由度为23的临界值1.711,通过了变量的显著性检验。故还须对模型进行计量经济学检验并作出修正。3.计量检验(1)多重线性检验对各解释变量进行多重共线性检验利用EVIEWS软件得到各变量间相关系数矩阵表:10.984833-0.407265-0.416781表2. X1、X2、X3相关系数矩阵表从系数矩阵表中看出,X1与X2之间的相关系数较高,可能存在多重共线性。修正多重共线性.用EVIEWS分别对Y与各解释变量X1、X2、X3做最小二乘回归: 14:1
8、1-1143.176559.4057-2.0435540.05170.1610650.00658424.463690.9599020.9582982023.59218.134321.02E+0818.23031-242.8134598.47240.170737图5Y = -1143.176 + 0.161065 * X1 (559.4057) (0.006584)=0.959902 DW=0.17073713-292.7317212.2144-1.3794150.18000.8925750.01434062.244310.9935890.993332809.161416.30106163685
9、5616.39705-218.06433874.3550.501126图6Y = -292.7317 + 0.892575 * X2 (212.2144) (0.014340)=0.993589 DW=0.5011261468011.8528622.302.3761840.0255-564.9916272.0256-2.0769790.04820.1471610.1130479332.43921.191572.18E+0921.28756-284.08624.3138430.1796870.048232图7Y = 68011.85 + 564.9916 * X3 (28622.30) (272
10、.0256)=0.147161 DW=0.179687以上3个方程根据经济理论和统计检验得出,财政支出X2是最重要的解释变量(t检验值=62.24431也最大),从而得出最优简单回归方程Y=f(X2)。.对模型进行逐步回归,在初始模型的基础上加入解释变量X1与X3,得到如下回归结果:加入X1,32-218.4640240.3033-0.9091180.3723-0.0105150.015337-0.6855710.49960.9489780.08353911.359650.9937120.993188817.877316.355741605415716.49972-217.80251896.3
11、450.526704图8Y = -218.4640 + -0.010515 *X1 + 0.948978 * X2(240.3033) (0.015337) (0.083539)=0.993712加入X3,37-6394.6562568.992-2.4891690.02010.9069500.01448062.6362756.7307423.815652.3820780.9948150.994383742.702716.162911323857416.30689-215.19932302.2120.652300图9Y = -6394.656 + 0.906950 * X2 + 56.73074
12、 * X3 (2568.992) (0.014480) (23.81565)=0.994815由以上数据构成表格如下:(X1)(X2)(X3)Y=f(X2)(212.2144)(0.014340)Y=f(X1,X2)(240.3033)(0.015337)(0.083539)Y=f(X3,X2)(2568.992)(0.014480)(23.81565)Y=f(X1,X2,X3)(2589.143)(0.014037)(0.076821)(24.00345)表3. 税收收入模型估计结果分析:在最优简单回归方程Y=f(X2)中引入X1,值略有提高。虽然X2与X1高度相关,在X1的引入对参数影响不
13、大,的符号不满意,可以是“多余变量”,暂时删除;模型中引入X3,使值由0.993589提升到0.994815,正号也合理,进行t检验,不显著。从经济理论分析,X3应该是重要变量,虽然X2与X3高度相关,但不影响的显著性和稳定性,因此,可能是“有利变量”,暂时保留;最后在Y=f(X3,X2)的基础上引入X1,=0.994954几乎没有增加,其他两个参数系数没有多大影响,可以确定X1是多余变量,应从模型中删除。得出最后回归模型是:Y = -6394.656 + 0.906950 * X2 + 56.73074 * X3由于剔除了变量X1,故模型已不存在多重共线性,且各解释变量前得系数均符合经济意义
14、,模型拟合度上升,各变量t检验值上升。在其他因素保持不变的情况下,财政支出每增加1亿元,商品零售物价指数增加1%,税收收入增加57.6377亿元。(2)邹氏检验考虑到1980-2006年时间跨度较大,政府财政支出及商品零售物价指数均发生了较大的变化,有必要对模型进行参数的稳定性检验。将数据分为1980-1992年和1993-2006年两组分别进行普通最小二乘回归结果如下:1980-1992年: 15:47 1980 1992 13-3271.7351116.480-2.9304020.01501.0799520.07083115.2469525.7728610.765052.3941240.03770.9650391855.6340.958047999.6892204.761613.68074419273.013.81112-85.92483138.01591.601545图10记此时的残差平方和为RSS1=4192731993-2006年: 16:10 1993 2006 14-10058.024408.677-2.2814140.04340.9409590.02693934.9291984.4832740.020972.1109750.05850.99285815019.010.99156010277.24944.187516.725949806391.Schwa