优秀毕业论文----我国电力消费因素分析.doc

。【实验作者】学号： 1032215227 pole【实验名称】 我国电力消费因素分析【实验目的】 通过查阅全国统计年鉴，分析总体的结构、变量的集中趋势、离散趋势及分布形态，对我国电力消费水平和其影响因素作出总结分析。【实验内容】 进入21世纪以来，中国经济一直保持了良好的发展势头，电力工业是国民经济和社会经济发展的基础产业，对经济发展影响程度日益突出。近年来，伴随着国民经济的持续快速发展我国能源消费快速增长，电力供应日趋紧张。由于电力短缺，有些地区陆续出现了拉闸限电的情况，政府也出台了推动电力供应的措施，但这并不能从根本上解决电力短缺问题。本文主要研究的是我国电力消费的因素分析，预测未来用电需求提供依据从而对我国电力中长期发展战略提出一定的意见。数据主要来源于2010年中国统计年鉴，拟通过GDP，人口总量，工业总产值，电力出厂价价格指数等数据，建立全国电力消费总量的计量经济模型。一、 影响因素的选取根据本文的研究课题，在模型中以我们年电力消费总量作为因变量，选择解释变量如下：1全国GDP总额：一国经济增长水平由这个指标明确的反映出来。电力工业是国民经济和社会经济发展的基础产业，对电力的依赖性越来越高，所以我们将GDP作为电力消费量的一个影响因素。2全国人口总量：在电力消费上，我们不仅要关注工业用电，也要关注城乡居民的用电用电需求，每个人的正常工作和生活都需要不断的消费电力。3工业总产值：总体上来说，工业部门多数属于高密度用电部门。随着工业生产现代化的发展，工业产品的生产效率大大提高，与此同时，对于生产过程中电力需求也大大提高，我们需要更多的考虑工业总产值变动的因素。4电力出厂价格指数：价格总能影响需求，电价波动会导致电力消费量的变动，所以我们引入价格因素，又因为各地区电价存在差异，直接引用较难，所以我们用电力出厂价指数来估计电价。二、 模型设定、根据我们以上分析，回归模型这里如下：（i=1,2.15）分别对应自1992年起至2006年的全国电力消费总量（单位：亿千瓦时）（i=1,2.15）分别对应自1992年起至2006年的全国GDP(单位：亿元)（i=1,2.15）分别对应自1992年起至2006年的全国人口总量（单位：万人）（i=1,2.15）分别对应自1992年起至2006年的全国工业总产值（单位：亿元）（i=1,2.15）分别对应自1992年起至2006年的全国电力出厂价指数（i=1,2.15）随机扰动项序列三、 样本数据的收集 年份电力消费总量（亿千瓦时）GDP(亿元)人口总量工业总产值（亿元）电力出厂价指数1992年7859.20 26923.50 117171.00 27724.21 108.80 1993年8426.50 35333.90 118517.00 39693.00 147.86 1994年9260.40 48197.90 119850.00 51353.03 206.26 1995年10023.40 60793.70 121121.00 54946.86 225.86 1996年10764.30 71176.60 122389.00 62740.16 255.45 1997年11284.40 78973.00 123626.00 68352.68 291.21 1998年11598.40 84402.30 124761.00 67737.14 307.23 1999年12305.20 89677.10 125786.00 72707.04 309.99 2000年13471.40 99214.60 126743.00 85673.66 317.43 2001年14633.50 109655.20 127627.00 95448.98 324.73 2002年16331.50 120332.70 128453.00 110776.48 327.33 2003年19031.60 135822.80 129227.00 142271.22 330.27 2004年21971.40 159878.30 129988.00 201722.19 338.20 2005年24940.40 184937.40 130756.00 251619.50 352.41 2006年28588.00 216314.40 131448.00 316588.96 362.27 数据来源：中华人民共和国统计年鉴/ （中华人民共和国国家统计局）四、 数据平稳性检验由于所用数据为时间序列数据，需要检验其平稳性，分别对电力消费总量（Y），GDP、人口总量工业总产值，电力出厂价指数进行单位根检验，Null Hypothesis: D(Y) has a unit rootExogenous: Constant, Linear TrendLag Length: 3 (Automatic based on SIC, MAXLAG=3)t-StatisticProb.*Augmented Dickey-Fuller test statistic-5.3769210.0091Test critical values:1% level-5.2953845% level-4.00815710% level-3.460791*MacKinnon (1996) one-sided p-values.Warning: Probabilities and critical values calculated for 20 observationsand may not be accurate for a sample size of 10Augmented Dickey-Fuller Test EquationDependent Variable: D(Y,2)Method: Least SquaresDate: 11/18/11 Time: 17:00Sample (adjusted): 1997 2006Included observations: 10 after adjustmentsVariableCoefficientStd. Errort-StatisticProb.D(Y(-1)-0.6928200.128851-5.3769210.0058D(Y(-1),2)0.5921300.1797463.2942640.0301D(Y(-2),2)-0.4691190.155523-3.0163870.0393D(Y(-3),2)0.7347890.2137403.4377740.0263C-1059.811208.9839-5.0712570.0071TREND(1992)231.096132.496727.1113670.0021R-squared0.967452Mean dependent var290.6700Adjusted R-squared0.926766S.D. dependent var397.4776S.E. of regression107.5643Akaike info criterion12.47776Sum squared resid46280.33Schwarz criterion12.65932Log likelihood-56.38882Hannan-Quinn criter.12.27860F-statistic23.77885Durbin-Watson stat1.880893Prob(F-statistic)0.004485 表（1）对Y 进行一阶差分的ADF检验，从表（1）看，检验t统计量是-5.38，比显著性水平为1%的临界值都小，所以拒绝原假设，序列不存在单位根，是平稳的。Null Hypothesis: D(X1,2) has a unit rootExogenous: Constant, Linear TrendLag Length: 1 (Automatic based on SIC, MAXLAG=2)t-StatisticProb.*Augmented Dickey-Fuller test statistic-3.6813310.0706Test critical values:1% level-5.1248755% level-3.93336410% level-3.420030*MacKinnon (1996) one-sided p-values.Warning: Probabilities and critical values calculated for 20 observationsand may not be accurate for a sample size of 11Augmented Dickey-Fuller Test EquationDependent Variable: D(X1,3)Method: Least SquaresDate: 11/18/11 Time: 17:17Sample (adjusted): 1996 2006Included observations: 11 after adjustmentsVariableCoefficientStd. Errort-StatisticProb.D(X1(-1),2)-1.4929930.405558-3.6813310.0078D(X1(-1),3)0.2814780.2661951.0574130.3254C-8262.3613035.276-2.7221120.0297TREND(1992)1178.287363.14723.2446530.0142R-squared0.698108Mean dependent var598.7364Adjusted R-squared0.568726S.D. dependent var3957.365S.E. of regression2598.859Akaike info criterion18.83882Sum squared resid47278462Schwarz criterion18.98351Log likelihood-99.61351Hannan-Quinn criter.18.74761F-statistic5.395708Durbin-Watson stat2.389043Prob(F-statistic)0.030770 表（2）对X1进行二阶差分ADF检验，从表（2）看，检验t统计量是-3.68，比显著性水平为10%的临界值都小，所以拒绝原假设，序列不存在单位根，是平稳的。Null Hypothesis: D(LNX2) has a unit rootExogenous: NoneLag Length: 2 (Automatic based on SIC, MAXLAG=3)t-StatisticProb.*Augmented Dickey-Fuller test statistic-1.8616590.0622Test critical values:1% level-2.7921545% level-1.97773810% level-1.602074*MacKinnon (1996) one-sided p-values.Warning: Probabilities and critical values calculated for 20 observationsand may not be accurate for a sample size of 11Augmented Dickey-Fuller Test EquationDependent Variable: D(LNX2,2)Method: Least SquaresDate: 11/25/11 Time: 16:44Sample (adjusted): 1996 2006Included observations: 11 after adjustmentsVariableCoefficientStd. Errort-StatisticProb.D(LNX2(-1)-0.0471070.025304-1.8616590.0997D(LNX2(-1),2)0.3651140.3704990.9854660.3533D(LNX2(-2),2)-0.1505890.353985-0.4254120.6817R-squared0.184705Mean dependent var-0.000479Adjusted R-squared-0.019118S.D. dependent var0.000313S.E. of regression0.000316Akaike info criterion-13.05755Sum squared resid7.97E-07Schwarz criterion-12.94903Log likelihood74.81653Hannan-Quinn criter.-13.12596Durbin-Watson stat1.282587 表（3）对X2进行二阶差分ADF检验，从表（3）看，检验t统计量是-1.86，比显著性水平为10%的临界值小，所以拒绝原假设，序列不存在单位根，是平稳的。Null Hypothesis: D(X3,2) has a unit rootExogenous: Constant, Linear TrendLag Length: 2 (Automatic based on SIC, MAXLAG=2)t-StatisticProb.*Augmented Dickey-Fuller test statistic-3.7668560.0681Test critical values:1% level-5.2953845% level-4.00815710% level-3.460791*MacKinnon (1996) one-sided p-values.Warning: Probabilities and critical values calculated for 20 observationsand may not be accurate for a sample size of 10Augmented Dickey-Fuller Test EquationDependent Variable: D(X3,3)Method: Least SquaresDate: 11/18/11 Time: 21:07Sample (adjusted): 1997 2006Included observations: 10 after adjustmentsVariableCoefficientStd. Errort-StatisticProb.D(X3(-1),2)-3.8534851.022998-3.7668560.0131D(X3(-1),3)1.9669990.7613922.5834230.0492D(X3(-2),3)0.7431400.5730451.2968260.2513C-51359.5216197.08-3.1709120.0248TREND(1992)7208.6172015.6543.5763170.0159R-squared0.885336Mean dependent var1087.268Adjusted R-squared0.793604S.D. dependent var17183.14S.E. of regression7806.434Akaike info criterion21.07014Sum squared resid3.05E+08Schwarz criterion21.22143Log likelihood-100.3507Hannan-Quinn criter.20.90417F-statistic9.651394Durbin-Watson stat2.112956Prob(F-statistic)0.014306 表（4）对X3进行二阶差分ADF检验，从表（4）看，检验t统计量是-3.77.，比显著性水平为10%的临界值都小，所以拒绝原假设，序列不存在单位根，是平稳的。Null Hypothesis: D(X4,2) has a unit rootExogenous: Constant, Linear TrendLag Length: 1 (Automatic based on SIC, MAXLAG=2)t-StatisticProb.*Augmented Dickey-Fuller test statistic-5.2856580.0081Test critical values:1% level-5.1248755% level-3.93336410% level-3.420030*MacKinnon (1996) one-sided p-values.Warning: Probabilities and critical values calculated for 20 observationsand may not be accurate for a sample size of 11Augmented Dickey-Fuller Test EquationDependent Variable: D(X4,3)Method: Least SquaresDate: 11/18/11 Time: 21:11Sample (adjusted): 1996 2006Included observations: 11 after adjustmentsVariableCoefficientStd. Errort-StatisticProb.D(X4(-1),2)-1.9815450.374891-5.2856580.0011D(X4(-1),3)0.4540200.1877142.4186770.0462C-17.913959.333267-1.9193650.0964TREND(1992)1.5138230.9117961.6602640.1408R-squared0.883625Mean dependent var3.131818Adjusted R-squared0.833750S.D. dependent var18.78307S.E. of regression7.658571Akaike info criterion7.184816Sum squared resid410.5760Schwarz criterion7.329505Log likelihood-35.51649Hannan-Quinn criter.7.093609F-statistic17.71677Durbin-Watson stat1.030766Prob(F-statistic)0.001194 表（5）对X4进行二阶差分ADF检验，从表（5）看，检验t统计量是-5.26，比显著性水平为5%的临界值都小，所以拒绝原假设，序列不存在单位根，是平稳的。 结论，从检验结果可知Y,X1，X2，X3，X4都为平稳序列，可直接使用，建立模型。五、模型的估计 运用最小二乘法（OLS）对模型进行回归分析，得到结果如下Dependent Variable: YMethod: Least SquaresDate: 11/18/11 Time: 21:52Sample: 1992 2006Included observations: 15VariableCoefficientStd. Errort-StatisticProb.C-48288.7423646.18-2.0421370.0684X10.0579140.0416551.3903180.1946X20.4806250.2064602.3279360.0422X30.0304280.0164421.8505860.0940X4-22.791234.089523-5.5730760.0002R-squared0.998785Mean dependent var14699.31Adjusted R-squared0.998298S.D. dependent var6288.756S.E. of regression259.4133Akaike info criterion14.21592Sum squared resid672952.4Schwarz criterion14.45194Log likelihood-101.6194Hannan-Quinn criter.14.21341F-statistic2054.400Durbin-Watson stat1.414818Prob(F-statistic)0.000000我们可以看到R-squared和Adjusted R-squared都接近于1，表明模型拟合的效果非常好，解释变量系数的t检验除X1以外都显著，Y = C(1) + C(2)*X1 + C(3)*X2 + C(4)*X3 + C(5)*X4Y = -48288.7438148 + 0.0579136581183*X1 + 0.480624618597*X2 + 0.0304278955754*X3 - 22.7912259877*X4若改变模型设定形式，采用对数形式后进行回归结果如下Dependent Variable: LNYMethod: Least SquaresDate: 11/20/11 Time: 14:16Sample: 1992 2006Included observations: 15VariableCoefficientStd. Errort-StatisticProb.C-35.720166.931618-5.1532210.0004LNX10.3724900.1056743.5249070.0055LNX23.4421880.6326535.4408800.0003LNX30.2972710.0432516.8731410.0000LNX4-0.4979090.059549-8.3613940.0000R-squared0.999357Mean dependent var9.518860Adjusted R-squared0.999099S.D. dependent var0.396157S.E. of regression0.011890Akaike info criterion-5.764960Sum squared resid0.001414Schwarz criterion-5.528943Log likelihood48.23720Hannan-Quinn criter.-5.767474F-statistic3882.636Durbin-Watson stat2.532065Prob(F-statistic)0.000000我们发现，采用对数形式后模型的R2进一步提高， lnx1和lnx2的检验变量t都很小，lnx3和lnx4的检验变量t为0，所以修改模型为对数形式。 LNY = C(1) + C(2)*LNX1 + C(3)*LNX2 + C(4)*LNX3 + C(5)*LNX4 LNY = -35.720161358 + 0.372489956351*LNX1 + 3.44218763279*LNX2 + 0.29727132501*LNX3 - 0.497908956484*LNX4 六 回归结果的检验 （一）统计检验 1、从回归的结果看，方程样本的判定系数R2在调整前和调整后都非常的高，表明方程拟合效果非常好。2、系统显著性检验，调整前x1和x3的p值大于0.05，没有通过t检验，但是在调整后，所以的系数都通过了t检验。（二）计量经济学检验1、异方差检验WHITE检验结果如下：Heteroskedasticity Test: WhiteF-statistic1.205946Prob. F(10,4)0.4639Obs*R-squared11.26389Prob. Chi-Square(10)0.3373Scaled explained SS3.146636Prob. Chi-Square(10)0.9778Test Equation:Dependent Variable: RESID2Method: Least SquaresDate: 11/20/11 Time: 15:02Sample: 1992 2006Included observations: 15Collinear test regressors dropped from specificationVariableCoefficientStd. Errort-StatisticProb.C-0.0174910.223523-0.0782530.9414LNX1-0.0187290.047364-0.3954190.7127LNX12-0.0237390.029903-0.7938650.4717LNX1*LNX20.0026530.0099940.2654770.8038LNX1*LNX30.0388530.0341801.1367130.3191LNX1*LNX40.0150400.0426220.3528620.7420LNX2*LNX3-0.0019060.006202-0.3072840.7740LNX2*LNX40.0025660.0063250.4057570.7057LNX32-0.0139560.010663-1.3088760.2607LNX3*LNX4-0.0181830.015037-1.2092240.2932LNX420.0004110.0245610.0167140.9875R-squared0.750926Mean dependent var9.43E-05Adjusted R-squared0.128240S.D. dependent var0.000109S.E. of regression0.000102Akaike info criterion-15.39566Sum squared resid4.17E-08Schwarz criterion-14.87642Log likelihood126.4674Hannan-Quinn criter.-15.40119F-statistic1.205946Durbin-Watson stat3.598213Prob(F-statistic)0.463866从表中我们可以看出，检验的伴随概率是0.3373，Obs*R-squared为11.26389，由white检验知，在=0.05下，查X2分布表，的临界值为（x2 10）18.30711.26389，表明模型不存在异方差， ARCH检验结果如下：Heteroskedasticity Test: ARCHF-statistic0.055602Prob. F(3,7)0.9814Obs*R-squared0.256022Prob. Chi-Square(3)0.9681Test Equation:Dependent Variable: RESID2Method: Least SquaresDate: 11/25/11 Time: 19:52Sample (adjusted): 1996 2006Included observations: 11 after adjustmentsVariableCoefficientStd. Errort-StatisticProb.C6.45E-055.91E-051.0918260.3111RESID2(-1)-0.1043340.374956-0.2782560.7889RESID2(-2)0.1011380.3716450.2721350.7934RESID2(-3)0.0312320.3694140.0845450.9350R-squared0.023275Mean dependent var6.60E-05Adjusted R-squared-0.395322S.D. dependent var0.000121S.E. of regression0.000143Akaike info criterion-14.59501Sum squared resid1.43E-07Schwarz criterion-14.45033Log likelihood84.27258Hannan-Quinn criter.-14.68622F-statistic0.055602Durbin-Watson stat2.003818Prob(F-statistic)0.981375从上表可以看出，Obs*R-squared=0.256022，在=0.05下，查X2分布表，的临界值=7.8150.256022,表明模型不存在异方差。由于WHITE检验和ARCH检验同时表明没有异方差，所以我们认定模型不存在异方差。也可以从残差的散点图看出，是同方差图形。2自相关检验 先利用残差数据给出残差如下所示的相关图，我们初步判断序列存在负相关。 给定显著性水平0.05，差DW表，当n=15，k=4，dl为0.82，du为1.75而实验的DW值为2.532大于du，4-duDW4所以序列存在负相关Breusch-Godfrey Serial Correlation LM Test:F-statistic0.903745Prob. F(1,9)0.3666Obs*R-squared1.368793Prob. Chi-Square(1)0.2420Test Equation:Dependent Variable: RESIDMethod: Least SquaresDate: 11/25/11 Time: 18:45Sample: 1992 2006Included observations: 15Presample missing value lagged residuals set to zero.VariableCoefficientStd. Errort-StatisticProb.C0.3984046.9778170.0570960.9557LNX10.0142750.1072420.1331080.8970LNX2-0.0371910.636922-0.0583910.9547LNX3-0.0059400.043908-0.1352840.8954LNX4-0.0101650.060785-0.1672300.8709RESID(-1)-0.3085920.324610-0.9506550.3666R-squared0.091253Mean dependent var-2.22E-14Adjusted R-squared-0.413607S.D. dependent var0.010049S.E. of regression0.011948Akaike info criterion-5.727315Sum squared resid0.001285Schwarz criterion-5.444095Log likelihood48.95486Hannan-Quinn criter.-5.730332F-statistic0.180749Durbin-Watson stat2.164899Prob(F-statistic)0.962923通过LM的检验，表明模型存在着一阶序列相关，既自相关。Dependent Variable: D(LNY)Method: Least SquaresDate: 11/25/11 Time: 18:54Sample (adjusted): 1993 2006Included observations: 14 after adjustmentsVariableCoefficientStd. Errort-StatisticProb.C0.1287640.0389193.3084960.0091D(LNX1)0.3342950.1010573.3079720.0091D(LNX2)-11.285024.607982-2.4490150.0368D(LNX3)0.1122990.0665031.6886460.1256D(LNX4)-0.1529140.107488-1.4226080.1886R-squared0.916183Mean dependent var0.092236Adjusted R-squared0.878931S.D. dependent var0.037549S.E. of regression0.013065Akaike info criterion-5.565293Sum squared resid0.001536Schwarz criterion-5.337059Log likelihood43.95705Hannan-Quinn criter.-5.586421F-statistic24.59421Durbin-Watson stat2.082526Prob(F-statistic)0.000073通过一阶差分，建立出差分模型，得到DW为2.08，duDW4-du，没有自相关。Y=?我们也可以运用Durbin两步法来估计自相关系数，【实验小结】模型系数的经济意义分析（一） 系数经济意义分析 1 全国人口总数。从回归系数3.442188可以