主模型的構(gòu)建與回歸分析

1、計(jì)量經(jīng)濟(jì)學(xué)結(jié)課論文趙嘉祺林經(jīng)12-15主模型的構(gòu)建與回歸分析5.1 模型的設(shè)定影響我國(guó)電子商務(wù)發(fā)展的因素是非常復(fù)雜的，我們力求建立經(jīng)濟(jì)模型把所要研究的主要因素之間的關(guān)系，用適當(dāng)?shù)臄?shù)學(xué)關(guān)系式近似地、簡(jiǎn)化地表達(dá)出來(lái)。為此，我們基于收集到的數(shù)據(jù)，利用Eviews軟件，觀測(cè)yx1x2x3x4Inylnx1lnx2lnx3lnx4等數(shù)據(jù)及數(shù)據(jù)處理結(jié)果的趨勢(shì)線，最終得到的回歸模型為：Lny=c+a1lnX1+a2lnX2+a3X3+a4X4+a5X5+u其中y為被解釋變量，表示人均網(wǎng)絡(luò)購(gòu)物在線交易規(guī)模交易（元/人），其余為解釋變量x1表示PC互聯(lián)網(wǎng)用戶(hù)（億人）x2表示手機(jī)上網(wǎng)人數(shù)x3社會(huì)物流總費(fèi)用（億元）

2、x4表示城鎮(zhèn)居民可支配收入（元）x5表示B2C、C2c企業(yè)數(shù)量（家）。5.2 模型的估計(jì)利用Eviews軟件生成所需對(duì)數(shù)數(shù)據(jù)，并進(jìn)行普通最小二乘參數(shù)估計(jì):表格0-1第一次回歸結(jié)果VariableCoefficientStd.Errort-StatisticProb.C10,7211817664336.05939200000LNX10.3810210302207126079402255LNX209534510256005374777600018X3-5.28E-051.88E-Q5-2B0677600127X4-0.0003730000215-173222101025X5-5.19E-059.7

3、0E-05-0.53501906000R-squared0964126Meandependentvar5162339AdjustedR-squared0.952916S.D.dependentvar1.241937SEofregression0.269489Akaikeinfocriterion0442424Sumsquaredresid1161991Schwarzcriterion0739381LugJikelihoad1.133332F-statsstfc36.00007Durbin-Watsonstat1210574Prob(F-statistic0000000根據(jù)表中數(shù)據(jù)，模型的估計(jì)結(jié)

4、果為L(zhǎng)nyA=10.721+0.381lnX1+0.959lnX2-(5.28E-5)X3-3.728X4-(5.19E-5)lnX5計(jì)量經(jīng)濟(jì)學(xué)結(jié)課論文趙嘉祺林經(jīng)12-15.3 模型的檢驗(yàn)5.3.1 經(jīng)濟(jì)意義檢驗(yàn)從回歸結(jié)果來(lái)看，回歸系數(shù)表明在其他因素不變時(shí)，每一個(gè)單位的單個(gè)解釋變量變化對(duì)于被解釋變量變化的影響，其中，x1、x2、x3的回歸系數(shù)與預(yù)測(cè)符號(hào)相同x4、x5的回歸系數(shù)與預(yù)測(cè)符號(hào)相反，說(shuō)明模型可能存在多重共線。5.3.2 統(tǒng)計(jì)推斷檢驗(yàn)從回歸結(jié)果來(lái)看，可決系數(shù)R2=0.964126,調(diào)整后的R2=0.952915,說(shuō)明模型整體擬合程度顯著；給定a=0.05,查t分布表，在自由度=n-k=2

5、2-5=17時(shí)的臨界值為2.1098,各解釋變量對(duì)于被解釋變量的影響不顯著。5.3.3 多重共線檢驗(yàn)利用eviews軟件計(jì)算各解釋變量的相關(guān)系數(shù)，可以看出，各解釋變量相互之間的相關(guān)系數(shù)較高，證實(shí)確實(shí)存在多重共線。表格0-2解釋變量相關(guān)系數(shù)矩陣LNX1LNX2X3X4X5LNX11000000096988106388620.9544690893139LNX20.96938110000000.7394260.9848610.925117X30.63886207394261.0000000.7169960820179X40.9544650.9848610.71699610000000.946004X

6、50.8931390.9291170.8201790.9460041000000采用逐步回歸法，首先得到一元回歸估計(jì)結(jié)果表格0-3解釋變量逐步回歸結(jié)果lnx1lnx2x3x4x5參數(shù)估計(jì)值1.2119930.5531178.39E-050.0005230.000364t統(tǒng)計(jì)量13.6193111.675802.6727559.2522685.705318R20.9026700.8720610.2631790.8106140.619415調(diào)整R20.8978030.8656640.2263380.8011450.600386其中加入lnx1的方程調(diào)整R2最大，以lnx1為基礎(chǔ)，順次加入其他變量逐

7、步回歸。計(jì)量經(jīng)濟(jì)學(xué)結(jié)課論文趙嘉祺林經(jīng)12-1表格0-4以解釋變量LNX1為基礎(chǔ)加入變量Lnx1lnx2X3X4X5調(diào)整R2Lnx1lnx20.9540700.1234750.895274(2.579691)(0.719049)Lnx1x31.341383-2.6E-050.908913(12.28309)(-1.854597)Lnx1x41.300794-4.23E-050.892947(4.260394)(-0.304717)Lnx1x51.55854-0.0001410.913111(8.542966)(-2.2126837)分別加入x3、x5之后，調(diào)整R2有所增加，x5對(duì)調(diào)整R2改進(jìn)最大

8、且lnx1-與x5的t檢驗(yàn)顯著，保留x5,繼續(xù)加入其余變量回歸。表格0-5以解釋變量LNX1、X5為基礎(chǔ)加入變量Lnx1lnx2X3X4X5調(diào)整R2Lnx1x5lnx20.8595120.468835-0.0002550.937367(2.993279)(2.891065)(-3.710661)Lnx1x5x31.520987-1.04E-05-0.0001040.909622(7.613775)(-0.516404)(-1.071921)Lnx1x5x41.2269400.000279-0.0002430.920580(4.641516)(1.669431)(-2.758753)分別加入ln

9、x2、x4之后，調(diào)整R2有所增加，lnx2對(duì)調(diào)整R2改進(jìn)最大且lnx1、lnx2主x5的t檢驗(yàn)顯著，保留lnx2,繼續(xù)加入其余變量回歸。表格0-6以解釋變量LNX1、X5、LNX2為基礎(chǔ)加入變量Lnx1lnx2X3X4X5調(diào)整R2Lnx1x5lnx20.5305340.604186-3.52E-05-0.0001640.947374x3(1.732714)(3.729922)(-2.103022)(-2.162728)Lnx1x5lnx20.8613860.504660-4.64E-05-0.0002460.933865x4(2.918044)(2.148408)(-0.216393)(-3.

10、056679)加入變量x3,模型的整體擬合程度提高，但是單個(gè)解釋變量的t統(tǒng)計(jì)值不顯著，可以判斷變量x3與其余變量存在多重共線，加入變量x4后，調(diào)整R2下降，且影響lnx1、lnx2與x5的t檢驗(yàn)顯著。故在模型中剔除解釋變量x3、x4o對(duì)經(jīng)過(guò)多重共線修正后的模型進(jìn)行普通最小二乘，得到的模型為:LnyA=7.103462+0.859512lnX1+0.468835lnX2-0.000255X5計(jì)量經(jīng)濟(jì)學(xué)結(jié)課論文趙嘉祺林經(jīng)12-1表格0-7LNX1、X5、LNX2與LNY回歸結(jié)果VariableCoefficientStdErrorStatisticProbC71034620.566B1S12.53

11、2170.0000LNX108595120.2871472.99327900078LNX20468S350.1621672.89106600097X5-0.0002556.86E-05-3.7106610.0016R-squared0.946314Meandependentvar6162339AdjustedRsquared0.937367S.D.dependentvar1.241937SE.ofregression0.310815Akaikeinfocntenon0663725Sumsquaredresid1738903Schwarzcriterion0062097Loglikelihood

12、-3.300979F-statistic1057620Durbin-Watsonstat0.856103Prob(F-statistic)00000005.3.4 自相關(guān)檢驗(yàn)由多重共線修正后的最小二乘表可知，DW=0.856103在n=22,k'=3時(shí)，查詢(xún)?cè)陲@著水平0.05的條件下，臨界值dl=1.053,du=1.664DW<dl,模型存在一階正自相關(guān)。生成殘差序列e,對(duì)e進(jìn)行滯后一期的自回歸e=ae(-1)+v得a=0.568228表格0-8殘差序列的自回歸VariableCoefficientStd.Errort-StatisbcProb.E(-1)0.568228016

13、69653.0392250.0D65-squared0315538Meandependentvar0006872AdjustedR-squared0.315538S.D.dependentvar0.2930093Eofregression0242413Akaikeinfocriterion0050098Sumsquaredresid1175278Schwar2criterion0099837likelihood0473974DurbinWatsonstat1857874對(duì)模型進(jìn)行廣義差分法Lslny-0.568228*lny(-1)clnx1-0.568228*lnx1(-1)lnx2-0.5

14、68228*lnx2(-1)x5-0.568228*x5(-1)計(jì)量經(jīng)濟(jì)學(xué)結(jié)課論文趙嘉祺林經(jīng)12-1表格0-9廣義差分結(jié)果VariableCoefficientStd.Errort-StatisticProb.C2.80102902999689.33806100000LNXIO568228X1(-111.2972440.5457852.3768410.0295LNX2-O.568228*LNX2C1)02640340.211BB41.2461290.2296X5-0.568228-0.0001154.73E-05-2.44190200258R*squared0811781Meandepende

15、ntvar2356173AdjustedRsquared0778566SDdependentvar0.508640SEofregression0239302Akaikeinfocnterion0147466Sumsquaredresid0973616Schwarzcriterion0346422Loglikelihood2.451611F-statistic24.44018Durbin-Watsonstat1908356Prob(F-statistic)0.000002差分后的DW=1.908356,在n=21,k'=3時(shí)，查詢(xún)?cè)陲@著水平0.05的條件下，臨界值dl=1.026,du=

16、1.669DW>du.,模型不存在自相關(guān)。C=CA/(1-a)=2.801029/(1-0.568228)=6.487287272a1=a1人a2=a2人5.3.5 異方差檢驗(yàn)對(duì)修正后的*II型進(jìn)行white檢驗(yàn)。計(jì)量經(jīng)濟(jì)學(xué)結(jié)課論文趙嘉祺林經(jīng)12-1表格0-10white檢驗(yàn)結(jié)果ViE'AProc：Object：PrintName:FreezeStirrimte；orecastResideWhiteHeteroskedasticityTestF-statrstic0612149Probability0766090Obs*R-sgjared6.922335Probability06

17、45205TestEquation:DependentVariabSeRESIDA2MethodLeastSquaresDate:12/28/14Time:22:30Sample:2000S12010S2Includedobservations.22VariableCciMfici巳MStdEuort-StatisticProbC2.9270892.693061108690002984LNX1-16306202.367961-068861805042LNX1*2-07591330779262-0.97416903492LNXVLNX20.15775105026110-313B640.7590L

18、NX1*X50.0006020000552103962502973LNX212407061428651Q06844604022LNX2空0.0674960.1676070402702口6942LNX2*X5-00002580000208-12422880.2379X5Q口006900.000551-125253202342X5A2265E-092B3E-0B0093674Q9269R-sqjared0.314652Meandependentvar0.079041AdjustedR*squared-0.199360SD.dependentvar0.095016SEofregression0104057Akaikeinfocriterion-1384808Sumsqua