对《多元线性回归模型在物流成本预测中的应用》的解读

你好对多元线性回归模型在物流成本预测中的应用的解读一. 内容简介本文以国家宏观物流成本数据为研究对象，应用多元回归分析理论，对影响物流成本的相关指标进行分析，建立了线性回归模型，通过物流成本的统计分析，使企业可以从全局的角度了解自身的物流运作现状，明确目前关键的瓶颈问题以及突破口，提出解决的方法，以提高企业整体的运作绩效。二研究过程（一）变量选择全国宏观物流成本总额()由运输成本（）、保管成本（）、包装及货损成本（）、信息及管理成本（）组成，根据19962012年统计数据（见表1）进行多元线性回归分析。表1 1996-2012年我国宏观物流成本19963320.27831.81863.98755.213775.2719973479.921073.15923.43800.0784279.57819983835.471080.251084.51920.77492319994484.621180.791324.61046.46037.4120004814.341484.351599.781169.67068.0720015332.082093.072341.1616579422.3120026284.412506.52815.152065.811669.8620036181.032699.62908.682191.811978.1120045909.082593.623372.542676.8612548.120056726.752150.533361.872631.3912865.5420067293.91828.344083.12839.0714038.4120077719.871934.324500.93010.1415158.2320088125.481886.165119.063419.1216541.8220098759.882088.966340.672921.8818102.39201010165.862316.257794.663208.1620474.92201110879.822510.538776.163379.2322847.54201211955.272667.649684.273864.2924231.14（二）数据分析输入数据到EViews，进行回归分析，回归分析结果（见表2）。表2 Y-X1、X2、X3、X4最小二乘法回归结果Dependent Variable: YMethod: Least SquaresSample: 1996 2012Included observations: 17VariableCoefficientStd. Errort-StatisticProb.C-1757.210768.0691-2.2878280.0411X10.9284330.2854243.2528270.0069X20.9276050.2062804.4968340.0007X30.7672290.2216083.4621040.0047X41.4770410.2397656.1603660.0000R-squared0.998212Mean dependent var12703.63Adjusted R-squared0.997616S.D. dependent var6362.436S.E. of regression310.6257Akaike info criterion14.55498Sum squared resid1157860.Schwarz criterion14.80005Log likelihood-118.7174Hannan-Quinn criter.14.57934F-statistic1675.155Durbin-Watson stat2.156717Prob(F-statistic)0.000000 由运行结果看出，可决系数与均接近于1，综合度量回归模型对样本观测值拟合优较好；取=0.05，F检验值为1675.155，P值约等于0，说明结果显著；再通过自变量、的pro(t-Statistic)值看它们的拟合效果，分别是0.0411、0.0069、0.0007、0.0047和0.0000，其pro(t-Statistic)值均小于，结果显著，拒绝原假设，说明它们与显著相关，通过t检验。回归方程为： (1)（三）统计检验1.异方差性检验对上述模型（1）利用怀特（White）检验方法：首先，利用OLS估计模型（1）获得残差；其次，作残差平方对所有原始变量、变量平方、变量交叉乘积的回归分析，得到辅助方程；最后，检验的零假设是:不存在异方差，对辅助方程的作检验。回归模型（1）的残差平方与、及其平方项与交叉项作辅助回归，得表3：Dependent Variable: E2Method: Least SquaresSample: 1996 2012Included observations: 17VariableCoefficientStd. Errort-StatisticProb.C2236829.2269530.0.9855910.4282X1-2075.6631820.925-1.1398950.3724X2568.8256908.44190.6261550.5951X32479.7782007.1091.2354970.3421X473.031671326.3220.0550630.9611X120.3622480.3260361.1110660.3822X22-0.1626500.191098-0.8511340.4843X320.6307420.3725181.6931840.2325X42-1.7698930.320301-5.5257150.0312X1*X2-0.1030070.326672-0.3153220.7824X1*X3-1.2776650.661778-1.9306540.1933X1*X40.7448500.3512652.1204750.1681X2*X3-0.1287330.362658-0.3549700.7565X2*X40.4785990.4161101.1501730.3690X3*X40.7784900.2140713.6365940.0680R-squared0.993744Mean dependent var68109.41Adjusted R-squared0.949948S.D. dependent var152879.0S.E. of regression34202.52Akaike info criterion23.34263Sum squared resid2.34E+09Schwarz criterion24.07781Log likelihood-183.4123Hannan-Quinn criter.23.41571F-statistic22.69058Durbin-Watson stat2.335532Prob(F-statistic)0.042982由此，可以得到：，怀特统计量,在5%的显著性水平下，自由度为14的分布的相应临界值 16.8936，因此接受原假设，即不存在异方差。2.序列自相关性检验对模型（1）采用D.W.检验法检验序列的自相关性，在上述的表1结果中已经显示：，当n=17，k=4（不包含常数项），查表得 ，由于，得该模型无自相关性。3.比较预测结果与真实值 表4 模型预测结果与真实值比较YY219963775.2703875.37719974279.5784359.35319984923.0004997.89619996037.4106063.61420007068.0707044.41920019422.3109378.475200211669.8611613.62200311978.1111954.63200412548.10126765412948.97200614038.4114036.76200715158.2315103.78200816541.8216514.04200918102.3917493.96201020474.9220548.55201122847.5422397.34201224231.1424954.74图1 模型预测结果与真实值比较 图2 模型预测结果与真实值的相对误差 合图1和图2，真实值