《计量经济学实验(5):模型设定误差检验.doc》由会员分享,可在线阅读,更多相关《计量经济学实验(5):模型设定误差检验.doc(6页珍藏版)》请在淘文阁 - 分享文档赚钱的网站上搜索。
1、计量经济学实验(计量经济学实验(5):模型设定误差):模型设定误差以引子中所提出的问题为例,分析影响中国进口量的主要因素(数据如表 9.3 所示) 。 表 9.3 单位:人民币亿元、亿美元年份GDP进口总额IM(人民币)进口总额IMdollar(美元)汇率EXCHANGE19804517.8298.8000200.17149.840019814862.4375.3800220.15170.510019825294.7364.9900192.85189.260019835934.5422.6000213.90197.570019847171.0637.8300274.10232.70001985
2、8964.41257.800422.52293.6600198610202.201498.300429.04345.2800198711962.501614.200432.16372.2100198814928.302055.100552.75372.2100198916909.202199.900591.40376.5100199018547.902574.300533.45478.3200199121617.803398.700637.91532.3300199226638.104443.300805.85551.4600199334634.405986.2001039.59576.200
3、0199446759.409960.1001156.14861.8700199558478.1011048.101320.84835.1000199667884.6011557.401388.33831.4200199774462.6011806.501423.70828.9800199878345.2011626.101402.37827.9100199982067.5013736.401656.99827.8300200089468.1018638.802250.94827.8400200197314.8020159.202435.53827.70002002105172.324430.3
4、02951.70827.70002003117251.934195.604127.60827.7000数据来源:中国统计年鉴 2004中国统计出版社设定如下的模型。12tttIMGDPu(9.50)其中,IMt是进口总额,tGDP是国内生产总值。为了分析此模型是否有变量设定误差,进行变量设定误差检验。有人认为,货物与服务的进口量受到一国的生产规模、货物与服务的进口价格、汇率等其他影响因素,而不能只仅用 GDP 来解释商品进口的变化。因此,设定的回归模型应该为:123ttttIMf GDPg Exchangeu(9.51)其中:GDP 为国内生产总值,f GDP为 GDP 的线性函数,Excha
5、nge 为美元兑换人民币的汇率,g Exchange为 Exchange 的线性函数。如果是这样,显然设定的回归模型(9.50)式中可能遗漏了变量 GDP、Exchange 以及两者的线性组合。那么GDP、Exchange 以及两者的线性组合是否被遗漏的重要变量呢?依据表 9.3 的数据,录入到 EViews 响应的数据表中,考证 IM=f(GDP)基本关系图:05000100001500020000250003000035000020000400006000080000100000 120000 GDPIM对(9.50)进行回归,有回归结果1067.3370.2307iiiimGDPe se
6、= (792.2620) (0.0142)t= (-2.0288) (16.2378)20.9230R 20.9195R DW=0.5357 F=263.6657 并作(9.50)回归的残差图:-6000-4000-20000200040006000800010000808284868890929496980002IM Residuals显然,存在自相关现象,其主要原因可能是建模时遗漏了重要的相关变量造成的。1、DW 检验检验:检验自相关检验自相关模型1067.3370.2307iiiimGDPe 的 DW 统计量表明,存在正的自相关自相关,由于遗漏变量 exchange 或 GDP 已经按从
7、小到大顺序排列,因此,无需重新计算 d 统计量。对n=24 和1k ,5%的德宾-沃森 d-统计量的临界值为1.273Ld 和1.446Ud,0.53571.273Ld,表明存在显著的遗漏变量现象。 为此,进行如下的校正:Dependent Variable: IMMethod: Least SquaresDate: 07/08/05 Time: 15:40Sample (adjusted): 1981 2003Included observations: 23 after adjustmentsVariableCoefficientStd. Errort-StatisticProb. C-2
8、24.36321892.132-0.1185770.9069GDP1.1482590.1514337.5826060.0000GDP(-1)-0.8224440.147359-5.5812130.0000EXCHANGE-4.2907468.348744-0.5139390.6135EXCHANGE2-0.0186370.008353-2.2311620.0386R-squared0.978691 Mean dependent var8434.222Adjusted R-squared0.973956 S.D. dependent var9025.326S.E. of regression14
9、56.525 Akaike info criterion17.59515Sum squared resid38186370 Schwarz criterion17.84200Log likelihood-197.3443 F-statistic206.6799Durbin-Watson stat1.962659 Prob(F-statistic)0.000000其中,exchange 的系数在统计意义上不显著,可以剔除,则有:Dependent Variable: IMMethod: Least SquaresDate: 07/08/05 Time: 15:43Sample (adjusted
10、): 1981 2003Included observations: 23 after adjustmentsVariableCoefficientStd. Errort-StatisticProb. C-1159.179511.0396-2.2682760.0352GDP1.1428970.1481197.7160700.0000GDP(-1)-0.8158420.143928-5.6684200.0000EXCHANGE2-0.0225690.003291-6.8578440.0000R-squared0.978378 Mean dependent var8434.222Adjusted
11、R-squared0.974965 S.D. dependent var9025.326S.E. of regression1428.041 Akaike info criterion17.52277Sum squared resid38746720 Schwarz criterion17.72024Log likelihood-197.5118 F-statistic286.5846Durbin-Watson stat2.047965 Prob(F-statistic)0.000000可以认为,这时模型设定无变量设定误差。2、LM 检验按照 LM 检验步骤,首先生成残差序列ie(用 EE 表
12、示) ,用 EE 对全部解释变量(包括遗漏变量)进行回归,有:Dependent Variable: EEMethod: Least SquaresDate: 07/08/05 Time: 15:45Sample (adjusted): 1981 2003Included observations: 23 after adjustmentsVariableCoefficientStd. Errort-StatisticProb. C448.1584511.03960.8769540.3915GDP0.9122010.1481196.1585680.0000GDP(-1)-0.8158420.1
13、43928-5.6684200.0000EXCHANGE2-0.0225690.003291-6.8578440.0000R-squared0.727360 Mean dependent var-37.56085Adjusted R-squared0.684312 S.D. dependent var2541.624S.E. of regression1428.041 Akaike info criterion17.52277Sum squared resid38746720 Schwarz criterion17.72024Log likelihood-197.5118 F-statistic16.89632Durbin-Watson stat2.047965 Prob(F-statistic)0.000014再计算223 0.7273616.72928nR ,查表 2 0.02527.37776,显然, 16.729287.37776,拒绝0H:受约束回归模型,接受1:H无约束回归模型的假设,即 确实存在遗漏变量。因此,在本章的引子中不能判断虽然简单但遗漏了重要变量的方程 (1)比复杂的方程(2)更好
限制150内