ARIMA模型建立与应用

1、 #聽EViews-Equation:UNTITLEDWorkFile:实鑿一UJntitL=旦11FileEditObjectViewProcQuickOEltionsWindowHelp_n1XViewPb匚Obj已匚t|FTintNam已FreezeEstimateFore匚已st|宝已乜|尺巴引日生DependentVariable:DEUROMethod:LeastSquaresDate:04/12/11Time:10:48Sample(adjusted):1993M032007M12Includedobservations:178afteradjustmentsConvergenc

2、eachievedafter3iterationsVariableCoefficientStd.Errort-StatisticProb.C-0.0009220.002097-0.4398900.6606AR0.3082790.0714844.3125520.0000R-squared0.095572Meandependentvar-.886AdjustedR-squared0.090433S.D.dependentvar0.020291S.E.ofregression0.019352Akaikeinfocriterion-5.040911Sumsquaredresid0.065909Schw

3、arzcriterion-5.005160Loglikelihood450.6411F-statistic18.59810Durbin-Watsonstat1.871573Prob(F-statistic)0.000027InvertedARRoots.31Path=匚:do匚uiti已itsandsettingsadministrdtormydo匚umentsDB=non已WF常数c的概率太大（0.6606），接受C=0的假设，所以模型应该去掉常数。QuickEstimateLS(NLSandARMA)在对话框输入d(euro)ar(1)EViews-Equation:UNTITLEDWor

4、kFile:实豊一UJntitL.I!FileEditObjectViewProcQuickOptionsWindowHelp_n1XViewProc|Obie匚PrintPJameFreezeEstimateForecast|Stats|ResidsDependentVariable:DEUROMethod:LeastSquaresDate:04/12/11Time:10:52Sample(adjusted):1993M032007M12Includedobservations:178afteradjustmentsConvergenceachievedafter2iterationsVar

5、iableCoefficientStd.Errort-StatisticProb.AR0.3095220.0712654.3432280.0000R-squared0.094579Meandependentvar-.0886AdjustedR-squared0.094579S.D.dependentvar0.020291S.E.ofregression0.019307Akaikeinfocriterion-5.051050Sumsquaredresid0.065982Schwarzcriterion-5.033174Loglikelihood450.5434Durbin-Watsonstat1

6、.871488InvertedARRoots.31模型为：w二0.309522wtt1t=(4.343228)p=(0.0000)w二,-(1L)xttt从p值看，系数是显者的。从InvertedARRoots（自回归特征方程根的倒数）是0.31，在单位圆之外，说明模型是平稳的。但还要对残差进行白噪声检验：在QuickEstimateLS（NLSandARMA）在对话框输入d（euro）ar（1）OK出结果的页面上ViewResidualTestsCorrelogram-Q-statistics选K=13（由178/10或178平方根来）Date:04/11/11Time:10:51Sampl

7、e:1993M032007M12Includedobservations:178Q-statisticprobabilitiesadjustedfor1ARMAterm(s)AutocorrelationPartialCorrelationACPACQ-StatProb.|.|.|.|10.0650.0650.7661*|.|*|.|2-0.165-0.1705.73060.017*|.|*|.|3-0.104-0.0847.72760.021.|.|.|.|4-0.021-0.0387.80870.050.|*|.|.|50.0810.0569.01900.061.|.|.|.|60.000

8、-0.0279.01910.108.|.|.|.|70.0310.0519.19580.163.|*|.|*|80.1000.10611.0720.136.|.|.|.|9-0.0020.00111.0730.198.|*|.|*|100.0990.14312.9480.165.|.|.|.|110.0230.03413.0540.221.|.|.|.|12-0.0270.01113.1960.281.|.|.|.|130.0100.03013.2160.354从K=13一行找到Q统计量值为13.216,相伴概率（记为p-Q）为0.3540.05,接受序列不相关的假设，即认为残差序列是白噪声。

9、类似地，对模型ARIMA（2，1，0）、ARIMA（0，1，1）、ARIMA（1，1，1）、ARIMA（2，1，1）进行估计与检验。ARIMA（1，1，0），ARIMA（2，1，0）、ARIMA（0，1，1）三个检验都通过参数显者性检验，模型平稳性和可逆性检验，残差序列白噪声检验。但是模型ARIMA（1,1，1）、ARIMA（2,1，1）没有通过检验。模型评价与比较模型e1e1e1ARIMA(1,1,0)0.3100.0950.340ARIMA(2,1,0)0.373-0.2020.1260.698ARIMA(0,1,1)0.3850.1220.727RA2和p-Q两指标越大越好，ARIMA(1,1,0)不好，在一样好的两模型ARIMA(2,1,0)和ARIMA(0,1,1)中，ARIMA(2,1,0)用自回归信息预测，所以在预测方面ARIMA(2,1,0)明显好。最终选择ARIMA(2,1,0)：w二0.373w-0.202wtt1t2w二,-(1L)xttt即(1L)x=0.373(1L)x0.202(1L)xtt1t2x=1.373x0.575x+0.202xtt1t2t3(五)模型外推应用已知2007：12,2007：