时间序列分析报告及VAR模型

《时间序列分析报告及VAR模型》由会员上传分享，免费在线阅读，更多相关内容在工程资料-天天文库。

1、实用文档Lecture66.Timeseriesanalysis:Multivariatemodels6.1Learningoutcomes·Vectorautoregression(VAR)·Cointegration·Vectorerrorcorrectionmodel(VECM)·Application:pairstrading6.2Vectorautoregression(VAR)向量自回归Theclassicallinearregressionmodelassumesstrictexogeneity;hence,thereisnos

2、erialcorrelationbetweenerrortermsandanyrealisationofanyindependentvariable(leadorlag).Aswediscovered,serialcorrelation(orautocorrelation)isverycommoninfinancialtimeseriesandpaneldata.Furthermore,weassumedapre-definedrelationofcausality:explanatoryvariableaffectthedependentv

3、ariable.传统的线性回归模型假设严格的外生性，误差项与可实现的独立变量之间没有序列相关性。金融时间序列及面板数据往往都有很强的自相关性，假定解释变量影响因变量。WenowrelaxbothassumptionsusingaVARmodel.VARmodelscanberegardedasageneralisationofAR(p)processesbyaddingadditionaltimeseries.Hence,weenterthefieldofmultivariatetimeseriesanalysis.VAR模型可以当作是在一般

4、的自回归过程中加入时间序列。Let’slookatastandardAR(p)processfortwovariables(ytandxt).(1)yt=α1+i=1pβ1iyt-i+ε1t(2)xt=α2+i=1pβ2ixt-i+ε2tThenextstepistoallowthatlaggedvaluesofxtcanaffectytandviceversa.Thismeansthatweobtainasystemofequationsfortwodependentvariables(ytandxt).Bothdependentvaria

5、blesareinfluencedbypastrealisationsofytandxt.Bydoingthat,weviolatestrictexogeneity(seeLecture2);however,wecanuseamorerelaxedconcept,namelyweakexogeneity.Asweuselaggedvaluesofbothdependentvariables,wecanarguethattheselaggedvaluesareknowntous,asweobservedtheminthepreviousperi

6、od.Wecallthesevariablespredetermined.Predetermined(lagged)variablesfulfilweakexogeneityinthesensethattheyhavetobeuncorrelatedwiththecontemporaneouserrortermint.WecanstilluseOLStoestimatethefollowingsystemofequations,whichiscalledaVARinreducedform.(3)yt=α1+i=1pβ11iyt-i+i=1pβ

7、12ixt-i+ε1t(4)xt=α2+i=1pβ21iyt-i+i=1pβ22ixt-i+ε2t标准文案实用文档Thebeautyofthismodelisthatwedon’tneedtopredefinewhetherxoryareendogenous(thedependentvariable).Infact,wecantestwhetherx(y)isendogenousorexogenoususingGrangercausalitytests.TheideaofGrangercausalityisthatpastobservatio

8、ns(laggeddependentvariables)caninfluencecurrentobservations–butnotviceversa.Sothei