vector autoregressive models i

《vector autoregressive models i》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、c09-VectorAutoregressModelsPage321Thursday,October26,20062:07PMCHAPTER9VectorAutoregressiveModelsInthisandthefollowingtwochapterswediscussvectorautoregressiveImodels.HereweprovidetheformalbackgroundofVARmodelsanddiscussingtheirstatisticalproperties.Thenextchapteraddressesth

2、eestimationofVARmodels.VARMODELSDEFINEDVectorautoregressive(VAR)modelsare,assuggestedbytheirname,modelsofvectorsofvariablesasautoregressiveprocesses,whereeachvariabledependslinearlyonitsownlaggedvaluesandthoseoftheothervariablesinthevector.Thismeansthatthefuturevaluesofthep

3、rocessareaweightedsumofpastandpresentvaluesplussomenoise(and,possibly,exogenousvariables).Forexample,itisknownthatthereareequityprice“leaders”andequityprice“laggards”inthesensethatthereturnsofsomeportfoliosoflarge-capstocksanticipatethereturnsoflargeportfoliosofsmall-capsto

4、cks.1Ananalystwhowantstoexploitthisrelationshipforaspecificpairofleader-laggardportfoliosmightfitabivariateVARtomodelthereturnsoftheleaderandlaggardportfolios.Suppose,forexample,thatportfolioAisaleaderportfolioandportfolioBalaggardportfolio.Theanalystcanwritethefollowingmodel

5、forreturns:1SeeJohnY.Campbell,AndrewW.Lo,andA.CraigMacKinlay,TheEconometricsofFinancialMarkets(Princeton,NJ:PrincetonUniversityPress,1997);andAngelosKanasandGeorgeP.Kouretas,“ACointegrationApproachtotheLead-LagEffectAmongSized-SortedEquityPortfolios,”WorkingPaper,Department

6、ofEconomics,UniversityofCrete,2001.321c09-VectorAutoregressModelsPage322Thursday,October26,20062:07PM322FINANCIALECONOMETRICSRA()t+1=RA()εt+A()t+1RB()t+1=aRA()εt+B()t+1whereRA(t)andRB(t)arethereturnsofthetwoportfolios,respectively,andεAandεBareindependentwhite-noiseterms(II

7、Dzero-meanvari-ables).Thefirstequationstatesthattheprice-leaderportfoliofollowsarandomwalk;thesecondequationstatesthatthelaggardportfoliotendstofollowtheleaderwithadelayofoneperiod.Theaboveisasimpleexampleofamultivariateextensionoftheautoregressive(AR)model.Avectorautoregres

8、sivemodeloforderp[VAR(p)]hasthefollowinggeneralform:xt=A1xt–1++A2xt–2…++Apxtp–st+ε