已阅读5页,还剩49页未读, 继续免费阅读
版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
ChrisBrooks2002,陈磊2004,6-1,Chapter6,Multivariatemodels,ChrisBrooks2002,陈磊2004,6-2,1Motivations,Allthemodelswehavelookedatthusfarhavebeensingleequationsmodelsoftheformy=X+uAllofthevariablescontainedintheXmatrixareassumedtobeEXOGENOUS.由系统外因素决定的变量yisanENDOGENOUSvariable.既影响系统同时又被该系统及其外部因素所影响的变量.Anexample-thedemandandsupplyofagood:(1)(2)(3)、=quantityofthegooddemanded/suppliedPt=price,St=priceofasubstitutegoodTt=somevariableembodyingthestateoftechnology,ChrisBrooks2002,陈磊2004,6-3,Assumingthatthemarketalwaysclears,anddroppingthetimesubscriptsforsimplicity(4)(5)ThisisasimultaneousSTRUCTURALFORMofthemodel.Thepointisthatpriceandquantityaredeterminedsimultaneously(priceaffectsquantityandquantityaffectsprice).PandQareendogenousvariables,whileSandTareexogenous.WecanobtainREDUCEDFORMequationscorrespondingto(4)and(5)bysolvingequations(4)and(5)forPandforQ.,SimultaneousEquationsModels:TheStructuralForm,ChrisBrooks2002,陈磊2004,6-4,SolvingforQ,(6)SolvingforP,(7)Rearranging(6),(8),ObtainingtheReducedForm,ChrisBrooks2002,陈磊2004,6-5,Multiplying(7)throughby,(9)(8)and(9)arethereducedformequationsforPandQ.,ObtainingtheReducedForm,ChrisBrooks2002,陈磊2004,6-6,Butwhatwouldhappenifwehadestimatedequations(4)and(5),i.e.thestructuralformequations,separatelyusingOLS?BothequationsdependonP.OneoftheCLRMassumptionswasthatE(Xu)=0,whereXisamatrixcontainingallthevariablesontheRHSoftheequation.Itisclearfrom(8)thatPisrelatedtotheerrorsin(4)and(5)-i.e.itisstochastic.WhatwouldbetheconsequencesfortheOLSestimator,ifweignorethesimultaneity?,2SimultaneousEquationsBias,ChrisBrooks2002,陈磊2004,6-7,RecallthatandSothatTakingexpectations,IftheXsarenon-stochastic,E(Xu)=0,whichwouldbethecaseinasingleequationsystem,sothat,whichistheconditionforunbiasedness.,SimultaneousEquationsBias,ChrisBrooks2002,陈磊2004,6-8,But.iftheequationispartofasystem,thenE(Xu)0,ingeneral.Conclusion:ApplicationofOLStostructuralequationswhicharepartofasimultaneoussystemwillleadtobiasedcoefficientestimates.IstheOLSestimatorstillconsistent,eventhoughitisbiased?No-Infacttheestimatorisinconsistentaswell.Henceitwouldnotbepossibletoestimateequations(4)and(5)validlyusingOLS.,SimultaneousEquationsBias,ChrisBrooks2002,陈磊2004,6-9,SoWhatCanWeDo?Takingequations(8)and(9),wecanrewritethemas(10)(11)WeCANestimateequations(10)whatwewantedweretheoriginalparametersinthestructuralequations-,.,3AvoidingSimultaneousEquationsBias,ChrisBrooks2002,陈磊2004,6-10,CanWeRetrievetheOriginalCoefficientsfromthes?Shortanswer:sometimes.wesometimesencounteranotherproblem:identification.*ConsiderthefollowingdemandandsupplyequationsSupplyequation(12)Demandequation(13)Wecannottellwhichiswhich!BothequationsareUNIDENTIFIEDorUNDERIDENTIFIED.Theproblemisthatwedonothaveenoughinformationfromtheequationstoestimate4parameters.Noticethatwewouldnothavehadthisproblemwithequations(4)and(5)sincetheyhavedifferentexogenousvariables.,4IdentificationofSimultaneousEquations,ChrisBrooks2002,陈磊2004,6-11,Wecouldhavethreepossiblesituations:1.Anequationisunidentifiedlike(12)or(13)wecannotgetthestructuralcoefficientsfromthereducedformestimates2.Anequationisexactlyidentifiede.g.(4)or(5)cangetuniquestructuralformcoefficientestimates3.Anequationisover-identifiedExamplegivenlaterMorethanonesetofstructuralcoefficientscouldbeobtainedfromthereducedform.,WhatDetermineswhetheranEquationisIdentifiedornot?,ChrisBrooks2002,陈磊2004,6-12,Howdowetellifanequationisidentifiedornot?Therearetwoconditionswecouldlookat:-Theorder阶condition-isanecessarybutnotsufficientconditionforanequationtobeidentified.-Therank秩condition-isanecessaryandsufficientconditionforidentification.在G个内生变量、G个方程的联立方程组模型中,某一方程是可识别的,当且仅当该方程没有包含的变量在其他方程中对应系数组成的矩阵的秩为G-1。对于相对简单的方程系统,这两个规则将得到同样的结论。事实上,大多数经济和金融方程系统都是过度识别的。,WhatDetermineswhetheranEquationisIdentifiedornot?,ChrisBrooks2002,陈磊2004,6-13,StatementoftheOrderConditionLetGdenotethenumberofstructuralequations.AnequationisjustidentifiedifthenumberofvariablesexcludedfromanequationisG-1.IfmorethanG-1areabsent,itisover-identified.IflessthanG-1areabsent,itisnotidentified.ExampletheYsareendogenous,whiletheXsareexogenous.Determinewhethereachequationisover-,under-,orjust-identified.(14)-(16),Statementoftheordercondition,ChrisBrooks2002,陈磊2004,6-14,SolutionG=3;If#excludedvariables=2,theeqnisjustidentifiedIf#excludedvariables2,theeqnisover-identifiedIf#excludedvariables2,theeqnisnotidentifiedEquation14:NotidentifiedEquation15:JustidentifiedEquation16:Over-identified如果模型中每个结构方程都是可识别的,则称结构型联立方程组模型是可识别的。,Exampleoftheordercondition,ChrisBrooks2002,陈磊2004,6-15,5外生性的定义,Leamer(1985):p310变量X对变量Y是外生的,如果变量Y关于X的条件分布不随产生X的过程的变化而改变。外生性的两种形式:前定变量:与方程中的当前和未来误差项独立。严格外生变量:与方程中任何时期的误差项独立。前定变量的通常定义:包括外生变量和滞后的内生变量,ChrisBrooks2002,陈磊2004,6-16,Howdowetellwhethervariablesreallyneedtobetreatedasendogenousornot?Consideragainequations(14)-(16).Equation(14)containsY2andY3-butdowereallyneedequationsforthem?WecanformallytestthisusingaHausmantestasfollows:1.Obtainthereducedformequationscorrespondingto(14)-(16).Thereducedformsturnouttobe:(17)-(19)Estimatethereducedformequations(17)-(19)usingOLS,andobtainthefittedvalues,5TestsforExogeneity,ChrisBrooks2002,陈磊2004,6-17,2.Runtheregressioncorrespondingtoequation(14).3.Runtheregression(14)again,butnowalsoincludingthefittedvaluesasadditionalregressors:(20)4.UseanF-testtotestthejointrestrictionthat2=0,and3=0.Ifthenullhypothesisisrejected,Y2andY3shouldbetreatedasendogenous.,TestsforExogeneity,ChrisBrooks2002,陈磊2004,6-18,Considerthefollowingsystemofequations:(21-23)Assumethattheerrortermsarenotcorrelatedwitheachother.CanweestimatetheequationsindividuallyusingOLS?Equation21:Containsnoendogenousvariables,soX1andX2arenotcorrelatedwithu1.SowecanuseOLSon(21).Equation22:ContainsendogenousY1togetherwithexogenousX1andX2.WecanuseOLSon(22)ifalltheRHSvariablesin(22)areuncorrelatedwiththatequationserrorterm.Infact,Y1isnotcorrelatedwithu2becausethereisnoY2terminequation(21).SowecanuseOLSon(22).,6RecursiveSystems,ChrisBrooks2002,陈磊2004,6-19,Equation23:ContainsbothY1andY2;werequirethesetobeuncorrelatedwithu3.Bysimilarargumentstotheabove,equations(21)and(22)donotcontainY3,sowecanuseOLSon(23).ThisisknownasaRECURSIVEorTRIANGULARsystem.Wedonothaveasimultaneityproblemhere.Butinpracticenotmanysystemsofequationswillberecursive.,RecursiveSystems,ChrisBrooks2002,陈磊2004,6-20,IndirectLeastSquares(ILS)CannotuseOLSonstructuralequations,butwecanvalidlyapplyittothereducedformequations.Ifthesystemisjustidentified,ILSinvolvesestimatingthereducedformequationsusingOLS,andthenusingthemtosubstitutebacktoobtainthestructuralparameters.However,ILSisnotusedmuchbecause1.Solvingbacktogetthestructuralparameterscanbetedious.2.Mostsimultaneousequationssystemsareover-identified.,7EstimationproceduresforSystems,ChrisBrooks2002,陈磊2004,6-21,Infact,wecanusethistechniqueforjust-identifiedandover-identifiedsystems.Twostageleastsquares(2SLSorTSLS)isdoneintwostages:Stage1:ObtainandestimatethereducedformequationsusingOLS.Savethefittedvaluesforthedependentvariables.Stage2:Estimatethestructuralequations,butreplaceanyRHSendogenousvariableswiththeirstage1fittedvalues.,EstimationofSystemsUsingTwo-StageLeastSquares,ChrisBrooks2002,陈磊2004,6-22,Example:Sayequations(14)-(16)arerequired.Stage1:Estimatethereducedformequations(17)-(19)individuallybyOLSandobtainthefittedvalues,.Stage2:ReplacetheRHSendogenousvariableswiththeirstage1estimatedvalues:(24)-(26)Nowandwillnotbecorrelatedwithu1,willnotbecorrelatedwithu2,andwillnotbecorrelatedwithu3.,EstimationofSystemsUsingTwo-StageLeastSquares,ChrisBrooks2002,陈磊2004,6-23,TSLS是比较经济、易用的方法。如果在第一阶段估计时所得到的R2非常高,那么古典OLS估计量与TSLS估计量将非常接近;如果在第一阶段估计时所得到的R2非常低,TSLS估计量将没有太大的实际意义。TSLS估计量是有偏估计量,但却是一致估计量。ItisstillofconcerninthecontextofsimultaneoussystemswhethertheCLRMassumptionsaresupportedbythedata.Ifthedisturbancesinthestructuralequationsareautocorrelated,the2SLSestimatorisnotevenconsistent.ThestandarderrorestimatesalsoneedtobemodifiedcomparedwiththeirOLScounterparts,butoncethishasbeendone,wecanusetheusualt-andF-teststotesthypothesesaboutthestructuralformcoefficients.,EstimationofSystemsUsingTwo-StageLeastSquares,ChrisBrooks2002,陈磊2004,6-24,RecallthatthereasonwecannotuseOLSdirectlyonthestructuralequationsisthattheendogenousvariablesarecorrelatedwiththeerrors.OnesolutiontothiswouldbenottouseY2orY3,butrathertousesomeothervariablesinstead.Wewanttheseothervariablestobe(highly)correlatedwithY2andY3,butnotcorrelatedwiththeerrors-theyarecalledINSTRUMENTS.SaywefoundsuitableinstrumentsforY2andY3,z2andz3respectively.Wedonotusetheinstrumentsdirectly,butrunregressionsoftheform(27)E(u1tu2t)=0.TheanalysiscouldbeextendedtoaVAR(g)model,orsothattherearegvariablesandgequations.,8VectorAutoregressiveModels,ChrisBrooks2002,陈磊2004,6-28,OneimportantfeatureofVARsisthecompactnesswithwhichwecanwritethenotation.Forexample,considerthecasefromabovewherek=1.Wecanwritethisasoryt=0+1yt-1+utg1g1ggg1g1,VectorAutoregressiveModels:NotationandConcepts,ChrisBrooks2002,陈磊2004,6-29,VAR模型还具有灵活性和易于一般化的重要特点.例如,模型可以扩展到包含移动平均误差项,即VARMA。Thismodelcanbeextendedtothecasewherethereareklagsofeachvariableineachequation:yt=0+1yt-1+2yt-2+.+kyt-k+utg1g1ggg1ggg1ggg1g1Wecanalsoextendthistothecasewherethemodelincludesfirstdifferencetermsandcointegratingrelationships(aVECM).,VectorAutoregressiveModels:NotationandConcepts,ChrisBrooks2002,陈磊2004,6-30,AdvantagesofVARModelling-Donotneedtospecifywhichvariablesareendogenousorexogenous-allareendogenous-Allowsthevalueofavariabletodependonmorethanjustitsownlagsorcombinationsofwhitenoiseterms,somoregeneralthanARMAmodelling-Providedthattherearenocontemporaneoustermsontherighthandsideoftheequations,cansimplyuseOLSseparatelyoneachequation,因为方程右边的变量都是前定变量。-Forecastsareoftenbetterthan“traditionalstructural”models.,VARModelsComparedwithStructuralEquationsModels,ChrisBrooks2002,陈磊2004,6-31,ProblemswithVARs-VARsarea-theoretical(asareARMAmodels)。VAR模型较少用于理论分析和政策建议。-Howdoyoudecidetheappropriatelaglength?-Somanyparameters!Ifwehavegequationsforgvariablesandwehaveklagsofeachofthevariablesineachequation,wehavetoestimate(g+kg2)parameters.e.g.g=3,k=3,parameters=30-DoweneedtoensureallcomponentsoftheVARarestationary?-Howdoweinterpretthecoefficients?,VARModelsComparedwithStructuralEquationsModels,ChrisBrooks2002,陈磊2004,6-32,ChoosingtheOptimalLagLength,Cross-EquationRestrictionsInthespiritof(unrestricted)VARmodelling,eachequationshouldhavethesamelaglengthSupposethatabivariateVAR(8)estimatedusingquarterlydatahas8lagsofthetwovariablesineachequation,andwewanttoexaminearestrictionthatthecoefficientsonlags5through8arejointlyzero.ThiscanbedoneusingalikelihoodratiotestDenotethevariance-covariancematrixofresiduals(givenbyE),as.Thelikelihoodratiotestforthisjointhypothesisisgivenby,ChrisBrooks2002,陈磊2004,6-33,ChoosingtheOptimalLagLength,whereisthevariance-covariancematrixoftheresidualsfortherestrictedmodel(with4lags),isthevariance-covariancematrixofresidualsfortheunrestrictedVAR(with8lags),andTisthesamplesize.Theteststatisticisasymptoticallydistributedasa2withdegreesoffreedomequaltothetotalnumberofrestrictions.IntheVARcaseabove,wearerestricting4lagsoftwovariablesineachofthetwoequations=atotalof4*2*2=16restrictions.InthegeneralcasewherewehaveaVARwithgequations,andwewanttoimposetherestrictionthatthelastqlagshavezerocoefficients,therewouldbeg2qrestrictionsaltogetherDisadvantages:ConductingtheLRtestiscumbersomeandrequiresanormalityassumptionforthedisturbances.,ChrisBrooks2002,陈磊2004,6-34,InformationCriteriaforVARLagLengthSelection,Multivariateversionsoftheinformationcriteriaarerequired.Thesecanbedefinedas:whereallnotationisasaboveandkisthetotalnumberofregressorsinallequations,whichwillbeequaltog2k+gforgequations,eachwithklagsofthegvariables,plusaconstanttermineachequation.Thevaluesoftheinformationcriteriaareconstructedfor0,1,lags(uptosomepre-specifiedmaximum).,ChrisBrooks2002,陈磊2004,6-35,DoestheVARIncludeContemporaneousTerms?,Sofar,wehaveassumedtheVARisoftheformButwhatiftheequationshadacontemporaneousfeedbackterm?WecanwritethisasThisVARisinprimitive/structuralform.,ChrisBrooks2002,陈磊2004,6-36,PrimitiveversusStandardFormVARs,WecantakethecontemporaneoustermsovertotheLHSandwriteorByt=0+1yt-1+utWecanthenpre-multiplybothsidesbyB-1togiveyt=B-10+B-11yt-1+B-1utoryt=A0+A1yt-1+etThisisknownasastandardformVAR,whichwecanestimateusingOLS.,ChrisBrooks2002,陈磊2004,6-37,VAR模型的识别,粗略地讲,结构型VAR模型的识别问题是指,能否从一个简约模型的估计值反导出原来的结构模型的系数。结构型VAR模型是不可识别的,因为两个方程的等号右边具有相同的前定变量为了解决这个问题,需要加入一定的约束条件。即同期项的一个系数12或22须设为0,使距阵B为三角形。最好是依据经济理论加入约束条件,ChrisBrooks2002,陈磊2004,6-38,BlockSignificanceandCausalityTests,Itislikelythat,whenaVARincludesmanylagsofvariables,itwillbedifficulttoseewhichsetsofvariableshavesignificanteffectsoneachdependentvariableandwhichdonot.Forillustration,considerthefollowingbivariateVAR(3):Wemightbeinterestedintestingthefollowinghypotheses,andtheirimpliedrestrictionsontheparameters:,ChrisBrooks2002,陈磊2004,6-39,BlockSignificanceandCausalityTests,EachofthesefourjointhypothesescanbetestedwithintheF-testframework.ThesetestscouldalsobereferredtoasGrangercausalitytests.Grangercausalitytestsseektoanswerquestionssuchas“Dochangesiny1causechangesiny2?”Ify1causesy2,lagsofy1shouldbesignificantintheequationfory2.Ifthisisthecase,wesaythaty1“Granger-causes”y2.Ify2causesy1,lagsofy2shouldbesignificantintheequationfory1.Ifbothsetsoflagsaresignificant,thereis“bi-directionalcausality”.Ify2causesy1,buty1doesnotcausesy2,theny2isstrongexogenous(intheequationfory1).Ifneithery2causesy1,nory1causesy2,theny1y2areindependent此处因果性并不意味着一个变量的变动引起另一个变量的变动,ChrisBrooks2002,陈磊2004,6-40,ImpulseResponses,VARmodelsareoftendifficulttointerpret:onesolutionistoconstructtheimpulseresponsesandvariancedecompositions.ImpulseresponsestraceouttheresponsivenessofthedependentvariablesintheVARtoshockstotheerrorterm.Aunitshockisappliedtoeachvariableanditseffectsarenoted.ConsiderforexampleasimplebivariateVAR(1):Achangeinu1twillimmediatelychangey1.Itwillchangey2andalsoy1duringthenextperiod.Wecanexaminehowlongandtowhatdegreeashocktoagivenequationhasonallofthevariablesinthesystem.eg.p341,ChrisBrooks2002,陈磊2004,6-41,多变量VAR模型也可改写为这里yt是一个k维内生变量向量,t是协方差矩阵为的扰动向量。,一般VAR模型的脉冲响应函数,假如VAR(p)可逆,我们可以得到VMA()的表达式:,ChrisBrooks2002,陈磊2004,6-42,VMA表达式的系数可按下面的方式给出:VAR的系数A和VMA的系数必须满足下面关系:其中,。关于的条件递归定义了VMA系数:,从而可知VMA的系数可以由VAR的系数递归得到。,一般VAR模型的脉冲响应函数,ChrisBrooks2002,陈磊2004,6-43,考虑VMA()的表达式设,y的第i个变量可以写成:其中k是变量个数。仅考虑2个变量(k=2)的情形:,现在假定在基期给一个单位的脉冲,即:21012345t,ChrisBrooks2002,陈磊2004,6-44,由的脉冲引起的的响应函数:,由上述推导可知由的脉冲引起的的响应函数序列是由VMA()中系数矩阵第2行,第1列的元素组成,q=1,2,。因此,一般地,由的脉冲引起的的响应函数可以求出如下:其中,代表着对第j个变量的单位冲击引起第i个变量的第q期滞后反映。,ChrisBrooks2002,陈磊2004,6-45,VarianceDecompositions,VariancedecompositionsofferaslightlydifferentmethodofexaminingVARdynamics.Theygivetheproportionofthemovementsinthedependentvariablesthatareduetotheir“own”shocks,versusshockstotheothervariables.Thisisdonebydetermininghowmuchofthes-stepaheadforecasterrorvarianceforeachvariableisexplainedbyinnovationstoeachexplanatoryvariable(s=1,2,).ThevariancedecompositiongivesinformationabouttherelativeimportanceofeachshocktothevariablesintheVAR.脉冲响应函数和方差分解常常提供一定程度上的相似信息.,ChrisBrooks2002,陈磊2004,6-46,TheOrderingoftheVariables,Butforcalculatingimpulseresponsesandvariancedecompositions,theorderingofthevariablesisimportant.Themainreasonforthisisthatabove,weassumedthattheVARerrortermswerestatisticallyindependentofoneanother.Thisisgenerallynottrue,however.Theerrortermswilltypicallybecorrelatedtosomedegree.Therefore,thenotionofexaminingtheeffectoftheinnovationsseparatelyhaslittlemeaning,sincetheyhaveacommoncomponent.Whatisdoneisto“orthogonalise”theinnovations.InthebivariateVAR,thisproblemwouldbeapproachedbyattributingalloftheeffectofthecommoncomponenttothefirstofthetwovariablesintheVAR.Inthegeneralcasewheretherearemorevariables,thesituationismorecomplexbuttheinterpretationisthesame.,ChrisBrooks2002,陈磊2004,6-47,由以上讨论可知,对第i个变量的冲击不仅直接影响第i个
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 2026年新能源车设计创新解决方案报告
- 智造赋能未来 2026年华东生物质能发电可行性研究报告
- 2026年人形机器人工业装配协作方案
- 竞品分析报告撰写框架及案例
- 【基于交通流预测的单交叉口无模型自适应控制分析6800字(论文)】
- 2026年数字化转型下的组织变革与人才激励
- 大型制造企业设备维护与预防性保养标准操作程序
- 医药代表拜访记录表
- 建筑工程造价预算表
- 健康食品功能成分研发及市场准入指南
- 2026年中小学生安全知识竞赛试题(附答案)
- 2026年安全管理人员安全培训考试题附答案
- 加速康复外科中国专家共识
- 2026年人教版七年级下册政治期末综合测评卷(含答案可下载)
- 2026年全国新高考1卷英语试卷(含答案及详解)
- 市场监督管理局特种设备安全监察工作手册(标准版)
- 护理个案查房:糖尿病足的预防与护理
- 高中数学必修一2.2基本不等式常见题型(含答案)
- 2026年衡阳市应急管理系统事业单位人员招聘考试备考试题及答案详解
- 口腔材料调拌方法
- 某锻造厂供配电系统设计
评论
0/150
提交评论