在计量经济学的Endogeneity : 有助易变的估计 (1).pptx
《在计量经济学的Endogeneity : 有助易变的估计 (1).pptx》由会员分享,可在线阅读,更多相关《在计量经济学的Endogeneity : 有助易变的估计 (1).pptx(32页珍藏版)》请在冰豆网上搜索。
EndogeneityinEconometrics:
InstrumentalVariableEstimationMingLUEndogeneityOmittingvariablebiasSimultaneityMeasurementerrorCanweignoretheomittedvariablesbias?
Itcanbesatisfactoryiftheestimatesarecoupledwiththedirectionofthebiasesforthekeyparameters.Canweuseproxytoeliminateomittedvariablebias?
Sometimes.CanFEestimationsolveomittingvariableproblem?
Firstdifferencingorfixedeffectsestimationeliminatestime-constantvariables.Inaddition,thepaneldatamethodsdonotsolvetheproblemoftime-varyingomittedvariablesIdeaofIVEstimationExogenousvariable.IndirecteffectsofIV.ExampleWhatcanserveasIVforedu?
Motherseducation?
Numberofsiblings?
Thereportofothers?
Adummyvariablethatisequalto1ifamanisborninthefirstquarteroftheyear.AngristandKrueger(1991).(Problematic.)InChina,theyearsofprimaryedu?
IVforskippedclass?
Thedistancefromhometoschool.OtherexamplesofIVIVforinstitution:
Language?
History?
Mauro(1995)使用人口的种族和语言构成作为腐败的工具变量,HallandJones(1999)用距离赤道的距离和以西欧语言为第一语言的程度作为制度质量的工具变量,LaPortaetal.(1997,1998,1999)把法律的起源作为各种法律结构的工具变量。
Acemoglu,Johnson,andRobinson(2001,2002)使用殖民地时代(1500年前后)的死亡率和人口密度作为制度的工具变量IVforschoolchoice:
Numberofsteams?
IdentificationReferto(15.9)and(15.10)The(asymptotic)standarderrorofSSTisthetotalsumofsquaresofthexiSelf-selectionAngrist(1990)studiedtheeffectthatbeingaveteranintheVietnamwarhadonlifetimeearnings.DraftlotterynumberisagoodIVcandidateforveteran.SomeadditionalwordsaboutnaturalexperimentandDIDPropertiesofIVwithaPoorInstrumentalVariablePoorIVcancauseseriousbias.R2MostregressionpackagescomputeanR-squaredafterIVestimation,usingthestandardformula:
R2=1-SSR/SST,whereSSRisthesumofsquaredIVresiduals,andSSTisthetotalsumofsquaresofy.R2canbenegativeinthiscase.IVESTIMATIONOFTHEMULTIPLEREGRESSIONMODELstructuralequationEstimationEfficientIVEquation(15.26)isanexampleofareducedformequation,whichmeansthatwehavewrittenanendogenousvariableintermsofexogenousvariables.TWOSTAGELEASTSQUARES2SLSinwordsThefirststageistoruntheregressionin(15.36),whereweobtainthefittedvaluesy2.ThesecondstageistheOLSregression(15.38).Becauseweusey2inplaceofy2,the2SLSestimatescandiffersubstantiallyfromtheOLSestimates.Anotherinterpretation:
MultipleEndogenousExplanatoryVariablesORDERCONDITIONFORIDENTIFICATIONOFANEQUATION:
Weneedatleastasmanyexcludedexogenousvariablesasthereareincludedendogenousexplanatoryvariablesinthestructuralequation.IVSOLUTIONSTOERRORS-IN-VARIABLESPROBLEMSOnepossibilityistoobtainasecondmeasurementonX*1,say,z1,asIV.AnalternativeistouseotherexogenousvariablesasIVsforapotentiallymismeasuredvariable.TESTINGFORENDOGENEITYANDTESTINGOVERIDENTIFYINGRESTRICTIONSThe2SLSestimatorislessefficientthanOLSwhentheexplanatoryvariablesareexogenous;aswehaveseen,the2SLSestimatescanhaveverylargestandarderrors.Howtotestendogeneity?
1.ComparingtheOLSand2SLSestimatesanddeterminingwhetherthedifferencesarestatisticallysignificant.(Hausman,1978)2.Aregressiontest:
Anotherinterpretationof2SLSIncludingv2intheOLSregression(15.51)clearsuptheendogeneityofy2.Wecanalsotestforendogeneityofmultipleexplanatoryvariables.Foreachsuspectedendogenousvariable,weobtainthereducedformresiduals.Then,wetestforjointsignificanceoftheseresidualsinthestructuralequation,usinganFtest.TestingOveridentificationRestrictionsIfwehavemorethanoneinstrumentalvariable,wecaneffectivelytestwhethersomeofthemareuncorrelatedwiththestructuralerror.UseoneIVandgetthepredictedresidual,thentestthecorrelationbetweenotherIVsandtheresidual.TESTINGOVERIDENTIFYINGRESTRICTIONS:
(i)Estimatethestructuralequationby2SLSandobtainthe2SLSresiduals,u1.(ii)Regressu1onallexogenousvariables.ObtaintheR-squared,sayR12.(iii)UnderthenullhypothesisthatallIVsareuncorrelatedwithu1,nR12X2(q),whereqisthenumberofinstrumentalvariablesfromoutsidethemodelminusthetotalnumberofendogenousexplanatoryvariables.IfnR12exceeds(say)the5%criticalvalueintheX2(q)distribution,werejectH0andconcludethatatleastsomeoftheIVsarenotexogenous.IsitbettertohavemoreIVs?
Addinginstrumentstothelistimprovestheasymptoticefficiencyofthe2SLS.Butthisrequiresthatanynewinstrumentsareinfactexogenous.Withthetypicalsamplesizesavailable,addingtoomanyinstrumentsthatis,increasingthenumberofoveridentifyingrestrictionscancauseseverebiasesin2SLS.2OmittedTopics2SLSWITHHETEROSKEDASTICITYAPPLYING2SLSTOTIMESERIESEQUATIONSAPPLYING2SLSTOPOOLEDCROSSSECTIONSANDPANELDATAForpooledcrosssectionsdata:
addtimedummy.Forpaneldata:
Inthefirststage,usethedifferencedIVtogetanestimateoftheendogenousvariable.Question:
IfthepanelmodelisaFEone,howtochecktheefficiencyofIViftheIVistimeinvariant?
STATAcommandsTocompareOLSand2SLSivregy(x=iv)x2eststoref2regyxx2hausmanf2Thesequenceisimportant.STATAcommandsTocompareFEandIV-FExtivregy(x=iv)x2,feeststoref2xtregyxx2,fehausmanf2Theend.