【文档说明】斯坦福大学计量经济学应用lecture1.pptx，共(28)页，354.249 KB，由精品优选上传

转载请保留链接：https://www.ichengzhen.cn/view-323201.html

以下为本文档部分文字说明：

NickBloom,AppliedEconometrics,Winter2010APPLIEDECONOMETRICSLecture1-IdentificationNickBloom,AppliedEconometrics,W

inter2010DefiningIdentificationExperimentsNaturalExperimentsInstrumentalvariablesEconometricIdentificationNickBloom,AppliedE

conometrics,Winter2010WHATISIDENTIFICATION?GraduateandprofessionaleconomicsmainlyconcernedwithidentificationinempiricalworkConceptofunderstandin

gwhatisthecausalrelationshipbehindempiricalresults:Thisisessentialforlearningfromempiricalresearch–Time-seriesexample:InterestratesandGDP–Cross-

sectionexample:Management&ProductivityNickBloom,AppliedEconometrics,Winter2010WHATISDRIVINGTHISRELATIONSHIP?05101520Fe

deralfundsrate-.15-.1-.050.05.1196019701980199020002010ym...DetrendedlogindustrialoutputFederalfundsrateC

orrelation=0.233NickBloom,AppliedEconometrics,Winter2010REASONSFORCORRELATIONImaginevariablesYtandXtarecorrelated:Therecanbethreer

easonsforthis,whicharenotmutuallyexclusive:•Cause:ChangesinXtdrivechangesinYt•ReverseCause:ChangesinYtdrivechangesinXt•Correlatedvariable:ChangesinZ

tdrivesXtandYtNickBloom,AppliedEconometrics,Winter2010WHATISDRIVINGTHISRELATIONSHIP?246812345ManagementNickBloom,AppliedEconometric

s,Winter2010SOHOWDOWEGETIDENTIFICATIONFourbroadapproachesforidentification•Experiments–yougeneratethevariation•NaturalExperiments–youknowwhatge

neratedthevariation•Instrumentalvariables–youhaveavariablethatcanprovideyouvariation•EconometricIdentification–yourelyon(testable)econometricassu

mptionsforidentificationNickBloom,AppliedEconometrics,Winter2010DefiningIdentificationExperimentsNaturalExperimentsInstrumentalvariablesEconometricId

entificationNickBloom,AppliedEconometrics,Winter2010EXPERIMENTS(1)ExperimentsaretotallystandardinScience&MedicineForexample:•Setupatreatme

ntandcontrolgroupforanewdrug,makingsurethesearecomparable(orrandomlyselected)•Ensurethesamplesizesarelargeenoug

htoobtainstatisticalsignificance•Ensuretheexperimentisunbiased–i.e.thedrugandtheplaceboareassimilaraspossible•RuntheexperimentNic

kBloom,AppliedEconometrics,Winter2010EXPERIMENTS(2)EconomistsliketousethelanguageofScienceForexampletheUKconsideredintroducinganEducationM

aintenanceAllowance,topaykidstostayonatschool.Butwanttotestfirsttoseeifthiswouldthiswork.•Setupatreatmentandcontrolregionst

omatchtheseincharacteristics•Selectenoughregionstogetlargesamplesizes•Observeagentsactionstoevaluateimpact(ra

therthanselfreportedoutcomes)•RuntheexperimentNickBloom,AppliedEconometrics,Winter2010EXPERIMENTS(3)Experimentsarerareineco

nomicsbecausetheyareexpensive,althoughtheybecomingmorepopular:Typicalareasforrunningexperimentsinclude:•Developmenteconomics–cheapertorunexperimentsi

nthethirdWorld(watersupplyormanagementpractices)•Consumereconomics–smallstakesexperimentsthatareeasytoadminister(creditcards)•

Individualbusinessapplications–firmscanfinancethese(retailstorelayout)Butsomefieldswillneverhaveexperiment–forexamplemacroeconomicsNickBloom,Applied

Econometrics,Winter2010DefiningIdentificationExperimentsNaturalExperimentsInstrumentalvariablesEconometricIdentificationNic

kBloom,AppliedEconometrics,Winter2010NATURALEXPERIMENTS(1)Naturalexperimentsarewherefortunatesituationscreategoodunderlying

identification:Typicallyseveralapproaches:•Taxe.g.ResponseofR&Dtothecostofcapital(Bloom,Griffith&VanReenen,200

2),(ChettyandSaez,2003)•Discontinuity(seeover)•Shock-financialcrisisandKibutzim(Abramitzky,2007)•Disasters-EthiopianJewsairlift(Gould,Lev

y&Passerman,2004)NickBloom,AppliedEconometrics,Winter2010NATURALEXPERIMENTS(2)Naturalexperimentsarealmostthehollygrailofmodernappliede

conomicsIntheabsenceoftrueexperimentstheyprovidethebestwaytoprovidesimpleidentificationCoupleofstandardwaytousenaturalexperimentsinpract

ice–Discountinunityanalysisand/or–DifferenceindifferencesNickBloom,AppliedEconometrics,Winter2010DISCON

TINUITYANALYSIS–example1RegionA(notax)RegionB(50%tax)Imaginea50%taxisleviedoninvestmentintherichcoastalregionAbutnotinthep

oorinlandregionB.Ifyousawthegraphbelowcouldyousaywhattheimpactofthetaxisoninvestment?InvestmentEstimatedimp

actofthetaxNickBloom,AppliedEconometrics,Winter2010DISCONTINUITYANALYSIS–example2Impactoftelephonesonpriceoffishin

Kerala(India)NickBloom,AppliedEconometrics,Winter2010DIFFERNCESINDIFFERENCESt0denotespre-treatmentperiodsforwhi

chdataareavailablet1denotespost-treatmentperiodsforwhichdataareavailableAveragechangeinoutcome(preandpost-treatment)fortre

atmentgroupminusaveragechangeinoutcomeforcontrolgroup()()ControltControltTreattTreattRRRR0101Impact−−−=Identificationcomesfromthedifferentia

lchangebetweenthetwogroupspreandpost-treatment–differenceoutunobservedfixedeffects–differenceoutcommontimeeffectsKeyassumptionofcommontimee

ffectsforthetwogroupsNickBloom,AppliedEconometrics,Winter2010POLICYEXAMPLEOF“DIFF-IN-DIFF”•SmallfirmsR&Dtaxcreditintr

oducedin2000forfirmswith250orlessemployees•Socouldlookatfirmsbeforeandaftercredit–Butotherthingsalsochanging(2000peak

ofdotcomboometc…)–Soneedtosetupacontrolgroupofcompanieslooksimilartofirmsgettingthecreditexceptdon’tgetthec

redit•Comparefirmswith240employeestothosewith260•Thisisdouble-diff(ordiffindiffs)tocomparedifferences:–Betweenpreandpostthecredit(1999ve

rsus2001)–Betweenthetreated(240employees)anduntreatedfirms(260employees)NickBloom,AppliedEconometrics,Wi

nter2010DefiningIdentificationExperimentsNaturalExperimentsInstrumentalvariablesEconometricIdentificationNickBloom,AppliedEconometrics,Winter2010IN

STRUMENTALVARIABLES(1)Wanttolookateffectofschooling(Si)onearnings(Yi)Assumethetruemodelis:Yi=α+β1Si+β2Ai+vi

whereAiis(unobserved)abilitywhichispositivelycorrelatedwithSi,andviisrandomindependentnoiseWhatwouldhappenifwees

timatedthefollowinginstead?Yi=a+b1Si+eiwhereei=β2Ai+viNickBloom,AppliedEconometrics,Winter2010INSTRUMENTALVARIABLES(2)------Background

AssumeestimatingequationbelowinOrdinaryLeastSquaresY=α+βX+eTheestimateofβ=E(Y’X)/E(X’X)=E((βX+e)’X)/

E(X’X)=β+E(e’X)/E(X’X)=βonlyifeandXareindependentButifeandXarecorrelatedthentheestimatedisbiased,andXiscalle

d“endogenous”(correlatedwiththeerror)---------------------NickBloom,AppliedEconometrics,Winter2010INSTRUM

ENTALVARIABLES(3)Thus,estimationofthefollowingwouldbebiased:Yi=a+b1Si+eibecauseSiandeiarecorrelatedaseiisafunctionofa

bilityE[b1]=E[Y’S]/E[S’S]=E[(β1Si+β2Ai+vi)’S]/E[S’S]=β1+E[(β2Ai+vi)’S]/E[S’S]=β1+β2E[Ai’S]/E[S’S]>β1So

becauseignoreability,whichiscorrelatedwithschooling,weoverestimatetheimpactofschoolingonearningsNickBloom,AppliedEconometrics,Winter2010INSTRUMEN

TALVARIABLES(4)Imaginewehadavariable–calledaninstrumentZ–thatwascorrelatedwithschoolingbutnotability.Wecouldthenusethistoexplainvaria

tioninschoolingasitisnotcorrelatedwithabilityOneexampleofthiswouldbeiftheGovernmentpaideveryonebornonevendaystostayinschoolThen“bornonanevenday”woul

dbeaninstrumentforschooling–correlatedwithschoolingbutnotabilityInpracticeinstrumentsareoftenhardtofindNickBloom,App

liedEconometrics,Winter2010INSTRUMENTALVARIABLES(5)AssumethatZiscorrelatedwithSbutnotA.Thenthefollowinginstrumentalvariableestim

atorisconsistentE[b1IV]=E[Y’Z]/E[S’Z]=E[(β1Si+β2Ai+vi)’Z]/E[S’Z]=E[β1Si’Z+β2Ai’Z+vi’Z]/E[S’Z]=β1+(β2E[Ai’S]+E[vi’Z])/E[S’Z]=β1Statawillcalculatethis

foryou.AllyouneedtofindisavariablethatonlyaffectsyourdependentvariableviathevariableyouareinterestedinNickBl

oom,AppliedEconometrics,Winter2010INSTRUMENTALVARIABLESAnyquestionsonthis?Imagineyouwantedtoevaluatetheimpactofcropyieldsonfarmersbehavior–cananyones

uggestagoodinstrumentNickBloom,AppliedEconometrics,Winter2010DefiningIdentificationExperimentsNaturalExperimentsInstrumentalvariablesEco

nometricIdentificationNickBloom,AppliedEconometrics,Winter2010ECONOMETRICIDENTIFICATIONAnotherwaytoobtainidentificationistrytomodeleveryt

hing•Forexample,weclaimweknowhowabilityiscorrelatedwithschoolingandsomodelthewholesystemTheproblemwiththisis:•Itisalotmorecomplicated•It

requiresstrongassumptionsThus,thisisusuallyonlyundertakenwhenthereisnoobviousinstrumentornaturalexperimentNickBloom,AppliedEconome

trics,Winter2010SUMMARYIdentification–understandingthecausalityinaregression–isessentialforgeneratingmeaningfulresultsTherearearangeofap

proaches–buttheyallneedsomeprioreconomicthought(i.e.istheiranaturalexperiment?)