The Multiplier Effect of Education Expenditure

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Multiplier Effect of Education Expenditure Maarten De Ridder, Simona M. Hannon, Damjan Pfajfar 2020-058 Please cite this paper as: De Ridder, Maarten, Simona M. Hannon, and Damjan Pfajfar (2020). “The Multiplier Effect of Education Expenditure,” Finance and Economics Discussion Series 2020-058. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2020.058. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

The Multiplier Effect of Education Expenditure* MaartenDeRidder‡ SimonaM.Hannon§ DamjanPfajfar¶ UniversityofCambridge FederalReserveBoard FederalReserveBoard May28,2020 Abstract Thispaperexaminestheshort-runeffectsoffederaleducationexpendituresonlocalincome. Weexploitcity-levelvariationinexposuretonationalchangesinthe$30-billionFederalPell Grant Program, which is the largest program to help low-income students attend college in theU.S.,tocalculatefiscalmultipliersofeducationexpenditures.AnincreaseinPellgrantsby 1percentofacity’sincomeraiseslocalincomeby2.4percentoverthenexttwoyears. This multipliereffectislargerthanestimatesformilitaryspending(1.5onaverage).Multipliersare higherwhengrantsareawardedtostudentsatnon-profitcolleges,asfor-profitcollegesabsorb mostofthegrantincreaseswithraisesintuition.Multipliersarealsohigherduringrecessions thaninexpansions: Pellgrantscanbeaneffectivetoolforcountercyclicalpolicythataddsto alreadyestablishedbenefits,suchas,increasingtheaffordabilityofcollegeandfosteringlongruneconomicgrowth. Keywords:FiscalExpenditure,PellGrants,EducationPolicy,FiscalMultipliers. JELclassification:H52,E62 *WethankKyleCoombs,RachaelBeer,HannahFarkasandWinstonLinforexcellentresearchassistance,andRobert (Bob)Adamsforhelpfulsuggestionsinreferencetothemetropolitanareageography. Wealsothanktheparticipants attheFederalReserveBoard,2018CEFconferenceinMilan,andAEFPconferenceinWashingtonfortheircomments andsuggestions. TheviewsinthispaperarethoseoftheauthorsanddonotnecessarilyreflectthoseoftheBoardof GovernorsoftheFederalReserveSystemoritsstaff. ‡Address: University of Cambridge, Faculty of Economics, Sidgwick Ave, Cambridge, CB3 9DD, U.K. E-mail: mcd58@cam.ac.uk.Web:http://www.maartenderidder.com/. §Address: BoardofGovernorsoftheFederalReserveSystem,20thStreetandConstitutionAveN.W.,Washington, D.C.20551,U.S.A.E-mail:simona.m.hannon@frb.gov. ¶Address: BoardofGovernorsoftheFederalReserveSystem,20thStreetandConstitutionAveN.W.,Washington, D.C.20551,U.S.A.E-mail:damjan.pfajfar@frb.gov.Web:https://sites.google.com/site/dpfajfar/.

1. Introduction Investmentsineducationmakeupasignificantpartofgovernmentspendinginadvancedeconomies. IntheUnitedStates,educationalspendingmeasured6percentofnationalincomein2019,which exceedsdefensespendingandspendingonwelfareprograms. Theseinvestmentsareusuallymotivated by the well-documented effects that education has on well-being and economic growth in the long run (see, e.g., Barro, 1991, Benhabib and Spiegel, 1994, Bils and Klenow, 2000, and ManuelliandSeshadri,2014). Likeanyotherformofgovernmentspending,however,educational investmentsalsohavethepotentialtostimulateeconomicactivityintheshortrun.Programsthat reducethecostoftuitionorthatinvolvedirecttransferstostudentscould,forexample,increase purchasingpowerandthereforeraiseconsumptionandgrowth. Suchprogramscouldbeusedto stimulate economic activity during recessions and serve as a tool for macroeconomic stabilization.Empiricalevidenceontheshort-runeffectsofeducationalinvestmentsoneconomicactivity isneededtoassesswhetherthisisthecase. Wequantifytheeffectofeducationalinvestmentsoneconomicgrowthintheshort-run.Specifically, we measure the impact of the Federal Pell Grant Program on economic activity at the city (metropolitanstatisticalarea–MSA)level. Pellgrantsareneed-basedgrantstolow-incomeundergraduateandselectpost-baccalaureatestudents,designedtoenablethemtoaccesspost-secondary education. It is the largest program to help low-income students attend college in the United States: total awards exceeded 30 billion U.S. Dollars in 2015 (Figure 1). We measure the effect ofthesegrantsoneconomicactivityusingcity-levelvariationinthedisbursementsofPellgrants. Inparticular,wequantifytheeffectofarelativeincreaseinPellgrantdisbursementsontherelative increase of a city’s aggregate income. To assure a causal interpretation of our results, we instrument city-level changes with changes in the national-level generosity of the Pell Grant Program. ThisfollowstheapproachbyNakamuraandSteinsson(2014),whoestimatetheeffectofdefense spendingongrowthattheU.S.statelevel.Ityieldsacausalinterpretationaslongaschangestothe national-levelgenerosityoftheprogramdonotdependontherelativeeconomicgrowthofcities. Figure1.ThePellGrantProgram:ExpendituresandRecipients Billions (2016 $) 40 Annual 30 20 10 0 1979 1985 1991 1997 2003 2009 2015 Notes:ThefigureplotsthetotalvalueofPellgrantsin2016USD.DataisobtainedfromtheTitleIVProgramVolume ReportsbytheDepartmentofEducation. 1

WefindthatPellgrantshaveasignificantlypositiveeffectoneconomicactivity.Ourmainresult isthatthefiscalmultiplierofPellgrants—thepercentageincreaseinacity’srelativeincomefrom arelativeincreaseinPellgrantsbyonepercentofinitialincome—is2.4onaverage. Thismeans thatadollarspentonPellgrantscreatesmorethantwiceasmuchrelativeeconomicactivity. This coefficient is robust to the inclusion of city and time fixed effects, city-time trends, controls for spendingbystategovernmentsandvariousothercontrolsforeconomicperformanceofacity.We findthatschoolsdonotincreaseexpenditureswhenthePellgrantprogrambecomesmoregenerous,whichsuggeststhatconsumerspendingratherthaneducationalspendingisthesourceofthe short-runeconomicgains.WhenwecomparetheeffectofPellgrantsthatarereceivedbystudents at for-profit institutions to Pell grants for students at non-profit institutions, we find that multipliersareloweratfor-profitcolleges. Thisispotentiallybecausefor-profitschoolsrespondtoan increaseinPellgrantgenerositybyproportionallyraisingtuitionfees.ItthereforeappearsthatPell grantsareimplicitlyactingassubsidiesforthefor-profituniversitysector.1 Wealsofindthat2-year institutionshavesignificantlylargermultipliersthat4-yearinstitutions.Multipliersat4-yearinstitutionsarearound1.6,whileat2-yearinstitutionsarearound4. Thus,Pellgrantshaveespecially high multipliers if granted to students attending public community colleges. Finally, we assess whetherthemultiplierofPellgrantsishigherduringrecessions,ashaspreviouslybeenfoundfor militaryexpenditure.2 WefindthattheeffectofPellgrantsonincomeislargerinrecessionsthanin expansions.Whilethateffectisnotstatisticallysignificant,thepointestimateislargeandsuggests thatPellgrantscanserveasamacroeconomicstabilizerduringrecessions. OurestimatesofthemultiplierofPellgrantsaddtoavastliteraturethatusesgeographiccrosssectionalvariationinfiscalspendingtoestimateitsshort-runeconomiceffects. TheuseofgeographicvariationinspendingbecameincreasinglypopularintheaftermathoftheGreatRecession. Theadvantageofusinggeographicalcross-sectionaldataisthatthereismuchgreatervariationin spendingatthesub-nationallevel, andmoreofthisvariationisplausiblyexogenousthanvariationatthenationallevel. Likeothersub-nationalestimationsofmultipliers, ourresultsdohave a particular interpretation: they measure the effect of Pell grants in one city on that city’s relativeeconomicperformance,ratherthantheeffectofPellgrantsoneconomicperformanceatthe nationallevel. Thisisbecauseacity-levelincreaseinPellgrantstypicallydoesnotinvolveanincreaseincity-levelfiscaldeficits(andsubsequenttaxation),suchthatPellgrantsdonotcrowdout privatespending.Chodorow-Reich(2019),however,arguesontheoreticalgroundsthatthekindof geographicalcross-sectionalmultiplierweestimateremainsinformative.Itmeasuresthenationallevelmultiplieroffiscalspendingwhenitisdeficitfinancedandwhenmonetarypolicydoesnot respondtothefiscalexpansion,forexamplebecauseinterestratesareconstrainedbytheeffective lowerbound. Astheseconditionsoftenapplyduringrecessions, ourresultsgiveinsightintothe effectivenessofPellgrantsasatooltostimulatedemandduringdownturnsatthenationallevel. 1Notethatsince2010the“GainfulEmployment”regulationhaslimitedthePellgrantsatcertainfor-profitcolleges (seeCellinietal.,2016).Ingeneral,Turner(2017)estimatesthat11-20%ofPellgrantspassesthroughtoschools. 2See,e.g.,NakamuraandSteinsson(2014),AuerbachandGorodnichenko(2012),andBergeetal.(2020). 2

The multiplier of 2.4 for Pell Grants is higher than most estimates based on cross-sectional geographicalvariationofotherformsofgovernmentspending.EarlyexamplesincludeNakamura andSteinsson(2014),whoestimatethestate-levelresponseofoutputtodefensespendingandfind anaveragemultiplierof1.5.Acconciaetal.(2014)estimatemultipliersfromreductionsinspending due the expulsion of mafia-infiltrated city council members, and find a multiplier of 1.9. CrosssectionalestimatesofthemultiplierwerealsofrequentlyusedtoassesstheeffectoftheAmerican Recovery and Reinvestment Act (ARRA) during the Global Financial Crisis (see, e.g., Chodorow- Reichetal.2012,Chhabraetal.2019,ConleyandDupor2013,DuporandMehkari2016,Feyrerand Sacerdote2011). Chodorow-Reich(2019)summarizestheliteratureoncross-sectionalmultipliers on both ARRA-based and other studies and finds that the mean estimated multiplier is 2.1 and the median estimated multiplier is 1.9. This suggests that the multiplier for Pell grants is high comparedtothemultiplierofotherformsofgovernmentspending, andisthereforeaneffective tooltostimulateshort-runeconomicactivity. In addition to providing evidence on the magnitude of the fiscal multiplier, this paper contributestotheliteratureonthePellGrantProgram. Previousworkhasdocumentedseveralother positiveeffects,inparticularinrelationtoeducationoutcomes.Bettinger(2004)showsthatreceivingaPellgrantreducescollegedrop-outbehavior. MarxandTurner(2018)showthatPellgrants substantiallyreduceborrowingaseveryadditionaldollarofPellgrantscrowdsout1.80dollarsof potentialborrowing,thereforereducingstudentdebt. Pellgrantsalsoincreaseeducationalattainment, the probability of attending college, credit accumulation and has positive effects on students persistence and degree completion (Dynarski, 2003, Castleman and Long, 2016, and Fack andGrenet,2015). Denningetal.(2019)showsthateligibilityforanadditionalPellgrantsignificantlyincreasesthelikelihoodofdegreereceiptandraisesearningsfouryearsafterthereceiptof thedegree.Ashigherearningsincreasetaxpayments,theyestimatethatthegovernmentexpendituresarefullyrepaidwithin10years. Dinersteinetal.(2014)lookattheshort-termeconomicbenefitsofPellgrantsaspartofvarious federal transfers to post-secondary education during the Global Financial Crisis. They find thatcountieswhichbenefitedfromincreasesinthegenerosityofthePellGrantProgramdidnot haveasignificantincreaseinlocalincome. Theyarguethatonereasonforthismaybethatstudentsdonotspendtheirgrantsintheimmediatevicinityoftheiruniversity. Ouranalysisdiffers fromDinersteinetal.’sbecauseweestimatethePellgrant’smultiplierfromvariationinPellgrants thatisdrivenbychangestonationalgenerosityinthePellGrantProgram,interactedwithcity-level dummiestoobtainvariationbyyear. Thelatterenablesustocontrolforstateandtime-fixedeffects, suchthatourestimateofthemultiplieriscausalifnationalgenerosityisnotdrivenbythe relative performance of cities. We furthermore consider a sample from 1990 to 2015 rather than 2006-2009,andconductourananalysisatthecity(MSA)ratherthanthecountylevel. 3

The remainder of this paper proceeds as follows. We begin by providing an overview of the PellGrantPrograminSection2,inwhichwealsoexplainourempiricalapproach. InSection3we discussourmainresultswhileinSection4wediscusshowmultipliersvaryoverthebusinesscycle andcomparemultipliersatdifferenttypesofcolleges.Section5concludes. 2. EmpiricalApproach This section outlines the empirical strategy to estimate the short-term economic effects of Pell grants. We start with a brief summary of the Pell grant program and how grants are allocated to students in Section 2.1. Section 2.2 summarizes the dataset while the estimation equations are presentedinSection2.3. 2.1. PellGrants:Background TheFederalPellGrantProgramwasinitiatedin1974astheBasicEducationalOpportunityGrant, toprovideaneed-basedgranttoenablelow-incomestudentstoattendcollege.Itwasrenamedthe PellGrantProgramafterSenatorClaibornePellin1980.Theevolutionoftheprogramisplottedin Figure2a. Itstartedoffasaprogramfor280thousandstudentsin1974withatotalappropriation of122milliondollars,whichincreasedtoover9millionrecipientsanda30billionappropriation by 2015. The program’s size depends on the size of the cohort receiving Pell grants and on the maximumgrantamountdeterminedbythelaw. Theprogramexpandedparticularlyrapidlyfrom theearly2000sto2010. Since2000,theU.S.haswitnessedasubstantialincreaseinenrolmentat post-secondary institutions and a marked increase in college tuition, both reflected in the nonprofitandthefor-profiteducationsectors. Federalsupportforhighereducationwasexpandedin ordertocompensateforthisincreasingcosts, whichleadtoanincreaseinboththeaverageand themaximumawardsforPellgrants(Figure2b). Thesewerepart,forexample,oftheCollegeCost ReductionandAccessActof2007andoftheAmericanRecoveryandReinvestmentActof2009.3 The size of individual grants primarily depends on a student’s family earnings. The largest shareofPellgrantdisbursementsaretypicallyreceivedbystudentsfromfamilieswithanadjusted grossincomeoflessthan$60,000.4 Thegrantamountsareconditionalonthestudent’sexpected family contribution (EFC), the institutional cost of attendance, the student’s enrollment status, andwhetherornottheyattendafullacademicyearorless.5 Afull-timestudentiseligibleforthe followingPellGrantawardifthemaximumPellGrant(PellMAX)higherthantheEFC: Pell =max (cid:169) (PellMAX −EFC ),PellMIN(cid:170) , (1) i,t t i,t t 3AfullsummaryoflegislativechangesisfoundinAppendixA. 4Forexample,96.6%ofPellGrantrecipientsin2011-12hadanincomeof$65,995orless(seeDelisle,2017). 5FinancialneedisdeterminedbytheDepartmentofEducationusingastandardformulaestablishedbyCongressto evaluatethetodeterminetheEFC.Theformulareliesonthestudent’sincome(andassetsforindependentstudents),the parents’incomeandassets(fordependentstudents),thefamily’shouseholdsize,andthenumberoffamilymembers (excludingparents)attendingpost-secondaryeducation. 4

Figure2.EvolutionoftheFederalPellGrantProgram Billions (2016 $) Millions Thousands (2016 $) 40 10 Annual Annual Total Pell Grant Expenditures (left axis) Maximum Pell Grant 6 Total Pell Grant Recipients (right axis) Average Pell Grant 30 Minimum Pell Grant 4 20 5 2 10 0 0 0 1979 1985 1991 1997 2003 2009 2015 1979 1985 1991 1997 2003 2009 2015 (a) AggregateAwards (b) IndividualAwards Notes: Leftfigureplotstheevolutionofthedollar-valueofPellgrantawards(leftaxis,solid)andthenumberof PellGrantrecipients(right-axis,dashed). Rightfigureplotstheminimum(dashed),average(dotted)andmaximum(solid)valueofanindividualPellgrant. DataisobtainedfromtheTitleIVProgramVolumeReportsbythe DepartmentofEducation. wherePellMIN istheminimumPellGrant.6 Oncethegrantamountisdetermined,theinstitution atwhichthestudentisenrolledeithercreditsthegrantfundstothestudent’saccount,paysthestudentdirectlybycheck,orcombinesthesemethods. Grantrecipientscanenrollatvarioustypesof institutions,rangingfrom4-yearcollegestothosespecializedinoccupationaltraining. Currently, about 5,000 post-secondary institutions participate in the program and more than 40 percent of allundergraduatesarerelyingonthistypeofaid. Themajorityofgrantrecipientsareenrolledat public2-yearschoolsandasignificantshareisenrolledatfor-profitinstitutions.Pellgrantsdonot typicallycovertheentirecostofattendanceand,asresultmostrecipientssupplementthistypeof aidwithfundsfromothersources,suchasfederaland/orprivatestudentloans,personalsavings, 529plansavings,andothersources. 2.2. Data Toestimatetheshort-runeconomiceffectsofthePellGrantProgramweanalyzeasampleof367 metropolitanareaswithdatafrom1990to2015.7 Dataonpersonalincomeisobtainedfromthe Bureau of Economic Analysis (BEA). We retrieve detailed employment data at the sector level by MSAfromtheBLS.WecombineannualdatafromDeltaCostwithquarterlyPellGrantinformation retrievedfromtheTitleIVProgramVolumereportspublishedbytheDepartmentofEducation.8 Thedataallowsustocontrolforasetofinstitutionallevelcharacteristics,suchasthenumberof undergraduate students and the average tuition fee, a dummy on whether the institution is forprofit,andavariableonwhethertheinstitutionprimarilyoffers2or4yeardegrees. Bothdatasets covernearlytheuniverseofAmericanhighereducationinstitutions. Weaggregatethedatatothe 6Awardsareroundedtothenearest$100.Part-timestudentsawardsarescaledbyafactorof0.5;scalefactorisused foralldeterminantsinEq.(1).Part-yearstudentsreceiveaproratedPellGrant. 7Thisislessthantheuniverseof382MSAs,asthosethatneverreceivePellgrantswereexcluded,aswellasMSAs thatreceivemorethan5%ofPelltotalPellgrants. 8DeltaCostProjectisanindependent,nonprofitorganization,whichputtogetherit’snamesakedataset,whichis basedondatafromtheIntegratedPostsecondaryEducationDataSystems(IPEDS). 5

Table1:SummaryStatistics Mean SD Obs. Min. Max. Source DependVariable ∆PersonalIncome(Biannual) 0.077 0.046 9355 -0.393 0.697 BEA Pellgrants GrowthinExpenditure-Metro 0.024 0.066 9355 -0.722 1.143 DeltaCost GrowthinExpenditure-National 0.175 0.311 9355 -0.210 1.235 DeltaCost ControlVariables Students(log) 9.606 1.271 9355 1.386 13.78 DeltaCost Students(%ofPopulation) 6.508 5.672 9355 0.004 112.5 DeltaCost Students(logchange) 0.033 0.164 9355 -4.222 4.425 DeltaCost Tuitionfee(log) 8.542 0.868 9355 4.605 10.77 DeltaCost ForProfit(%) 18.95 20.172 9355 0 100 DeltaCost Black(%ofPopulation) 11.011 14.686 9355 0.076 52.67 Census Hispanic(%ofPopulation) 11.109 27.405 9355 0.277 95.75 Census BachelorsDegree(%ofPop.) 9.868 8.531 9355 3.200 139.5 Census CreditCardUtilizationRate 27.363 6.257 6110 8.004 65.02 CCP(post1999) Age(Median) 46.574 4.79 6110 27.50 63.00 CCP(post1999) EquifaxRiskScore(Median) 700.8 34.43 6110 583.9 787.9 CCP(post1999) MortgageDelinq.(%) 5.527 4.807 6110 -7.601 64.4 CCP(post1999) TotalDebt(%ofIncome) 86.096 31.956 6110 11.247 299.2 CCP(post1999) StudentDebt(%ofIncome) 5.035 3.961 6110 0.127 67.25 CCP(post1999) Notes: Summarystatisticsforthemergedsample. Datafrom1990to2015covering376metropolitanareas. CCP standsforFederalReserveBankofNewYork/EquifaxConsumerCreditPanel. metropolitanarealevel.9 Financialcontrolvariablesareobtainedfroma10%sampleofadataset thatcovers5%oftheuniverseofEquifaxdataintheFederalReserveBankofNewYork’sConsumer CreditPanel(CCP).10 Weusethisdatasettocontrolforstudentandoveralldebt,medianEquifax RiskScore, mortgagedelinquencyandcreditcardutilization. Dataisavailableforthepost-1999 periodatquarterlyfrequency,whichweannualizebytakingaverages. DemographiccontrolvariablesforraceandeducationlevelsareretrievedfromtheCensusBureau. Summarystatisticsare providedinTable1. WeuseDeltaCostasourprimarysourceforPellgrantdatabecauseitcoverstheperiodbetween 1987and2015,whileofficialdatafromtheDepartmentofEducation(DoEd)onthefederalfunding programs is available only from 2000. One issue with Delta Cost data is that a small fraction of observationsisadjustedorimputed.11 TovalidatetheDeltaCostdatawecomparetheMSA-level DeltaCostdatawiththeavailableDoEddataaggregatedatthesamelevel.Thiscomparisonreveals 16areaswherePellgrantsfromDeltaCostdiffererraticallyfromtheDoEddata,whichweaddress intwowaysonacase-by-casebasis.First,forthecaseswhenoneyearofdataweremissingorone MSA-yearobservationwasconsideredsuspicious,weusedlinearinterpolationbasedontheDelta 9Weallocatebetween80%and87%ofallPellgrantsinagivenyeartotheMSAs. Theremainingisdistributedto ruralareas.Thissharehasbeenincreasingovertime. 10Weuseda10%sampleoftheCCPinordertoincreaseourgeographicalcoverage. Formoreinformationonthe CCP,seeLeeandderKlaauw(2010). 11Formoreinformation,seethedocumentationpostedontheirwebsite: https://www.deltacostproject.org/deltacost-project-database. 6

Figure3.RegionalVariation:ChangestoNationalandLocalPellGranttoIncomeRatio 3 Annual 2 1 0 −1 1995 1999 2003 2007 2011 2015 Notes: LinepresentsnationalpercentagepointchangeofPellGranttoGDPratio. Confidenceintervalscapture 90thto10thpercentileofpercentagepointchangeMSA-levelPellGranttoIncomeRatio,whilesquarespresent themedianchange. Costdata.Second,forthecaseswhenmultipleMSA-yearobservationswereeithermissingorwere questionable,weappliedthegrowthrateobservedintheDoEddatatoDeltaCostdata. Fromour sampleof367MSAswecorrectthepathofPellgrantsfor9usinginterpolationand10usingthe DoEdgrowthrate.Datafortheremaining348MSAswasnotsubjecttothisadjustment. 2.3. Strategy WhilethegenerosityandconditionalityofPellgrantsaredeterminedatthenationallevel,thereis significantvariationintheextenttowhichsub-nationalareasbenefitfromanincreaseinnationallevelPellgrantawards.Thisvariationisdrivenbythefactthatareasdifferinthenumberofeligible studentsthatareenrolledinpost-secondaryeducation. Acitywithalargenumberofuniversities benefitsmorefromanincreasethanacitywithoutuniversities,whileacitywhereasmallfraction ofitsstudentpopulationiseligible(e.g. becauseoftheleveloffamilyincome)benefitslessthan a city where a greater fraction is eligible, even if both cities inhibit a similar number of students overall. TheregionalvariationinPellgrantdisbursementsisillustratedinFigure3. Thesolidline plots the biannual growth rate of Pell grant disbursements at the national level, which is highly correlatedwiththemediangrowthofdisbursementsatthelocallevel,markedbysquares.12 The figure’sconfidenceintervals,whichplotthe90thand10thpercentileofthegrowthinPellGrants, showthatthereisconsiderableregionalvariation,andthatthisvariationisparticularlyhighinthe last10yearsofthesample. WeusethisvariationtoestimatetheeffectofanincreaseinthegenerosityofthePellGrantProgramonincomegrowth.Toperformthisanalysisatthenationalleveliscomplicated,becausePell grantsincreaseinresponsetoeconomicfluctuations. Enrolmentinuniversityiscounter-cyclical, tendingtoincreasewheneconomicperformanceispoor, forexample, causinganendogenously negativerelationshipbetweengrowthandthesizeofthePellGrantProgram. Infact,the2009in- 12Thecorrelationcoefficientbetweenthemedianmetropolitan-levelgrowthrateandthenationalgrowthrateis0.99. 7

creaseinthelevelofindividualPellgrantsaspartoftheAmericanRecoveryandReinvestmentAct wasexpresslyinresponsetopooreconomicperformanceduringtheGlobalFinancialCrisis. We overcomethisnational-levellimitationbyanalysingtheeffectofanincreaseingenerosityofthe Pell Grant Program at a the city (Metropolitan Statistical Area–MSA). Metropolitan areas are the appropriatelevelofanalysisbecauseavastmajorityofU.S.collegestudentsresideslocallywhere theirschoolislocated.13 Additionally,thereismorevariationinspendingacrossmetropolitanareasthanatotherlevelscommonlyusedintheestimationofmultipliers,likeatthestatelevel.14 ToobtainacausalestimateoftheeffectofPellgrantsonametropolitanarea’seconomicgrowth, wemuststilladdressthepossibilityofendogeneityinchangestolocalPell-grantawards.Increases inPellgrantawardsatthelevelofametropolitanareamayrespond,forexample,toanincreasein localcollegeenrolmentthatisdrivenbyadeteriorationoflocaleconomicconditions. Thisputsa downwardbiasontheestimates.Tosolvethis,weexploittwocharacteristicsofPellgrants:theprogramhasbecomesubstantiallymoregenerousduetonation-widelegislativechanges,andthereis strongheterogeneityintheextenttowhichtheadditionalgrantswereawardedacrossmetropolitan areas. Following Nakamura and Steinsson (2014)’s state-level approach for defense spending, we use these characteristics by instrumenting metropolitan changes in Pell Grant spending bychangesinnationalspending,withseparatecoefficientsforeacharea. WeestimatetheeffectofanincreaseinPellgrantdisbursementsalong: Y m,t −Y m,t−2 =β· E m,t −E m,t−2 +φ +ψ +γ(cid:48) X +µ , (2) m t m,t m,t Y m,t−2 Y m,t−2 whereY isthemacroeconomicvariableofinterestobservedinmetropolitanareaminyeart, m,t E isthetotaltransferofPellgrantstostudentsenrolledatschoolsinmetropolitanaream,X isa m,t vectorofcontrolvariablesthatincludesanarea-specifictimetrend,whileφ andψ ,respectively, m t denotefixedeffectsformetropolitanareasandyears.Allvariablesareinpercapitaterms.OurdependentvariablesforY ispersonalincome,whichisameasurethatcorrelateshighlywithGDP.15 Biannualchangesinvariablesareconsideredtomitigatethenoiseassociatedwithdifferencesbetweencalendaryearsandacademicyears(forwhichPellgrantsareassigned),andtoaccountfor thefactthataone-yearshocktospendingtendstoprecipitateinthesecondyear.16 To address the endogeneity concern that changes to local Pell grant disbursements are endogenous to local economic growth, we instrumentlocal Pell grant disbursements with changes innational-levelgenerosity,multipliedwithadummyforeachmetropolitanarea’sinordertocap- 13Also,accordingtothe2015DigestofEducationStatisticsTable309.10coveringstudentresidenceandmigration, 82percentoffirst-timedegree-seekingundergraduatestudentsattendcollegewithintheirstateofresidence. 14Overthecompletesample,theratioofPellgrantspendingtoGDPis0.16%acrossMSAswithastandarddeviation of0.17%,whilethatratiois0.12%atthestatelevelwithastandarddeviationofjust0.08%. 15MSA-levelGDPisonlyavailableafter2001whileourothervariablesstartin1990.Foryearsinwhichbothvariables areavailable,theircorrelationcoefficientequals0.997. 16RobustnesschecksusingalternativehorizonsareprovidedintheAppendix. 8

Figure4.LocalSensitivitytoNationalChangesinSupplyofPellgrants Notes:Sensitivitylevelreferstocoefficientϕ mfromequation(3)forthemetropolitanarea.Greyareasfalloutside metropolitanareasor,inrarecases,areareasthatneverreceivePellgrants. turedifferentialregionalsensitivitytochangesinPellgrantsatthenationallevel. Thefirst-stage equationreads: E m,t −E m,t−2 = (cid:88) (cid:49) i=m ·ϕ m · E N,t −E N,t−2 +ζ m +η t +τ(cid:48) X m,t +ε m,t , (3) Y m,t−2 i∈M Y N,t−2 wheresubscripts N identifynationalvaluesofthevariablesand(cid:49)denotestheindicatorfunction whichequalsoneifanobservationbelongstoMSAm. Coefficientsϕ capturethesensitivityof m changes to MSA-level Pell grants to changes in national-level grants, which gives cross-sectional variation to the otherwise constant within-period national changes in Pell grant disbursements. Thegeographicaldistributionofϕ acrossmetropolitanareasismappedinFigure4.17 Darkareas m attainthelargestincreaseinspendingwhennationalspendingincreases,whilelightareasattain the smallest increase in spending. The figure shows that the distribution of sensitivity does not exhibitclustering,andareaswithhighandlowsensitivityarefrequentlyneighbors. Thereissubstantialvariationintheestimatedcoefficientsϕ . Inmetropolitanareaswiththe m lowestsensitivitytonationalspendingthereisamodestdeclineinPellgrantspendingasnational spendingincreases. The 10th percentile ofϕ is -0.38 while the coefficientreads 1.8 at the 90th m percentile.ThismeansthataonepercentagepointincreaseinthePellgranttoGDPratioleadstoa 1.8percentagepointincreaseinthePellGranttoincomeratiointhatMSA.Theinterquartilerange is-0.10to0.89. Table2presentsmeanvaluesforanumberofcovariatesforMSAswithsensitivity 17Thecoefficientsareestimatedwhilecontrollingforchangesinstatespending,thepercentageofinhabitantsthatis astudent,anddemographiccontrols.Thissetofcontrolsisthelargestsetofcontrolsthatisavailableforthefullsample. AdiscussionisprovidedinSection3.1. 9

Table2:MeanValuesofObservablesforMSAswithLoworHighSensitivitytoNationalSpending LowSensitivity HighSensitivity Variable Mean St.Dev. Mean St.Dev. OutcomeVariables ∆Incomepercapita .079 .05 .078 .044 ∆Employmentrate .005 .031 .006 .03 EducationVariables Ratioundergr.studentstopop. .039 .021 .088 .069 Forprofit(%students) .065 .146 .036 .077 Tuition(average,log) 8.28 .892 8.11 .724 Undegraduatestudents(log) 9.36 1.57 9.75 1.03 DemographicControls Ratioblacktopop. .093 .089 .114 .119 Ratiohisp.topop. .091 .11 .095 .165 Ratiocollegedegreetopop. .097 .029 .087 .024 FinancialControls CreditCardutil.rate .268 .058 .285 .065 Age(median) 46.46 4.67 45.5 4.55 EquifaxRiskScore(median) 708.905 31.778 690.48 34.828 MortgageDel.30day .056 .051 .057 .048 Totaldebttoincome .903 .332 .842 .315 Studentdebttoincome .042 .035 .048 .037 Notes: TablepresentstheaverageandstandarddeviationofselectvariablesacrossMSAswithbelow(left)orabove (right)-averagesensitivityoflocaltonationalPellgrantdisbursements. Datafrom1990to2015foroutcome,educationanddemographicvariables. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/Equifax ConsumerCreditPanelandisavailablefrom1999to2015. levelsbelow(leftcolumns)orabove(rightcolumn)themediansensitivityϕ . Resultsshowthat m therearenomeaningfuldifferencesbetweentheaveragegrowthrateofpercapitaincomeratesof bothgroups. Themetropolitanareasarealsosimilarintermsofage,creditscores,creditcardutilizationandmortgagedelinquencyrates,aswellasracialcomposition.Theshareofthepopulation withaBachelor’sdegreeisslightlygreaterinareaswithlowsensitivityoflocaltonationalspending, whiledebt-to-incomeratiosandthefractionofschoolsoperatingforprofitarehigher. Thelargest differencebetweenbothgroupsofmetropolitanareasisintheratioofundergraduatestudentsto thepopulation. InareaswithabovemediansensitivitytonationalPellgrants,8.8%oftheinhabitantsareenrolledinanundergraduatedegree. Inareaswithbelowmediansensitivityonly3.9% ofpopulationisenrolledinanundergraduatedegree.Wecontrolforthesedifferencesinourmain analysis,andalwaysincludeMSAfixedeffectstocontrolforanytime-invariantdifferencesacross stateswithloworhighϕ . m OurestimatesofthemultiplieroftheFederalPellGrantProgramcorrespondtocoefficientβ in equation (2). This coefficient measures the relative increase in metropolitan area m’s income whenitachievesarelativeincreaseinPellgrantsasanincreaseinPellgrantsof1%oflocalincome. The estimate has a causal interpretation under the assumption that national changes to the Pell GrantProgramareorthogonaltotherelative economicperformanceofmetropolitanareas. This 10

is because we exploit two sources of variation for changes in local-level Pell disbursements: (1) thenational-levelchange,towhichtheidentificationrestrictioncorresponds,and(2)theaverage local-level Pell grant sensitivity to national changes. Any endogeneity in the latter, for example becauselow-growthcitieshavemorestudentsthatqualifyforPellgrants,iscontrolledforthrough theinclusionofmetropolitanareafixedeffectsintheregression. Weadditionallycontrolforthe share of students in a metropolitan area’s population. This assures that differences in local Pell grant spending are not driven by changes in a city’s student population. Students usually have below-averageincome,suchthatnotcontrollingforchangesinstudentpopulationcouldcausea negativebiasonmultiplierestimates. 3. Results Wenowproceedwiththemainestimationexercise. Section3.1presentsresultsforthefirststage andshowsthatnationaltrendsarearelevantinstrumentforlocalchangesinPellgrantdisbursements. Italsoexplainsthesequenceofcontrolvariablesthatweaddinthemainestimation. Section 3.2 presents the estimation of the multiplier of Pell grants along equation (2), as well as a numberofrobustnesschecks. 3.1. FirstStageandControlVariables Wefirstassesswhetherchangesinnation-widegenerosityofthePellGrantprogramarevalidinstrumentsforlocalgrantreceiptsbytestingtheinstrumentrelevancecondition. WedosobytestingthesignificanceoftherelationshipbetweenchangesinlocalPellgrantdisbursementandthe fittedvaluefromtheinteractionoftheestimatedϕ (seeeq.3)andchangesinnationalPellgrant m disbursements: (cid:88) (cid:49) ·ϕ · E N,t −E N,t−2 i=m m i∈M Y N,t−2 Results in Table 3 present the F-statistic for this term.18 Each column relies on a different set of controlvariables,whichmatchthesequenceofcontrolvariablesthatareusedinthemainmultiplierestimationpresentedinSection3.2. AllestimationscontrolforMSA-levelfixedeffects,year fixedeffects,alinearMSAtimetrend,andacontrolforchangesinstateappropriationsforhigher education. AnMSAtimetrendisincludedbecauseitincreasesstabilityoftheestimatedmultipliersoverthesampleinthesecondstage. Stateappropriationsareincludedbecausetheyinteract withPellgrants. Duringthe2009recession, forexample, stateappropriationsfellby29centsfor every dollar increase in federal research funds (Dinerstein et al. 2014). Some states even reduce appropriationsproportionallytoincreasesinPellgrants. Thishasanegativeeffectoneconomic 18Notethatthedegreetowhichnationalchangespredictlocalchangesnaturallydiffersacrosscities,asthisisthe sourceofvariationonwhichtheidentificationofourmultipliersforPellgrantsrelies. 11

Table3:FirstStage:EffectofNationalSpendingonLocalSpending FullSample Post1999 Income I II III IV V VI VII F-Statistic 26.39 26.21 26.40 26.40 25.31 25.43 25.43 P-Value 0 0 0 0 0 0 0 Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes StateSpending Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9355 9355 9355 8,997 6110 6110 6110 No.MSAs 367 367 367 367 365 365 365 Notes:MSAcontrols:numberofundergraduatestudents(log),changeinundergraduatestudents(log)last2years,averagetuition fee(log),for-profitpenetration,percentageofpopulationblack,percentageHispanic,percentagewithatleastabachelorsdegree. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/EquifaxConsumerCreditPanelandisavailablefrom1999to 2015. ItincludesmedianEquifaxRiskScore,age,debt-to-incomeratio,creditcardutilizationand30-daymortgagedelinquency rate.WeexcludeMSA-yearsreceivingmorethan5%ofTotalPellgrantsinagivenyear. activityandnotcontrollingforappropriationswouldthereforeleadtoanunderestimationofthe ceterisparibuseffectofPellgrantsonshort-termeconomicgrowth.19 Thefirstfourcolumnsincludecontrolvariablesthatareavailablefortheentire1990-2015sample. ColumnIpresentstheF-statisticofthefittedvalueforPellgrantsfromthespecificationwith thebasicsetofcontrols,whichishighlysignificant. Thenextcolumnsadditionallycontrolforthe fractionofthepopulationthatareundergraduatestudents;MSAcontrolsincludingaveragetuition andthepercentageofschoolsthatisfor-profit,demographiccontrolsconsistingofthepercentage of the population that is black, Hispanic, and the fraction that at least have a Bachelor’s degree; andthelagofbiannualgrowth. F-statisticsarehighlysignificantandsimilarinallspecifications. ThelastthreecolumnsofTable3arebasedonestimationsforashorter,post-1999sample. Thisis thesampleforwhichwehavecontrolvariablesonthefinancialpositionofhouseholdsintheMSA. Thesecontrols,whichweobtainfromEquifax,includethemedianEquifaxRiskScore,averageage, theaverageratiooftotaldebttoincome,creditcardutilizationratesand30-daymortgagedelinquencyrates. Thesecontrolsarelikelytoaffectlocalgrowthandarethereforerelevant, butthey reducethesampleto11years. 12

Table4:EffectofPellGrantsonLocalIncomePerCapita FullSample Post1999 FullSample Income I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS Multiplier 2.269* 2.441** 2.366** 2.144* 2.904** 2.617** 2.490** 0.906 (1.190) (1.207) (1.172) (1.258) (1.244) (1.144) (1.226) (0.890) Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes StateSpending Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,355 9,355 9,355 8,997 6,110 6,110 6,110 8,641 No.MSAs 366 366 366 366 364 364 364 366 Notes: Dependentvariableisbiannualgrowthofpercapitaincome. MultiplierequalscoefficientβinEquation2. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending. StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and***denotesignificanceatthe10,5,and1%level,respectively. Statespendingisinstrumentedexploitingthestate-levelsensitivitytonationaltrendsinstateappropriation, analogoustotheinstrumentsforPellgrants. MSAcontrols: numberofundergraduatestudents(log), changein undergraduatestudents(log)last2years,averagetuitionfee(log),for-profitpenetration,percentageofpopulation black,percentageHispanic,percentagewithatleastabachelorsdegree. DataonfinancialcontrolsisfromFederal ReserveBankofNewYork/EquifaxConsumerCreditPanelandisavailablefrom1999to2015. Itincludesmedian EquifaxRiskScore,age,debt-to-incomeratio,creditcardutilizationand30-daymortgagedelinquencyrate. WeexcludeMSA-yearsreceivingmorethan5%ofTotalPellgrantsinagivenyear. 3.2. MultiplierEstimates TheresultsforthemainestimationofthemultiplierofPellgrantsarepresentedinTable4. Multipliersrepresenttheestimatedcoefficientβfromequation(2).ThecolumnsareorderedasinTable 3. All columns include fixed effects for metropolitan areas and years, as well as an area-specific timetrendandchangestostateappropriations.Standarderrorsareclusteredtocorrectforwithinpanelcorrelationandheteroskedasticity. Thefirstcolumnshowsthattheestimatedmultiplieris 2.27inthebasespecification,whichmeansthatwhenlocalspendingincreasesby1percentofthe MSA’sincome,theMSA’sincomeincreasesby2.27percent. ColumnIIaddsacontrolfortheshare ofthepopulationthatareundergraduatestudentswhichraisestheestimateto2.44. ColumnIIIpresentsthepreferredspecification,whichincludesDeltaCostcontrolvariablesfor tuition,thelevelandchange(inlog)ofthenumberofstudents,thepercentageoffor-profitschools inthearea,andcontrolsforraceandeducationlevels.Byaddingcontrolsforthechangeinnumber ofstudent,ourresultsarenotaffectedbychangesinthenumberofstudentsreceivingPellgrantsin 19Reductionsinstateappropriationstendtohavenegativeeffectsonstudents. Webber(2017)showsthatforevery $1000perstudentstatebudgetcut,theaveragestudentpays$257moreintuitionandfees. Webber(2017)alsoshows thatthistrendhasincreasedovertime.Stateappropriationsforhighereducationarealsoshowntohaveanimpacton enrollmentandborrowing.GoodmanandVolz(2019)findthatchangesinappropriationsinducestudentstosubstitute betweenpublicandfor-profitcollegesandhavecorrespondingeffectsonstudentborrowing. 13

ametropolitanarea. Instead,theestimatescapturetheeffectofhavingahighershareofstudents that receive Pell grants. The estimated multiplier is 2.37. That is within the range of 1.3 to 2.5% estimated for military expenditure based on state-level data by Nakamura and Steinsson (2014). It is higher than the median (1.9) and the average (2.1) of multipliers found in previous studies thatrelyongeographiccross-sectionalvariationinotherformsoffiscalspending,assurveyedby Chodorow-Reich(2019).ColumnIVaddsalaggeddependentvariable,whichsomewhatlowersthe estimatedmultiplierto2.14,closetotheaverageinthestudiessummarizedbyChodorow-Reich.20 ColumnsVtoVIIareontheshorterpost-1999sampleforwhichwehavefinancialcontrolsfrom theFederalReserveBankofNewYork/EquifaxConsumerCreditPanel.InColumnVwereproduce ourpreferredspecification(III)fortheshortersample,andfindamultiplierof2.9. InColumnVI weaddthefinancialcontrolvariableswhichreducesthemultiplierto2.6,stillexceedingtheestimatesinColumnIII.ThehighermultipliermaybeexplainedbythefactthattheGreatRecession occurredinthesecondhalfofthesample; inSection4weshowthatPellgrantsgenerallyhavea largereffectonlocaleconomicactivityduringrecessions. InColumnVIIweaddlaggedgrowthas inspecification(IV),whichmodestlyreducesthemultiplierto2.5. Ourprimarydependentvariableisthechangeinlocalpersonalincome.Largelyforreasonsof dataavailability,apartofthefiscalmultiplierliteratureinsteadstudieshowspendingaffectsemployment. WealsoestimatetheeffectofPellgrantdisbursementsonlocalemployment,butfind smallereffects. Table5repeatstheanalysisofTable4usingthebiannualchangesintheemployment rate as the dependent variable. Most estimates of the fiscal multiplier on employment are positive,withthepreferredestimatearound0.75,butquitefarfrombeingstatisticallysignificant. Forcomparison,thesamespecificationformilitaryexpenditureatthestatelevelinNakamuraand Steinsson(2014)givesanemploymentratemultiplierof1.3. Ourestimateimpliesthatthecostof creatingajobthroughPellgrantsisaround$40,000.21 Pellgrantshavealargershort-runeffectonincomethantheliteratureusuallyfindsforother fiscalexpenditures,whiletheireffectonemploymentissmaller. Thewedgebetweentheeffectof Pell grants on income and employment can be explained by the fact that the direct effect of Pell grantsdoesnotworkthroughemployment. Thegrantsareatransferandcauseaone-for-oneincrease in personal income as measured by the BEA. Other forms of government spending, such asinfrastructuralinvestmentsordefensespending,usuallycauseadirectincreaseinemployment (e.g. thehiringofconstructionworkersormilitarypersonnel). Pellgrantsareacashtransfer, so theeffectonemploymentworksthroughtheconsumerspendingbystudentsthatreceivegrants.22 While this can create additional employment as a second round effect, the effect is more uncertainandfallsshortofthoseformsofspendingthatdirectlyaffectemployment(inlinewithour— 20OurpreferredspecificationdoesnotincludealaggeddependentvariablebecauseboththeAkaikeandBayesian InformationCriterionsuggestitshouldbeomitted. 21Thisnumberisfoundfromequation(2)usingthechangeinemploymentratesasthedependentvariable. The effectofPellgrantsonemploymentcountisgivenby∂Lt/∂Et =β(cid:98)·Lt−2/Yt−2. Insertingaveragepercapitapersonal incomeinthesample($29.700)andβ(cid:98)=0.745gives0.25jobsper$10,000. 22Weprovidesomeevidenceonthemechanisminthenextsection. 14

Table5:EffectofPellGrantsonEmploymentRate FullSample Post1999 FullSample Employment I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS Multiplier 0.746 0.694 0.745 0.534 0.753 0.417 0.002 -0.208 (0.911) (0.923) (0.893) (0.914) (0.943) (0.909) (0.925) (0.680) Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,354 9,354 9,354 8,995 6,109 6,109 6,107 9,355 No.MSAs 366 366 366 366 364 364 364 366 Notes: Dependentvariableisbiannualgrowthofemployment. MultiplierequalscoefficientβinEquation2. 2SLS regressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending.StandarderrorsclusteredbyMSAandgiveninparentheses.*,**,and***denotesignificanceatthe10,5,and1%level,respectively.MSA controls:numberofundergraduatestudents(log),changeinundergraduatestudents(log)lastyear,averagetuition fee(log),for-profitpenetration,percentageofpopulationblack,percentagehispanic,percentagewithatleastabachelorsdegree. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/EquifaxConsumerCreditPanel andisavailablefrom1999to2015.ItincludesmedianEquifaxRiskScore,age,debt-to-incomeratio,creditcardutilizationand30-daymortgagedelinquencyrate.WeexcludeMSA-yearsreceivingmorethan5%ofTotalPellgrantsin agivenyear. insignificant—pointestimateof0.75onemployment). Thisisalsoinlinewithpreviousevidence fromFeyrerandSacerdote(2011)thattransfersforeducationhavemodesteffectsonemployment. Ourresultssofarhavestudiedtheeffectofbiannualchangesinspendingonbiannualchanges inlocalincome. ThisispreferredoveraoneyearhorizonasPellgrantsaremeasuredbyacademic year,whileoutputismeasuredbycalendaryear. BiannualchangesthereforecapturealargerportionofactualchangestolocalPellgrants.Multiplierestimatesoverdifferenthorizonsareprovided inAppendixB.TablesA1andA2respectivelycalculateone-yearand4-yearmultipliersofPellgrants tolocalincome.One-yearmultipliersarepositiveandaround1forthefullsampleandaround2for thepost-1999sample,butgenerallynotsignificant.Conversely,wefindverylargemultipliersover 4-yearhorizons: estimatesrangefrom4.7to5.5,wellbeyondusualestimates. Tounderstandthe latter,considerthepathofPellgrantsattheMSAlevelafteraone-timeincreasealongthefollowing localprojection: E m,t+h −E m,t−1 =α h E m,t −E m,t−1 +φ m,h +ψ t,h +ε m,t+h , (4) Y m,t−1 Y m,t−1 forh=0,1,2,3,4. TheimpulseresponseisplottedinFigure5. APellgrantshockattime0isslightly amplifiedinthefirstyear:oneyearafteraone-percentagepointshockinthePellGranttoincome ratio,theratiohasincreasedby1.5percentagepoints.Theimpulseresponsefunctionsuggeststhat 15

Figure5.EffectofContemporaneousSpendingShockoverTime Spending 1.5 1.0 0.5 0 −0.5 0 1 2 3 4 Years from Shock Notes:y-axisplotscoefficientsα .Shadedareaidentifies95%confidenceinterval.Estimatesobtainedusing2SLS h instrumentingthefirstargumentof(4)withnationalvaluesinteractedwithMSAdummyvariables. theeffectofincreasedPellgrantsissignificantforthreeyears.Thispatternisinlinewiththe2to4 yeardurationofundergraduateprogramsatU.S.(community)colleges. Figure5impliesthatanincreaseinPellgrantswearsoffafteraroundfouryears. Ifeconomic activityrespondsmorepersistentlytotheinitialincrease,thecumulativeincreaseinoutputvastly exceedsthecumulativeincreaseinPellgrants(whichisnegligible). Thisleadstolargeestimates forthemultiplierovera4-yearhorizon. Weconcludethatourbaselinebiannualapproachisappropriatefortheanalysisoftheeffectsofeducationexpendituresoneconomicactivity. 4. WhenArePellGrantsMostEffective? WenextassessunderwhatconditionstheeffectofanincreaseinPellgrantdisbursementsonlocal economicactivityisthelargest. Todoso,welookathowtheeffectofPellgrantsdependsonthe stateoftheeconomywhendisbursementsareincreased,andwhethertheeffectofgrantsdepends onthetypeofinstitutionthatstudentsattend. 4.1. MultipliersinRecessionsandExpansions OurfirstestimationcomparesthemultiplierofPellgrantsduringepisodeswhentheeconomyisin expansiontowhenitisinrecession. Recentevidencesuggeststhatfiscalspendinggenerallyhasa greatereffectonoutputwhentheeconomyisinrecession.23 IfthisholdsforPellgrants,theycould form a particularly effective tool to stabilize macroeconomic activity. We estimate the following equationtotestthis: Y m,t −Y m,t−2 = F(z m,t−1 ) (cid:183) β R E m,t −E m,t−2 (cid:184) +[1−F(z m,t−1 )] (cid:183) β E E m,t −E m,t−2 (cid:184) Y m,t−2 Y m,t−2 Y m,t−2 +φ +ψ +γ(cid:48) X +µ , (5) m t m,t m,t 23ExamplesincludeAuerbachandGorodnichenko(2012),Corsettietal.(2012),Ilzetzkietal.(2013),Blanchardand Leigh(2013),JordàandTaylor(2016),andBergeetal.(2020). RameyandZubairy(2018)donotfindstate-dependence inahistoricalsamplewithnewsshocksaboutdefensespending. 16

whereβ andβ respectivelycapturethemultiplierinrecessionsandexpansions,whileF(z )is R E m,t acontinuousfunctionthatstrictlydecreaseswithlaggedbiannualgrowthz . t1 Thisequationisalsoknownasasmoothtransitionmodel,whichweborrowfromtheliterature onthestate-dependenteffectoffiscalandmonetarypolicyoneconomicactivity.24 Thespecificationassignsweighttoobservationsbasedonwhethertheeconomyisinrecessionorexpansion. Iflastyear’sgrowthwasrelativelyhigh,theobservationweighstowardsβ whileitweightsmore E towardsβ .FollowingTenreyroandThwaites(2016),F(z )isalogisticfunction: R m,t exp (cid:179) θ[zm,t −µ m] (cid:180) σ F(z )= m , (6) t 1+exp (cid:179) θ[zm,t −µ m] (cid:180) σ m where µ determines the fraction of the sample in which the metropolitan area is in recession, m σ givesthestandarddeviationofbiannualgrowthinwhileθdetermineshowstarkthedemarcam tionbetweenrecessionsandexpansionsare(e.g.,foralowerθ,theweightofobservationsismore equallysplitbetweenβ andβ ). µ iscalibratedsuchthateachareaisinrecession20%ofthe E R m sample, which matches the percent of quarters that the economy is in recession at the national levelaccordingtotheNBER.Wecalibrateθto3inlinewithTenreyroandThwaites(2016). ResultsarepresentedinTable6. Recessionmultipliersrepresentβ inequation(5)whileex- R pansionmultipliersrepresentβ . Therecession(expansion)shouldbeinterpretedasthe2-year E effectofarelativeincreaseinPellgrantsonrelativeincomegrowthifgrowthisinitiallyatitslowest (highest)levelinthedataset. TheactualmultiplierofanincreaseinPellgrantdisbursements depends on how close growth is to either of these levels. The estimations of columns I to V are estimatedwithtwo-stageleastsquaresregressions. PellgrantdisbursementsattheMSAlevelare instrumented with disbursements at the national level, multiplied by F(z m,t−1 ) for the first term and1−F(z m,t−1 )forthesecondterm. ColumnIcontainsthebasespecificationthatcontrolsfor metropolitanandyearfixedeffectsaswellasanarea-specifictimetrend. ColumnIIaddscontrol fortheshareofstudentsinthepopulationwhilecolumnIIIaddstheothernon-financialcontrol variables. ColumnIVrepeatstheregressionincolumnIIIonthepost-1999sample,whilecolumn VIaddsthefinancialcontrolvariables. Columnswithlaggeddependentvariablesareomittedas theyarecorrelatedwiththeassignmentfunction(F(z m,t−1 )). Estimatesforthemultiplierduring recessionsvaryfrom2.9to3.2(3.8whencontrollingforfinancialvariables),whileestimatesduring expansionsliebetween1.4to1.7(2.7forthepost-1999sample). Onlymultipliersinrecessionare statisticallysignificant. The difference between the multiplier when the economy is in expansion or recession rate rangesfromamodest0.5toaneconomicallyrelevant2.4,althoughthedifferenceisnotsignificant attheconventionallevels. Whileourestimatessuggestthattheeffectisnoisy,thelargeestimates 24SimilarspecificationsareusedbyAuerbachandGorodnichenko(2012),RameyandZubairy(2018),Tenreyroand Thwaites(2016),andDeRidderandPfajfar(2017). 17

Table6:State-DependenceofEducationSpendingMultiplier FullSample Post1999 FullSample Income I II III IV V VI 2SLS 2SLS 2SLS 2SLS 2SLS OLS RecessionMultiplier 2.889*** 3.092*** 3.165** 3.153** 3.784** 2.056 (1.473) (1.492) (1.490) (1.567) (1.524) (1.325) ExpansionMultiplier 1.609 1.728 1.539 2.701 1.416 0.131 (1.866) (1.880) (1.820) (1.984) (1.863) (1.359) Difference -1.280 -1.364 -1.626 -0.452 -2.367 -1.925 (2.370) (2.376) (2.351) (2.538) (2.503) (2.010) Controls MSAF.E. Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes StateSpending Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes FinancialControls Yes Observations 8,997 8,997 8,997 6,109 6,109 8,641 No.MSAs 366 366 366 364 364 366 Notes:Dependentvariableisbiannualgrowthofpercapitaincome.MultipliersfollowfromSmoothTransitionestimates. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending. StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and***denotesignificanceatthe10,5,and1%level, respectively. MSAcontrols: numberofundergraduatestudents(log),changeinundergraduatestudents(log)last2 years,averagetuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentageHispanic,percentagewithatleastabachelorsdegree. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/Equifax ConsumerCreditPanelandisavailablefrom1999to2015. ItincludesmedianEquifaxRiskScore, age, debt-toincomeratio,creditcardutilizationand30-daymortgagedelinquencyrate. WeexcludeMSA-yearsreceivingmore than5%ofTotalPellgrantsinagivenyear. oftherecessionmultipliersuggestthatPellgrantsareparticularlyeffectivewhengrowthislowand thattheycouldthereforebeatoolforcountercyclicalfiscalpolicy. 4.2. Institutions:For-ProfitvsNon-Profit Wenextassesswhethermultipliersdependonthetypeofinstitutionsthatthebeneficiaryattends. TheprevioussectionshaveshownthatPellgrantshavesubstantialmultipliers, especiallyduring recessions. OneobjectiontousingPellgrantsforcountercyclicalpolicymaybe,however,that15– 20%ofgrantsisspentatfor-profitcolleges(Figure6).25 Iffor-profitcollegeshavemarketpower, theymaybeabletochargehighertuitionfeesinresponsetohighergenerosityofPellgrants. Pell grantscanthereforeoperateasanimplicitsubsidy. Aspubliccompaniesownalargefractionof for-profitcolleges,notallofthesesubsidieswillbespentwithinthecollege’smetropolitanarea.26 25Thereductionafter2013istheresultof"GainfulEmployment"regulation.Thisregulationrestrictsfederalstudent aidatseveralfor-profitinstitutions(see,forexample,Cellinietal.,2016). 26Examplesofpubliclylistedcompaniesthatownfor-profitcollegesareGrandCanyonUniversity(LOPE),Adtalem (ATGE,previouslyDeVry),AmericanPublicUniversitySystem(APEI)andBridgepointEducationInc.(BPI). 18

Figure6.PercentageofPellgrantsAwardedtoFor-ProfitSchools Share 0.3 Annual 0.2 0.1 0.0 1987 1991 1995 1999 2003 2007 2011 2015 Notes: Figureplotsthefractionofnational-levelPellgrantsthatisawardedtostudentswhoareenrolledatforprofitinstitutions.DataisobtainedfromDeltaCost. Totestwhetherfor-profitschoolsindeedresponddifferentlytoariseinPellgrantswecompare multipliersatfor-profitandnon-profitinstitutions. Wefirstassesshowbothtypesofinstitutions changetheirtuitionfees.Weuseschool-levelmicrodataonenrollment,expendituresandrevenue sourcesfromDeltaCosttoperformthisanalysis.27 Wedefineaschool’stuitionfeeastheamount oftuitionreceiveddirectlyfromstudents,netofanygrantsor(institutional)studentaid,divided bythenumberoffull-timeequivalentstudents.Theestimationequationreads: τ i,t −τ i,t−2 =Γ Pell i,t −Pell i,t−2 +φ +ψ +µ , (7) τ τ i t i,t i,t−2 i,t−2 where τ is the average tuition rate at school i during academic year t, while Pell denotes the it amount of Pell grants received per full-time equivalent student. We look at biannual changes to matchthehorizonoverwhichweestimatethemultiplier. Theestimationof(7)issubjecttoendogeneitybecauseanincreaseindemandforschooling mayincreasebothtuitionfeesandthenumberofPellgrantsaschoolreceives.Toaddressthis,we instrumentchangesinschool-levelPellGrantwiththeinteractionbetweenaschooldummyand national changes in Pell grants analogous to our first-stage approach at the MSA level described in equation (3). In combination with the year and MSA fixed effects in (7), this enables a causal estimate ofΓ if the federal government does not change the national generosity of Pell grants in responsetoanindividualschool’stuitionfees. ResultsarepresentedinTable7. WhenPellgrantsincreaseasapercentageofthetotaltuition revenue,bothnon-profitandfor-profitschoolsincreasetheiraveragetuitionfees,buttheresponse ofthefor-profitschoolsismuchhigherthaninthecaseofnon-profitschools. Non-profitschools increasetheirtuitionfeesbyabout0.08%whenPellgrantsshareinthetotaltuitionincreaseby1 percentagepoint. For-profitschools,ontheotherhand,increasetheiraveragetuitionfeesby1.3 to1.6%. Thatmeansthatfor-profitschoolsraisetuitionfeesmorethanproportionallywhenthe grants increase. This confirms that Pell grants implicitly subsidize for-profit schools and do not increasethepurchasingpowerofthestudentsthatareawardedagrant. 27RelevantsummarystatisticsareprovidedinAppendixTableA10. 19

Table7:EffectofPellgrantsonTuitionFees For-Profit Non-Profit ∆Tuition (I) (II) (III) (IV) ∆Pellgrants(%Tuition) 1.558*** 1.258** 0.0795*** 0.0861*** (0.537) (0.504) (0.030) (0.023) SchoolF.E. Yes Yes Yes Yes SchoolTime-Trend No Yes No Yes YearF.E. Yes Yes Yes Yes Observations 16,408 16,408 75,873 75,873 No.Schools 1,574 1,574 3,580 3,580 Notes: Dependentvariableisbiannualgrowthinper-capitatuitionreceivedfromstudents. 2SLSregressionsuse nationalspendinginteractedwithschool-dummiestoinstrumentschool-levelgrants. Standarderrorsclusteredby schoolandgiveninparentheses. *, **, and***denotesignificanceatthe10, 5, and1%level, respectively. Both regressionscontrolforschoolandyearfixedeffects. ColumnsIandIIcontainresultsfromtheestimationof(7)on thesampleoffor-profitschools,whilecolumnsIIIandIVconducttheestimationonthesampleofnon-profitschools. Becausefor-profitschoolspreventstudentsfromgainingpurchasingpowerwhenPellgrants disbursements increase, grants may have a smaller effect on economic activity. To test this, we comparethemultiplierofPellgrantsthatareawardedtofor-profitschoolstothemultiplierofPell grantsthatareawardedtonon-profitschools. BecausethesemaybecorrelatedattheMSAlevel, weestimatebothmultipliersjointlyalong: Y m,t −Y m,t−2 =βFP E m FP ,t −E m FP ,t−2 +βNP E m NP ,t −E m NP ,t−2 +φ +ψ +γ(cid:48) X +µ , (8) m t m,t m,t Y m,t−2 Y m,t−2 Y m,t−2 whereEFP denotesthetotalamountofPellgrantsawardedtofor-profitschoolsinmetropolitan m,t area m in year t, while ENP denotes the amount awarded to non-profit schools. The first-stage m,t equationisaugmentedtoincludeboththenationaltrendsinnon-profitandfor-profitawards. Results are presented in Table 8. Control variables follow the same sequence as in Table 4. The for-profit multiplier of Pell grants estimates the multiplier effects of grants awarded to privatefor-profitschools,whilethenon-profitmultiplierestimatestheeffectsofPellgrantsatother schools. Byincludingbothestimatesinthesamespecificationwecontrolforthecorrelationbetweenawardsatbothtypesofschools. AswecanseeinTable8,themultipliersareconsiderably higherfornon-profitschoolsthanforfor-profitschoolswhereinmostcasesthedifferencebetween themultipliersishigherthen1.5,althoughisnotstatisticallysignificantatconventionallevelsdue to high standard errors. The multiplier for non-profit schools ranges from 3.6 to 4.1, while for for-profitschoolsfrom1.6to3.4. Multiplierinitiatedfromgrantstofor-profitschoolsare—except inColumnIII—significantonlyforthepost-1999sample. Thisimpliesthattherearenotabledifference in the policy transmission of the education spending depending on profit orientation of recipientschools.Multipliersinfor-profiteducationsectorareconsiderablysmaller. 20

Table8:EffectofPellGrantsonLocalIncomePerCapita:For-ProfitvsNon-Profit FullSample Post1999 FullSample Income I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS Non-ProfitMult. 3.682*** 3.815*** 3.576*** 4.093*** 4.006*** 3.600*** 4.008*** 1.109 (1.374) (1.388) (1.368) (1.463) (1.547) (1.366) (1.474) (0.991) For-ProfitMult. 1.614 1.925 2.625** 2.272 2.842** 3.363** 2.460** 0.0131 (1.294) (1.337) (1.162) (1.440) (1.130) (1.591) (1.176) (2.088) Difference 2.068 1.890 0.951 1.821 1.165 0.237 1.548 1.090 (1.666) (1.685) (1.537) (1.799) (1.607) (1.917) (1.617) (2.238) Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes StateSpending Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,348 9,348 9,348 9,348 6,103 6,103 6,103 9,348 No.MSAs 366 366 366 366 364 364 364 366 Notes: Dependentvariableisbiannualgrowthofpercapitaincome. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending.StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and***denotesignificanceatthe10,5,and1%level,respectively. MSAcontrols: numberofundergraduate students(log),changeinundergraduatestudents(log)last2years,averagetuitionfee(log),for-profitpenetration, percentageofpopulationblack,percentageHispanic,percentagewithatleastabachelorsdegree. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/EquifaxConsumerCreditPanelandisavailablefrom1999to 2015. ItincludesmedianEquifaxRiskScore,age,debt-to-incomeratio,creditcardutilizationand30-daymortgage delinquencyrate.WeexcludeMSA-yearsreceivingmorethan5%ofTotalPellgrantsinagivenyear. To understand the lower effect of for-profit Pell grant disbursements, we look at how educational expenditures respond when Pell grants increase. Pell grants can have a positive effect on growth because consumer spending increases or because schools spend additional income productively.Thetuitionhikeatfor-profitinstitutionssignalsthatthetransmissionthroughconsumer spendingwillbelow,whichcouldexplainthedifferenceinmultipliersinTable8. Toseehowexpenditures by colleges respond to a change in Pell grants, Table 9 estimates the effect on overall collegeexpenditures.28 Thedependentvariableisthebiannualchangeinoverallexpenditureas a percentage of aggregate personal income in the MSA, analogous to equation (2). Increases in Pellgrantsgenerositydonotsignificantlyincreaseoverallexpendituresatbothfor-profitandnonprofitinstitutions. Whilemultipliersfornon-profitsectorarepositive,andinsomecasescloseto beingsignificant,themultipliersatfor-profitsectorarevirtuallyzero.29 28TablesA3andA4inAppendixBconducttheestimationseparatelyforeducation-relatedexpendituresandother expenditures. 29Wealsofindnoevidenceofanincreaseinemploymentintheeducationsector.Resultsareavailableonrequest. 21

Table9:EffectofEducationSpendingonCollegeExpenditures:For-ProfitvsNon-Profit FullSample Post1999 FullSample OverallExpenditures I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS Non-ProfitMultiplier 0.426 0.421 0.423 0.417 0.261 0.225 0.324 0.500* (0.275) (0.278) (0.280) (0.293) (0.281) (0.290) (0.269) (0.274) For-ProfitMultiplier 0.024 0.007 0.025 -0.021 0.010 0.039 -0.015 0.113 (0.117) (0.118) (0.124) (0.102) (0.110) (0.119) (0.100) (0.130) Difference 0.775 0.784 0.772 0.867 0.631 0.531 0.675 0.381 (0.437) (0.437) (0.434) (0.437) (0.426) (0.409) (0.407) (0.493) Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes StateSpending Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,348 9,348 9,348 8,987 6,103 6,103 6,099 9,349 No.MSAs 366 366 366 365 364 364 363 366 Notes: Dependentvariableisannualgrowthoftotalexpendituresasapercentageofaggregatepersonalincomein theMSA.MultiplierequalscoefficientβinEquation2.2SLSregressionsusenationalspendinginteractedwithMSAdummiestoinstrumentlocalspending. StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and*** denotesignificanceatthe10,5,and1%level,respectively. MSAcontrols:numberofundergraduatestudents(log), changeinundergraduatestudents(log)lastyear,averagetuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentagehispanic,percentagewithatleastabachelorsdegree. Dataonfinancialcontrolsisfrom FederalReserveBankofNewYork/EquifaxConsumerCreditPanelandisavailablefrom1999to2015. Itincludes medianEquifaxRiskScore,age,debt-to-incomeratio,creditcardutilizationand30-daymortgagedelinquencyrate. WeexcludeMSA-yearsreceivingmorethan5%ofTotalPellgrantsinagivenyear. Dinerstein et al. (2014) found that public universities during the Great Recession increased their educational expenditures with the increase in the maximum Pell grants that occurred during2009/2010.Ourresults,however,suggestthatmostofthe“transmission"oftheincreaseofPell grantsdonothappenthroughcollegespending,butthroughstudentspendingasconsumers. To confirmthishypothesis,wealsochecktheeffectonemploymentineducationalsectorandfindno effectonemploymentinthissector. 4.3. Institutions:Two-YearvsFour-Year Analternativecharacteristicthatdiffersbetweencollegesiswhethertheyprimarilyoffer2-yearor 4-yeardegrees. Theshareoftheformerhassteadilyincreasedovertime: whileonly25%ofallPell grants were disbursed to 2-year institutions in 1987, this share has increased in the 80’s and 90’s andhasfluctuatedandbetween35and40%(Figure7). Theseinstitutionsareusuallycommunity collegesthatofferpost-secondaryeducationtolocalstudents,whoarelikelytospendtheirgrants 22

Table10:EffectofPellGrantsonLocalIncomePerCapita:Two-YearvsFour-YearSchools FullSample Post1999 FullSample Income I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS 4-yearMultiplier 1.468 1.610 1.434 1.644 2.030 1.604 1.663 1.225 (1.307) (1.323) (1.269) (1.303) (1.393) (1.174) (1.302) (1.232) 2-yearMultiplier 4.072** 4.357** 4.629** 4.592* 5.810*** 5.231** 5.879** 0.208 (2.025) (2.026) (2.017) (2.346) (2.176) (2.096) (2.380) (1.978) Difference -2.604 -2.746 -3.196 -2.948 -3.780 -3.628 -4.216 1.017 (2.345) (2.338) (2.300) (2.585) (2.449) (2.310) (2.630) (2.617) Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes StateSpending Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,402 9,402 9,402 9,044 6,145 6,145 6,145 9,403 No.MSAs 367 367 367 367 365 365 365 366 Notes: Dependentvariableisbiannualgrowthofpercapitaincome. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending.StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and***denotesignificanceatthe10,5,and1%level,respectively. MSAcontrols: numberofundergraduate students(log),changeinundergraduatestudents(log)last2years,averagetuitionfee(log),for-profitpenetration, percentageofpopulationblack,percentageHispanic,percentagewithatleastabachelorsdegree. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/EquifaxConsumerCreditPanelandisavailablefrom1999to 2015. ItincludesmedianEquifaxRiskScore,age,debt-to-incomeratio,creditcardutilizationand30-daymortgage delinquencyrate.WeexcludeMSA-yearsreceivingmorethan5%ofTotalPellgrantsinagivenyear. inthemetropolitanoftheirschool. Figure7.PercentageofPellgrantsAwardedat2-yearInstitutions Share Annual 0.4 0.2 0.0 1987 1991 1995 1999 2003 2007 2011 2015 Notes:Figureplotsthefractionofnational-levelPellgrantsthatisawardedtostudentswhoareenrolledat2-year institutions.DataisobtainedfromDeltaCost. 23

Table11:EffectofPellGrantsonCollegeExpenditures:Two-YearvsFour-Year FullSample Post1999 FullSample OverallExpenditures I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS 4-yearMultiplier 0.577* 0.569* 0.569* 0.581* 0.410 0.362 0.470 0.551** (0.334) (0.337) (0.337) (0.351) (0.325) (0.323) (0.316) (0.263) 2-yearMultiplier -0.197 -0.215 -0.203 -0.287 -0.221 -0.168 -0.206 0.171 (0.231) (0.233) (0.237) (0.238) (0.227) (0.212) (0.217) (0.462) Difference 1.092 1.096 1.084 1.151 0.909 0.801 0.910 0.528 Error 0.509 0.508 0.504 0.523 0.452 0.441 0.451 0.509 Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes StateSpending Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,402 9,402 9,402 9,040 6,145 6,145 6,141 9,403 No.MSAs 367 367 367 366 365 365 364 366 Notes: Dependentvariableisannualgrowthoftotalexpenditures. MultiplierequalscoefficientβinEquation2. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending.Standarderrors clusteredbyMSAandgiveninparentheses.*,**,and***denotesignificanceatthe10,5,and1%level,respectively. MSAcontrols: numberofundergraduatestudents(log),changeinundergraduatestudents(log)lastyear,average tuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentagehispanic,percentagewithatleast abachelorsdegree. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/EquifaxConsumerCredit Panelandisavailablefrom1999to2015.ItincludesmedianEquifaxRiskScore,age,debt-to-incomeratio,creditcard utilizationand30-daymortgagedelinquencyrate.WeexcludeMSA-yearsreceivingmorethan5%ofTotalPellgrants inagivenyear. To test whether multipliers are different, we estimate equation (8) with Pell grant disbursements to two and 4-year institutions rather than for-profit and non-profit institutions. Table 10 presentstheresults. Thereisaconsiderabledifferencebetweentheestimatedmultipliersacross 2-year and 4-year institutions. Most estimates of the multiplier for the full sample are around 4 for the 2-year institutions and around 1.6 for 4-year institutions, while for the short sample the 2-year institutions multiplier increases to above 5 and 4-year institutions multiplier to about 1.6 to2.0. Thedifferencebetweenthe4-yearand2-yearinstitutionsmultipliersissignificantforthe post-1999sample. Themultiplierfor4-yearinstitutionsisinsignificant,whichislikelyduetothe positivecorrelationbetweentheinstrumentsforthetwoand4-yearPellgrants(0.26). Thisraises thestandarderrorsintheestimation. Thepoint-estimateofthemultiplierinourpreferredspecification(III)is1.43,0.94pointsbelowtheaveragemultiplieracrossinstitutions. The analysis in this section leads us to conclude that Pell Grants for attending public 2-year institutions are likely to generate the largest positive short-run effects on economic activity. We nextassesswhetherthisisduetodifferencesinthechangeineducationalspendingbythesetypes 24

ofinstitutions. Table11presentstheresults,whichisanalogoustoTable9forprofitvsnon-profit. The table shows that at 4-year colleges, an increase in Pell grants leads by 1% of local personal incomeleadstoanincreaseineducationalspendingby0.3to0.6%oflocalpersonalincome. In contrast,2-yearinstitutionsdonotincreasetheirexpendituresinresponsetotheincreaseofPell grants. This suggests that the positive multipliers that we estimated in Table 10 are not due to the increase of college expenditures, but likely due to a consumer spending effect. In Appendix B’s Tables A5 and A6 we further look at the education and non-education expenditures and we observethatmostoftheincreaseinoverallexpendituresinTable11areduetotheincreaseofnoneducationexpendituresat4-yearinstitutions,althoughinthepost-1999samplethishassomewhat shiftedtowardeducationexpenditures. 5. Conclusion ThispaperestimatestheeffectoftheFederalPellGrantProgramonshort-runeconomicactivity. Specifically,weassesshowarelativeincreaseinPellgrantdisbursementsatthemetropolitanarea raises the area’s relative income. To do so, we exploit the fact that areas differ in the degree to which their disbursements respond to changes in the national-level generosity of the Program. ThisgivesacausalestimateofthemultiplierofPellgrants,undertheassumptionthatthenationallevelgenerositydoesnotrespondtotherelativeperformanceofmetropolitanareas. Wefindanaveragemultiplierofaround2.4inthemainspecification. Thisimpliesthata1% increaseinPellgrantsasafractionoflocalincomeraiseslocalincomeby2.4%.Thisishigherthan the average (2.1) and median (1.9) estimate of the multiplier from geographical cross-sectional data of other forms of fiscal spending found in the literature. We also find that multipliers are higher when the economy is in recession. Our results imply that having beneficial effects in the longrun,educationalinvestmentscanalsobeusedforcountercyclicalfiscalpolicy. Our findings also have implications for education policy. We find that for-profit institutions raisetuitionfeesinresponsetoanincreaseinPellgrantdisbursementsandthatPellgrantsdonot increaseeithereducationornon-educationexpendituresattheseschools. Conversely,non-profit schoolsonlyincreasetuitionbyasmallfractionoftheincreaseinPellgrants.Thisvalidatesrecent restrictionsimposedontheeligibilityofstudentsatfor-profitinstitutionsforPellgrants. Finally, weshowthat2-yearinstitutionshavesignificantlylargermultipliersthat4-yearinstitutions. Pell grantsarethereforeparticularlyeffectiveasatoolforcountercyclicalpolicyifgrantedtostudents attendingpubliccommunitycolleges. 25

References Acconcia, A., Corsetti, G., and Simonelli, S. (2014). Mafia and public spending: Evidence on the fiscalmultiplierfromaquasi-experiment. AmericanEconomicReview,104(7):2185–2209. Auerbach, A. J. and Gorodnichenko, Y. (2012). Measuring the output responses to fiscal policy. AmericanEconomicJournal:EconomicPolicy,4(2):1–27. Barro, R. J. (1991). Economic Growth in a Cross Section of Countries. The Quarterly Journal of Economics,106(2):407–443. Benhabib,J.andSpiegel,M.M.(1994). Theroleofhumancapitalineconomicdevelopmentevidencefromaggregatecross-countrydata. JournalofMonetaryEconomics,34(2):143–173. Berge,T.J.,DeRidder,M.,andPfajfar,D.(2020). Whenisthefiscalmultiplierhigh? acomparison offourbusinesscyclephases. FEDSWorkingPaper. Bettinger, E. (2004). How financial aid affects persistence. In College choices: The economics of wheretogo,whentogo,andhowtopayforit,pages207–238.UniversityofChicagoPress. Bils, M. and Klenow, P. J. (2000). Does schooling cause growth? American Economic Review, 90(5):1160–1183. Blanchard,O.J.andLeigh,D.(2013). Growthforecasterrorsandfiscalmultipliers. TheAmerican EconomicReview,103(3):117–120. Castleman, B. L. and Long, B. T. (2016). Looking beyond enrollment: The causal effect of needbased grants on college access, persistence, and graduation. Journal of Labor Economics, 34(4):1023–1073. Cellini,S.R.,Darolia,R.,andTurner,L.J.(2016). WhereDoStudentsGowhenFor-ProfitColleges LoseFederalAid? NBERWorkingPapers22967,NationalBureauofEconomicResearch,Inc. Chhabra,Y.,Levenstein,M.C.,andOwen-Smith,J.(2019). Thelocaleconomicimpactofscience spending:Evidencefromtheamericanrecoveryandreinvestmentact.WorkingPaper1383,Ross SchoolofBusiness. Chodorow-Reich,G.(2019). Geographiccross-sectionalfiscalspendingmultipliers:Whathavewe learned? AEJ:Policy,11(2):1–34. Chodorow-Reich, G., Feiveson, L., Liscow, Z., andWoolston, W.G.(2012). Doesstatefiscalrelief during recessions increase employment? evidence from the american recovery and reinvestmentact. AmericanEconomicJournal:EconomicPolicy,4(3):118–45. 26

Conley,T.G.andDupor,B.(2013). Theamericanrecoveryandreinvestmentact: Solelyagovernmentjobsprogram? JournalofMonetaryEconomics,60(5):535–549. AggregateImplicationsof LocalPublicFinance. Corsetti,G.,Meier,A.,andMüller,G.J.(2012).Whatdeterminesgovernmentspendingmultipliers? EconomicPolicy,27(72):521–565. DeRidder, M.andPfajfar, D.(2017). Policyshocksandwagerigidities: Empiricalevidencefrom regionaleffectsofnationalshocks. WorkingPaper. Delisle, J.(2017). Thepell grantproxy: Aubiquitousbutflawedmeasureoflow-incomestudent enrollment. EvidenceSpeaksReports,2(26):1–12. Denning,J.T.,Marx,B.M.,andTurner,L.J.(2019). Propelled: Theeffectsofgrantsongraduation, earnings,andwelfare. AmericanEconomicJournal:AppliedEconomics,11(3):193–224. Dinerstein,M.F.,Hoxby,C.M.,Meer,J.,andVillanueva,P.(2014). DidtheFiscalStimulusWorkfor Universities?,pages263–320. UniversityofChicagoPress. Dupor,B.andMehkari,M.S.(2016).The2009recoveryact:Stimulusattheextensiveandintensive labormargins. EuropeanEconomicReview,85:208–228. Dynarski,S.M.(2003). Doesaidmatter?measuringtheeffectofstudentaidoncollegeattendance andcompletion. AmericanEconomicReview,(93):179–88. Fack,G.andGrenet,J.(2015). Improvingcollegeaccessandsuccessforlow-incomestudents:Evidencefromalargeneed-basedgrantprogram. AmericanEconomicJournal:AppliedEconomics, 7(2):1–34. Feyrer, J.andSacerdote, B.(2011). Didthestimulusstimulate? realtimeestimatesoftheeffects oftheamericanrecoveryandreinvestmentact. WorkingPaper16759,NationalBureauofEconomicResearch. Goodman,S.andVolz,A.H.(2019). Attendancespilloversbetweenpublicandfor-profitcolleges: Evidence from statewide variation in appropriations for higher education. Education Finance andPolicy,0(ja):1–67. Ilzetzki,E.,Mendoza,E.G.,andVégh,C.A.(2013). Howbig(small?)arefiscalmultipliers? Journal ofmonetaryeconomics,60(2):239–254. Jordà,Ò.andTaylor,A.M.(2016). Thetimeforausterity: estimatingtheaveragetreatmenteffect offiscalpolicy. TheEconomicJournal,126(590):219–255. Lee,D.andderKlaauw,W.V.(2010). AnintroductiontotheFRBNYConsumerCreditPanel. Staff Reports479,FederalReserveBankofNewYork. 27

Manuelli,R.E.andSeshadri,A.(2014). Humancapitalandthewealthofnations. AmericanEconomicReview,104(9):2736–62. Marx, B. M. and Turner, L. J. (2018). Borrowing trouble? human capital investment with opt-in costsandimplicationsfortheeffectivenessofgrantaid. AmericanEconomicJournal: Applied Economics,10(2):163–201. Nakamura, E. and Steinsson, J. (2014). Fiscal Stimulus in a Monetary Union: Evidence from US Regions. AmericanEconomicReview,104(3):753–92. Ramey,V.A.andZubairy,S.(2018). GovernmentSpendingMultipliersinGoodTimesandinBad: EvidencefromU.S.HistoricalData. JournalofPoliticalEconomy,126(2). Tenreyro,S.andThwaites,G.(2016). Pushingonastring: Usmonetarypolicyislesspowerfulin recessions. AmericanEconomicJournal:Macroeconomics,8(4):43–74. Turner,L.J.(2017). TheEconomicIncidenceofFederalStudentGrantAid. mimeo,Universityof Maryland. Webber,D.A.(2017). Statedivestmentandtuitionatpublicinstitutions. EconomicsofEducation Review,60(C):1–4. 28

Appendix A. MainLegislativeChangestothePellGrantProgram First,in1978,theMiddleIncomeStudentAssistanceAct(MISAA)expandedstudenteligibilityby limitingtherateatwhichparentaldiscretionaryincomewasassessedundertheEFCformula.This actwasrepealedtwoyearslater, in1980. In1990, theOmnibusBudgetReconciliationActeliminatedstudentaideligibilityathighdefaultschools. In1992, theHigherEducationActwasreauthorizedandchangedthedefinitionofanindependentstudent.In1994,theViolentCrimeControl andLawEnforcementActeliminatedPellgrantsforprisoners. In2007CongresspassedtheCollegeCostReductionandAccessAct(CCRAA),whichsupplementedthegrantfundingandchanged PelleligibilitybyincreasingtheamountandtypesofincomeexcludedfromtheEFCformula.Arenewedsetoflegislativemeasurespairedwiththecountercyclicalityoftheenrollmenteffectcaused asignificantincreaseinPellGrantdisburements. Theselegislativemeasuresinclude: theHigher EducationOpportunityAct(HEOA)of2008,whichauthorizedyear-roundPellgrantsandlimited eligibility to 18 full-time semesters or the equivalent; the American Recovery and Reinvestment Act (ARRA) of 2009, which provided additional funding to the Pell Grant program (ARRA raised themaximumPellGrantbymorethan$400);theHealthCareandEducationReconciliationActof 2010,whichincreasedthemaximumPellgrantbyover$600andexpandedeligibilitybyincreasing theincomethreshold(from$20,000to$30,000)foranautomaticEFCofzero. PellGrantdisbursements started to decline in 2011, once the economy gained momentum and undergraduate enrollmentreturnedtopre-crisislevels.30 Congresseliminatedtheyear-roundPellGranteligibility established in 2008, whenitprovidedsupplemental funding to the programandlowered the incomethresholdforanautomaticEFCofzeroto23,000. In2012,theConsolidatedAppropriations ActprovidedadditionalfundingtothePellGrantprogramandreducedPelllifetimeeligibilityto12 semesters. 30Duringeconomicrecovery,fewerindividualsqualifytoreceivePellgrants.Enrollmentdecreasesaspeopleoptfor employmentinsteadofeducation. 29

B. AdditionalTables TableA1:EffectofPellGrantsonLocalIncomePerCapitafor1YearHorizon FullSample Post1999 FullSample Income I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS Multiplier 1.043 1.126 1.084 1.077 2.115* 1.952* 2.093* 0.418 (1.067) (1.083) (1.077) (1.049) (1.128) (1.083) (1.083) (0.765) Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,354 9,354 9,354 9,354 6,109 6,109 6,109 9,355 No.MSAs 367 367 367 367 365 365 365 367 Notes: Dependentvariableisannualgrowthofpercapitaincome. MultiplierequalscoefficientβinEquation2. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending.Standarderrors clusteredbyMSAandgiveninparentheses.*,**,and***denotesignificanceatthe10,5,and1%level,respectively. MSAcontrols: numberofundergraduatestudents(log),changeinundergraduatestudents(log)lastyear,average tuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentagehispanic,percentagewithatleast abachelorsdegree. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/EquifaxConsumerCredit Panelandisavailablefrom1999to2015.ItincludesmedianEquifaxRiskScore,age,debt-to-incomeratio,creditcard utilizationand30-daymortgagedelinquencyrate.WeexcludeMSA-yearsreceivingmorethan5%ofTotalPellgrants inagivenyear. 30

TableA2:EffectofPellGrantonLocalIncomePerCapitafor4YearHorizon FullSample Post1999 FullSample Income I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS Multiplier 5.048*** 5.498*** 4.974*** 0.838 5.346*** 4.773*** 0.898 0.641 (1.627) (1.642) (1.545) (1.231) (1.678) (1.533) (1.112) (1.281) Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 8,993 8,993 8,993 7,552 6,105 6,105 6,105 8,994 No.MSAs 366 366 366 366 364 364 364 366 Notes:Dependentvariableis4-yeargrowthofpercapitaincome. MultiplierequalscoefficientβinEquation2with 4-yearPellGrantchanges. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrument localspending. StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and***denotesignificanceatthe 10,5,and1%level,respectively. MSAcontrols: numberofundergraduatestudents(log),changeinundergraduate students(log)lastyear,averagetuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentage hispanic,percentagewithatleastabachelorsdegree.DataonfinancialcontrolsisfromFederalReserveBankofNew York/EquifaxConsumerCreditPanelandisavailablefrom1999to2015.ItincludesmedianEquifaxRiskScore,age, debt-to-incomeratio,creditcardutilizationand30-daymortgagedelinquencyrate.WeexcludeMSA-yearsreceiving morethan5%ofTotalPellgrantsinagivenyear. 31

TableA3:EffectofPellGrantonNon-EducationExpendituresbyColleges FullSample Post1999 FullSample Non-educationexp. I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS Non-ProfitMultiplier 0.167 0.172 0.171 0.180 0.0449 0.0335 0.0837 0.122 (0.132) (0.132) (0.134) (0.119) (0.167) (0.159) (0.142) (0.131) For-ProfitMultiplier -0.00440 0.00690 0.0181 -0.0161 0.0251 0.0287 0.0107 0.00887 (0.0713) (0.0718) (0.0738) (0.0578) (0.0755) (0.0702) (0.0610) (0.0562) Difference 0.171 0.165 0.153 0.196 0.0198 0.00487 0.0730 0.113 Error 0.153 0.152 0.153 0.138 0.187 0.178 0.161 0.144 Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes StateSpending Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,297 9,297 9,297 8,927 6,097 6,097 6,085 9,298 No.MSAs 363 363 363 361 363 363 361 363 Notes: Dependentvariableisbiannualgrowthofnon-educationalexpenditures. Multiplierequalscoefficientβin Equation2. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending. StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and***denotesignificanceatthe10,5,and1% level,respectively. MSAcontrols: numberofundergraduatestudents(log),changeinundergraduatestudents(log) lastyear,averagetuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentagehispanic,percentagewithatleastabachelorsdegree.DataonfinancialcontrolsisfromFederalReserveBankofNewYork/Equifax ConsumerCreditPanelandisavailablefrom1999to2015. ItincludesmedianEquifaxRiskScore, age, debt-toincomeratio,creditcardutilizationand30-daymortgagedelinquencyrate. WeexcludeMSA-yearsreceivingmore than5%ofTotalPellgrantsinagivenyear. 32

TableA4:EffectofPellGrantsonEducationExpendituresbyColleges FullSample Post1999 FullSample Educationexp. I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS Non-ProfitMultiplier 0.259 0.249 0.253 0.240 0.216 0.192 0.241 0.381 (0.216) (0.218) (0.221) (0.240) (0.194) (0.205) (0.192) (0.282) For-ProfitMultiplier 0.0280 0.000107 0.00987 -0.00899 -0.0154 0.0101 -0.0302 0.108 (0.0760) (0.0798) (0.0826) (0.0771) (0.0617) (0.0745) (0.0669) (0.106) Difference 0.232 0.249 0.243 0.249 0.231 0.182 0.272 0.273 Error 0.238 0.237 0.238 0.253 0.203 0.223 0.203 0.301 Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes StateSpending Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,297 9,297 9,297 8,927 6,097 6,097 6,085 9,298 No.MSAs 363 363 363 361 363 363 361 363 Notes:Dependentvariableisbiannualgrowthofeducationalexpenditures. MultiplierequalscoefficientβinEquation2. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending. StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and***denotesignificanceatthe10,5,and1%level, respectively. MSAcontrols: numberofundergraduatestudents(log),changeinundergraduatestudents(log)last year,averagetuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentagehispanic,percentagewithatleastabachelorsdegree. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/Equifax ConsumerCreditPanelandisavailablefrom1999to2015. ItincludesmedianEquifaxRiskScore, age, debt-toincomeratio,creditcardutilizationand30-daymortgagedelinquencyrate. WeexcludeMSA-yearsreceivingmore than5%ofTotalPellgrantsinagivenyear. 33

TableA5:EffectofPellGrantsonNon-EducationExpendituresbyColleges FullSample Post1999 FullSample Non-educationexp. I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS 4-yearMultiplier 0.371** 0.378** 0.373** 0.366** 0.222 0.208 0.256* 0.310** (0.161) (0.160) (0.159) (0.160) (0.162) (0.150) (0.151) (0.128) 2-yearMultiplier -0.219* -0.206 -0.199 -0.218 -0.176 -0.175* -0.196* -0.290** (0.131) (0.130) (0.134) (0.140) (0.116) (0.102) (0.114) (0.132) Difference 0.590 0.584 0.572 0.585 0.398 0.383 0.452 0.600 Error 0.225 0.224 0.224 0.232 0.217 0.202 0.220 0.185 Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes StateSpending Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,351 9,351 9,351 8,980 6,139 6,139 6,127 9,352 No.MSAs 364 364 364 362 364 364 362 364 Notes: Dependentvariableisbiannualgrowthofnon-educationalexpenditures. Multiplierequalscoefficientβin Equation2. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending. StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and***denotesignificanceatthe10,5,and1% level,respectively. MSAcontrols: numberofundergraduatestudents(log),changeinundergraduatestudents(log) lastyear,averagetuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentagehispanic,percentagewithatleastabachelorsdegree.DataonfinancialcontrolsisfromFederalReserveBankofNewYork/Equifax ConsumerCreditPanelandisavailablefrom1999to2015. ItincludesmedianEquifaxRiskScore, age, debt-toincomeratio,creditcardutilizationand30-daymortgagedelinquencyrate. WeexcludeMSA-yearsreceivingmore than5%ofTotalPellgrantsinagivenyear. 34

TableA6:EffectofPellGrantsonEducationExpendituresbyColleges:2YearHorizon FullSample Post1999 FullSample Educationexp. I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS 4-yearMultiplier 0.485 0.476 0.478 0.472 0.455* 0.425 0.447* 0.362* (0.303) (0.307) (0.307) (0.331) (0.263) (0.266) (0.266) (0.207) 2-yearMultiplier -0.0178 -0.0373 -0.0307 -0.0905 -0.0565 0.00725 -0.0161 0.439 (0.172) (0.173) (0.176) (0.176) (0.161) (0.165) (0.167) (0.497) Difference 0.503 0.513 0.509 0.563 0.512 0.418 0.463 -0.0767 Error 0.343 0.342 0.339 0.354 0.306 0.300 0.300 0.481 Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes StateSpending Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,351 9,351 9,351 8,980 6,139 6,139 6,127 9,352 No.MSAs 364 364 364 362 364 364 362 364 Notes:Dependentvariableisbiannualgrowthofeducationalexpenditures. MultiplierequalscoefficientβinEquation2. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending. StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and***denotesignificanceatthe10,5,and1%level, respectively. MSAcontrols: numberofundergraduatestudents(log),changeinundergraduatestudents(log)last year,averagetuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentagehispanic,percentagewithatleastabachelorsdegree. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/Equifax ConsumerCreditPanelandisavailablefrom1999to2015. ItincludesmedianEquifaxRiskScore, age, debt-toincomeratio,creditcardutilizationand30-daymortgagedelinquencyrate. WeexcludeMSA-yearsreceivingmore than5%ofTotalPellgrantsinagivenyear. 35

TableA7:EffectofPellGrantsonOverallExpendituresbyColleges FullSample Post1999 FullSample Allexp. I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS Multiplier 0.206 0.191 0.198 0.227 0.166 0.171 0.197 0.333 (0.184) (0.189) (0.192) (0.200) (0.163) (0.164) (0.160) (0.239) Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 9,303 9,303 9,303 8,933 6,103 6,103 6,091 9,304 No.MSAs 366 366 366 365 364 364 363 366 Notes: Dependentvariableisbiannualgrowthoftotalexpenditures. MultiplierequalscoefficientβinEquation2. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending.Standarderrors clusteredbyMSAandgiveninparentheses.*,**,and***denotesignificanceatthe10,5,and1%level,respectively. MSAcontrols: numberofundergraduatestudents(log),changeinundergraduatestudents(log)lastyear,average tuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentagehispanic,percentagewithatleast abachelorsdegree. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/EquifaxConsumerCredit Panelandisavailablefrom1999to2015.ItincludesmedianEquifaxRiskScore,age,debt-to-incomeratio,creditcard utilizationand30-daymortgagedelinquencyrate.WeexcludeMSA-yearsreceivingmorethan5%ofTotalPellgrants inagivenyear. TableA8:EffectofPellGrantsSpendingonNon-EducationExpendituresbyColleges FullSample Post1999 FullSample Non-educationexp. I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS Multiplier 0.100 0.107 0.111 0.155 0.0462 0.0387 0.0788 0.0471 (0.0984) (0.0992) (0.101) (0.0952) (0.122) (0.120) (0.107) (0.0843) Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 8,595 8,595 8,595 7,868 6,103 6,103 6,091 8,596 No.MSAs 363 363 363 361 363 363 361 363 Notes: Dependentvariableisbiannualgrowthofnon-educationalexpenditures. Multiplierequalscoefficientβin Equation2. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending. StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and***denotesignificanceatthe10,5,and1% level,respectively. MSAcontrols: numberofundergraduatestudents(log),changeinundergraduatestudents(log) lastyear,averagetuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentagehispanic,percentagewithatleastabachelorsdegree.DataonfinancialcontrolsisfromFederalReserveBankofNewYork/Equifax ConsumerCreditPanelandisavailablefrom1999to2015. ItincludesmedianEquifaxRiskScore, age, debt-toincomeratio,creditcardutilizationand30-daymortgagedelinquencyrate. WeexcludeMSA-yearsreceivingmore than5%ofTotalPellgrantsinagivenyear. 36

TableA9:EffectofPellGrantsonEducationExpendituresbyColleges FullSample Post1999 FullSample Educationexp. I II III IV V VI VII VIII 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS OLS Multiplier 0.183 0.170 0.179 0.225 0.166 0.171 0.196 0.307 (0.180) (0.184) (0.187) (0.204) (0.163) (0.163) (0.160) (0.246) Controls MSAF.E. Yes Yes Yes Yes Yes Yes Yes Yes YearF.E. Yes Yes Yes Yes Yes Yes Yes Yes MSATrend Yes Yes Yes Yes Yes Yes Yes Yes %Student Yes Yes Yes Yes Yes Yes Yes MSAControls Yes Yes Yes Yes Yes Yes LaggedGrowth Yes Yes FinancialControls Yes Yes Observations 8,595 8,595 8,595 7,868 6,103 6,103 6,091 8,596 No.MSAs 363 363 363 361 363 363 361 363 Notes:Dependentvariableisbiannualgrowthofeducationalexpenditures. MultiplierequalscoefficientβinEquation2. 2SLSregressionsusenationalspendinginteractedwithMSA-dummiestoinstrumentlocalspending. StandarderrorsclusteredbyMSAandgiveninparentheses. *,**,and***denotesignificanceatthe10,5,and1%level, respectively. MSAcontrols: numberofundergraduatestudents(log),changeinundergraduatestudents(log)last year,averagetuitionfee(log),for-profitpenetration,percentageofpopulationblack,percentagehispanic,percentagewithatleastabachelorsdegree. DataonfinancialcontrolsisfromFederalReserveBankofNewYork/Equifax ConsumerCreditPanelandisavailablefrom1999to2015. ItincludesmedianEquifaxRiskScore, age, debt-toincomeratio,creditcardutilizationand30-daymortgagedelinquencyrate. WeexcludeMSA-yearsreceivingmore than5%ofTotalPellgrantsinagivenyear. TableA10:AggregatedSchool-LevelSummaryStatisticsfromDeltaCost Variable Obs. Mean St.Dev. Min Max Source PellGrantDisbursements 10,729 26,868,982 82,315,079 0 1,645,212,928 DeltaCost PellGrants:ForProfit 10,725 4,397,998 30,588,941 0 1,299,494,912 DeltaCost Avg.TuitionFee 10,451 6,711 5,273 0 47,480 DeltaCost School(count) 10,758 14.4 56.5 1 1,096 DeltaCost For-ProfitSchool(count) 10,758 3.22 8.08 0 127 DeltaCost TotalUndergraduates 10,758 38,409 120,232 0 2,372,351 DeltaCost TotalEnrollment 10,758 44,709 137,400 0 2,638,180 DeltaCost StateAppropriation 10,758 135,386,046 411,576,994 0 8,576,051,452 DeltaCost SchoolRevenue 10,758 836,372,660 2641825108 -3,420,080,033 49,300,000,000 DeltaCost SchoolSpending 10,758 252,154,297 738,396,323 0 13,900,000,000 DeltaCost EducationSpending(frac.) 10,613 0.859 0.15 0.223 1 DeltaCost 37