FinanceandEconomicsDiscussionSeries

FederalReserveBoard,Washington,D.C.

ISSN1936-2854(Print)ISSN2767-3898(Online)

LostinAggregation:GeographicMismeasurementofIncomeand

Spending

SinemHacioglu-Hoke,LeoFeler,JackChylak

2025-050

Pleasecitethispaperas:

Hac?o?gluHoke,Sinem,LeoFeler,andJackChylak(2025).“LostinAggregation:Ge-ographicMismeasurementofIncomeandSpending,”FinanceandEconomicsDiscus-sionSeries2025-050.Washington:BoardofGovernorsoftheFederalReserveSystem,

/10.17016/FEDS.2025.050

.

NOTE:Sta?workingpapersintheFinanceandEconomicsDiscussionSeries(FEDS)arepreliminarymaterialscirculatedtostimulatediscussionandcriticalcomment.Theanalysisandconclusionssetfortharethoseoftheauthorsanddonotindicateconcurrencebyothermembersoftheresearchsta?ortheBoardofGovernors.ReferencesinpublicationstotheFinanceandEconomicsDiscussionSeries(otherthanacknowledgement)shouldbeclearedwiththeauthor(s)toprotectthetentativecharacterofthesepapers.

LostinAggregation:GeographicMismeasurement

ofIncomeandSpending☆

SinemHac?o?glu-Hoke*

FederalReserveBoard

CEPR

LeoFelert

Numerator

JackChylak?Numerator

July3,2025

Abstract

Usingzip-codemedianincomeasaproxyforhouseholdincomeiscommonineconomicsbutcanmaskheterogeneityandyieldmisleadingconclusions.Usingzip-codemedianincomeandself-reportedhouseholdincomesfromarepresenta-tivepanelof150,000U.S.households,wedecomposeaverageretailspendingfor2018-2024.Whenusingself-reportedincomes,weobservesubstantialdivergenceinspendingbetweenlow-andhigh-incomehouseholdsstartinginmid-2021.Whenusingzip-codeaggregatesasaproxy,thisdivergencedisappears.Ourfindingsindi-catea35to75percentdiscrepancybetweenzip-codeaggregatesandself-reportedincomes,highlightingthelimitationofzip-codeaggregatesasaproxyforhouseholdincomes.

Keywords:Spending,Income,Heterogeneity,Zip-codeAverageIncome

JELClassification:E01,E2,E32

*

sinem.haciogluhoke@

Web:

t

leo.feler@

Web:

?

jack.chylak@

☆TheauthorswouldliketothankTomazCajner,RyanDecker,AaronFlaaen,ChrisKurzandMichaelPalumbo.TheanalysisandconclusionssetforthhereinarethoseoftheauthorsanddonotindicateconcurrencebyothermembersoftheresearchstaffortheBoardofGovernorsoftheFederalReserveSystem.Theuseofcommerciallyprovideddataisforresearchpurposesonlyanddoesnotimplyendorsement,recommendation,orfavoringofanybrand,product,service,orcompanybytheBoardofGovernorsortheFederalReserveSystem.TheviewsexpressedinthispaperaretheresponsibilityoftheauthorsandshouldnotbeinterpretedasreflectingtheviewsofNumerator.Allerrorsremainourown.

2

1Introduction

Itiscommoninsocialsciencestocategorizehouseholdsaslow-,middle-orhigh-incometoassesstheheterogeneityin,forexample,consumers’spendingbehavior.Whenhouseholdincomesarenotobserved,acommonpracticeistouseasaproxytheaverageormedianincomeinzipcodeswherehouseholdslive,generallyobtainedfrompopulationsurveys.Thereasonsforsuchanapproximationofhouseholdincomesusingzip-codeaggregatesarebecause,first,wherehouseholdsliveiscorrelatedwiththeirincomes,andsecond,itisraretohaveaccesstodatasetsthatprovideinformationondisaggregatedhouseholdincomes.Inthispaper,weassesswhetherzip-codeaggregatesareavalidproxyfordisaggregatedhouseholdincomesthataccuratelyreflectsheterogeneityinconsumerbehavior,usingarepresentativepanelofU.S.householdswithdisaggregatedinformationonhouseholdincomes.

Thispaperhastwoobjectivesthatleadtonovelcontributionstoexistingstudiesthatdocumentheterogeneityinspendingpatterns.First,usingadetailedmicrodataset,weconstructameasureofrealaverageretailspendingforlow-,middle-andhigh-incomehouseholdsusinghouseholds’self-reportedincomestoconstructincomegroups.Second,weuseourmicro-datatotesttheimplicationsofusingzip-codeaggregates.Ourre-sultsindicatethatusingzip-codeaggregatesasaproxyforhouseholdincomesmasksheterogeneityinconsumerbehaviorandleadstomisleadingconclusionsaboutchangesinconsumerbehaviorduringthepost-pandemicperiod.

Weuseapaneldatasetof150,000representativeU.S.householdsfromNumerator,aconsumerdataandsurveycompany.NumeratorobtainsphysicalandonlinereceiptsfromthisrollingstaticpanelofU.S.householdsforwhomwehavedetailedinformationonhouseholdattributes.Wefirstcompareouroverallmeasureofmonthlyaverageretailspendingbetween2018and2024againsttheCensusBureau’sAdvanceMonthlySalesforRetailandFoodServicesreport(MARTS)toestablishourmeasure’sreliability.

HavingestablishedthattheaggregatedNumeratorretailspendingseriescloselymatchestheCensusBureau’sretailsalesseries,weexaminespendingpatternsbasedonhouse-holdincome.Weanalyzetheaveragemonthlyhouseholdretailspendingforlow-incomehouseholds(§0-§60Kinannualhouseholdincome),formiddle-incomehouseholds(§60K-

3

§100K),andforhigh-incomehouseholds(§100K+)from2018to2024.Publishedmea-suresdonotprovidedetailsonwhichconsumers’spendinghasremainedresilientinthepost-pandemicperiod,andhencefallshortofidentifyingvulnerabilitiesintheeconomyoriginatingfromspecificgroups,soouranalysisfillsthisgap.Ourresultssuggestthatretailspendinggrowthevolvessimilarlyforallhouseholdsbeforethepandemic.However,startinginmid-2021,thespendingofhigh-incomehouseholdsdivergesfromthespend-ingoflow-andmiddle-incomehouseholds.High-incomehouseholdscontinuetospendstronglywhilelow-andmiddle-incomehouseholds’spendinglagsbehind.Thisfindingissupportedbymanyanalysesandcanberationalizedbylow-incomehouseholdsdepletingtheirpandemic-erasavings(

Abdelrahmanetal.

,

2024

)andbytheexpirationofgov-ernmentsupportprogramsthathelpedlow-andmiddle-incomehouseholdsduringandfollowingthepandemic.

1

Moreover,weshowthatthisdivergenceisarobustfindingwhenwedecomposespendingbyeducationandsimulatescenarios,usingahouseholdsurvey,wherewecontrolforthehighermobilitybetweenincomegroupsin2022and2023.

Insteadofusinghouseholds’self-reportedincomestoconstructincomegroups,wenextusemedianzip-codeincomesreportedintheAmericanCommunitySurvey(ACS),following

Chettyetal.

(

2023

),toclassifyhouseholdsaslow-,middle-andhigh-income.OuranalysisusingNumeratorself-reportedincomedataandreplicatingzip-codemedianincomeaggregationisuniqueinallowingadirectcomparisonofspendingmeasuresbasedondisaggregatedself-reportedincomeversusaggregatedproxiedincome.Usingeithermethod,wedocumentsimilarpre-pandemicspendingdynamics.However,thediver-genceweobserveinthepost-2021spendingdisappearswhenweuseaggregatedzip-codemedianincomeasaproxy.Usingaggregatedzip-codemedianincome,itappearsthatspendingevolvedsimilarlyforallhouseholdincomegroupsduringthepost-pandemic

1Suchanalysisofdocumentingthediscrepanciesinspendinggrowthissparseatbestinacademicplatforms,mostlyduetodataavailabilitytoassessspendingbyincomegroups.OnesuchanalysisisMoody’sAnalytics’,whichshowthattopearners–thetop10%ofUShouseholdsintermsofearnings

–drivenearlyhalfofallconsumerspending.Forthedetailsofthatanalysis,see

https://www.wsj.

com/economy/consumers/us-economy-strength-rich-spending-2c34a571?st=2odEgM&reflink=

article_copyURL_share

.TheMoody’sAnalyticsanalysispostdatesourinitialanalysiswhichshowsthedivergenceinspendingbetweenincomegroupsin

HaciogluHokeetal.

(

2024

).AnotherrelatedstudybytheBankofAmericaInstitutedocumentstheslowingpaceoflower-incomehouseholds’spending:

https:

///economic-insights/consumer-checkpoint-march-2024.html

.

Finally,MorningConsultanalysisalsodocumentsslowerspendingbylow-andmiddle-incomehouseholds:

/analysis/consumer-spending-september-2024

.

4

period.Thisresultcontradictsourpreviousfinding,usingdisaggregatedself-reportedhouseholdincome,thathigher-incomehouseholdsweretheonesdrivingconsumerspend-ingintheU.S.inthepost-pandemicperiodwhilelow-andmiddle-incomehouseholdspulledback.

Whenwelookintothesharesoflow-,middle-andhigh-incomehouseholdswholiveinlow-,middle-andhigh-incomezipcodes,wefindlargediscrepancies.In2024,only59%oflow-incomehouseholdslivedinzip-codesclassifiedashavinglowmedianhouseholdincome,whereastherestofthehouseholdslivingintheselow-incomezip-codeswereactuallymiddle-andhigh-incomehouseholds.Similarly,only32%ofhouseholdsthatlivedinhigh-incomezipcodesreportedhighincomesinNumeratordata;therestwerelowandmiddle-incomehouseholdsin2024.

Theuseofzipcodeincomeorotherhouseholdcharacteristicsasaproxyforindividualhouseholdattributeshasbeensubjecttoscrutiny,particularlyinthehealthliterature.Severalstudieshavehighlightedthepotentiallimitationsofthisapproach.Forinstance,

Geronimusetal.

(

1996

)exploresthechallengesassociatedwithusingaggregatecensus-basedvariablesasproxiesformicro-levelsocioeconomiccharacteristics.Similarly,

Hanley

andMorgan

(

2008

)utilizesCanadianadministrativedatatodemonstratethesignificantvariabilitybetweenhousehold-levelincomeandarea-basedincomemeasures.Morere-cently,

Buajittietal.

(

2020

),alsodrawingonCanadiandata,advisescautionwhenem-ployingarea-leveldataasindicatorsofindividualsocioeconomicstatus.Inthefieldofeconomics,

Geronimusetal.

(

1996

)standsoutasoneofthefewstudiestocriticallyex-aminetheuseofzip-codelevelmeasuresasproxiesforhouseholdcharacteristics.Theirfindingssuggestthatusingaggregateproxiesmayoverstatetheimpactofsocioeconomicfactorsonhealthoutcomeswhileinadequatelyaddressingconfoundingvariablesbetweensocioeconomiccharacteristics.Thesestudiescollectivelyunderscoretheimportanceofcarefullyconsideringthelimitationsandpotentialbiaseswhenusingaggregatedatatorepresentindividual-levelsocioeconomicinformation.

Ouranalysisbuildsuponrecentstudiesthatexamineconsumerspendingpatternsduringandafterthepandemicusingprivate-sectordata(see

Vavra

(

2021

)and

Brodeur

etal.

(

2021

)andthereferencestherein).Attheonsetofthepandemic,theEconomicTrackerdevelopedby

Chettyetal.

(

2023

)providedimportant,innovative,andtimely

5

high-frequencyestimatesofspendingcontributionsfromdifferentincomegroupsfrom

2020onward.Theiranalysis,whichusescreditcardspendingdata,illuminatedpotentialeconomicrisksandproxiedincomeusingmedianzip-codeincomefromtheACS.Anothernotablestudyby

Coxetal.

(

2020

)documentstheheterogeneouseffectsofthepandemiconhouseholdsusingbankaccountdata.Theiranalysisofspendingbyincomequartilesoffersinsightsintoexpendituredifferencesduringthepandemic’searlystages.However,astheauthorsacknowledge,theirstudywaslimitedbytheunavailabilityofmicro-dataonincomeatthetimeofwriting,necessitatingtheuseofpubliclyavailabledatatosimulateincomechangesinthepandemic’sinitialmonths.

Ourresearchcontributestothisgrowingbodyofliteratureinseveralways.First,weutilizeanewmicrodatasetthatdirectlycaptureshouseholds’self-reportedincomesandobservedspending.Second,weexplorespendingchangesacrossincomegroupsoveranextendedperiod,encompassingbothpre-andpost-pandemictimeframes.Third,wecomparetheimplicationsofusingdisaggregatedself-reportedhouseholdincomestoconstructincomegroupsversususingproxiedincomebasedonzipcodedatatoassesshowspendingchangesforlow-,middle-,andhigh-incomehouseholds.Furthermore,ouranalysiscanbeupdatedweekly,similarto

Chettyetal.

(

2023

)’sapproach,allowingforanearreal-timeassessmentofspendingdynamics.Theseuniquefeaturesenableustoreadilydetectchangesinspendingpatternsbydifferentincomegroupswithgreateraccuracyduringatimewhentheeconomyandconsumerspendingbehaviorsarebeingaffectedbyvariousshocks.

Therestofthepaperproceedsasfollows.Section

2

providesthedetailsoftheNu-meratordataandthereliabilityofourspendingmeasure.InSection

3

,weconstructtheseasonally-adjustedrealaverageretailspendingmeasuresforlow-,middle-,andhigh-incomehouseholdsbyself-reportedincomegroupsbetween2018and2024,andprovideadditionalevidencefortherobustnessofthedivergenceinspending.Section

4

insteaduseszip-codelevelaggregationtodecomposespendingbyincome.InSection

5

,wequan-tifythereportedversuszipcodeincomediscrepancyandSection

6

concludes.WeprovideadditionalanalysesandrobustnesschecksintheOnlineAppendix.

6

2Data

TheNumeratordatacontain150,000panelistswhoself-identifyastheprimaryshopperinthehouseholdandwhosetransactionsarecontinuousandcompleteforaperiodofatleast12months.These150,000panelistsareselectedfrommorethan1millionpanelistsinawaythatisdemographicallyandnationallyrepresentative.NumeratorprovidesweightsforhouseholdstoensureamatchwithCensusdemographicdataandtoensuretheirdetailedpurchases,whensummedbyretailerormanufacturer,alignwithquarterlyearningsreportsofmajorretailersandconsumerpackagedgoodscompanies.

Numeratorcollectsdatafromhouseholdsinseveralwaysusingamobilephoneapp.

First,consumerscanuploadpicturesoftheirpaperreceipts.Second,theycanallowNumeratortoscrapetheiremailsfordigitalreceipts.Third,userscanlinkloyaltyandmembershipaccountssotheirtransactionsareautomaticallyrecorded.

2

Numera-torcollectssurveyinformationonhouseholdincome,zipcode,age,education,householdsize,race/ethnicity,andvariousothermeasures.Panelists’identifiableinformationisanonymized.

TheNumeratordataallowfor16incomegroupingsbasedonpanelist’sself-reportincome.

3

Allpanelistsareresurveyedabouttheirhouseholdincomeapproximatelyevery12monthsandoftenmorefrequentlyifpanelistsreportlifeevents,suchasjobchanges,whenpromptedapproximatelyevery3months.Panelistsreportanychangestotheirincomegroupwhentheyarere-surveyed.Althoughtheremaybealagindetectingchangesinhouseholdincome,frequentsurveysprovideconfidencethatthemarginoferrorinaccountingforchangesinhouseholdincomeiswithinreasonablelimits.

ToestablishthereliabilityoftheaverageretailspendingmeasureweconstructusingNumeratordata,wecompareitagainstofficialstatisticspublishedforretailspending.Everymonth,theCensusBureaupublishestheMARTSreporttoprovideanestimateofretailsalesandfoodservicesspending.TopreparetheMARTSreport,theCensusBureaucollectsthedollarvalueofsalesdirectlyfromretailersbyfieldingestablishment-

2NumeratorcollectsthesedatafromhouseholdsusingamobilephoneappcalledReceiptHog.Pan-elistsarerewardeddirectlywithcoinsandwithvirtualspinsofaslotmachinethatalsorewardscoins,whichpanelistscanredeemforgiftcardsorforcashthroughPayPal.Onaverage,Numeratorrewardspanelistsapproximately$43peryearforprovidingtheirpurchaseinformationandcompletingsurveys.

3Theincomegroupsstartwith<$20Kandgoupto$100Kwith$10Kincrements.After$100K,theyincreasewith$25Kincrementsupto$250Kwiththelastgroupbeing$250K+.

7

levelsurveysandestimatingtotalresultsandtheresultsbytherelevantthree-digitNorthAmericanIndustryClassificationSystem(NAICS)categories.

TheNumeratorpaneldataincludeinformationonhouseholds’purchasesfrompaperreceipts,onlinereceipts,andtransactionrecords.Thepurchaseinformationcontainsthequantityandpriceofeachitempurchased,theoveralltotalonthereceipt,andwhereandwhenthepurchasetookplace.Thesedetailsallowustoclassifyhouseholds’spendingintotherelevantthree-digitNAICScategoriesforretailsalescategoriestoconstructaspendingmeasurecomparabletotheCensusBureau’sretailsalesmeasureintheMARTSreport.

4

Notethatourmeasureexcludesspendingonmotorvehiclesandparts,aspendingcategoryNumeratordatadonotcapture.

Figure

1

compareshowtheretailspendingmeasureweconstructusingNumeratordatacompareswiththeCensusBureau’spublishedseries.TheredlineindicatestheCensusBureau’spublishedseasonallyadjustedtotalretailsalesexcludingmotorvehiclesandparts.ThebluedashedlineistheretailspendingmeasureweconstructfromtheNumeratordata,whichweseasonallyadjustusingtheCensusBureau’sX13-ARIMA-SEATSpackage.

5

Thegraydash-dottedlineisthenon-seasonallyadjustedCensusseriesthatwethenseasonallyadjustusingthesameX13-ARIMA-SEATSmethodologyweuseforNumeratordata.ThepurposeofthegraylineistoshowthatourseasonaladjustmentmethodologyalignscloselywiththeCensusmethodology,andtherefore,thedifferencesbetweentheNumeratorseasonally-adjustedseriesandtheofficialCensusseasonally-adjustedseriesisnotduetodifferencesinseasonaladjustment.

6

Thecorrelationbe-

4These11spendingcategoriesareFurniture&homefurn.stores(NAICS442),Electronics&appliancestores(NAICS443),Buildingmaterial&gardeneq.&suppliesdealers(NAICS444),Food&beveragestores(NAICS445),Health&personalcarestores(NAICS446),Gasolinesta-tions(NAICS447),Clothing&clothingaccessoriesstores(NAICS448),Sportinggoods,hobby,musi-calinstrument,&bookstores(NAICS451),Generalmerchandisestores(NAICS452),Miscellaneousstoreretailers(NAICS453),Non-storeretailers(NAICS454),andFoodservices&drinkingplaces(NAICS722).AccordingtotheCensusBureau,theMARTSreportcoversfirmsclassifiedintheRe-tailTradeandFoodServicessectorsasdefinedbytheNorthAmericanIndustryClassificationSystem(NAICS).RetailTrade,asdefinedbyNAICSsectors44-45,includesestablishmentsengagedinsell-ingmerchandiseinsmallquantitiestothegeneralpublic,withouttransformation,andrenderingser-vicesincidentaltothesaleofmerchandise.Foodanddrinkingservicestofinalconsumers,subsector

722,isalsoincludedinthesurvey.Moreinformationonhowthedataarecollectedisavailableat

/retail/how_surveys_are_collected.html

.

5Inaddition,duetoachangeinNumeratormethodologystartinginApril2021,weapplyafixedlevel-correctiontotheNumeratorseriesforApril2021andonward.Thislevel-correctionisequivalenttoafixed-effectthataccountsforthechangeinNumerator’smethodology.

6ItisnotentirelypossibleforustoreplicatetheCensusBureau’sseasonaladjustmentmethodology.

8

Figure1:Realretailsales,totalexcludingmotorvehiclesandparts,constructedusing

CensusBureauestimatesandNumeratorpaneldata

125

115

Index,2019=100

105

95

85

2018201920202021202220232024

Censusseries,SACensusseries,usingauthors'SAmethodsNumeratorseries,usingauthors'SAmethods

Note:ThesolidredlineistheCensusBureau’sseasonallyadjustedretailsalesseriesexcludingmotorvehiclesandparts.ThedashedbluelineistheretailspendingmeasureweconstructusingtheNumeratordata,whichweseasonallyadjust.Thedashedgraylineisthenon-seasonallyadjustedCensusseriesthatwethenseasonallyadjust.ThedataaremonthlyfromJanuary2018throughDecember2024.Authors’seasonaladjustmentmethodusesX13-ARIMA-SEATS.InflationadjustmentusesthechainindexedPCEdeflatorforgoodsandfoodservicesexcludingmotorvehicles.Allthreelinesareindexedto100in2019.

tweentheNumeratorseriesandtheCensusBureau’sofficialseasonallyadjustedretailsalesseriesis0.94inlevelsand0.83whenlookingatmonthlypercentchanges.Figure

1

andthecorrelationofNumerator’sretailspendingmeasurewiththeofficiallypublishedretailsalesmeasureallowustoverifythatthebottom-upconstructionoftheNumera-tordata,capturedfromhouseholdpurchases,closelymatchesthetop-downconstructionoftheCensusBureau’sretailsalesseries,capturedfromsurveysofretailandfoodser-viceestablishments.Appendix

A

providesadditionalfiguresandreportscorrelationsforsub-spendingcategories.

Retailsalessubcategoriesareseasonallyadjustedseparately,sometimesevenatamoregranularlevel,andthencombinedtosumtothetotalretailsalesseriespublishedbytheCensusBureau.Thisisthemainreasonwhyweperformourownseasonaladjustment.Duetothetimespanofthedataandthereasonsexplainedabove,wecangetclosetobutcannotfullyreplicatetheofficiallypublishedseasonallyadjustedretailsalesseries.MoreinformationontheCensusBureau’sseasonaladjustmentcanbefoundat

/retail/marts/www/timeseries.html

.

9

3Spendingbyself-reportedincome

Thissectionreportsourfindingsonaverageretailspendingbyincome.Usingthecat-egoricalincomegroupsrecordedinNumeratordata,wedecomposeretailspending,thebluelineinFigure

1

,intospendingbylow-,middle-,andhigh-incomehouseholds.Wethenprovideadditionalevidenceforthevalidityofthedivergenceinspendingbetweenhigh-vslow-incomehouseholds.

3.1Averagehouseholdspendingbyincome

Figure

2

reportsgrowthsinceJanuary2018inaveragemonthlyhouseholdspending,adjustedforinflationandseasonality,overallandbythreeincomegroups:low(§0-§60Kinannualhouseholdincome),middle(§60K-§100K),andhigh(§100K+).

7

Figure2:Growthofaverageretailspendingdecomposedbyhouseholdincome

20

%Change,Jan2018=0

10

0

2018201920202021202220232024

AllGroupsLowIncome$0?60KMiddleIncome$60K?100KHighIncome$100K+

Note:ThedataaremonthlyfromJanuary2018throughDecember2024.AllseriesareadjustedforinflationusingthechainindexedPCEdeflatorforgoodsandfoodservicesexcludingmotorvehiclesandareshownasgrowthrelativetoJanuary

2018.AllseriesareseasonallyadjustedusingX13-ARIMA-SEATS.Solidpurple,dash-dottedgrayanddashedgreenlinesplotthespendingbylow-,middle-andhigh-incomehouseholds.

Thegrowthinaveragehouseholdretailspendingissimilaracrossallincomegroupsinthepre-pandemicera.Duringtheonsetofthepandemic,allhouseholds’spendingdeclined.Variousstudies,e.g.

Coxetal.

(

2020

),showthatspendingdeclineddueto

7Weselectthecutoffsforlow-,middle-,andhigh-incomegroupsbasedonthefollowingCensusBureaureportthatisbasedon2022incomes:

/library/publications/2023/demo/

p60-279.html

.Minormodificationstotheseincomegroupingsdonotchangeourqualitativeresults.

10

thedirecteffectsofpandemicratherthanalossofincome.Followingthedeclineinthe

beginningofthepandemic,averagehouseholdspendingforallincomegroupsbouncedback,reachinglevelshigherthanpre-pandemiclevelsashouseholdsspentpent-upsavingsandgovernment-providedstimuluspayments.

Duringthepandemicrecoveryperiodandshortlythereafter,low-incomehouseholdsappearedtobenefitmorefrompandemic-erastimulus.Inpercentageterms,low-incomehouseholdsincreasedtheirspendingbymorethanmiddleandhigh-incomehouseholdsin2020andearly2021.Thelargejumpintheaveragespendingoflow-incomehouseholdsinearly2021coincideswiththelaststimulusprogramwhichprovidedmorestimulus,asapercentageoftheirincome,tolow-incomeindividuals.Thefindingthatmiddle-andhigh-incomehouseholdsdidnotincreasetheirspendingasmuchaslow-incomehouseholdsaroundthetimeofthelaststimulusprograminMarch2021isconsistentwith

Chetty

etal.

(

2023

)’sfindings.

Despitesimilarpre-pandemicandearly-pandemictrends,thespendingbehavioroflow-andhigh-incomehouseholdsstartstodivergeinmid-2021.Betweenmid-2021andmid-2023,middle-andhigh-incomehouseholdsmaintainorincreasetheirrealaveragespending;low-incomehouseholdsreducetheirrealaveragespendingcomparedwithmid-2021levels.Sincemid-2023,low-,middle-,andhigh-incomehouseholdshaveallbeenincreasingtheirrealaveragespending.AsofDecember2024,realaveragespendingbylow-incomehouseholdsisup13.5%relativetoJanuary2018;spendingbymiddle-incomehouseholdsisup17.0%,andrealaveragespendingbyhigh-incomehouseholdsisup20.9%.

Whileouranalysisdoesnotofferevidenceforthedifferencesinspendingweobserveforlow-,middle-,andhigh-incomehouseholds,weofferafewhypothesesforwhythesedif-ferencesmayhaveemerged.First,low-incomehouseholdsdepletedpandemic-eraexcesssavingsearlierasdocumentedby

Abdelrahmanetal.

(

2024

).Second,governmentsup-portprograms–suchasSNAPemergencyallotments,enhancedunemploymentinsurance,andenhancedchildtaxcreditsthatdisproportionatelybenefitedlow-incomehouseholds–expiredandnolongerprovidedaboosttothesehouseholds’spending.Third,high-incomehouseholdsmighthaveexperiencedawealtheffectastheirhomesandinvestmentsin-creasedinvalue,whilealsoreceivingmoreinterestandinvestmentincomeduringperiodsofhigherinterestrates,providingastimulusforsustainedlevelsofspen