The two sources match up well on credit card charges and fairly well on account totals But the SCF always yields much lower estimates of revolving debt My estimated lower bound for the discrepancy in 2004 is half of the revolving credit card debt to ID: 15906
Download Pdf The PPT/PDF document "WHERE IS THE MISSING CREDIT CARD DEBT CL..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
WHEREISTHEMISSINGCREDITCARDDEBT?CLUESANDIMPLICATIONSDartmouthCollegeIcreatecomparableestimatesofaggregatecreditcardusebasedonhouseholddatafromtheSurveyofConsumerFinances(SCF)andindustrydata.Thetwosourcesmatchupwelloncreditcardchargesandfairlywellonaccounttotals.ButtheSCFalwaysyieldsmuchlowerestimatesofrevolvingdebt.Myestimatedlowerboundforthediscrepancyin2004ishalfoftherevolvingcreditcarddebttotalimpliedbyindustrydata.Thereisnoobvioussourceforthisremainingdiscrepancyandsomeevidencethatthediscrepancyhasgrownovertime.Suchgrowthisworrisomebecauseitparallelssubstantialchangesincreditcarduseandinthepoolofcreditcardusers,suggestingthatthediscrepancycouldbedrivenbyhouseholdunderreportingthatiscorrelatedwithunobservedheterogeneity.ThiscorrelationcouldconfoundinferenceontherelationshipbetweencreditcardborrowingandoutcomesofinterestlikehouseholdÞnancialcondition,consumptionpaths,andportfoliochoice.Giventhispossibilityitiscriticaltocontinuedevelopingevidenceonwhetherandwhyhouseholdsurveysundercountcreditcard1.IAcasualcomparisonofUnitedStatesdatasuggestsabigdiscrepancybetweenthebalancesheetsofcreditcardborrowersandlenders.LendersreportowningthreetimesasmuchconsumercreditcarddebtintheFederalReserveStatisticalReleaseG.19:ConsumerCredit(theÒG.19Ó)ashouseholdsreportowingintheFederalReserveBoardÕsSurveyofConsumerFinancesIn2004thediscrepancywas$537billion.UnderstandingthereasonsforthisÒmissingÓcreditcarddebtiscriticalforstudiesofhouseholdÞnancesuchasintertemporalandportfoliochoiceandofcreditmarket(in)efÞciency.Creditcardsarethelargestsourceofnon-mortgageborrowingintheU.S.andarerapidlyapproachingthatstatuselsewhere.CreditcardsarealsoplausiblythesourceofborrowingandconsumptionformostU.S.households.ResearchershaveincreasinglyturnedtoÞnancialmicro-datafromhouseholdsurveysliketheSCFforseveralreasons.Oneispractical:administrativedataisoftenhardtoobtain.Otherreasonsaresubstantive.The:IthankthePaymentCardCenterattheFederalReserveBankofPhiladelphiaforhostingmeasaVisitingScholar,AdeelIkramforresearchassistance,andMarkFurletti,KathleenJohnson,ArtKennickell,BillLang,DougStaiger,theeditorMarcelTimmer,twoanonymousreferees,andespeciallyChristopherOdyforcomments.*Correspondenceto:JonathanZinman,DepartmentofEconomics,DartmouthCollege,314RockefellerHall,Hanover,NH03755,USA(jzinman@dartmouth.edu).The2004SurveyofConsumerFinancesÞndsthat72percentofU.S.householdsheldageneralpurposecreditcard;incontrast,only12percenthadahomeequitylineofcredit.Averageutilizationofcreditcardcreditlimitswasonlyaround20percent(http://www.nationalscoreindex.com/NSI_Site/USScore.aspx;Nilson#705,1999).SeeSullivan(2008)forevidenceonconsumptionsmoothingwithcreditcards.ReviewofIncomeandWealthSeries55,Number2,June2009©2009TheAuthorJournalcompilation©2009InternationalAssociationforResearchinIncomeandWealthPublishedbyBlackwellPublishing,9600GarsingtonRoad,OxfordOX42DQ,UKand350MainSt,Malden,MA,02148,USA. SCFisÒthemostcomprehensivesurveyonhouseholdwealthÓ(Antoniewiczetal2005,p.7).Householdsholdcreditcards(andotherbalancesheetitems)fromseveraldifferentissuers,andhenceahouseholdsurveyisoftentheonlywaytogetacomprehensivepictureofthehouseholdbalancesheet.Householdsurveysmayalsobebettersuitedthanaggregatesoradministrativemicrodataforcollect-inginformationusedtostudyheterogeneityinÞnancialoutcomesandmodelWhilehouseholdsurveydataoncreditcardborrowingisusefulinprinciple,thelimitedavailableevidencesuggestsapotentiallycriticalpracticallimitation:severeundercounting.Twodouble-blindvalidationstudiesonotherformsofunsecuredconsumerborrowingshowthatmorethanhalfofborrowers,knownfromlender-sidedatatohaveborrowedrecently,donotreportrecentbor-rowing.ElliehausenandLawrence(2001)showunderreportingofpaydayloanborrowinginphonesurveys.KarlanandZinman(2008)Þndunderreportingofapayday-likeloanproductinaSouthAfricanin-personsurvey.ResearchersusingtheSCFtypicallyignorethepossibilityofunderreporting(orofundercountingmoregenerally),ortheyscaleupSCFborrowingwithanaverageundercountingfactorthatmatchestotal,weighted-upborrowingintheSCFtotheG.19aggregate.TheG.19iswidelyconsideredtobeaccurateapriori(becauselendershavestrongincentivestoreporttruthfully),andhasbeenvali-datedbyathirdparty(FurlettiandOdy,2006).ButformostapplicationstheseapproachestoSCFundercountingwillpermitvalidinferenceifandonlyiftwoconditionshold.First,onemustusethecorrectscalarforaverageundercounting.Second,anadjustedaverageundercountingfactorisonlybroadlyusefulifhouse-hold(under)countingishomogenousoreffectivelyso;e.g.ifeveryoneunderre-portsbythesamefactor,orifunderreportingisnotcorrelatedwithparametersandoutcomesofinterest.Unobservedheterogeneityinrespondentreportingofcreditcarddebtcouldpresentaseriousproblemforresearchersacrossarangeofliteraturesandmethods.Itwouldconfoundattemptstoestimatethedistribution,incidence,anddeterminantsofborrowingcosts;thisconcernapplies,forexample,tothelitera-tureexploringwhyhouseholdssimultaneouslyborrowexpensivelyoncardswhilelendinginlow-yieldingdemanddeposits.UnobservedheterogeneityinreportingmightalsobiasestimatesofparametersormodelÞt;theseconcernsapply,forexample,toliteraturesestimatingtheeffectofdebtburdenonÞnancialdistress,theeffectofhousepricesonconsumption,andtheÞtofalternativemodelsofinter-temporalandportfoliochoice.Idiscussimplicationsofreportingheterogeneitymorein-depthinSection5.ThechallengeinassessingwhetherunobservedThemainsourcefortheG.19revolvingcreditestimate,thequarterlyReportofConditionandIncome(ÒCallReportsÓ),iswidelyusedbyresearchersandconsideredtobeaccurate.Commercialbanks(whoissueandownover80percentofcreditcardoutstandings)arecloselyscrutinizedbyregulators;thissupervisionpresumablyconstrainstheabilityofissuerstomisreportonascalelargeenoughtoexplainthewedgebetweentheG.19andtheSCF.Moretothepointthereisnoapriorireasonwhy,afteradjustingforthedeÞnitionaldifferencesasIdobelow,onewouldexpectissuerstoreport.Ifanythingtherelativelysteepcapitalchargeoncreditcardassetswouldpushbankstoreport.Chargeoffsordelinquenciescouldexplainasmallfractionofthewedge(theyareequaltoabout4percentofoutstandingsonaverage)ifborrowersdonotreporttheirbad/latedebtandissuersdo.ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 heterogeneityisactuallyaproblemisthatmeasuringhouseholdreportingbe-haviorisdifÞcultwithoutdirectvalidationfromlenderdata.InthispaperIexplorewhetherunobservedheterogeneityisaproblemusingindirectmethodsfeasiblegiventheavailabledata.TheÞrstÞndingisthatthediscrepancybetweenSCFandindustrymeasuresofaggregatecreditcardborrow-ingcannotbeexplainedbyobviousdeÞnitionalorcoveragedifferences.TheG.19measureofincludesseveraltypesofcreditcardusethatareexcludedfromtheSCFmeasuresofcreditcardbydesign.Theseincludeßoat(asotherresearchershaverecognized),businessuseofpersonalcards,andnon-creditcardlinesofcredit.Afteradjustingforseveraldifferencesmyestimatedboundonthediscrepancyfor2004is$200billion.ThissuggeststhattheSCFmissesabouthalfofrevolvingcreditcarddebt.AsecondÞndingissomeevidencethatthedeÞnition/coverage-adjusteddebtdiscrepancyhasbeenrisingovertime(e.g.mypreferredestimatessuggestthattheSCFundercountedbyafactorof1.3or2.0in1989,andby1.9or2.4in2004).Thisisworrisomegiventhatcreditcarduseandthepoolofcreditcarduserswasalsochangingoverthistimeperiod.Useshiftedtowardgeneralpurposecards,and26millionhouseholds(generallylesscreditworthyones)enteredthemarket.suggeststhatanySCFundercountingmightbedrivenbychangesinunobservedhouseholdcharacteristicsthatcouldconfoundinference.AthirdÞndingisthatincontrasttodebt,theSCFcreditcardchargesandaccounttotalsmatchuprelativelywellwithindustrysources.Thiscouldbeduetodifferencesbetweenindustrysources(hereNilsonReportsinsteadoftheG.19)ratherthananythingaboutSCF.ButifchargesandaccountsdoultimatelyprovetobereportedmorecompletelythandebtitwilllimitthesetoflikelyexplanationsforSCFundercountingofcreditcarddebt.Therestofthepaperproceedsasfollows.Section2describesthestrengthsandweaknessesofthedatasourcesIusetomeasurecreditcardactivity.Section3providesestimatesofthediscrepancybetweenSCFandindustrysourcesthatattempttoadjustfordeÞnitionalandcoveragedifferences.Section4discussespossiblesourcesofthediscrepancy.Section5exploreswhethertheremainingdiscrepancymakestheSCFcreditcardborrowingdatatooproblematicforbroadusebyresearchers.Section6concludes.2.D Thissectionprovidesanoverviewdescriptionofmythreemaindatasourcesformeasuringcreditcardaccountholding,charging,andborrowing.Section3providesdetailsondeÞnitionalandcoveragedifferences,andontherelatedadjust-mentsImaketofacilitatecomparisons.See,e.g.GrossandSouleles(2002),Johnson(2007),andLaibsonetal.SeeEdelberg(2006)andCalemetal.(2006)fordetailsonindustryinnovationsthatdrovethiscreditexpansion.Forexample,Table1suggeststhatthemarginalcreditcardholderswerepoorerandmorelikelytobefemalethaninframarginalones.KarlanandZinman(2008)Þndevidencesuggestingthatfemalesandrelativelylow-incomeborrowersaremorelikelytounderreporttheirexpensive,unsecureddebt.ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 TheprimaryindustrysourceoncreditcarddebtistheFederalReserveSta-tisticalReleaseG.19:ConsumerCredit(G.19).TheG.19takesasnapshotestimateofoutstandings(amountsowedtocreditcardissuers)onallconsumercreditcardaccountsatmonth-end,makingtheappropriateadjustmentssothatsecuritizedreceivablesdonotgetdouble-counted.G.19creditcardoutstandingsarecom-prised(thoughnoteasilydisaggregatedinto)threecategoriesofcards:thosethatcanbeusedatmanydifferentmerchants(generalpurposecards),thosethatcanbeusedonlyataparticularstoreorchain(storecards),andthosethancanbeusedatparticularfuelingstations(gascards).TheG.19ÕsrevolvingÒtotalÓalsoincludesasmallamountofnon-creditcardlinesofcredit,asdetailedinSection3.4.Althoughoneneedstobecautiousaboutusingadministrativedataforsta-tisticalresearch(see,e.g.Brackstone,1987),boththeoryandevidencesuggestthattheG.19producesanaccurateestimateofoutstandingcreditcarddebt.Apriori,itseemslikelythatmostcreditcardissuershaveincentivestoreportaccurately.ThemainsourcedatafortheG.19,thequarterlyReportofConditionandIncome(ÒCallReportsÓ),isscrutinizedbybanksafetyandsoundnessregu-lators(commercialbanksissueandownmorethan80percentofcreditcardThissupervisionpresumablyconstrainstheabilityofissuerstomisreportonascalelargeenoughtoexplainalargediscrepancybetweentheG.19andtheSCF.Moretothepointthereisnoapriorireasonwhy,afteradjustingforthedeÞnitionaldifferencesasIdobelow,onewouldexpectissuerstoIfanythingtherelativelysteepcapitalchargeoncreditcardassetswouldpushbankstoreport.Chargeoffsordelinquenciescouldexplainasmallfractionofthediscrepancy(theyareequaltoabout4percentofoutstandingsonaverage)ifborrowersdonotreporttheirbad/latedebtandissuersdo.GeneralpurposecardsareofferedbytheVisa,MasterCard,Discover,andAmericanExpressbrands/networks,andmosthaverevolvingcreditfeatures.AsmallfractionofgeneralpurposecardactivityisonÒchargecardsÓ(issuedprincipallybyAmericanExpress)thatdonothaverevolvingfeaturesÑeachmonthÕsbalancemustbepaidinfull.TheCallReportsarealsowidelyusedbyresearchers;tomyknowledgenoonehasuncoveredseriousreportingbiases.TABLE1 SCFM 35NonwhiteFemaleUnmarriedRenter19890.610.490.210.140.260.390.290.550.4619920.630.500.230.160.260.420.320.530.4619950.650.560.220.180.290.470.350.560.5019980.620.530.210.180.260.470.350.540.4820010.670.540.210.190.300.490.340.540.4520040.680.510.230.230.320.500.310.590.50p-valueontestofwhether2004mean1989mean0.010.550.400.000.020.000.430.140.27 :ProportionsweightedwithSCFvariablex42001.Educationisyearscompleted.Education,age,race,gender,jobtenurewithcurrentemployer,andmaritalstatusarethoseofthesurveyrespondents.ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 Inpractice,FurlettiandOdy(2006)detailandcheckthesamplingproceduresusedtoconstructtheG.19monthlyrevolvingcreditestimate.AlthoughtheysuggestsomeminormodiÞcationsoveralltheyÞndtheG.19tobeÒhighlyButtheG.19doesnotprovideanyinformationoncreditcardchargevolumeoraccounttallies,soIcombedvariousissuesofTheNilsonReport(Nilson)forissuer-sidedataonthesemargins.UnliketheG.19andSCF,Nilsondatavariesincontentandformatacrossyears.ConsequentlyIreviewedeveryissueintheyearfollowingthecomparisonyearsinmysample(e.g.2005issuesfor2004data)fordatathatmightbeinformative.Nilsondoesnotprovidedetailsonitssamplingproceduresandhenceitsaccuracyhasnotbeenvalidated(tomyknowledge).TheSurveyofConsumerFinances(SCF)isthehousehold-leveldatasourceusedinmycomparisons.ItisthemostcomprehensivenationallyrepresentativesourceofdataoncreditcarduseandhouseholdÞnancemoregenerally.TheSCFisconductedeverythreeyearsandsurveysaround4,000householdseachwave.Ifocusonthe1989Ð2004SCFsbecauseearlierG.19sdidnotcapturelargeswathsofoutstandings.ThereisnopanelcomponenttotheSCFduringthisperiod.ForeachtypeofcreditcarduseIconstructaggregateestimatesbyinßatingeachhouseholdbyitsSCF-assignedpopulationweight(variableTheSCFquestionsandpromptsarenearlyidenticalacrossthesixdifferentsurveysduringmystudyperiod.Thesurveycollectsthedataofintereststartingwithaquestiononwhetheranyoneinthehouseholdhasanycreditcardsorchargecards(withaprompttodistinguishthesefromdebitcards).ForrespondentsthatanswerÒyes,ÓthesurveyorthenasksÞveyes/noquestionsaboutwhetheranyoneinthehouseholdhasanyofÞvedifferenttypesofcards(generalpurposerevolving,store,gas,generalpurposecharge,other).FollowingeachÒyesÓanswerrespon-dentsareaskedhowmanytheyhaveofthattype,andarespeciÞcallyinstructedtoÒnotcountduplicatecardsonthesameaccountoranybusinessorcompanyaccounts.ÓRespondentsreportingoneormoreaccountsforagivencardtypearethenasked:ÒOnyourlastbill(s),roughlyhowmuchwerethenewchargesmadetothis/theseaccount(s)?ÓandÒAfterthelastpaymentsweremadeonthis/theseaccount(s),roughlywhatwasthebalancestillowedontheseaccounts?ÓTheseproduceaccounttype-levelmeasuresofwhatIlabelÒrecentchargesÓandÒrevolvingdebt.ÓTheBoardofGovernorsusesNilsondataasonesourceinconstructingthestorecardandgascardcomponentsofG.19outstandings.SeeBucksetal.(2006)formoredetailsontheSCF.ThemodernSCFstartedin1983andbeganitstriennialrepeatedcross-sectionsamplingin1989;assuch,theonlycandidateforapre-1989comparisonwouldbe1983.Butthe1983G.19omitsgeneralpurposecardsissuedbyÞnancialinstitutionsotherthancommercialbanks,manyimportantissuersofstorecards,andchargecards(Nilson#339).The1989Ð2004SCFsusemultipleimputationandprovideÞveimplicatesforeachhousehold.Iadjustpointestimatesforthisbydividingeachweighted-upestimateby5,andcorrectstandarderrorsperthestandardproceduredetailedinSCFcodebooks.ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 3.C OverviewofApproachSCFmeasurementofcreditcarditemsÑcreditcardaccountholding,charg-ing,andborrowingÑdiffersinpotentiallyimportantwaysfromindustrysources.ForeachitemtheSCFplausiblyproducesnarrowercoverage,bydesign,thantheG.19orNilson.TheSCFsandindustrymeasuresmayalsobecollectedatdifferenttimesofyear.ThissectionestimateshowmuchdiscrepancyremainsbetweenSCFandindustrysourcesafteradjustingfordifferencesindeÞnitionorcoverage(ÒdeÞ-nitionÓbelow)andseasonalityorothertiming(ÒtimingÓbelow).ForeachcreditcarditemandeachSCFsurveyyear1989Ð2004IstartbycomparingtheunadjusteddiscrepancybetweentheSCFandthemainindustrysourceforthatitem.ThisrawestimatedoesadjustfordeÞnitionortimingdifferences(thisistheÒcasualÓapproachmentionedatthebeginningofthepaper):itsimplytakesthenumberfeaturedintheG.19release(e.g.therevolvingdebtÒtotalÓ)oraNilsonreport(e.g.thetotalnumberofcreditcardaccounts)andsubtractstheweighted-upSCFaggregatefortherelateditem.TheratiooftheunadjustedindustrynumbertotheSCFnumberthengivestheunadjustedÒSCFUndercountingFactor.ÓIthenadjustfordeÞnitionandtimingdifferencesandrecalculatethedis-crepancyandSCFUndercountingFactor.Tablesreportestimatesforofadjustmentstosavespace.Iaccountforimprecisionintheaggregatesandtheadjustmentstheretousingrangesofpointestimates(minimal,moderate,andmaximal)becausetheindustrysourcesdonotreportstandarderrors,perstandardpracticewithmacroÞnancialaccounts.TherelevantSCFstandarderrorsaresmallandIreportsomeinthetexttogivethereaderasenseoftheSCFÕsprecision.Table2,Column1reportsestimatesoftheunadjusteddiscrepancybetweenthetotalnumberofaccountsreportedinNilsonandcalculatedfromtheSCF,andtheresultingSCFUndercountingFactor.TheunadjustedUndercountingFactorhererangesfromalowof1.6in1989toahighof2.4in2001and2004.TheSCFestimatesareprecise;forexample,the2004accounttotalhasanimputation-correctedstandarderrorof3million(whichisabout1/64oftheSCFpointestimate).Columns2Ð4ofTable2thenpresentestimatesofthediscrepancyandUndercountingFactorthatadjustforthreepotentiallyimportantdeÞnitionaldifferencesbetweentheG.19andSCF.Comparingcreditlimitsandpricingmightalsobeofinterest,butIcouldnotÞndindustrydataoncreditlimits,andpricingcomparisonsaregreatlycomplicatedbydatalimitationsonbothsides.Formoredetailsoneachindividualadjustment,pleaseseethetechnicalworkingpaperversionathttp://www.dartmouth.edu/~jzinman/Papers/Zinman_MissingCardDebt_sep07.pdf.AsAntoniewiczetal.(2005)note,ÒthecomplexstructureofthemacroÞnancialaccountsandthevastdisparatesourcesthatareusedasinputsmakecalculatingeventhemostsimplisticstandarderroradauntingtask.ÓIreporttotalsforgeneralpurpose(a.k.a.ÒbankÓ)creditcardsonlytoconservespaceandbecausetheygeneratethelionÕsshareofchargesanddebt(Table3).ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 TheÞrstadjustmentisforSCFtop-coding.Ineachofthe1995Ð2004surveysabout1percentoftheweightedsamplehastheirnumberofaccountscensoredat10.Myminimaladjustment(Column2)assumesthattheaveragetop-codedhouseholdhasoneadditionalaccountonaverage;themoderateandmaximaladjustments(Columns3and4)assume5and10additionalaccounts.Thisadjustmentaddsbetween1and12millionaccountstotheunadjustedSCFThesecondadjustmentisforinactiveaccounts.TheunadjustedNilsontotalincludesthem;SCFhouseholdsmaynotreportinactiveaccountstheyhavefor-gotten,closed,orperceiveasclosed.NilsonreportstotalsforinactiveaccountsseparatelyandIusethistomakeanadjustment.TheminimaladjustmentIssuersreportanindeterminatenumberofclosedaccountsasactiveduetoclericalerrors.TABLE2 DiscrepancyAfter:(1)(2)(3)(4)19895953261.61.51.31.019928070311.71.61.30.91995162146711.91.81.41.01998222200962.32.11.50.920012732371162.42.21.61.02004300249972.42.21.50.7 :Discrepanciesinmillionsofaccounts.Italics:SCFUndercountingFactor(Nilsonestimate/SCFestimate).Columns2Ð4adjustforSCFtop-coding,inactiveaccounts,andbusiness-relatedaccounts.SeeSection3.2oftextfordetailsandthetechnicalworkingpaperonlineformoredetails.TABLE3 (1)(2)(3)(4)WithAnyCreditCardProportionWithAnyGeneralPurposeCreditCardProportionofChargesonGeneralPurposeCardsProportionofRevolvingonGeneralPurposeCards19890.700.570.590.7119920.720.630.640.7419950.740.670.670.8419980.730.680.700.8720010.760.730.710.8720040.750.720.800.89 :AllestimatesfromtheSCFandweightedusingvariablex42001.ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 assumesthathouseholdsreportallinactiveaccounts(andhencedoesnotsub-tractanythingfromtheunadjustedNilsontotal),themoderateadjustmentsub-tracts50percentofinactiveaccountsfromtheunadjustedNilsontotal,andthemaximaladjustmentsubtractsallinactiveaccountsfromtheunadjustedNilsontotal(i.e.itassumesthatSCFhouseholdsdonotreportanyinactiveaccounts).Theeffectoftheadjustmentforinactiveaccountscanbelarge,andisincreasingovertime;forexample,Nilsonreported47millioninactiveaccountsin1989(29percentoftotalaccounts),and230millionin2004(45percentoftotalaccounts).Thethirdadjustmentisforbusiness-relatedaccounts.WhiletheSCFpromptsrespondentstoÒnotcount...anybusinessorcompanyaccounts,ÓtheunadjustedNilsontotalincludesthem.Accordinglymyminimaladjustmentheresubtractsanestimateofthecountofopencommercialaccounts(fromNilson)fromtheunad-justeddiscrepancy;thisadjustmentrangesfrom6millionin1989to50millionin2004.ThemoderateadjustmentalsosubtractsaconservativeestimateofpersonalaccountsusedforbusinesspurposesbasedonNilson;thisrangesfrom4millionin1989to32millionin2004.Themaximaladjustmentusesamoreaggres-siveestimateofpersonalaccountsusedforbusinesspurposesthatrangesfrom8millionin1989to64millionin2004.TheneteffectontheestimatedSCFUndercountingFactorforcreditcardaccountsvariesacrossthethreedifferentsetsofadjustments.Thefactorafterminimaladjustmentsrisesfrom1.5in1989to2.2in2004,andisquitecloseinmagnitudetotheunadjustedUndercountingFactorineachyear.Thefactorismuchsmallerandmorestableafterthemoderateormaximaladjustments.Themoderate-adjustedUndercountingFactorrangesfrom1.3to1.6.Themaximal-adjustedUndercountingFactoris0.9or1.0ineveryyearexcept2004,whenitdropsto0.7.RecentChargeVolumeTable4,Column1reportsestimatesoftheunadjusteddiscrepancybetweenmonthlytotalcreditcardchargevolumefromNilson(annualtotal/12)andtheSCF(weighted-upchargesonlastbills),andtheresultingSCFUndercountingFactor.TheunadjustedUndercountingFactorhereissmallerthanforaccounts(orfordebt,asdetailedbelow).Itrangesfrom1.2(in1989)to1.8(in1998),andwas1.5in2004.AswiththeotheritemstheaggregateSCFestimatesareprecise;forexample,2004chargeshaveanimputation-correctedstandarderrorofabout$3.5billion(whichisabout1/30oftheSCFpointestimate).Columns2Ð4ofTable4thenpresentestimatesofthediscrepancyandUnder-countingFactorthatadjustforthreepotentiallyimportantdeÞnitionandtimingdifferencesbetweentheG.19andSCF.TheuncertaintyaboutthecorrectpointestimateforpersonaluseofNilsonbusiness-relatedaccountscomeinpartfromthefactthatNilsonreportsthenumberofbusinessusingpersonalcards,ratherthanthenumberofaccounts.Ireportchargevolumeacrossallcreditcardsforcomparabilitytotherevolvingdebtestimatesinthenextsub-section(theG.19doesnotdisaggregatedebtbycardtype,makingitdifÞculttofocusongeneralpurposecardsonly).ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 TheÞrstadjustmentisforseasonality.WhiletheNilsonestimateisbasedonannualtotalcharges,SCFsurveysareconductedMayÐDecember.GiventhattheSCFasksaboutchargesonpreviousbills,thismeansthattheSCFmissesmostoftheChristmasdrivenuptickincharges.AlthoughIcouldnotÞndmonth-to-monthchargedata,Nilsonshowsthatchargesaretypicallyabout10percenthigherinthefourthquarter(OctoberÐDecember)thanintheaverageoftheÞrstthreequarters.ThusifSCFdatamissedthefourthquarteruptickcompletelyitwouldunderestimateaveragemonthlychargesbyabout3percent.SoIinßatetheSCFnumberby3percentinmymoderateadjustment,andby1percentand5percentintheminimalandmaximaladjustments.Thesecondadjustmentisforbusiness-relatedcharges.WhiletheSCFpromptsrespondentstoexcludebusinessorcompanyaccounts,theNilsonyear-endtotalincludeschargesoncommercialaccounts(whichtheSCFclearlymeanstoexclude)andbusiness-relatedchargesonconsumercards(whichtheSCFmayonlypartiallyexclude,ifsomeaccountsareusedforbothpersonalandbusinesspurposes).IusesupplementaldatafromNilson(oncommercialaccountchargesorbusiness-relatedchargesacrossallcards)andtheSurveyofSmallBusinessFinances(onbusiness-relatedchargesonconsumercards)toestimatetheappro-priateadjustment.Themaximaladjustmentsubtractsanestimateofallbusiness-relatedchargesfromtheNilsonunadjustedtotal.Thisisasubstantialamountinallyears;e.g.$5billionmonthly14percentoftheNilsontotalin1989,and$4227percentoftheNilsontotalin2004.Theminimaladjustmentsub-tractsanestimateofchargevolumeforcommercialaccountsonly.Thistooissubstantial(atleastrelativetotheunadjusteddiscrepancy)inallyears;e.g.$2Eachcreditcardbillingcycle(thetimeperiodcoveredbytheSCFquestion)coversamonth,soseasonalityistheonlyadjustmentisneededtomakeNilsonandSCFchargescomparable.TABLE4 DiscrepancyAfter:(1)(2)(3)(4)1989521.21.11.00.91992131261.41.41.21.01995251891.51.41.21.0199841281401.81.51.21.020014930121.61.41.10.92004542971.51.31.10.9 :Discrepanciesinnominalbillionsofdollars.Italics:SCFUndercountingFactor(Nilsonestimate/SCFestimate).Columns2Ð4adjustforseasonality,business-relatedcharges,andcashadvances.SeeSection3.3oftextfordetailsandthetechnicalworkingpaperonlineformoredetails.ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 billionmonthlyin1989,and$24billionin2004.Themoderateadjustmentusesthemidpointofminimalandmaximaladjustmentsforbusiness-relatedThethirdadjustmentisforcashadvances.TheNilsontotalincludesthem;theSCFquestionsmayexcludethemifrespondentsdonotthinkofcashadvancesasÒcharges.ÓThemaximaladjustmentassumesthatSCFrespondentsdonotreportanycashadvancesandhencesubtractsanestimateofmonthlycashadvancesfromtheunadjustedNilsonchargetotal.Thecashadvanceesti-matealsocomesfromNilsonandis,forexample,$2billionin1989and$23billionin2004.ThemoderateandminimaladjustmentssubtracthalfandnoneofNilsonÕscashadvancevolume.OnnettheadjustmentseliminatemostorallofthediscrepancybetweenchargesreportedintheSCFandNilson.TheestimatedyearlySCFUndercountingFactorrangesfrom1.1to1.5withtheminimaladjustments,from1.0to1.2withmoderateadjustments,andfrom0.9to1.0withmaximaladjustments.Chargesseemtomatchupwellunderarangeofassumptions.Table5,Column1reportsestimatesoftheunadjusteddiscrepancybetweentherevolvingÒtotalÓintheG.19September30releaseandtherelatedSCFitem(balancesremainingafterlastpayment),andtheresultingSCFUndercountingFactor.TheunadjustedUndercountingFactorrangesfrom2.9(in1989)to3.8(in2001),andwas3.1in2004.AswiththeotheritemstheSCFestimatesareprecise;e.g.2004totalrevolvingbalanceshaveanimputationcorrectedstandarderrorof$8billion(whichisabout1/30oftheSCFpointestimate,and1/67oftheTABLE5 DiscrepancyAfter:(1)(2)(3)(4)1989$132$125$67$212.92.82.01.31992$179$170$103$472.92.82.11.51995$297$296$195$993.33.32.51.81998$390$376$269$1733.23.12.52.02001$522$495$372$2633.83.73.02.42004$537$503$362$2183.13.02.41.9 :Discrepanciesinnominalbillionsofdollars.Italics:SCFUndercountingFactor(G.19estimate/SCFestimate)Columns2Ð4adjustfortransactionbalances,seasonality,non-creditcardlinesofcredit,andbusiness-relatedrevolvingonpersonalcardsintheG.19.SeeSection3.4oftextfordetailsandthetechnicalworkingpaperonlineformoredetails.ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 unadjustedestimateddiscrepancybetweentheG.19andSCF).Columns2Ð4thenpresentestimatesofthediscrepancyandUndercountingFactorthatadjustforfourpotentiallyimportantdeÞnitionandtimingdifferencesbetweentheG.19andAÞrstadjustmenttotheG.19revolvingtotalisfortiming.MostSCFsurveysareconductedduringMaythroughDecember,butindividualsurveydatesarenotreportedandhenceonecannotseasonallyweightthereportedcreditcarditems.HenceIusedifferentdatesfortheG.19tohelpestablisharangeforcomparison:December30(Column2),September30(Column3),andJune30(Column4).TheupwardtrendinG.19outstandingsthroughouttheyear(December30outstandingsareonaverage6percenthigherthanSeptember30,and8percenthigherthanJune30)impliesthattheJune30numbertypicallyproducesthesmallestdiscrepancyfromtheSCF,holdingconstanttheotherthreeadjustmentsdetailedbelow.ThesecondadjustmentremovesrecentchargesfromtheG.19.WhereastheSCFrevolvingbalanceestimateincludesonlybalancesafterthelastbillswerepaid,theG.19takesasnapshotofcurrentdebtoutstanding.AssuchtheG.19includesnotonlytheÒrevolvingÓbalancescapturedbytheSCFborrowingques-tions,butalsoÒtransactionÓbalances(basically,chargesincurredbyhouseholdssincetheirpreviousbillingcyclesclosed).Toadjustforthis,oneÞrstneedsesti-matesofmonthlycharges.Iconstructthesebysimplyaddinganadjusteddis-crepancyestimatedinTable4tounadjustedSCFmonthlycharges.Sincetheminimal(maximal)adjustmentinTable4producesalarger(smaller)baseofadjustedmonthlycharges,Iusethatminimal(maximal)adjustmenttoconstructthemaximal(minimal)adjustmentfortransactionbalancesneededinTable5forrevolvingdebt.Nextweneedanestimateoftheproportionoftransactionbal-ancesthatwillshowupintheG.19snapshotofoutstandings.ThetechnicalworkingpapershowshowtoestimatesensibleupperandlowerboundsforthisIuse1.35fortheupperbound:multiplyingthisbythemaximalestimateoftransactionbalances,andthensubtractingtheproductfromtheunadjustedG.19revolvingtotal,givesamaximaladjustment.Thisisadjustmentisquitelarge(e.g.$48billionand36percentoftheunadjusteddiscrepancyin1989;$198billionand37percentin2004).Iuse1.0forthemoderateadjust-ment,and0.5forthelowerboundandtheminimaladjustment.Eventheminimaladjustmentreducestheunadjusteddiscrepancysubstantially(e.g.by10percentin1989and8percentin2004).AthirdadjustmentremovesamountsfromtheG.19revolvingtotalthatarenotissuedthroughcreditcardsbutratherthroughprearrangedoverdraftplansorcheck-accessedlinesofcredit.(TheSCFquestionsreferspeciÞcallyandonlytocreditcardsÑthereareseparatequestionsforotherlinesofcredit.)Non-creditcardlinesaredisaggregatedintheG.19releasesbutcanberemovedfor2001and2004,followingFurlettiandOdy(2006),byreferencingtheG.19ÕssourceIdentifyingamorepreciseestimateoftheimpactoftransactionbalancesonG.19outstandingswouldrequiredataonthewithin-monthdistributionofchargesandthedurationofßoatobtainedbyconvenienceusers(i.e.bythosewhodonotrevolvedebt).ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 Theresultingpointestimatesforthesenoncreditcardlinesare$32billionin2001and$33billionin2004(i.e.6percentoftheunadjusteddiscrepancy).anecdotalevidencesuggeststhatmanyoftheseÒnon-creditcardÓlinescaninfactbeaccessedbycard(aswellasbycheck),somyminimaladjustmentinColumn2assumesthathalfofnon-cardlinescanbeaccessedbycardsandhencesubtractsonlyhalfofthepointestimatefromtheG.19total.Column3subtractsthenon-cardpointestimate.Column4allowsforthepossibilitythatmypointesti-matesunderstatetheamountofnon-cardoutstandingsÑthisisparticularlyimportantforearlieryears,wheremyestimatesarenoisierandnon-cardlinesmayhavebeenmoreprevalentÑandsubtracts1.5timesthenon-cardpointestimatefromtheG.19total.AfourthandÞnaladjustmentattemptstoremovebusiness-relatedrevolvingonpersonalcardsfromtheG.19.RecallthatwhiletheG.19includescreditcards;incontrasttheSCFinstructsrespondentstoÒnotcount...anybusinessorcompanyaccounts.ÓThisimpliesthattheG.19includessomeout-standingsonpersonalcardsthatareusedforbusinesspurposesandexcludedbydesignfromtheSCF.Iconstructtheminimaladjustmentforsuchoutstandings(inColumn2)usingquestionsfromthe1998and2003SurveysofSmallBusinessFinances(SSBFs)thatask:ÒOnaverage,whatisthebalanceofbusinesschargesonallownersÕpersonalcreditcardsafterpaymentsaremade?ÓIthenlinearlyinterpolateorextrapolatetogetestimatesfortheotheryearsinmysample.TheSSBFnumbersarequitesmallÑonly$1Ð7billion,or1Ð3percentofunadjustedSCFrevolvingbalances.ThesemagnitudesaretooconservativebecausetheSSBFdoesnotrepresentmanytypesofbusinesseswherepersonalcardsareused.Nilson(variousissues)Þndsthatmanylargebusinessesprovidecardsthatareinemploy-eesÕnames;thesemaybecountedaspersonalcardsintheG.19.MoreovertheSSBFonlythe6.3millionsmallbusinessesinDunÕsMarketIdentiÞerÞle,whileNilson(#772)reportsthatover20millionsmallbusinessownerspersonalcardsforbusinesspurposesin2002.Nilson(#776)alsoreportsatotalof$51billioninbusiness-relatedoutstandingsacrossbothpersonalandcommercialcardsforyear-end2001.SincecommercialcardsareexcludedfromboththeSCFandG.19Iuse$51billiontoconstructthemaximaladjustment(Column4),linearlyinterpolatingorextrapolatingwithatriennialgrowthratethatmatchestheunadjustedG.19revolvingnumber.Themoderateadjustmentthenusesthemeanoftheminimalandmaximal.TheneteffectoftheadjustmentsonthemagnitudeandtimepatternofthediscrepancybetweentheG.19andSCFvariesacrossthethreedifferentsetsofadjustments.Theminimaladjustmentsrangefrom2.8to3.7andproducediscrepanciesthatarejustslightlylowerthantheunadjusteddiscrepancyineachyear.Themoderate-adjusteddiscrepanciesrangefrom2.0to3.0andareCommercialbanksreportnon-creditcardlinesseparatelyinCallReportScheduleRC-CunderRCONB539.Priorto2001thisdisaggregationdidnotexistintheCallReports.Financecompaniesdonoteverreportnon-creditcardlinesseparatelyduringmysampleperiod,soIassumethattheyappearinthesameproportionasforbanks.Non-creditcardunsecuredrevolvingloansissuedbythriftsorcreditunionsareincludedintherevolvingportionoftheG.19andthereforedonotneedtobesubtractedheretoproducecomparableestimates.SeeFurlettiandOdy(pp.23Ð34andappendixI)formoredetails.Ilinearlyinterpolatetogetestimatesfortheyearsbefore2001.ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 somewhathigheronaverageinthelateryears.Themaximal-adjusteddiscre-panciesrangefrom1.3to2.4andareagainsomewhathigheronaverageinthelateryears.4.D Whydohundredsofbillionsofdollarsincreditcarddebtremainmissing,evenafteradjustingfordeÞnitionaldifferencesbetweentheG.19andSCF?Answeringthisquestioniscriticalforassessingtheusefulnessofhouseholdsurveydataonunsecuredborrowing,andfordesigningandadministeringsurveysthatelicitusefuldata.OnepossiblesourceofthediscrepancyisthatresearchershaveoverlookedsomethingthatinducesissuerstooverreportcreditcardassetsintheG.19Õssourcedata.ButasitstandstheweightoftheevidencepointsmoretowardSCFunder-countingthanG.19overcounting.SampleframeandresponserateissuesarenotlikelyexplanationsforSCFundercounting,sinceasnotedabovetheSCFseemstomatchupwellwithindustrydataonotherbalancesheetitemsandcreditcardcharges,andrelativelywelloncreditcardcreditaccounts.Butitremainspossiblethatsampleselectionissuesaffectcreditcardbalancesmorethanotherbalancesheetitems,and/orthatundisclosedpropertiesoftheNilsondatamaketherelativelyclosealignmentbetweenSCFandindustrymeasuresofchargesandaccountsamirage.NordoesnonresponseontheSCFcreditcardborrowingseemtobealikelycandidate.SCFstaffreportthattheyhavescrutinizedthecreditcardmodulecloselyandfoundnoevidenceofunusual(non)responsepatterns.Thisisnotsurprisinggiventhatthecreditcardquestionscomerelativelyearlyinthesurvey,minimizingthelikelihoodthatsurveyfatigueleadsrespondentstosupplyeasyanswerslikeÒzero.ÓIntentionalunderreportingofbehaviorthatisviewedassociallyundesirableisalwaysaconcernforhouseholdsurveys(Wyner,1980;Meansetal.,1992).ButtheSCFisskilledatelicitingsensitiveÞnancialinformation,andevidencefromMichiganSurveysofConsumersin2001(reportedinDurkin,2002)castsdoubtonthehypothesisthatcreditcardborrowingiswidelystigmatized.Eighty-ninepercentofcreditcardusersreportedthatcreditcardsmakemanagingÞnancesÒeasierÓorÒnodifferent,Ówhileonly10percentreportedÒmoredifÞcult.ÓNinetypercentofgeneralpurposecardholdersreportedbeingÒverysatisÞedÓorÒsome-whatsatisÞedÓoverallwiththeircard(s).Moreoveritseemslikelythatifstigmaplayedaroleitwouldbedecliningovertimeascreditcardusebecomesmoreprevalent.YettheUndercountingFactorforrevolvingbalancesintheSCFishigherin2004than1989byanyofthefourestimatesinTable5.Soitseemslikelythatsometypeofunderreportingplaysanimportantrole.TheimpactofthesheercomplexityofaccountingforcreditcardborrowingthroughoutthehouseholdÑmosthouseholdshavemultiplecards,potentiallyusedbyseveraldifferenthouseholdmembersÑonreportingisimportanttoconsider.ButcomplexityisultimatelyunconvincingasanexplanationforAccountingforcreditcardchargesrequiresarithmeticthanaccountingforReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 currentbalances,yetSCFchargesseemtomatchupwelltotheindustrydata.Othercomplexbalancesheetitems(e.g.ontheassetside)alsomatchupwellwithindustry-sidedata.Andevenifrespondentsmakeunintentionalreportingduetotheprivatenatureofcreditcardborrowingbyotherhouseholdmembers,itishardtoimaginewhytheseerrorswouldbebiasedinthedirectionofunderre-porting.Whynotmean-zero,orreportingerrors?Amorenovelexplanationfocusesonwhyrespondentsmightsystematicallyunderestimatetheircreditcardbalances.Resultsfromalaboratoryexperi-mentsuggestthatconsumersunderestimateorforgetcreditcardpurchasesbecausetheactofpayingbycreditcardislessmemorableandpainfulthanpayingbycheck(Soman,2001).Creditcardsmaystimulatespendingasaresult(seealsoThaler,1999),raisingtheintriguingpossibilitythatapsychologicalbiasaffectsnotonlyreportedbutcreditcardborrowingbehavior.ButthisexplanationmightstruggletoaccountfortherelativelycompleteSCFreportingofcreditcardInalltheredoesnotseemtobeanyobvious(oratleastanysingle)explana-tionforSCFundercountingofcreditcardborrowing.5.ISCFUAsnotedattheoutset,SCFundercountingofcreditcardborrowingwillbeproblematicforresearchersifitiscorrelatedwithunobservedheterogeneityinpreferences,resources,oranotherfactorthatisinturncorrelatedwithoutcomesofinterestlikeÞnancialcondition,consumptionpaths,orportfoliochoice.Thevaria-tionovertimeinundercountingfoundinTable5,Columns(3)and(4)isworri-some,particularlygiventhechangesincreditcarduseandusersdocumentedinTables1and3.Table3,Column2showsthatmanyhouseholds(about26million)enteredthecreditcardmarketfrom1989to2004.Theproportionofchargesandrevolvingdebtongeneralpurposecardsrosesharplyaswell(Columns3and4).Table1suggeststhatmarginalcardholders(i.e.thosewithonlyonegeneralpurposecreditcard,andhencethosemostlikelytohaveenteredthemarketandchangedaggregatereportingbehavior)weremorelikelytobefemale,nonwhite,unmarried,andhavebelowmedianincomein2004thanin1989.Debtreportingbehaviormayvarywithdemographiccharacteristicssuchasthese;KarlanandZinman(2008)Þndevidencesuggestingthatfemales(andperhapsrelativelylow-incomeborrowers)arelesslikelytoreporttheirexpensive,unsecureddebt.Unobservedheterogeneityinrespondentreportingofcreditcarddebtcouldpresentaseriousproblemforresearchersacrossarangeofliteraturesandmethods.Itwouldconfoundattemptstoestimatethedistribution,incidence,anddeterminantsofborrowingcosts;thisconcernapplies,forexample,tothelit-eratureexploringwhyhouseholdssimultaneouslyborrowexpensivelyoncardswhilelendinginlow-yieldingdemanddeposits.StudiesinthisveinusingtheSCFincludeBertautandHaliassos(2009),GrossandSouleles(2002),Telyukova(2008),TelyukovaandWright(2008),andZinman(2007).Anexampleoftheinferenceprobleminthisliteratureisitsattempttoanswerthequestionofwhetherless-educatedconsumersaremorelikelytoborrowhighandlendlow.IfeducationiscorrelatedwithreportingbehavioritwillbeunclearwhethercostsReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 areattributabletoeducationperse,eveninanotherwisewell-identiÞedmodel.UnobservedheterogeneityinreportingmightalsobiasestimatesofparametersormodelÞt;theseconcernsapply,forexample,toliteraturesestimatingtheeffectofdebtburdenonÞnancialdistress.Calemetal.(2006)usetheSCFtoestimatetheeffectofcreditcardborrowingonÞnancialdistress,conditionalonarichsetofcontrolvariables.Theconcernhereisthatunderreportingofcreditcardborrowingthatiscorrelatedwithincompletelyobserveddriversofdistress(e.g.riskaversion,impatience,accesstoothersourcesofliquidity)willbiastheestimateofborrowingÕscausaleffectinanindeterminatedirection.Itmightalsoaffectstudiesoftheeffectofhousepricesonconsumption,andtheÞtofalternativemodelsofintertemporalandportfoliochoice,e.g.computationallife-cyclemodelsincreasinglyusecreditcardborrowingasakeyoutcometobeÞtandfocusonparticularsub-samples(cohorts,educationgroups,Þnancialmarketparticipants,etc).Inall,theavailableevidenceisquitefarfrompermittingdeÞnitiveinferenceonwhetherSCFdataoncreditcardborrowingisusefulforresearchers.ButthepatternofevidenceisconsistentwithchangesintheSCFUndercountingFactorbeingdrivenbychangesincreditcarduseand/orusersthatcouldwellbecorre-latedwithconfoundingunobservedheterogeneity.Thisdisconcertingpossibilityhighlightstheimportanceofidentifyingthesource(s)oftheremainingdiscrepancybetweenSCFandindustrymeasuresofcreditcarddebt.Thenextsectiondiscussespossibilitiesforfutureworkalongtheselines.6.CManygovernmentagencies,internationalorganizations,andresearchers(includingtheauthor)conducthouseholdsurveysthatincludecreditcardsorotherhigh-frequencyborrowing.AsinterestinhouseholdÞnancecontinuestogrow,demandforsurveydataontheseimportantsourcesofcreditislikelytoincrease.IcreatecomparablemeasuresofaggregatecreditcardusebasedonhouseholddatafromtheSurveyofConsumerFinances(SCF),andindustrydatafromtheG.19andNilsonReports.TheSCFandindustrysourcesmatchupwelloncreditcardcharges,fairlywellonaccounttotals,andnotespeciallywellonrevolvingdebt.Myestimatedlowerboundfor2004debtdiscrepancyis$200billion.TheavailableevidenceisfarfromdeÞnitivebutsuggeststhattheSCFmisseshalfofrevolvingcreditcarddebt.PerhapsmoreworrisomeisthatIalsoÞndsomeevidenceofgrowthovertimeinthedebtdiscrepancy.Thisgrowthparallelssubstantialchangesincreditcarduse(theshifttogeneralpurposecards)andcreditcardusers(moremarginalcreditsSeeKarlanandZinman(2008)forrelatedevidencefromadifferentsetting.Manystudiesreviewthegrowinginterestinthedeterminantsandeffectsofhouseholddebtburden;see,e.g.Debelle(2004),DynanandKohn(2007),andWhite(2007).Hereunbiasedmeasuresofliquidityconstraints(andhencepresumablyofavailablecreditundercreditcardlimits)areneededtoidentifyorinterpretparametersofinterest;see,e.g.CampbellandCocco(2007)andHurstandStafford(2004).Unobservedheterogeneityinreportingbehavioracrossthesesub-samplescouldproducemis-leadingestimatesofmodelÞtorcalibrated/inferredparameters.StudiesinthisveinusingtheSCFincludeAngeletosetal.(2001),Carroll(2001),Davisetal.(2006),andLaibsonetal.(2009).ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 enteringthepool).TheparallelsuggeststhatanySCFundercountingmaybecorrelatedwithunobservedheterogeneityinhouseholdcharacteristicsthatcouldconfoundinferenceontherelationshipbetweencreditcardborrowingandout-comesofinterestlikehouseholdÞnancialcondition,consumptionpaths,andportfoliochoice.IftheSCFÑagoldstandardforelicitingcomprehensiveandaccurateÞnancialinformationÑisproblematic,thenotherhouseholdsurveyslikelyfaceevengreaterproblems.Giventhisdisconcertingpossibilityitiscriticaltodevelopsharperevidenceonwhetherandwhyhouseholdsurveysactuallyundercountcreditcardborrowing.OneapproachistocontinuescrutinizingtheG.19andotherindustrysources.Forexample,partofthediscrepancycouldbeduetochangesinthecompositionofcreditcarddebtissuersandhenceintheirdatareporting(see,e.g.FurlettiandOdy,2006,appendixF).Anotherapproachistocollectricherdatawithinexistingreportinginstruments.ThereconciliationexercisesinthispaperhighlightseveraladditionalpiecesofsurveyandindustrydataÑe.g.oncreditcardsforbusinessuse,onmonthswithunusuallyhighcharges,onthewithin-monthtimepathofchargesÑthatcouldbeusedtoobtainmorepreciseestimatesofthemagnitudeandsource(s)ofthediscrepancybetweenhouseholdandindustrymeasuresofcreditcardborrowing.Futureworkwouldalsodowelltofocusonvalidationstudiesusinglenderdataand/orincentivecompatibleelicitationprocedures(e.g.agamethatrewardssubjectsforaccurateestimationofacreditcardbalances,wherethesubjectherselfisresponsiblefordocumentingtheactualbalance).Thesemethodswouldpermitsharpertestsofwhetherconsumersactuallyunderreportsystematically.Coupledwithsupplementaldatacollectionondecisioninputs(e.g.preferences,expecta-tions,problem-solvingability,liquidityconstraints)theywouldalsoshedlightonwhatdrivesanyunderreporting,andonwhetherthesedriversarelikelytocon-foundinferenceusingcreditcardborrowingdatafromhouseholdsurveys.Angeletos,George-Marios,DavidLaibson,AndreaRepetto,JeremyTobacman,andStephenWein-berg,ÒTheHyperbolicConsumptionModel:Calibration,Simulation,andEmpiricalEvaluation,ÓJournalofEconomicPerspectives,15(3),47Ð68,2001.Antoniewicz,Rochelle,RiccardoBonci,AndreaGenerale,GraziaMarchsese,AndreaNeri,KarenMaser,andPatrickOÕHagan,ÒHouseholdWealth:ComparingMicroandMacroDatainCanada,Italy,andUnitedStates,ÓWorkingPaper,BoardofGovernorsoftheFederalReserve,January18,2005.Bertaut,Carol,MichaelHaliassos,andMichaelReiter,ÒCreditCardDebtPuzzlesandDebtRevolversforSelfControl,ÓReviewofFinance,2009,forthcoming.Brackstone,Gordon,ÒStatisticalUsesofAdministrativeData:IssuesandChallenges,ÓProceedingsofStatisticsCanadaSymposium,5Ð16,1987.Bucks,Brian,ArthurKennickell,andKevinMoore,ÒRecentChangesinFamilyFinances:Evidencefromthe2001and2004SurveyofConsumerFinances,ÓFederalReserveBulletin,92,A1Ð38,February2006.Calem,Paul,MichaelGordy,andLorettaMester,ÒSwitchingCostsandAdverseSelectionintheMarketforCreditCards:NewEvidence,ÓJournalofBankingandFinance,30,1653Ð85,Campbell,JohnY.andJoaoCocco,ÒHowDoHousePricesAffectConsumption?EvidencefromMicroData,ÓJournalofMonetaryEconomics,54,591Ð621,2007.Carroll,ChristopherD.,ÒATheoryoftheConsumptionFunction,WithandWithoutLiquidityJournalofEconomicPerspectives,15(3),23Ð45,2001.ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009 Davis,Steven,FelixKubler,andPaulWillen,ÒBorrowingCostsandtheDemandforEquityAcrosstheReviewofEconomicsandStatistics,88,348Ð62,2006.Debelle,Guy,ÒHouseholdDebtandtheMacroeconomy,ÓBISQuarterlyReview,March2004.Durkin,Thomas,ÒConsumersandCreditDisclosures:CreditCardsandCreditInsurance,ÓReserveBulletin,88,201Ð13,2002.Dynan,KarenandDonaldKohn,ÒTheRiseinU.S.HouseholdIndebtedness:CausesandConse-quences,ÓFEDSWorkingPaper,2007.Edelberg,Wendy,ÒRisk-BasedPricingofInterestRatesforConsumerLoans,ÓJournalofMonetary,53,2283Ð98,2006.Elliehausen,GregoryandEdwardLawrence,ÒPaydayAdvanceCreditinAmerica:AnAnalysisofCustomerDemand,ÓCreditResearchCenter,GeorgetownUniversity,April2001.Furletti,MarkandChristopherOdy,ÒMeasuringU.S.CreditCardBorrowing:AnAnalysisoftheG.19ÕsEstimateofConsumerRevolvingCredit,ÓFederalReserveBankofPhiladelphiaPaymentCardsCenterDiscussionPaper,April2006.Gross,DavidandNicholasSouleles,ÒDoLiquidityConstraintsandInterestRatesMatterforCon-sumerBehavior?EvidencefromCreditCardData,ÓQuarterlyJournalofEconomics,117,149Ð85,Hurst,ErikandFrankStafford,ÒHomeIsWheretheEquityIs:MortgageReÞnancingandHouseholdJournalofMoney,Credit,andBanking,36,985Ð1014,2004.Johnson,Kathleen,ÒTheTransactionsDemandforCreditCards,ÓTheB.E.JournalofEconomicAnalysis&Policy:Topics,7(1),Article16,2007.Karlan,DeanandJonathanZinman,ÒLyingAboutBorrowing,ÓJournaloftheEuropeanEconomicAssociationPapersandProceedings,6(2Ð3),August2008.Laibson,David,AndreaRepetto,andJeremyTobacman,ÒEstimatingDiscountFunctionswithCon-sumptionChoicesOvertheLifecycle,ÓAmericanEconomicReview,2009,forthcoming.Means,Barbara,KatherineHabina,GarySwan,andLisaJack,ÒCognitiveResearchonResponseErrorinSurveyQuestionsonSmoking,ÓVitalHealth&Statistics,Series6,Number5,NationalCenterforHealthStatistics,Hyattsville,MD,1992.Soman,Dilip,ÒEffectsofPaymentMechanismonSpendingBehavior:TheRoleofRehearsalandImmediacyofPayments,ÓJournalofConsumerResearch,27,460Ð74,2001.Sullivan,JamesX.,ÒBorrowingDuringUnemployment:UnsecuredDebtasaSafetyNet,ÓJournalofHumanResources,43,383Ð412,2008.Telyukova,Irina,ÒHouseholdNeedforLiquidityandtheCreditCardDebtPuzzle,ÓWorkingPaper,UniversityofCaliforniaÐSanDiego,December2008.Telyukova,IrinaandRandallWright,ÒAModelofMoneyandCredit,withApplicationtotheCreditCardDebtPuzzle,ÓReviewofEconomicStudies,75,629Ð47,2008.Thaler,Richard,ÒMentalAccountingMatters,ÓJournalofBehavioralDecisionMaking,12,183Ð206,White,Michelle,ÒBankruptcyReformandCreditCards,ÓJournalofEconomicPerspectives,21,175Ð99,2007.Wyner,Gordon,ÒResponseErrorsinSelf-ReportedNumberofArrests,ÓSociologicalMethodsand,9,161Ð77,1980.Zinman,Jonathan,ÒHouseholdBorrowingHighandLendingLowUnderNo-Arbitrage,ÓWorkingPaper,DartmouthCollege,April2007.ReviewofIncomeandWealth,Series55,Number2,June2009©2009TheAuthorJournalcompilation©InternationalAssociationforResearchinIncomeandWealth2009