Cost of Banking for LMI and Minority Communities

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Cost of Banking for LMI and Minority Communities Marco Migueis, Michael Suher, and Jessie Xu 2022-040 Please cite this paper as: Migueis, Marco, Michael Suher, and Jessie Xu (2025). “Cost of Banking for LMI and Minority Communities,” Finance and Economics Discussion Series 2022-040r1. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2022.040r1. 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.

Cost of Banking for LMI and Minority Communities Marco Migueis*, Michael Suher†, and Jessie Xu‡ February 2025 Abstract: Wetestwhetherminimumaccountbalancestoavoidfees,maintenancefeeamounts, andnonsufficientfundschargesaresystematicallydifferentinLMIandmajority­minoritycommu­ nitiesrelativetoothercommunitiesandfindthattheyaregenerallyhigher. Theminimumaccount balancetoavoidfeesonanoninterestcheckingaccountisabout$45higheronaverageinLMICen­ sus tracts than in higher income tracts, and more than $70 higher on average in majority­minority tracts than in majority­white tracts. We investigate potential sources of these differences such as bank business models, competition among providers of financial services, and other demographic characteristicsofthetracts. Opportunitiesforlendingincome,bankoperatingcosts,andbanksize can largely account for the higher cost of retail banking services in LMI and majority­minority communities. Keywords: Bankfees;depositaccounts;LMI;majority­minority JELClassificationNumbers: G21,G50,L11,J15,R23,I30 *FederalReserveBoard. Correspondingauthor: mmigueis@gmail.com. †FederalReserveBoard. ‡DukeUniversity. The views expressed in this manuscript belong to the authors and do not represent official positions of the Federal ReserveBoardortheFederalReserveSystem. TheauthorsthankElizabethDuncanandNatalieKrattsforexcellent researchassistance. 1

1 Introduction Bankaccountsfacilitateparticipationinthemoderneconomyandtheiruseisanintegralpartof amiddle­classlifestyleinmodernsocieties;forexample,wagesinmostsalariedjobsaretypically paidthroughdirectdeposittoabankaccount. However,thecostsassociatedwithdepositaccounts can take a toll on low­income bank customers. Account maintenance fees are often charged to customers with low balances, making those with limited savings more susceptible to these fees. Low­incomebankcustomersarealsomorelikelytofrequentlyoverdrafttheiraccounts,whichcan resultinheftyfees(PewCharitableTrusts,2016). Thispaperinvestigateswhetherlow­andmoderate­income(LMI)andmajority­minoritycom­ munities face the double whammy of higher minimum account balances and higher bank account fees. We begin by assessing the relation between bank fees and the income of bank customers by testing whether the fees and minimum account balances charged by banks in LMI Census tracts are higher than those charged by banks in higher income areas. We find that banks in LMI tracts chargehighermaintenancefeesandrequirelargerbalancestoavoidthosefees. Forexample,banks inLMItractsrequire,onaverage,abalanceabout$45highertoqualifyforfreecheckingandoth­ erwisechargeamonthlymaintenancefeearound$0.30higher,bothofwhichareabout5%higher thantheminimumsandfeesinnon­LMItracts. We also investigate fees in majority­minority Census tracts (i.e., Census tracts with over 50 percent minority residents). Residents in majority­minority tracts face significantly higher fees andminimumaccountbalancesacrossthevarioustypesofbankaccounts. TheassociationofLMI tractswithhigherfeesandminimumsdecreasesafteraccountingforthemajority­minoritystatusof atract,butretainsstatisticalsignificanceinmostcases. Basiccheckingaccountsofferedbybanks in tracts that are both LMI and majority­minority require minimum balance and maintenance fees about8%higher,onaverage,thansimilaraccountsinhigherincome,majority­whitetracts. We consider potential drivers for such differences in bank account fees between LMI commu­ nities, majority­minority communities, and other communities, including banks’ other sources of revenue, bank costs, competition among banks, and other demographic characteristics of the Cen­ 2

sus tracts. Higher lending income is generally associated with lower fees and minimums. Higher operatingcostsaregenerallyassociatedwithhigherfeesandminimums. Andlargebanksgenerally charge higher fees and require higher minimums. Controlling for these and other factors substan­ tially explains the association between LMI and majority­minority tracts and bank account fees, thoughsomeresidualunexplainedhigherfeesandminimumsremainformajority­minoritytracts. OurresultsaregenerallyconsistentwithAdams(2017),whofindsthatlow­incomeandminor­ itypopulationsgenerallypayhigherbankfees. UnlikeAdams(2017),ouranalysisisfullyfocused on the cost of banking for LMI and majority­minority populations and we consider several addi­ tional indicators of the cost of retail banking services, such as minimum deposit balances to avoid account maintenance fees. Our results are also consistent with the findings of Faber and Friedline (2020) that minimum account balances to avoid fees are substantially higher in majority­minority communitiesthaninmajority­whitecommunities. The rest of this article is divided as follows: Section 2 provides background and discusses the literature on the cost of banking for LMI and minority populations; Section 3 describes our data; Section 4 describes our statistical methodology and presents our empirical results; Section 5 concludes. 2 Background Bankfeerevenueshavemorethandoubledinthepastthreedecadesandhavegrownasashare of banks’ total revenues (Pew Charitable Trusts, 2016).1 According to bank regulatory reports, service charges on deposit accounts were equivalent to 21% of the net income of US banks in 2015, the year of our study. Major fee types include account maintenance fees and overdraft and non­sufficientfunds(NSF)fees. Account maintenance fees have significant impacts on the financial health of low­income and 1Previousresearchpapershavearguedthathighretailbankingfeesmaybe,inpart,anunintendedconsequence oflegislation(Bernard,2011;Sarin,2019). Followingthe2008financialcrisis,theDurbinAmendmentofthe2010 Dodd­FrankActrequiredbankstocutfeeschargedondebitcardprocessing. Bankswereestimatedtolose$6.6billion inrevenuesfromthisact,whichthesepapersarguecausedmanybankstointroducenewmonthlymaintenancefees. 3

minority bank customers. LMI households are estimated to have paid $1.4 billion out of the $2.5 billion U.S. banks earned from checking and savings account maintenance fees in 2021 (Greene et al., 2021). Also, Faber and Friedline (2020) find that checking accounts have higher minimum account balances to avoid fees in majority­minority communities than in majority­white commu­ nities($810inmajority­minoritycommunitiesvs. $620inmajority­whitecommunities). U.S. banks with assets over $1 billion collected over $11.45 billion in overdraft and non­ sufficient funds (NSF) fees in 2017 (Smith, 2018). Overdraft fees are high – the median over­ draft fee was $35 as of 2016 (Pew Charitable Trusts, 2016) – and often not well understood by customers.2,3 Overdraft and NSF fee revenue is concentrated in a small share of customers.4 The mostfrequentoverdraftershavelowcreditscores(sub­600)orlackcreditscorescompletely,carry low account balances, and are more likely to use debit card transactions and ATM services. Over­ draftfeeshitlow­incomeconsumersparticularlyhard. Asof2014,sevenintenconsumerspaying over$100inoverdraftfeesbelongedtohouseholdswithlessthan$50,000inannualincome(Pew CharitableTrusts,2016). BankedhouseholdsaredefinedbytheFederalDepositInsuranceCorporation(FDIC)asthose that have at least one checking or savings account at a bank or credit union; unbanked households as those that have no bank or credit union account; and underbanked households as those that use both traditional banking services and alternative financial services (AFS), such as payday lending (Federal Deposit Insurance Corporation, 2018).5 According to a 2017 FDIC survey, 8.4 million UShouseholds(6.5%)are“unbanked”andanadditional24.2millionhouseholds(18.7%)are“un­ derbanked.” Thelackofarelationshipwithabankoftenresultsinlow­incomehouseholdskeeping 2Thefederal“opt­in”ruleprohibitsfinancialinstitutionsfromchargingoverdraftfeesonATMandone­timedebit­ card transactions without customer consent (Federal Reserve System, 2009). However, banks often fail to discuss alternatives with frequent overdrafters. Seven in ten overdrafters are not aware that they have the right to overdraft protectionondebitcardsforfree(PewCharitableTrusts,2017). 3Alanetal.(2018)findthatTurkishbankclientsweremorelikelytooverdraftwhenfeesarehighbutnotmentioned intheaccountpromotionalmaterialsthanwhenfeesarelowbutmentionedintheaccountpromotionalmaterials. 4A2017studyfromtheConsumerFinancialProtectionBureaufoundthat79%ofoverdraftandNSFfeesarepaid byjust9%ofaccountholders(Lowetal.,2017). AsurveydiscussedinStangoandZinman(2009)findsthat68%of respondentspaidnooverdraftfees, whilethe90thpercentileofthosewhohadsomepositiveoverdraftamountpaid $43permonthonaverage. 5AFSisabroadtermforawiderangeofservices,includingcheck­cashing,paydayloans,pawnshops,early­access wages,orhigh­feeprepaidcards. 4

cash at home (Brobeck, 2020). According to the 2016 Federal Reserve Survey of Consumer Fi­ nances, households in the lowest quintile of the income distribution held only $900 in financial assets on average (which is less than the average minimum balance, $971, to have a free non­ interest checking account according to 2015 RateWatch data) and only 31% of households in this quintileheldasavingsormoneymarketdepositaccount(Brobeck,2020). Low­incomeconsumerscitemultiplereasonsfornotusingtraditionalbankaccountsincluding lackofsufficientsavings,highfeesandminimumbalancerequirements,andhiddenfeesleviedby banks (Booz­Allen Hamilton and Shugoll Research, 1997; Berry, 2004; Pew Health Group, 2011; Sarin, 2019).6 Our study aims to (1) compare the bank account fees faced by LMI and minority communitiestothosefacedbyhigherincomeandwhitehouseholdsand(2)understandthefactors thatexplainanydifferences. 3 Data Data on retail banking fees are gathered from a 2015­16 survey by RateWatch. Our dataset includes 81,909 branches, accounting for about 90% of bank branches across the US in 2015. We focus our analysis on the minimum account balances to avoid account maintenance fees and the account maintenance fee amounts for non­interest checking deposit accounts, interest checking deposit accounts, and savings accounts. In addition, we consider the returned check fees due to NSFandthedailymaximumNSFchargesthatcanbeaccruedinadepositaccount. The unit of analysis in the main regressions of our study is the Census tract. To examine the bankfeesexperiencedbytheresidentsofaCensustract,weaveragetheaccountminimumbalances andthefeeamountsacrossthebankswithbrancheswithinthreekilometersofthetractcentroid.7,8 6Studieshavealsofoundthatbanksaresignificantlyunderrepresentedinlow­income,urbanneighborhoods. To­ getherwithhighfees,distancetobankbranchesincreasesAFSusebyhouseholdslocatedinlow­incomeareas(Caskey, 1994;GoodsteinandRhine,2017). AFSproviderstargettheirservicestowardslowtomiddle­incomeborrowers(Barr, 2004),andlow­incomeconsumersseeAFSprovidersasattractivefortheirabilitytoprovidecashquickly. 7InlessdenseareaswheretherearenobankbrancheswithinthreekilometersofaCensustractcentroid,theclosest branchtothetractcentroidisassumedtobetheneighborhood’slocalbankingoptionanditsassociatedmenuoffees andaccountminimumsareassignedtothetract. Ifthereisnobranchwithin15kilometersofatract’scentroid, the tractisdroppedfromthesample. 8Theweightofabank’sfeevariablesonatract’saverageisproportionaltothenumberofbranchesofthebankin 5

Figure 1 visualizes this process for an example neighborhood. In this instance, the highlighted reference tract in Washington, DC is linked to the two branches inside its border and seven more branches within a three kilometer radius, all represented by triangles. More distant branches de­ notedbycirclesareconsideredoutsidethisCensustractgeographicmarket. Figure1: DefiningNeighborhoodBankBranches Note: BranchlocationsfromFDICSummaryofDeposits. Theexampleshownillustrateshowwedefinethelocal bankingmarketforaWashington,DCCensustract. Table 1 presents descriptive statistics for the bank fee variables. We restrict the sample to tracts where all eight fee variables of interest are available. Interest­bearing checking accounts – which pay interest without the withdrawal limitations associated with savings accounts – require, on average, substantially higher account balances to avoid account maintenance fees than non­ thetract. 6

interest­bearing checking accounts ($7,485 vs. $992). Meanwhile, savings accounts require, on average, the smallest balances to avoid account maintenance fees ($279). The average account maintenancefeeamountslineupsimilarly($16.2forinterestbearingcheckingaccounts,$7.88for non­interest­bearing checking accounts, and $4.42 for savings accounts). For perspective, these statisticsimplythatfortheaveragenon­interestcheckingaccount,failuretomaintaina$992mini­ mumbalancewouldresultinannualmaintenancefeesofabout$95,spellingtheerosionofatleast 10% of the account’s value. Maximum daily NSF fees average $167, and the average fee for a returned check due to NSF is $33.8. Minimum account balances to avoid account maintenance feesandfeeamountsarepositivelycorrelatedacrosstheboard.9 Table1: AccountFeesandMinimumBalanceRequirementsbyCensusTract N Mean StdDev Minimum Maximum No­InterestCheckingMinimum 58,246 992 608 0 15,000 InterestCheckingMinimum 58,246 7,485 4,259 0 25,000 SavingsMinimum 58,246 279 102 0 6,700 No­InterestCheckingFee 58,246 7.88 2.88 0 20 InterestCheckingFee 58,246 16.2 4.77 0 37.5 SavingsFee 58,246 4.42 1.29 0 25 NSFDailyMax 58,246 167 38.0 15 500 NSFReturnedCheck 58,246 33.8 2.39 10 42.5 Note: BankfeedatafromRateWatchsurveyofU.S.banksfrom2015/2016. DataisaveragedbyCensustractas describedatthebeginningofsection3. Minimumreferstotheaccountbalancerequiredtoavoidmonthly maintenancefees. NSFstandsfornon­sufficientfunds. NSFDailyMaxisthelimitononeday’saccumulated overdraftfeesafterwhichadditionaloverdraftsaredenied. We obtain the main explanatory variables of our study, whether a Census tract is LMI or majority­minority,fromthe2015FederalFinancialInstitutionsExaminationCouncil(FFIEC)Cen­ susFile.10 LMItractsaredefinedbyamedianfamilyincomebelow80%oftheirmetroareamedian familyincome. Majority­minoritytractsaredefinedbytheminorityshareoftheirpopulationbeing above 50%. In addition, our analysis controls for multiple demographic factors obtained from the 9SeeTableA.1intheAppendix. 10Seehttps://www.ffiec.gov/censusproducts.htm. 7

AmericanCommunitySurvey(ACS)2013­2017five­yearsample. Demographicvariablesinclude the percentage of the population above 65, the percentage of the population that attained a bache­ lor’s degree, and the percentage of homes occupied by their owners. Tract are classified as urban, suburban,orruralaccordingtotheUrbanizationPerceptionsSmallAreaIndex(UPSAI)produced bytheDepartmentofHousingandUrbanDevelopment(HUD).11 In addition, our analysis includes several controls based on bank data. We obtain lending in­ come,operatingcosts,and“othercharge­offs”for2015andtotalassetsasof2015Q4fromabank’s callreport.12 Weobtainthenumberofbranchesabankhasinatract’smarketareaandbranchages from the FDIC Summary of Deposits data. We identify which banks are minority depository in­ stitutions (MDI) from the FDIC Minority Depository Institutions list.13 We construct tract level controls based on these variables as follows: (1) in calculating the operating costs variable for a tract, we first calculate the operating costs of each branch in a tract by dividing their bank’s total operating cost by its number of branches; then, we average the operating costs for branches in the tract (this procedure mirrors how we calculate the average fees for a tract); (2) in calculating the lending income and the other changeoffs variables for a tract, we first divide a bank’s lending in­ comeorotherchargeoffsbyitstotalassets;then,wecalculateaweightedaverageoftheseforbanks in a tract, where each bank is weighted in proportion to how many branches it has in the tract; (3) branchnetworkageistheaverageageofthebranchesinatract’smarketarea;(4)bigbankshareis theproportionofbankbranchesinatractthatbelongtoabankwithmorethan500U.S.branches; (5)theMDIvariablecorrespondstotheproportionofbankbranchesinatractthatbelongtoaMDI bank; and (6) the Herfindahl­Hirschman Index (HHI) of concentration is calculated as the sum of squareddepositsharesofthebanksinthetract.14 11HUDusedsurveyresponsesinthe2017AmericanHousingSurveywhereresidentsself­identifiedtheirtractas urban,suburban,orruralandstatisticallyextrapolatedtheselabelstoalltractsusingACSdata. 12Abank’scallreportincludescomprehensivefinancialdata,suchasbalancesheetandincomestatementdata. All nationalandstatecommercialbanksarerequiredtosubmitthisreportonceperquarter. Othercharge­offscorrespond to item 5.c of schedule RI­B of the call report. They include charge­offs on extensions of credit to individuals for household,family,andotherpersonalexpensescharged­off. Inparticular,thisitemincludescharge­offsonamounts over­draftedbyconsumers. 13Minorityownershipisdefinedas51%concentrationofownershipamongmembersofacertainminoritygroup. Seehttps://www.fdic.gov/minority­depository­institutions­program/minority­depository­institutions­list. 14Branchesofthesamebankshavetheirdepositsharesaggregated. Inthelimit,ifallbranchesinatractbelongto 8

Table2presentsdescriptivestatisticsfortheexplanatoryandcontrolvariablesusedinouranal­ ysis. Table2: DescriptiveStatistics N Mean StdDev Minimum Maximum LMItract 58,246 0.33 0.47 0 1 Maj. Min. tract 58,246 0.33 0.47 0 1 Lendingincome 58,246 0.024 0.0043 0.0054 0.058 %Owneroccupied 58,246 60.5 23.1 0 100 Operatingcosts 58,246 4,646 2,657 374 38,934 Otherchargeoffs 58,246 0.00011 0.00011 0 0.0016 Branchnetworkage 58,246 38.0 8.07 4.53 124 Bigbankshare 58,246 0.57 0.31 0 1 MDI 58,246 0.012 0.054 0 1 HHI 58,246 41.0 29.9 6.47 100 Rural 58,246 0.12 0.33 0 1 Urban 58,246 0.36 0.48 0 1 %Over65 58,246 15.1 7.87 0 91 %BA 58,246 31.1 19.6 0 96.4 Note: LMIandmajority­minoritytractsarefromthe2015FFIECCensusFile. Demographicvariablesarefromthe ACS2013­2017five­yearsample. Theaverageoperatingcosts,lendingincome,andotherchargeoffsofatractare calculatedbasedonbankcallreports. Classificationoftractsasurban,suburban,orruralcomesfromHUD’s UrbanizationPerceptionsSmallAreaIndex. BranchageisfromtheFDICSummaryofDepositsdata. Classification asaminority­depositoryinstitutions(MDI)fromtheFDICMinorityDepositoryInstitutionlist. Variableconstruction describedinsection3. 4 Empirical Analysis 4.1 LMI communities, majority­minority communities, and bank account fees WebegintoexplorethecostsofbankingforLMIcommunitiesbytestingwhetheraccountfees and minimum account balances are systematically different in LMI Census tracts relative to non­ asinglebank,theHHIis100. 9

LMI Census tracts. Because LMI designations are specific to each metropolitan statistical areas (MSA),weincludeMSAfixedeffectsinallspecifications.15 Y = β +γ +β 1(LMI) +ε . (1) ic 0 c 1 ic ic HereY isafeeorminimumvariable,iisatract,andcisametroarea. 1(LMI) isanindicator ic equal to one if tract i is considered low or moderate income in metro c, and zero if it is middle or upperincome. Table3presentstheresultsoftheseregressions. Table3: RegressionofFeesandMinimumsonLMItracts No­IntCheck IntCheck Savings No­IntCheck IntCheck Savings NSF NSF Minimum Minimum Minimum Fee Fee Fee DailyMax RetCheckFee LMItract 45.8*** 215*** 6.92*** 0.31*** 0.43*** 0.060*** ­1.45*** 0.13*** (3.96) (27.8) (0.70) (0.018) (0.031) (0.0090) (0.25) (0.015) Observations 58,246 58,246 58,246 58,246 58,246 58,246 58,246 58,246 R­squared 0.324 0.375 0.309 0.441 0.405 0.301 0.345 0.388 Robuststandarderrorsinparentheses ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01 Note: Allspecificationsincludemetroareafixedeffects. ”Int”meansinterest. ”No­int”meansno­interest. NSF meansnon­sufficientfunds. ”RetCheckFee”meansreturnedcheckfee. Bank accounts are generally more expensive in LMI communities, both in their maintenance feesandminimumaccountbalancesrequiredtoavoidthosefees. Forthemostbasicaccounts,non­ interest checking, the minimum balance required to not pay a maintenance fee was, on average, $46 higher in LMI Census tracts than in non­LMI Census tracts, and the maintenance fee was $0.31 higher. Returned check fees are also higher on average in LMI tracts. Meanwhile, banks in LMI tracts set, on average, a smaller daily limit on the accumulation of multiple NSF fees before overdraftsaredisallowedthanbanksinnon­LMItracts. Following prior research that documented disparities in retail banking for minority customers, we also test whether account fees are systemically different in majority­minority neighborhoods. 15Censustractsoutsidemetroareasaregroupedtogetherasasingleunitineachstate. 10

Table4presentstheresultsoftheseregressions. Table4: RegressionsofFeesandMinimumsonLMItractsandMajority­minoritytracts No­IntCheck IntCheck Savings No­IntCheck IntCheck Savings NSF NSF Minimum Minimum Minimum Fee Fee Fee DailyMax RetCheckFee LMItract 45.8*** 215*** 6.92*** 0.31*** 0.43*** 0.060*** ­1.45*** 0.13*** (3.96) (27.8) (0.70) (0.018) (0.031) (0.0090) (0.25) (0.015) Maj.Min.tract 71.6*** 414*** 15.1*** 0.49*** 0.71*** 0.062*** ­3.53*** 0.15*** (3.62) (30.4) (0.80) (0.018) (0.034) (0.0096) (0.27) (0.015) LMItract 18.6*** 42.3 0.29 0.13*** 0.15*** 0.043*** 0.15 0.079*** (4.65) (32.9) (0.86) (0.021) (0.036) (0.011) (0.30) (0.018) Maj.Min.tract 61.3*** 390*** 15.0*** 0.42*** 0.63*** 0.038*** ­3.61*** 0.10*** (4.32) (36.1) (0.99) (0.022) (0.040) (0.011) (0.32) (0.018) Observations 58,246 58,246 58,246 58,246 58,246 58,246 58,246 58,246 R­squared 0.326 0.376 0.312 0.444 0.407 0.301 0.346 0.388 Robuststandarderrorsinparentheses ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01 Note: Thetoprowsshowtheresultsfromunivariateregressions. Allspecificationsincludemetroareafixedeffects. ”Int”meansinterest. ”No­int”meansno­interest. NSFmeansnon­sufficientfunds. ”RetCheckFee”meansreturned checkfee. The results in the second row display the average differences in the cost of retail banking in majority­minorityCensustractsrelativetoothertracts. Majority­minoritytractsexperiencehigher minimumsandfeesacrossallaccounttypes. Forno­interestcheckingaccounts,minimumbalances to avoid fees are on average $72 higher and account maintenance fees are $0.49 higher. These results are consistent with the findings of Faber and Friedline (2020), who, using a different bank survey,alsofindthatbanksinminorityneighborhoodsrequiredhigherminimumbalancestoavoid accountmaintenancefees. The lower panel of Table 4 presents results when indicators for whether tracts are LMI or majority­minority are included simultaneously in the regressions. Coefficients on both indicators decrease relative to the univariate regressions, which is expected given that they have a correla­ tion of about 0.5. For example, the coefficient on the LMI tract indicator for the regression of the minimum account balance required to avoid a maintenance fee on a no­interest checking account 11

goes down from $46 to $19. Still, the coefficients associated with LMI tracts remain positive and statistically significant in most regressions. The coefficients on the majority­minority indicators are positive (with the exception of the daily limit on NSF fees) and of larger magnitude than the coefficients on the LMI indicators. This suggests that the factors underpinning higher bank fees are stronger in majority­minority Census tracts than in LMI tracts. Importantly, these differences add up for tracts that are both LMI and majority­minority. Banks in neighborhoods that are both LMI and majority­minority require, on average, a $80 higher minimum account balance to avoid maintenancefeesonno­interestcheckingaccountsthanbanksintractsthatarebothhigherincome andmajority­white. 4.2 Explanations for higher bank fees in LMI and majority­minority com­ munities To better understand the differences in the cost of banking services for LMI and majority­ minority Census tracts, we study the impact of several potential drivers of these differences, in­ cludingbanks’othersourcesofrevenue,bankcosts,otherbankcharacteristics,competitionamong banks,andotherdemographiccharacteristicsofthetracts. Table5displaysthecorrelationbetween the various additional variables we consider and the indicator variables for LMI and majority­ minoritytracts. A bank’s ability to earn revenue from sources besides retail fees may influence how high a bank sets those fees. For example, a bank more focused on earning lending income may set lower fees in its deposit accounts to attract customers to whom it can sell other products. Per Table 5, the lending income of banks in a tract is somewhat negatively correlated with whether a tract is LMI or majority­minority; therefore, the lower lending income of banks in LMI and majority­ minoritytractsmaypartlyexplainthedifferenceinaccountfeesinthesetracts. Asanotherproxyfor prospectivelendingincome,wealsoincludethehomeownershiprateinatractinourregressions. More owner­occupied housing in an area likely increases a bank’s ability to earn income from products such as mortgages, instead of fees on its deposit accounts. Like lending income, the 12

Table5: CorrelationAmongExplanatoryVariables LMI Maj­Minority LMItract 1.00 0.47 Maj. Min. tract 0.47 1.00 Lendingincome ­0.07 ­0.19 %Owneroccupied ­0.56 ­0.43 Operatingcosts 0.08 0.31 Otherchargeoffs 0.02 ­0.03 Branchnetworkage 0.05 0.03 Bigbankshare 0.07 0.19 MDI 0.08 0.22 HHI ­0.10 ­0.11 Rural ­0.15 ­0.20 Urban 0.43 0.39 %Over65 ­0.26 ­0.32 %BA ­0.44 ­0.35 13

owner­occupiedhousingrateinaCensustractisnegativelycorrelatedwithwhetheratractisLMI ormajority­minority. Bank operating costs are also a plausible driver of fee levels, as banks with higher costs likely charge their customers higher fees to recoup costs. Operating costs are positively correlated with whether a tract is LMI or majority­minority. Similarly, banks with high charge­offs on overdrafts ofdepositaccountsmaychoosetochargehigherfeestocompensateforthesecosts. Toaccountfor thispossibleeffect,weincludeinouranalysisthe“othercharge­offs”itemfrombankcallreports, whichincludescharge­offsduetounpaidoverdraftsofdepositaccounts.16 Certain other characteristics, such as age, size, or ownership, may also affect a bank’s retail fees. Age may affect a bank’s fees if, for example, older banks have a customer base who will remainloyaldespitehigherfees. Ontheflipside,newerbanksmayuselowerfeesorno­minimum checkingaccountstoenticenewcustomers. Totestthispotentialeffect,weincludetheaverageage ofthebranchesinagiventractinourregressions. Sizemayalsoplausiblyaffectbankfees,dueto economies of scale (which may make fees lower), market power (which may make fees higher), ortheconvenienceofmoreubiquitousbranches(whichmaymakefeeshigher). Theproportionof largebanksinaCensustractcorrelatespositivelywithwhetheratractisLMIormajority­minority.17 Largebanksown48%ofbranchesintheaveragetract,butown55%ofbranchesintheaverageLMI tract and 61% of branches in the average majority­minority tract. In addition, we explore whether bankownershipbyminorityindividualsinfluencesbankfees. Bankswithminorityownershipmay be able to charge higher fees due to having a more loyal customer base. Minority ownership is positivelycorrelatedwithLMIandmajorityminoritytracts. Standard economic theory holds that, all else equal, more competition should be associated withlowerpricesforconsumersofgoodsandservices. Weaccountforcompetitioninouranalysis byintroducingtheHHIofatractinourregressions. TheHHIismildlynegativelycorrelatedwith thetractbeingLMIormajority­minority,implyingamorecompetitivebankinglandscapeinthose 16Theuseof“othercharge­offs”asacontrolvariableforbankbusinessmodelfollowsMelzerandMorgan(2015). Theyfoundthatlossesonoverdraftsarehigherwhencreditunionsmustcompetewithpaydaylendersinthemarket forsmallconsumerloans,implyingthatcreditunionsrespondtocompetitionbyincreasingrisk­taking. 17Wedefineabankaslargeifithasabranchnetworkwithatleast500locations. 14

tractsonaverage. The density of a location may also affect bank fees, at least partly due to its effect on bank operatingcosts. Thus,weincludedummyvariablesforwhetheratractisruralorurban(suburban is the excluded category). Urban tracts are more likely to be LMI and majority­minority, while ruraltractsarelesslikelytobeLMIormajority­minority. Certain other demographic characteristic of tracts may also influence bank fees. In particular, theshareofseniorsorofindividualswithabachelor’sdegreemayplausiblyaffectbankfees. One possiblemechanismisfinancialliteracy. Researchhasshownthatseniorindividualsandindividu­ alswithlessschoolingscorelowerinfinancialliteracytests(BumcrotandLusardi,2013).18 Lower financialliteracymaymeanthatindividualsarelessawareofsometimesopaquefeeandminimum structures, or of the availability of lower cost options at other branches or online. On average, LMItractsandmajority­minoritytractshaveasmallerproportionofseniorindividualsandcollege educatedindividualsthanothertracts. Table 6 presents the results of these regressions. To facilitate comparisons of effects in our re­ gressionanalysis,westandardizedthefollowingcontrolvariablestotheirz­score: lendingincome, % owner occupied, operating costs, other charge­offs, bank age, big bank share, MDI share, HHI, %over65,and%collegebachelors. Z­scoresfortract­levelmeasuresarecalculatedbysubtracting the average for the variable across all tracts and then dividing by the standard deviation across all tracts. Bank­levelvariablesarefirstwinsorizedatthe1stand99thpercentilestoreducetheimpact of any outlier values, and then averaged across banks operating in each tract as described above. Z­scoresforthesevariablesarethencomputedsimilarlyacrossthetractdistribution.19 Some of the control variables we consider have regression coefficients in line with our pre­ dictions. Higher lending income is generally associated with lower fees and minimums. Higher operating costs are generally associated with higher fees and minimums. Bank age is generally associated with higher minimums and fees. Banks charge lower fees and require lower minimum 18Williams (2016) finds that median age is negatively associated with bank overdraft fees, and that the effect of educationonfeesisnotstatisticallysignificant. 19Thedistributionofbank­levelvariablespre­winsorizationareinTableA.2andthesummarystatisticsofthetract levelvariablespostZ­scoringareinTableA.3intheAppendix. 15

Table6: RegressionsofFeesandMinimumsonLMItracts,Majority­minoritytracts,andOther Controls No­IntCheck IntCheck Savings No­IntCheck IntCheck Savings NSF NSF Minimum Minimum Minimum Fee Fee Fee DailyMax RetCheckFee LMItract 18.6*** 42.3 0.29 0.13*** 0.15*** 0.043*** 0.15 0.079*** (4.65) (32.9) (0.86) (0.021) (0.036) (0.011) (0.30) (0.018) Maj.Min.tract 61.3*** 390*** 15.0*** 0.42*** 0.63*** 0.038*** ­3.61*** 0.10*** (4.32) (36.1) (0.99) (0.022) (0.040) (0.011) (0.32) (0.018) LMItract 1.42 ­17.0 ­1.89* 0.085*** 0.14*** 0.030** 0.55 0.083*** (5.49) (38.3) (0.99) (0.022) (0.040) (0.014) (0.35) (0.021) Maj.Min.tract 10.1** 20.9 7.17*** 0.054** 0.19*** 0.016 ­1.34*** ­0.061*** (4.80) (38.8) (1.15) (0.021) (0.041) (0.013) (0.34) (0.020) Lendingincome ­33.3*** ­457*** ­0.66 ­0.73*** ­0.97*** ­0.00023 7.72*** ­0.10*** (4.51) (25.3) (0.68) (0.019) (0.031) (0.013) (0.30) (0.021) %Owneroccupied ­3.57* ­5.13 ­3.14*** 0.027*** 0.0043 ­0.0074 ­0.50*** 0.040*** (2.11) (16.0) (0.45) (0.0094) (0.018) (0.0055) (0.15) (0.0090) Operatingcosts 101*** 839*** 2.50*** 0.78*** 0.75*** ­0.033*** ­10.8*** 0.045*** (3.32) (28.1) (0.66) (0.015) (0.033) (0.0092) (0.28) (0.014) Otherchargeoffs ­16.6*** 21.9 4.73*** ­0.027** ­0.32*** 0.035*** 6.29*** ­0.55*** (3.31) (25.9) (0.69) (0.012) (0.022) (0.0075) (0.39) (0.021) Branchnetworkage 36.3*** ­31.8 ­1.25** ­0.16*** 0.081*** ­0.0081 1.19*** 0.17*** (4.24) (23.5) (0.53) (0.015) (0.027) (0.0092) (0.28) (0.019) Bigbankshare 72.0*** 959*** 32.5*** 0.63*** 1.01*** 0.27*** 5.62*** 0.43*** (5.37) (33.3) (0.68) (0.020) (0.034) (0.013) (0.32) (0.021) MDI 22.3*** 79.9*** 15.1*** 0.18*** 0.29*** ­0.023*** ­0.31** 0.22*** (1.93) (15.3) (0.64) (0.012) (0.015) (0.0082) (0.14) (0.011) HHI ­16.0*** ­154*** ­8.65*** ­0.19*** ­0.32*** ­0.077*** ­0.50** ­0.041*** (3.43) (23.2) (0.47) (0.013) (0.024) (0.0078) (0.21) (0.012) Rural ­20.7* ­183** ­3.95** ­0.24*** ­0.19** ­0.016 ­0.71 ­0.27*** (12.4) (77.7) (1.79) (0.044) (0.080) (0.026) (0.71) (0.044) Urban ­0.40 ­98.4*** 0.99 ­0.052*** ­0.13*** 0.021** 0.98*** 0.027* (4.13) (28.4) (0.87) (0.017) (0.030) (0.010) (0.27) (0.016) %Over65 ­5.16*** 9.90 2.73*** ­0.023*** ­0.013 0.0093* 0.23* ­0.018** (1.91) (16.0) (0.32) (0.0088) (0.016) (0.0048) (0.13) (0.0078) %BA ­2.60 ­4.01 4.51*** 0.0040 0.093*** 0.025*** 1.65*** 0.025*** (2.26) (16.8) (0.42) (0.0098) (0.018) (0.0057) (0.15) (0.0091) Observations 58,246 58,246 58,246 58,246 58,246 58,246 58,246 58,246 R­squared 0.371 0.467 0.395 0.615 0.526 0.328 0.419 0.460 Robuststandarderrorsinparentheses ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01 Note: ThetoppanelreprintstheresultswhennocontrolsareincludedfromTable4foreaseofcomparison. The variableslendingincome,%owneroccupied,operatingcosts,otherchargeoffs,branchnetworkage,bigbankshare, MDI,HHI,%over65,and%BAenterasz­scores,meaningthosecoefficientscaptureadifferenceofonestandard deviationinthatvariableacrossthedistributionoverallCensustracts. Thevariableslendingincome,operatingcosts, otherchargeoffs,andbranchnetworkagewerealsowinsorizedatthe1stand99thpercentilesbeforetheirz­scores arecalculated. Allspecificationsincludemetroareafixedeffects. 16

balancesinruralareas. Bankownershipbyminorityindividualsisgenerallyassociatedwithhigher fees. Andwefindastrongpositiverelationshipbetweenfeesandthepresenceoflargebanks. Certain control variables have effects that we did not expect. In our regressions, less competi­ tionacrossbanks(measuredbyahigherHHIindex)isassociatedwithlowerfees. Thisunintuitive relationmayresultfromthenumberofbanksinaCensustractbeingcorrelatedwithanunidentified variable that influences fees. Also, we find that a higher proportion of individuals with bachelor’s degree is associated with higher fees. Other control variables had inconsistent or not statistically significanteffectsonbankminimumaccountbalancesandfees. Most of the difference in minimum account balances and fees between LMI Census tracts, majority­minorityCensustracts,andotherCensustractsareaccountedforbytheeconomicfactors weincludedintheseregressions. Inseveralcases,theinclusionofcontrolvariableseliminatedthe statistical significance of the LMI or majority­minority coefficients. These regression results sug­ gestthatstandardeconomicexplanationslargelyexplainthesystematicallyhigherfeesobservedin LMI and majority­minority Census tracts. Still, some differences remain unexplained. For exam­ ple, the LMI tract variable continues to be associated with approximately a $0.09 higher monthly maintenance fee for a no­interest checking account that cannot be accounted for by observable differences. 4.3 Bank­level analysis Given that most fee and minimum decisions are made at the bank level rather than the branch level, we turn to analogous regressions where the unit of observation is the bank instead of the Census tract. This alternative setup provides a more direct lens into bank decision­making, as opposedtothefocusontheimpactoncommunitiesthatisachievedthroughtheCensustractlevel regressions. Under this approach, bank fee and minimum account balance variables generally require no averaging, as most banks (95%) report the same fee and minimum account balances across their variousbranches. Thefewbanksforwhichsomedifferenceswerereportedtendtobelargebanks. 17

In those cases, we average fee and minimum values across the bank’s branch network with every link between a branch and a tract contributing equally. Other bank variables, such as lending in­ come, branch network age, and ”big bank,” are also directly used for each bank. To construct the demographic variables used in these regressions, we take averages of the Census variables across alltractsdeemedwithinagivenbank’sbranchnetwork(wherewhetherabankisinaCensustract is determined as described in section 3). The mean LMI tract concentration is 26% with a stan­ darddeviationof19%,andthemeanmajority­minoritytractconcentrationis16%withastandard deviationof22%.20 In assessing whether banks concentrated in LMI communities or in majority­minority com­ munities set higher minimum account balances to avoid fees or higher fee amounts, we estimate two regression specifications: a first one where the extent to which a bank’s branches are located in LMI or majority­minority neighborhoods are the only explanatory variables, and a second one wherethevariouscontrolvariablespreviouslydiscussedfortheCensus­tractlevelregressionsare also included. All explanatory variables, with the exception of the big bank and minority owner­ shipindicators,enterasZ­scores. Thismeansthecoefficientscaptureadifferenceofonestandard deviation in that variable across the distribution over all banks and that the constant gives the av­ erage of the dependent variables for a bank with the mean value of each variable and zeroes for theindicatorvariables. Notethatthelendingincome,operatingcosts,otherchargeoffs,andbranch network age control variables are also winsorized at 1st the 99th percentiles to reduce the impact ofanyoutliersinouranalysis. Table7presentstheresultsoftheseregressions. BanksmoreconcentratedinLMItractsrequirehigherminimumbalancestoavoidmaintenance fees on their interest­checking accounts, have higher NSF daily maximum accruals, and charge higher returned check fees. We find no statistically significant relation between a bank’s concen­ tration in LMI tracts and other fee variables. Banks more concentrated in majority­minority tracts requirehigheraccountbalancestoavoidfeesandchargehigherfeesacrossmostcategoriesconsid­ ered. Forexample,theminimumaccountbalancetoavoidamaintenancefeeonasavingsaccount 20Summary statistics for the bank­level explanatory variables are in TableA.4, in Table A.5 post Z­scoring, and theircorrelationwithLMIandmajority­minoritytractbankfootprintsinTableA.6intheAppendix. 18

Table7: RegressionsofFeesandMinimumsonLMItracts,Majority­minoritytracts,andOther Controls­Banklevel No­IntCheck IntCheck Savings No­IntCheck IntCheck Savings NSF NSF Minimum Minimum Minimum Fee Fee Fee DailyMax RetCheckFee LMItract 52.1 272** 4.01 0.12 0.13 ­0.12 5.20** 0.68*** (33.9) (123) (4.22) (0.13) (0.17) (0.098) (2.28) (0.21) Maj.Min.tract 42.2 244* 39.5*** 1.18*** 1.05*** 0.50*** 5.96*** 0.95*** (27.8) (125) (5.85) (0.15) (0.20) (0.11) (2.04) (0.18) LMItract ­30.8 142 ­2.70 ­0.11 ­0.077 ­0.30*** 1.55 0.26 (36.7) (107) (4.75) (0.17) (0.19) (0.11) (2.56) (0.21) Maj.Min.tract ­12.0 46.3 25.9*** 0.92*** 0.67*** 0.36*** 1.25 0.48** (39.1) (142) (6.26) (0.17) (0.22) (0.12) (2.33) (0.19) Lendingincome ­85.7* ­144 ­7.50 ­0.20 ­0.21 ­0.021 1.54 ­0.49*** (45.4) (111) (5.55) (0.14) (0.17) (0.11) (2.20) (0.18) %Owneroccupied ­96.7* 58.0 ­14.6 ­0.34 0.23 0.13 ­6.16* ­0.36 (58.3) (191) (13.0) (0.24) (0.33) (0.17) (3.30) (0.29) Operatingcosts 23.5 255 12.0*** 0.30** 0.44*** 0.022 ­1.65 ­0.098 (27.6) (159) (4.52) (0.13) (0.15) (0.091) (1.97) (0.12) Otherchargeoffs ­41.2* 58.8 ­1.57 ­0.098 ­0.16 ­0.074 0.52 ­0.39* (21.4) (120) (4.94) (0.11) (0.14) (0.079) (2.09) (0.23) Branchnetworkage ­42.9 15.7 ­11.6** ­0.26* ­0.076 ­0.35*** ­7.60*** ­0.48** (35.6) (118) (4.84) (0.14) (0.19) (0.13) (2.41) (0.19) Bigbank 339** 4,424*** 130*** 3.43*** 5.53*** 0.94 9.00 1.01 (153) (1,548) (33.6) (0.83) (1.71) (0.68) (15.2) (1.03) MDI 55.6 ­1,389*** 86.0 1.77 0.057 ­1.28** 31.8** 1.90** (154) (451) (63.6) (1.29) (1.04) (0.56) (16.2) (0.83) HHI ­1.46 101 1.29 0.22 0.12 ­0.25* ­0.87 0.12 (38.4) (185) (7.17) (0.20) (0.28) (0.15) (3.49) (0.31) Rural 10.0 ­89.7 ­4.39 0.52*** ­0.28 0.24 0.75 ­0.84*** (40.0) (187) (6.78) (0.20) (0.25) (0.20) (3.73) (0.32) Urban 52.4 294* ­5.02 0.36* 0.52** 0.36* 0.50 0.30 (59.0) (162) (12.6) (0.21) (0.26) (0.18) (3.12) (0.26) %Over65 ­22.7 23.4 3.44 ­0.093 0.043 ­0.086 0.70 0.38** (26.9) (118) (6.87) (0.16) (0.18) (0.11) (2.50) (0.19) %BA ­109** 343** 5.05 ­0.0018 0.47** 0.11 3.45 0.30 (54.8) (165) (6.32) (0.18) (0.23) (0.13) (2.50) (0.25) Constant 402*** 2,108*** 140*** 3.15*** 9.30*** 3.38*** 159*** 30.5*** (37.1) (128) (5.54) (0.13) (0.17) (0.10) (1.94) (0.16) Observations 776 776 776 776 776 776 776 776 R­squared 0.030 0.095 0.128 0.156 0.139 0.070 0.078 0.180 Robuststandarderrorsinparentheses ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01 Note: Thetoppanelshowsresultswhennocontrolsareincluded. Inthebottompanel,allexplanatoryvariables excepttheindicatorsforbigbankandMDIenterasz­scores. Thelendingincome,operatingcosts,otherchargeoffs, andbranchnetworkagevariablesarealsowinsorizedatthe1stand99thpercentilesbeforetheirz­scoresare calculated. 19

is,onaverage,$40higherforabankonestandarddeviationmoreconcentratedinmajority­minority tracts. ThehigherminimumaccountbalancesandfeesobservedatbanksthatareconcentratedinLMI or majority­minority Census tracts are somewhat accounted for by the economic factors we in­ cludedintheseregressions. Oncethecontrolvariablesareaddedtotheregressions,thecoefficients associated with banks being located in LMI tracts generally decrease and their statistical signifi­ cance disappears. Meanwhile, the coefficients associated with banks’ concentration in majority­ minoritytractsalsodecreaseacrossallregressions,butretainstatisticalsignificanceinmostcases. For example, the minimum account balance to avoid a maintenance fee on a savings account still remains, on average, $26 higher for a bank that is one standard deviation more concentrated in majority­minority tracts. These results raise the possibility that additional factors not included in theseregressionsdrivethesedifferences. 5 Discussion OuranalysisfindssubstantialevidencethatbankcustomersinLMIcommunitiesandmajority­ minority communities face higher fees and must maintain higher account balances to avoid main­ tenancefeesintheirdepositaccounts. Thesedifferencesinfeesarecompoundedforcommunities thatarebothLMIandmajority­minority. Banks’higherfeesandminimumsinLMIandmajority­ minoritycommunitiesarelargelyexplainedbyeconomicfactorssuchasbanks’alternativesources ofincome(suchasinterestincome)andoperatingcosts. Still,thesehigherbankingcostsarelikely tocompoundtheeconomicdisadvantagesfacedbyLMIandmajority­minoritycommunities. Reducing cost inequalities in access to financial services would likely increase the standard of livingforLMIandmajority­minoritycommunitiesandreduceeconomicinequality.21 Butbecause mostofthedisparityinthecostofretailbankingisdrivenbyeconomicfactors,reducingthedisad­ 21See Beck et al. (2009) and Mookerjee and Kalipioni (2010) for discussions of the relation between access to financialservicesandincomeinequality. Inastudyspecificallyaboutaccesstobankaccounts,Prina(2015)findsthat freeaccesstosavingsaccountsincreasedfinancialwell­beingofNepalesewomen. 20

vantagefacedbyLMIandmajority­minoritycommunitieslikelyrequirescreativepolicysolutions. Direct regulation of fees could be considered, but recent research suggests such regulation may have unintended consequences. For example, Dlugosz et al. (2021) find that the unbanked popu­ lation decreased in states where federal regulation preempted state­based regulations that limited bankfees. Still,otherpolicyoptionscouldbeconsidered. ArmstrongandVickers(2012)andSarin (2019)suggestthatregulatorscouldenhancewelfarebymakingfeesmoresalient;Sarin(2019)also adds that the cross­subsidization of products (e.g., low­income bank clients often cross­subsidize products for high­income bank clients) could be regulated. Initiatives such as FDIC’s Model Safe Accounts Pilot have also showed possible approaches to offer low­cost banking services for un­ banked,low­incomeindividuals. Future research should aim to confirm the findings of this paper using a longer sample pe­ riod. Also, other potential drivers of higher bank fees, such as bank size, likely merit additional investigation. References Adams,RobertM.(2017),“BankFees,Aftermarkets,andConsumerBehavior.”FinanceandEco­ nomicsDiscussionSeries2017­054,BoardofGovernorsoftheFederalReserveSystem. Alan, Sule, Mehmet Cemalcilar, Dean Karlan, and Jonathan Zinman (2018), “Unshrouding: Evi­ dencefromBankOverdraftsinTurkey.”JournalofFinance,73,481–522. Armstrong,MarkandJohnVickers(2012),“ConsumerProtectionandContingentCharges.”Jour­ nalofEconomicLiterature,50,477–93. Barr,Michael(2004),“BankingthePoor.”YaleJournalonRegulation,21,121–237. Beck,Thorsten,AsliDemirgüç­Kunt,andPatrickHonohan(2009),“AccesstoFinancialServices.” WorldBankResearchObserver,24,119–145. 21

Bernard,TaraSiegel(2011),“InRetreat,BankofAmericaCancelsDebitCardFee.”TheNewYork Times. Berry, Christopher (2004), “To Bank or Not to Bank? A Survey of Low­Income Households.” Workingpaperseries,JointCenterforHousingStudies,HarvardUniversity. Booz­Allen Hamilton and Shugoll Research (1997), “Mandatory EFT Demographic Study.” De­ partmentoftheTreasury,FinancialManagementService. Brobeck, Stephen (2020), “Do Big Banks Provide Affordable Access To Lower Income Savers?” Discussionpaper,ConsumerFederationofAmerica. Bumcrot,JudyLin,ChristopherandAnnamariaLusardi(2013),“TheGeographyofFinancialLit­ eracy.”Numeracy,6. Caskey,JohnP(1994),Fringebanking: Check­cashingOutlets,Pawnshops,andthePoor.Russell SageFoundation. Dlugosz, Jennifer, Brian Melzer, and Donald P Morgan (2021), “Who Pays the Price? Overdraft FeeCeilingsandtheUnbanked.”FRBofNewYorkStaffReport. Faber, Jacob William and Terri Friedline (2020), “The Racialized Costs of “Traditional” Banking in Segregated America: Evidence from Entry­Level Checking Accounts.” Race and Social Problems,12,344–361. Federal Deposit Insurance Corporation (2018), “2017 National Survey of Unbanked and Under­ bankedHouseholds.” Federal Reserve System (2009), “Electronic Fund Transfers.” Federal Register 12 CFR Part 205, 74,59033–59056. Goodstein,RyanMandSherrieLWRhine(2017),“TheEffectsofBankandNonbankProviderLo­ cationsonHouseholdUseofFinancialTransactionServices.”JournalofBanking&Finance, 78,91–107. 22

Greene,Meghan,HannahGdalman,ElaineGolden,StephenArves,andNecatiCelik(2021),“The FinHealthSpendReport2021—WhatFinanciallyCopingandVulnerableAmericansPayfor EverydayFinancialServices.”FinancialHealthNetwork. Low, David, Eva Nagypál, Leslie Parrish, Akaki Skhirtladze, and Corey Stone (2017), “Frequent Overdrafters.”Datapoint,ConsumerFinancialProtectionBureau. Melzer,BrianTandDonaldPMorgan(2015),“CompetitioninaConsumerLoanMarket: Payday LoansandOverdraftCredit.”JournalofFinancialIntermediation,24,25–44. Mookerjee, Rajen and Paul Kalipioni (2010), “Availability of Financial Services and Income In­ equality: TheEvidencefromManyCountries.”EmergingMarketsReview,11,404–408. PewCharitableTrusts(2016),“ConsumersNeedProtectionFromExcessiveOverdraftCosts.” Pew Charitable Trusts (2017), “Overdraft Does Not Meet the Needs of Most Consumers.” Issue Brief. PewHealthGroup(2011),“SlippingBehind: Low­IncomeLosAngelesHouseholdsDriftFurther fromtheFinancialMainstream.” Prina,Silvia(2015),“BankingthePoorviaSavingsAccounts: EvidencefromaFieldExperiment.” JournalofDevelopmentEconomics,115,16–31. Sarin, Natasha (2019), “Making Consumer Finance Work.” Columbia Law Review, 119, 1519– 1596. Smith,Peter(2018),“UnfairMarket—TheStateofHigh­CostOverdraftPracticesin2017.”Center forResponsibleLending. Stango, Victor and Jonathan Zinman (2009), “What Do Consumers Really Pay on their Checking and Credit Card Accounts? Explicit, Implicit, and Avoidable Costs.” American Economic Review: Papers Proceedings,99,424–29. 23

Williams,MarlonL(2016),“BankOverdraftPricingandMyopicConsumers.”EconomicsLetters, 139,84–87. 24

Appendix A Additional tables and figures TableA.1: CorrelationAmongAccountFeesandMinimumBalanceRequirements No­IntCheck IntCheck Savings No­IntCheck IntCheck Savings NSF NSF Minimum Minimum Minimum Fee Fee Fee DailyMax RetCheckFee No­IntCheckMinimum 1.00 IntCheckMinimum 0.50 1.00 SavingsMinimum 0.30 0.35 1.00 No­IntCheckFee 0.56 0.60 0.43 1.00 IntCheckFee 0.43 0.75 0.41 0.73 1.00 SavingsFee 0.17 0.20 0.42 0.28 0.30 1.00 NSFDailyMax ­0.04 ­0.03 0.03 ­0.19 ­0.13 0.03 1.00 RetCheckFee 0.14 0.08 0.21 0.29 0.28 0.27 0.05 1.00 TableA.2: Bankcharacteristicsummarystatistics N Mean StdDev Minimum p1 p99 Maximum Operatingcosts 6,090 10,844 192,467 ­89 345 39,015 7,401,172 Lendingincome 6,090 0.032 0.013 0 0.0028 0.058 0.45 Otherchargeoffs 6,090 0.000094 0.00075 0 0 0.0016 0.035 Branchnetworkage 6,090 46.5 33.2 0 4.53 131 223 Note: Thesestatisticsarepriortowinsorizationatthe1stand99thpercentiles. 25

TableA.3: TractlevelregressionsamplesummarystatisticswithZ­scores N Mean StdDev Minimum Maximum LMItract 58,246 0.33 0.47 0 1 Maj. Min. tract 58,246 0.33 0.47 0 1 Rural 58,246 0.12 0.33 0 1 Urban 58,246 0.36 0.48 0 1 Lendingincome 58,246 0 1.00 ­4.34 7.80 %Owneroccupied 58,246 0 1.00 ­2.61 1.71 Operatingcosts 58,246 0 1.00 ­1.61 12.9 Otherchargeoffs 58,246 0 1.00 ­0.97 13.3 Branchnetworkage 58,246 0 1.00 ­4.15 10.7 Bigbankshare 58,246 0 1.00 ­1.84 1.41 MDI 58,246 0 1.00 ­0.22 18.3 HHI 58,246 0 1.00 ­1.15 1.97 %Over65 58,246 0 1.00 ­1.92 9.64 %BA 58,246 0 1.00 ­1.59 3.33 Note: ThesestatisticsreflectthesampleusedfortheresultsinTable6includingz­scoringofnon­indicatorvariables. TableA.4: Banklevelsummarystatistics N Mean StdDev Minimum Maximum LMItract 776 0.26 0.19 0 1 Maj. Min. tract 776 0.16 0.22 0 1 Lendingincome 776 0.032 0.0078 0.0054 0.058 %Owneroccupied 776 61.6 11.0 26.5 91.6 Operatingcosts 776 2,083 2,081 374 38,934 Otherchargeoffs 776 0.000083 0.00021 0 0.0016 Branchnetworkage 776 37.2 22.9 4.53 131 BigBank 776 0.026 0.16 0 1 MDI 776 0.012 0.11 0 1 HHI 776 40.4 18.2 9.73 100 Rural 776 0.23 0.29 0 1 Urban 776 0.31 0.24 0 1 %Over65 776 15.9 3.15 7.65 31.8 %BA 776 29.6 11.6 6.20 79.0 26

TableA.5: BanklevelregressionsamplesummarystatisticswithZ­scores N Mean StdDev Minimum Maximum LMItract 776 0 1.00 ­1.36 3.86 Maj. Min. tract 776 0 1.00 ­0.76 3.89 Lendingincome 776 0 1.00 ­3.38 3.38 %Owneroccupied 776 0 1.00 ­3.19 2.73 Operatingcosts 776 0 1.00 ­0.82 17.7 Otherchargeoffs 776 0 1.00 ­0.40 7.24 Branchnetworkage 776 0 1.00 ­1.42 4.09 Bigbank 776 0.026 0.16 0 1 MDI 776 0.012 0.11 0 1 HHI 776 0 1.00 ­1.68 3.27 Rural 776 0 1.00 ­0.81 2.65 Urban 776 0 1.00 ­1.27 2.84 %Over65 776 0 1.00 ­2.62 5.03 %BA 776 0 1.00 ­2.02 4.26 Note: ThesestatisticsreflectthesampleusedfortheresultsinTable7includingz­scoringofnon­indicatorvariables. 27

TableA.6: Banklevel­CorrelationAmongExplanatoryVariables LMI Maj­Minority LMItract 1.00 0.35 Maj. Min. tract 0.35 1.00 Lendingincome ­0.03 ­0.05 %Owneroccupied ­0.55 ­0.42 Operatingcosts 0.09 0.16 Otherchargeoffs 0.04 0.11 Branchnetworkage ­0.18 ­0.26 Bigbank 0.09 0.16 MDI 0.11 0.34 HHI ­0.32 ­0.30 Rural ­0.31 ­0.36 Urban 0.55 0.44 %Over65 ­0.32 ­0.36 %BA 0.02 0.10 28