Enforcing Fair Lending: Evidence from Mortgage Market Litigation

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Enforcing Fair Lending: Evidence from Mortgage Market Litigation Matthew Maury, Michael Suher, and Jeffery Y. Zhang 2026-012 Please cite this paper as: Maury, Matthew, Michael Suher, and Jeffery Y. Zhang (2026). “Enforcing Fair Lending: Evidence from Mortgage Market Litigation,” Finance and Economics Discussion Series 2026-012. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2026.012. 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.

Enforcing Fair Lending: Evidence from Mortgage Market Litigation Matthew Maury*, Michael Suher†, and Jeffery Y. Zhang‡ January 2026 Abstract: Doesfairlendinglitigationimpactmortgagelenderdecisions? Usinganoveldataset of all fair lending legal actions from 1991 to 2023, we find that it does. In the wake of legal settlements for discrimination against Black borrowers, lenders significantly reduced denial rates forBlackapplicants. Thereductionsoffsetpre-litigationracialdisparitiesindenialratesbylitigated banks, relative to those banks’ competitors. Origination rates for Black applicants also increased post-litigation. Wefurtherobserveevidenceofaspillovereffectontheapprovaldecisionsofnonlitigatedbanksoperatinginthesamecityasalitigatedbank. Altogether,theevidencesuggeststhat the enforcement of fair lending laws is an effective tool to reduce racial discrimination in credit markets. *NewYorkUniversitySchoolofLaw †BoardofGovernorsoftheFederalReserveSystem ‡UniversityofMichiganLawSchool TheauthorsthankCarolinHoeltken,MichaelOhlrogge,CatherinePetrusz,JJPrescott,JoshSilver,aswellasparticipants at the AREUEA-ASSA Conference for their helpful comments and suggestions. The views expressed in this articledonotnecessarilyrepresenttheviewsoftheauthors’affiliatedinstitutions,includingtheBoardofGovernors oftheFederalReserveBoardandtheFederalReserveSystem. 1

1 Introduction An important factor that contributes to the long-run economic well-being of individuals is the locationwhereonelives(Chetty,Hendren,andKatz,2016). Familieswhocanpurchaseahomein theneighborhoodoftheirchoiceatafairprice—andseethevalueoftheirhomegrowovertime—do bettereconomicallyinthelongrun,assurveyedbyRouse,Bernstein,Knudsen,andZhang(2021). OverthecourseofU.S.history,numerouspolicieshavediscriminatedagainstracialminoritieswho wishtopursuethepathofhomeownership. Overthedecades,Congresspassed“fairlendinglaws” suchastheFairHousingActof1968tocombatthisproblem. Despite their decades-long existence, much remains unknown regarding the difference these laws make. And importantly, if they have made a difference, how much of a difference? Theory suggeststhathavingtheselawsonthebooksshouldmovetheneedleagainstdiscriminationincredit markets. Yet we still see evidence of discrimination in mortgage lending (Faber, 2018; Zhang and Willen, 2021) and other credit markets, such as the market for auto loans (Butler et. al., 2022). Perhaps these fair lending laws were less powerful than originally imagined. Or, perhaps these lawsarepowerfulbuttherehasn’tbeenenoughenforcement. Wefocusonenforcementthroughlitigationasaparticularmethodofenforcingthefairlending laws. Specifically,weassesshowmortgagelendersandborrowersrespondtofairlendinglitigation, usingadifference-in-differencesresearchdesignacrossarangeofoutcomevariablesandcontexts, including denial, origination, and securitization rates as well as national litigation and litigation targetedatspecificgeographies. We find that lenders significantly reduce denial rates for Black applicants in the wake of legal settlements,largelyeliminatingpre-litigationracialdisparitiesindenialratesatlitigatedbanksrelativetotheirlocalcompetitors. Pre-litigation,banksthatfacefairlendingenforcementinoursample deny mortgage applications from Black borrowers at a rate that is 3 percentage points higher than the denial rate at banks that are not subject to fair lending enforcement (a statistically significant difference). Post-litigation, banks subject to enforcement reduce their denial rate for Black borrowersbynearly4percentagepoints. Thisrepresentsaboutonethirdoftheabsolutepre-litigation 2

difference in Black-White denial disparity rates at litigated banks. The change persists through at least four years post-litigation. Moreover, on average, origination rates for Black applicants increasepost-litigation. We also break down our results by allegation type, and report circumstantial evidence that the additionalloanstoBlackborrowersgotowardcreditworthyborrowers,meaningthatlitigatedbanks mayhavebeenrefusingtolendtocreditworthyBlackborrowerspriortolitigation. Finally,wefind evidenceconsistentwiththeexistenceofa“spillovereffect”whereinnon-litigatedbanks“exposed” toafairlendingsuitagainstalocalcompetitorthemselvesreducedenialratesforBlackapplicants in the wake of litigation. After lenders face enforcement for alleged fair lending violations within a given CBSA, the denial rate for Black borrowers in that CBSA—for lenders that do not face enforcement—decreasesbyastatisticallysignificant1percentagepointpost-litigation. Toappreciatethesignificanceofthequestionoffairlendingenforcement,onemustunderstand the history of discrimination that culminated in the fair lending laws. It wasn’t until 1948 that the SupremeCourtheldinShelleyv. Kraemerthatraciallyrestrictivecovenantsarelegallyunenforceable, reversing its own decision from 1926 in Corrigan v. Buckley. Because the Shelley decision wasbasedontheEqualProtectionClauseoftheU.S.Constitution,however,itonlyappliedtogovernment facilitation of housing discrimination and said nothing of discrimination among private actors, such as banks, landlords, and home-sellers. Nor did Shelley’s holding extend beyond the context of express, racially restrictive covenants. As discussed extensively in the literature, race became the primary determinant of mortgage eligibility in the 1930s. Using standardized evaluation forms, officials from the Federal Housing Administration determined which homes it would insure (Baradaran, 2019). The most important determination on each form was the percentage of “negro”or“foreignborn”residentsineachneighborhood,aswellasthelikelihoodof“infiltration” of each race. Race became a proxy for credit risk in government underwriting. These maps had fourcolorcategoriesbasedonperceivedrisk: A(green),B(blue),C(yellow),andD(red). Green wasthemostdesirableandredtheleast—hencethename“redlining”(Rothstein,2017). The economic impact of discrimination in mortgage lending has been significant. Appel and 3

Nickerson(2016)studythelong-termeffectsofredliningpoliciesthatrestrictedaccesstocreditin urban communities. They find that redlined neighborhoods had 4.8 percent lower home prices in 1990 relative to adjacent areas. Aaronson, Hartley, and Mazumder (2021) show that the redlining mapsreducedhomeownershiprates,housevalues,andrentsandalsoincreasedracialsegregation in later decades. The authors’ results suggest the maps had meaningful and lasting effects on the developmentofurbanneighborhoodsthroughreducedcreditaccessandsubsequentdisinvestment. Relatedly,Aaronson,Faber,Hartley,Mazumder,andSharkey(2021)estimatethelong-runeffects ofthe1930sredliningmapsoncensustract-levelmeasuresofsocioeconomicstatusandeconomic opportunity. They find that the maps had large and statistically significant causal effects on a wide variety of outcomes measured at the census tract level for cohorts born in the late 1970s and early1980s. Similarly,Li(2022)arguesthatearlyconstraintsonBlackhouseholds’neighborhood choicesexplainthepersistenceinsegregationacrosscitiesbetween1960and2010. In1968,aspartoftheCivilRightsAct,CongresspassedtheFairHousingAct,whichinitially prohibited discrimination in residential real estate transactions based on race, color, religion, sex, or national origin (Horowitz, 2018). However, before the 1970s, it was unclear whether federal law prohibited lenders from discriminating against prospective borrowers based on the perceived race of the borrower or the borrower’s neighborhood. In 1974, Congress passed the Equal Credit Opportunity Act (“ECOA”), which (including subsequent expansions) prohibited discrimination in any aspect of a credit transaction (Rohner, 1978). And, in 1977, Congress passed the CommunityReinvestmentAct(“CRA”),whichencouragedfederallyinsuredbanksandthriftstomeetthe creditneedsoftheentirecommunitiesthattheyserve,includinglow-andmoderate-incomeareas, consistentwithsafeandsoundbankingpractices(Barr,2005). By1980,itwassettled thatfederal lawprohibitedsuchdiscriminatorylendingpractices,atleastwhencarriedoutovertlywithrespect torace,color,nationalorigin,religion,sex,familialstatus,ordisability(Nash,2008). In the decades since the passage of these statutes, we still witness discrepancies with respect to lenders providing loans to racial minorities. Munnell, Tootell, Browne, and McEneaney (1996) showthatminoritiesweremorethantwiceaslikelytobedeniedamortgageaswhitesintheBoston 4

area. Zhang and Willen (2020) show that discrimination exists even when accounting for a menu problemwhereinmortgageborrowerscanchoosetoeitheravoidclosingcostsandpayahighinterestrateorcontributetoclosingcostsandgetalowerrate. Thesediscrepanciesacrossracialgroups evenpersistwithrespecttoFinTechlenders(Bartlett,Morse,Stanton,andWallace,2022). Given the continued existence of discrimination in lending markets, is it possible that the enforcementoffairlendinglawsisnoteffective? Orhavethelawssimplynotbeenenforcedenough? The academic literature has yet to fully address this question empirically. We are aware of only two related working papers. First, An, Bushman, Kleymenova, and Tomy (2022) explore whether banking supervision plays a role in improving access to credit for minorities by investigating enforcement decisions and orders (EDOs) executed as part of the bank supervisory process. They note that regulators bring enforcement actions against banks as a measure of last resort and exercisesomediscretioninissuingEDOs. WhileafewEDOsdirectlyreferencefairlendingpractices, EDOsaregenerallynotconcernedwithbanks’adherencetofairlendinglaws,asfairlendinglaws are overseen via a separate and distinct supervisory process (as we discuss below). The authors find that mortgage lending to minority borrowers does significantly increase post EDO and that thispositiveeffectincreaseswiththeseverityofanEDO. Ourarticledoesnotexamineactionstakenbybanksupervisorsbutratherlitigationincourt. In additiontoquantifyingtheimpactoffairlendinglitigationonabank’slendingbehavior,weanalyze whethertherearespillovereffectstootherbanksinthearea. Todoso,weassembleanoveldataset ofallfairlendinglegalactionsfromtheearly1990sthrough2023andpairitwithregulatoryHome Mortgage Disclosure Act (“HMDA”) mortgage application data. We use this combined dataset to analyzepatternsinmortgageapplicationdecisionsatlitigatedlenders,relativetotheirnon-litigated localcompetitors. Inasecondrelatedproject,BallewandPears(2023)analyzethesignificantdropinfairlending enforcement actions brought by the Department of Justice between the Obama administration and the first Trump administration. The authors note that the administration change in 2017 led to an immediate 99 percent drop in Consumer Financial Protection Bureau (“CFPB”) and Department 5

of Justice (“DOJ”) fair lending enforcement actions. The authors find that following the administration change, banks charged higher interest rate spreads to Black or Hispanic borrowers and borrowers in lower- and middle-income areas. The authors’ results were particularly substantial amongbanksthathadpreviouslybeenthesubjectoffairlendingenforcement. Whileweanalyzesimilarmortgageloandatainourproject,ouranalysisisbasedonindividual casesandthusservesasa“bottomup”complementtothe“topdown”analysisofBallewandPears (2023). In particular, Ballew and Pears (2023) analyze how racial disparities across the mortgage market—as well as specifically among banks that had previously been the subject of fair lending enforcement—changefollowingaggregatereductionsinfairlendingenforcement. Bycontrast,our analysis looks at individual enforcement actions and how the lending behavior of both the targets of those actions and of competitor banks in the locations where discrimination allegedly occurred changes in the immediate aftermath of litigation. We break down our results by allegation type andprimarilyfocusondenialrates. Oursamplealsoincludescasesbroughtbymunicipalitiesand privatepartiesasclassactions,inadditiontofederalgovernmentenforcement. The rest of our paper proceeds as follows: Section 2 presents the institutional background and economic theory. Section 3 describes various datasets used in the analysis. Section 4 explains in detail the regression specifications. Section 5 contains our main regression results as well as robustnesschecks. Section6concludes. 2 Background, Theory, and Hypothesis Development 2.1 Institutional and Theoretical Background on the Mortgage Market and Mortgage Discrimination The U.S. mortgage market consists of myriad credit suppliers, forms of credit, and regulators. Banks and non-bank mortgage lenders alike offer mortgage loans for home purchases, investment properties, mortgage refinances, and home improvements. Whereas “conventional” mortgages’ arenotinsuredbythefederalgovernment,specialmortgagecreditprogramsthatinsureagainstthe 6

risk of borrower default are available for target, priority groups and offered by the Fair Housing Administration (“FHA”), United States Department of Agriculture (“USDA”), and United States DepartmentofVeterans’Affairs(“VA”).Originatedmortgagescanbesecuritizedandsoldtogovernmentsponsoredentities(“GSEs”)suchasFannieMaeandFreddieMacortoprivatemortgage investors. The regulatory landscape, in turn, involves a host of regulators, including the Federal Reserve,FederalDepositInsuranceCorporation(“FDIC”),OfficeoftheComptrolleroftheCurrency (“OCC”), CFPB, and DOJ. While the Federal Reserve, FDIC, and OCC primarily focus on promoting safety and soundness through supervisory exams and stress testing rather than direct fair lending enforcement, An et al. (2022) show how EDOs stemming from the supervisory process reduce racial disparities in mortgage terms among the targets of such EDOs. Moreover, evidence of potential fair lending violations uncovered during supervisory examinations by the Federal Reserve, FDIC, or OCC are often referred to the CFPB or DOJ, who may then conduct a further investigationand/orbringenforcementaction.1 The CFPB, by contrast, is directly responsible for enforcing the ECOA and Fair Housing Act and to that effect has the power to issue rulemakings and conduct supervisory exams and enforcement actions. Frequently, CFPB enforcement actions are brought in conjunction with the DOJ, whichlikewisehasthepowertoenforcetheECOAthroughlitigation. Privatepartiesarearguably also part of the regulatory mix. Particularly with class action litigation, litigation vindicates the privateinterestsofnamedplaintiffsandclassmembersbutalsothebroaderpublicinterestinlegal compliancebydefendants. In principle, credit suppliers will offer loans in accordance with the expected profitability and riskoflendingtoagivenborrower,accountingforfactorssuchasincome,creditscore,delinquency history, and the degree of borrower leveraging. As with all markets, mortgage pricing terms will vary based on the level of competition in the market, and rational lenders will extend credit offers so long as the risk-adjusted expected returns meet or exceed the expected sum of the cost and 1See,e.g.,Complaint,USv. AlbankFederalSavingsBank(N.D.N.Y1998). 7

opportunitycostassociatedwithofferingtheloan(Dobbieetal.,2021). Lenders can vary interest rates as well as “discount points” (upfront fees paid in exchange for lowerinterestrates)toadjustfortherisksassociatedwithofferingagivenloantoagivenmortgage applicant (Zhang and Willen, 2021), including by applying “overage” charges above the levels thatobjectivemodelsofcreditworthinessrecommendand“underages”thatreduceloanchargesto below the levels that creditworthiness models recommend.2 Lenders can also adjust risk through non-price credit rationing, such as by denying a loan application altogether, refusing to operate in neighborhoodstheydeemhigh-risk,or(closelyrelated)avoidingmarketingandoutreachtocertain classesofprospectiveborrowers(StiglitzandWeiss1981). Yet mortgage lenders are not impervious to discrimination on the basis of legally protected classifications. Asdiscussedabove,thereisahistoryofrace-baseddisparatetreatmentofborrowers intheU.S.mortgagelendingprocess. A key worry with mortgage lending is that, in discriminating against mortgage applicants on the basis of permissible characteristics related to creditworthiness and risk, lenders will also discriminatebasedonimpermissiblecharacteristicssuchasrace,sex,ornationalorigin. The contemporary theoretical literature in economics on discrimination began with Gary Becker’s seminal work, “The Economics of Discrimination,” published in 1957. Becker’s theory provides that lenders with overt bias against persons of a certain group will only extend credit tosuchpersonsatacost-orcreditworthiness-premium. Saiddifferently,lenderswilleithercharge higher interest rates to, or expect superior creditworthiness from, persons of a targeted group in ordertoovercometheiraversiontosuchpersons. Accordingly,Becker’stheorypredictsthatloans extended to members of a targeted group will perform better than loans to other borrowers—in termsofrisk-adjustedexpectedprofits—giventhemorestringentstandardsappliedtothetargeted group. A corollary of this theory is that, ceteris paribus, borrowers from a discriminated group will be offered higher interest rates and subject to higher denial rates than other similarly situated mortgageapplicants. 2SeeUSv. ChevyChaseBank(E.D.Va. 2013). 8

A second theory of discrimination, presented by Arrow (1973), looks beyond overt aversion to certain groups and instead involves “statistical discrimination.” Arrow’s theory maintains that lendersmayrelyongroup-widestereotypesaboutthecreditworthinessofmembersofcertainraces and use those stereotypes to resolve uncertainty when make lending decisions on marginal applicants (Arrow 1973). Arrow’s theory offers the same predictions about how discrimination will manifestinlendingdecisionsasBecker’s,aswellasthesamecorollaryimplications. AmorerecenttheorybyDobbieetal. (2021)examinesprincipal-agentproblemswithinlending institutions. Specifically,theauthorstheorizethatmismatchedincentivesbetweeninstitutions(i.e., theprincipals)andagentloanofficersleadloanofficerstodiscountthelong-termreturnsassociated withextendingaloantoagivencreditapplicantinfavoroftheloanofficer’sownshort-termgains. Rather than neutrally assessing an applicant’s creditworthiness and the long-term profitability of originating a loan to the applicant, loan officers instead employ their own overt or “statistical” biases to vet applicants, since they do not share in the long-term benefits that principals receive from originated loans. Dobbie et al. (2021) offer empirical support for this theory, and other studies present evidence consistent with their theory. While Bhutta et al. (2022) offer compelling evidencetosuggestthataverageracialdifferencesincreditworthiness,ratherthanprejudicebyloan officers, explain mortgage market discrimination, Butler et al. (2022) demonstrate that in places where discrimination is more prevalent, racial minorities face access barriers to and differential treatment in the automotive lending market, where they tend to be vetted in in-person, face-toface interactions, while escaping such problems when applying for credit cards, for which credit decisionstendtobeautomated. The allegations advanced by the litigation actions analyzed in this study involve elements of eachofthesetheoriesanddifferentialtreatmentalongbothprice-andnon-priceterms. Oursample includesfairlendingactionsallegingdiscriminatorypricing,discriminatoryorigination,redlining, and reverse redlining. The actions over discriminatory pricing allege that racial minorities were offered credit on less favorable terms—higher interest rates—on average than White applicants withthesamecreditscoresandlevelsofcreditworthiness. Actionsoverdiscriminatoryorigination 9

allege the analogous with respect to mortgage application denial rates. Consistent with Becker (1957)’s and Dobbie et al. (2021)’s theories, the complaints in such actions frequently allege that banks gave loan officers too much discretion in the underwriting process, including by failing to implement uniform, objective criteria for assessing applicant creditworthiness and permitting too many opportunities for loan officers to override centralized underwriting systems (to the extent targeted banks have them) and alter recommended mortgage terms on an applicant-by-applicant basis. In turn, the redlining and reverse redlining actions center around banks allegedly failing to provide credit to majority-Black or Hispanic neighborhoods or deliberately targeting such neighborhoods with loans featuring supra-competitive interest rates and other unfavorable terms, respectively. The targets of redlining litigation, especially, may be engaging in overt discrimination wherein they are actively averse to lending to minorities (Becker, 1957). The targets of redlining andreverse redlining enforcement may alsobe statistically discriminating; the targetsofredlining actions may be relying on the perception that racial minorities are less creditworthy to deny credit to residents of majority-minority neighborhoods altogether (Arrow 1973). In turn, the targets of reverseredliningactionsmayperceivesuchresidentsaslessfinanciallysavvyorhavinglessaccess to credit on competitive terms and therefore more susceptible to accepting the unfavorable credit optionswithwhichdefendantsallegedlytargetresidentsofsuchneighborhoods. Becker (1968) theorizes that enforcement, whether through criminal or civil sanctions, incentivizes legal compliance by forcing potential targets of enforcement to internalize the cost of such sanctions. In the case of overtly biased lenders, a la Becker (1957)’s theory, the expected cost of enforcement shifts the calculus of whether to lend to a member of a disfavored racial minority by counteracting the “benefit”—the satisfaction gained from not lending to such a person—with the cost of enforcement, should the lender be caught illegally discriminating. As for the “statistical” and principal-agent theories of discrimination by Arrow (1973) and Dobbie et al. (2021), respectively,theexpectedcostofenforcementincentivizesthefindingofalternativewaysforevaluating mortgageapplicants(inthecaseofstatisticaldiscrimination)andtheestablishmentofpolicies,pro- 10

cesses, and practices for evaluating mortgage applicants that minimize the risk of racial prejudice byloanofficersaffectingtheoutcomeofcreditdecisions. Thus, the “stick” of monetary awards associated with enforcement actions serves as one potential channel through which such actions may redress and deter illicit mortgage discrimination. Injunctive-like relief is a second potential channel. In particular, fair lending settlements often enjoinbanksfromcontinuingwiththepoliciesorpracticesthatallegedlyleadtoracialdiscrimination to begin with. Thus, fair lending settlements may also reduce discrimination by tackling and enjoininghead-onsuchpurportedrootcausesofdiscrimination. Wedonotattempttoassesswhichoftheforegoingtwochannelsexplainourfindingsregarding the effectiveness of fair lending litigation. It could well be a combination. Nevertheless, the preceding discussion offers potential theoretical insight as to how and why mortgage discrimination observablydecreasesfollowingfairlendingenforcement,astheresultspresentedbelowindicate. 2.2 Hypothesis Development Our institutional and theoretical discussion yields several hypotheses. First, we hypothesize that: Denial rates (origination rates) for racial minorities at litigated banks will be higher (lower) prior to discriminatory pricing and origination litigation and lower (higher) followingsuchlitigation,relativetocorrespondingratesatnon-litigatedbanks. It is obvious why we would expect these results with discriminatory origination litigation: such litigationallegesthatdefendantshavehigherdenialratesandloweroriginationratesforracialminorities, relative to their competitors. Accordingly, if such litigation is effective, we would expect it to ameliorate those disparities, with denial rates decreasing and origination rates increasing followinglitigation. Itislessstraightforwardwhythishypothesisappliestodiscriminatorypricinglitigation,should it be effective. Our sample does not include interest rates or rate spreads for originated mortgages 11

before 2018, so we primarily analyze the effects of discriminatory pricing cases by looking at denialratesandoriginationratesatlitigatedbanksbeforeandafterlitigation. Weneverthelessexpect racialdisparitiesininterestratestodecreasefollowinglitigation,assumingsuchlitigationiseffective. As a result, we should also expect origination rates for minority applicants to increase after discriminatory pricing litigation (and disparities in origination rates to narrow) because borrowers wouldbemorelikelytoacceptmortgagecreditoffersiftheyaremadeonmorefavorableterms. As fordenialrates,weexpectthemtodecreaseifdiscriminatorypricinglitigationeffectivelyincreases origination rates; originated applications by definition are not denied, so an increased probability of origination should mean a decreased probability of denial.3 Moreover, to the extent banks chargehigherinterestratestominorityborrowers—whetherbecauseofovertprejudiceorstatistical discrimination—itispossiblethattheyalsosubjectminorityapplicantstomorestringentcriteriain whethertoofferminorityapplicantscredittobeginwith. Accordingly,wewouldexpectbanksthat offerminorityborrowersdisfavorableloantermstoalsodenytheirmortgagecreditapplicationsat higherrates—andforthoseheighteneddenialratestodecreaseaftereffectivelitigation. Second,wehypothesizethefollowingwithrespecttoredlininglitigation: Origination rates (denial rates) for racial minorities at litigated banks will be lower (higher)priortoredlininglitigationandhigher(lower)followingsuchlitigation,relativetocorrespondingratesatnon-litigatedbanks. This hypothesis follows from the fact that redlining involves under-provision of mortgage credit to majority-minority neighborhoods. A common remedy in redlining litigation, as discussed, is to requiredefendantbankstoopennewbranchesinsuchpreviously-under-servedneighborhoodsand to increase outreach and marketing to the target consumer group. Accordingly, if such measures are effective, we would expect mortgage applications from majority-minority neighborhoods to 3Itisneverthelesspossiblefororiginationratestoincreasesolelybecauseofreducedratesatwhichapplications are accepted but not originated, in which case effective litigation would not affect denial rates while still increasing origination rates. It is also possible for origination rates and denial rates to both increase post-litigation, but only if (1)increasedoriginationratesexclusivelyresultfromreducedratesofbeingacceptedbutnotoriginatedand(2)that increaseissufficientlylargethatitoffsetstheheightenedprobabilityofdenial. 12

increase following litigation. To the extent that minorities are more likely to live in such neighborhoods, we would expect applications from minority borrowers to increase as well. Along with those increases, we might also expect overall origination rates for minority borrowers to increase, anddenialratestomechanicallydecreaseinconsequence. Third,withrespecttoreverseredliningwehypothesizethat: Origination rates (denial rates) for racial minorities at litigated banks are higher (lower) prior to reverse redlining litigation and lower (higher) following such litigation, relative to corresponding rates at non-litigated banks. In turn, while our data setdoesnotcontaindataoninterestrates,weexpectinterestratesofferedtominority borrowers—andWhite–non-Whitedisparitiesininterestrates—todecreasefollowing reverseredlininglitigation,relativetocorrespondingratesatnon-litigatedbanks. Thishypothesisfollowsfromthefactthatreverseredlininglitigationinvolvestheallegedtargeting of non-White neighborhoods with unfavorable mortgage loans. Sometimes such litigation alleges thatbanksengaginginreverseredliningoffercredittoborrowersthatarenotcreditworthy,onlyto createhighratesofmortgagedefaultandforeclosurefollowingorigination. Asbanksbecomemore discerningandbegincontrollingriskthroughnon-pricecreditrationing,moreover,originationrates would be expected to decrease (and, as a mechanical result, denial rates be expected to increase). In turn, by concentrating the extension of credit on more creditworthy borrowers, interest rate disparitieswoulddecrease. Fourth, with respect to securitization rates for originated loans, there are two competing hypotheses. The first, which occurs if observed White–non-White disparities result from accurate differentiationbetweencreditworthyandnon-creditworthyborrowers,isthat: Non-GSE securitization rates among non-White borrowers increase after litigation and GSE securitization rates among this group decrease after litigation, relative to non-litigatedbanks. 13

As An et al. (2022) observe, GSEs have stringent requirements surrounding borrower creditworthiness for the securitized mortgages that they purchase. As a result, GSE purchases of securitized loans can serve as a proxy for borrowers being more creditworthy. Accordingly, if observed pre-litigation racial disparities in lending are actually the product of accurate discernment of borrower creditworthiness, then increases in post-litigation origination rates would entail originating mortgages to less creditworthy borrowers. That, in turn, would reasonably lead banks to increase securitization rates for new non-White borrowers, in order to mitigate default risk. Given that the securitization and selling of mortgages to GSEs functions as a proxy for a borrower being more creditworthy,however,wewouldalsoexpectsecuritizationstoGSEstodecrease;theincreasesin securitizationtomitigateagainsttherisksofreducedborrowercreditworthywouldprimarilybefor salestonon-GSE purchasers. The corollary of the foregoing, therefore, is that observed White–non-White disparities result from impermissible race-based discrimination rather than accurate differentiation between creditworthyandnon-creditworthyborrowers. Ifthatisthecase,thenthealternative,secondhypothesis regardingdiscriminationisthat: Non-GSE securitization rates among non-White borrowers decrease after litigation and GSE securitization rates among this group increase after litigation, relative to non-litigatedbanks. Thelogicbehindtheforegoingissimplytheinverseofthelogicfortheinitialhypothesissurroundingsecuritization. Ifrace-baseddisparitiesarebasedonrace-baseddiscriminationthathasnothing to do with discrimination, then effective litigation would increase the amount of loans originated to creditworthy borrowers; accordingly, non-GSE securitization rates would decrease and GSE securitizations,asproxyforpositivelycreditworthyborrowers,wouldincrease(Anetal.,2022). Finally, with respect to our estimates of the spillover effects on local competitors of litigated banks,wehypothesizethat: Denialrates(originationrates)forracialminoritiesat“exposed”bankswillbelower 14

(higher) following litigation, relative to corresponding rates at banks that are not exposedtothethreatoflitigationviaenforcementagainstalocalcompetitor. This final hypothesis is premised on the idea that litigation against a local competitor serves as a reminder of the risk of enforcement if banks do not comply with fair lending laws. As Ballew andPears(2023)demonstrate,decreasesinaggregatebankenforcementareassociatedwithhigher racial disparities in mortgage interest rates and thus, presumably, lower levels of compliance with fair lending laws. By contrast, litigation against local competitors may increase the expected cost of enforcement, as perceived by exposed banks, by increasing the perceived probability that noncompliancewithfairlendinglawwillleadtoenforcement. 3 Data Toconductouranalysis,wemakeuseofsixseparatedatasets: (1)ahandcodedoriginaldataset of 38 separate fair-lending litigation actions, (2) the Home Mortgage Disclosure Act (“HMDA”) dataset, (3) “The Avery File,” a government dataset for cross walking bank respondent IDs in HMDA with Federal Financial Institutions Examination Council (“FFIEC”) financial institution RSSD codes, (4) the FFIEC financial institution “Relationships” dataset, (5) the FFIEC financial institution“Transformations”dataset,and(6)ahierarchicaldatasetofCensusgeographicunits. 3.1 Litigation Sample WedrawfromtheCivilRightsLitigationClearinghouserunbytheUniversityofMichiganLaw School and cases listed and brought by the DOJ to construct our sample of fair lending litigation actions.4 WesearchedtheClearinghouseandDOJwebsitesforfairlendinglitigationbroughtunder theEqualCreditOpportunityActagainstbanksforallegeddiscriminationinthemortgagemarket, and then hand coded the details of each case.5 In particular, we identified case names, the year 4See https://clearinghouse.net/search/case/?case_type=5048ordering=-summary_approved_date and https://www.justice.gov/crt/housing-and-civil-enforcement-section-cases-prior-2018#lending. 5DuetolimitedloandataonIndianreservations,werestrictedoursampletolitigationforallegeddiscrimination withintheUnitedStatesbutnotonIndianreservations. 15

thateachcomplaintwasfiled,thenamesofdefendantbanks,andthetypeofdiscriminationalleged (i.e.,mortgageinterestratediscrimination,applicationapprovalratediscrimination,redlining,and reverse redlining). For each complaint, we also took note of the geographic location(s) where defendantsallegedlydiscriminated,althoughmanycomplaintsallegethatdiscriminationoccurred inall geographieswherethebankdoesbusiness. Because our analysis looks at the effect that litigation concerning discrimination in the home purchasemortgagemarkethasondiscriminationwithinthatmarket,ourlitigationsampleexcludes fairlendinglitigationoverformsofcreditthatdonotpertaintohomepurchases. Thus,weexclude litigation over home equity lines of credit, the secondary mortgage market (e.g., the market for mortgage-backedsecurities),businessloans,andotherformsofunsecuredcredit. Wefurtherrestrictoursampletoactionsbroughtbythefederalgovernment,stategovernments, municipal governments, and private actions in which plaintiffs successfully obtained class action status. The core insight behind limiting our analysis to government enforcement actions and certifiedclassactionsisthatthoseactions,contrarytoprivateindividuallitigation,pertaintoalleged company-widediscriminatorypractices,asopposedtoone-offinstancesofdiscriminatorymistreatment.6 The frequency of bank mergers and acquisitions presents a key difficulty in constructing our sample. Several actions in our sample were brought against banks after they were acquired by a different bank. Accordingly, our main analysis excludes those enforcement actions as well, as we could not measure how an entity that no longer exists changes its practices after facing enforcement.7 In other cases, enforcement is brought before a merger or acquisition, but the merger 6Ofcourse,individuallitigantscouldbringprivate,non-classactionenforcementchallengingbank-widediscriminatory practices. However, given the high volume of private, individual fair lending lawsuits, and the feasibility challengesinvolvedinseparatingsuchactionsthatconcernanindividual’streatmentalonefromprivateactionsover allegedlycommon,bank-widediscriminatorypoliciesandpractices,werestrictoursampletogovernmentactionsand classactions. 7Onealternativeanalyticalpossibilitywouldbetoassesstheextenttowhichmergedoracquiringentitieschange theirpractices–andobservedracialdisparitiesinlending–afterpurchasingormergingwithbanksthatfaceenforcement actions. That alternative possibility presents myriad problems for our empirical strategy, however, as it would be unclearhowtoestablishrelevant”pre-”and”post-”timeperiodsoverwhichtoassessdifferentialtreatmentbetween White and non-White mortgage applicants, as well as how any differential treatment changes from before to after enforcement. For example, if Bank A acquires Bank B shortly before B faces enforcement, we could analyze B’s treatment of mortgage applications pre-acquisition and pre-enforcement, but analyzing post-acquisition (and post- 16

or acquisition occurs within four years of the filing of the lawsuit. In those cases, we include pre-enforcement mortgage applications in our sample—and measure lending patterns during that window—butexcludepost-enforcementmortgageapplications.8 Thebasisforthisdecisionisthat, aftercontrollingforotherrelevantvariables,thepre-enforcementlendingpatternscanassistincalculatingthedifferentialbetweenpre-andpost-enforcement. Lastly,onesignificantpointofnoteis that some supervisory examinations for fair lending compliance occur in response to proposed or recentmergersandacquisitions.9 Thisinpartexplainstherelativelysubstantialnumberofmergers andacquisitionsofdefendantbanksinourlitigationsample. Wefurtherexcludedenforcementactionsagainstbanksforwhichthesamplesizeofmortgage applications in HMDA was so small that we could not run regressions estimating pre- and postlitigationtrends. Allofsuchexcludedactionsinvolvedbankswithfewerthan20applicationsfrom Blackborrowersoveran8-yearperiodsurroundingthedateoflitigationfiling(i.e.,4yearsbefore to4yearsafterfiling). In our analysis of the spillover effects of fair lending litigation, however, we nevertheless included cases that were otherwise excluded because of an M&A or limited sample size, as just described. The basis for this decision is that, with respect to both, the econometric problems with estimatingtheeffectsoflitigationonthedefendantbankdonotcarryovertoestimatingthespillover effectsonlocalcompetitors. Appendix 8.1 presents a table of every litigation that we identified but ultimately excluded, as wellasthebasisforourdecisiontoexclude. InTable2,wepresenttheresultingsampleofenforcementactions,whichconsistsofthirty-nine separate lawsuits. Twenty-eight cases are brought by the federal government, seven by municipal enforcement)differentialtreatmentwouldentailpoolingtogetherapplicationstothenew,combinedentity. Wecould alternativelymeasureA’soriginationpracticesbeforeandafterBissued,butanypost-acquisitionanalysiswouldnot controlfororganizationalchangesthatresultfromthatacquisition. Eitherway,therewouldnotbeaclose”applesto apples”comparisonoftreatmentversuscontrolgroups,andforthatreasonwedroptheselitigations. 8IntheAppendix,weneverthelessmeasurethetreatmenteffectoftheselitigationactions. Ourprimaryreasonfor excludingtheactionsinquestioninourmainspecificationisthatwedefineourtreatmentwindowasthefouryears thatfollowthefilingofeachlawsuit. 9See,e.g.,USv. ChevyChaseBank(E.D.Va. 2013)(notingthattheOCCconductedafairlendingexamination ofdefendantinconnectionwithitsacquisitionbyCapitalOne). 17

governments,andfourareprivateclassactionlawsuits. Elevenofthelawsuitsinoursampleinvolve allegednationwidediscriminationbythedefendant,whilethecomplaintsintheothertwenty-seven cases allege discrimination in specific markets or regions. In turn, eighteen cases allege discriminatory pricing, one alleges discriminatory origination rates, thirteen allege illegal redlining, and sevenallegeillicitreverseredlining. Table1outlinesthesesummarystatistics. Monetary relief in successful fair lending settlements tends to encompass compensatory relief for harmed consumers and/or remediation, such as consumer education programs, loan subsidies for non-White applicants, and credit provision targets to minority borrowers. Settlements tend to also include injunctive-like relief. In discriminatory pricing and origination litigation, such relief might include requisite internal policy changes to ensure uniform, objective criteria for vetting mortgageborrowers,withlimitsondiscretionaryinputfromloanofficers. Injunctive-likereliefin discriminatory pricing and origination settlements often also requires banks to provide their loan officers with anti-bias training. In turn, injunctive-like relief in redlining actions often require banks to open new branches in majority-minority neighborhoods, as well as to conduct marketing andoutreachtosuchneighborhoods. Table1: LitigationSummaryStatistics Count Percentage AllegationType Disc. Pricing 18 46% Disc. Origination 1 3% Redlining 13 33% ReverseRedlining 7 18% PlaintiffType Federal 28 72% Municipal 7 18% PrivateClassAction 4 10% Total 39 100% DefendantAcquired NumberofTransactions 5 Avg. YearsAfterComplaintFiled 0.5 18

Table2: LitigationSample: CaseList Case Allegation Outcome Locationof Discrimination Payaresv. J.P.Morgan Discriminatory $300/classmember; National Chase(C.D.Cal. 2007) Pricing $1.965min (privateclassaction) attorneys’fees USv. J.P.Morgan Discriminatory $53mcompensatory National Chase(S.D.N.Y2017) Pricing fund USv. WellsFargo Discriminatory $175mcompensatory National (D.D.C.2012) Pricing fund USv. Suntrust Discriminatory $21mcompensatory National Mortgage Pricing fund (E.D.Va. 2012) Puellov. Discriminatory $200/classmember; National CitifinancialServices, Pricing $400,000in Inc. (D.Mass. 2008) attorneys’fees (privateclassaction) USv. C&F Discriminatory $140,000compensatory National MortgageCorp. Pricing fund (E.D.Va. 2011) Allenv. DecisionOne Discriminatory $6.5mcompensatory& National MortgageCo. Pricing remediationfund& (D.Mass. 2007) attorneys’fees (privateclassaction) USv. Primelending Discriminatory $2mcompensatory National (N.D.Tex. 2010) Pricing fund USv. PlazaHome Discriminatory $3mcompensatory National Mortgage Pricing fund (S.D.Cal. 2013) CFPBv. Provident Discriminatory $9mcompensatory National FundingAssociates Pricing fund (N.D.Cal. 2015) †Jacksonv. Novastar Discriminatory Litigationstayedafter National (W.D.Tenn. 2006) Pricing defendantfiledfor (privateclassaction) bankruptcy USv. ChevyChase Redlining $11mremediation DCMSA Bank(D.D.C.1994) fund CFPBv. Northern Discriminatory $700,000 Chicago TrustCo. Origination compensatory MSA (N.D.Ill. 1995) fund USv. Huntington Discriminatory $420,000 Cleveland MortgageCo. Pricing compensatory MSA (N.D.Ohio1995) fund 19

Case Allegation Outcome Locationof Discrimination USv. Fleet Discriminatory $4m Westbury, MortgageCorp. Pricing compensatory& NY& (E.D.N.Y.1996) remediation Woodbridge, fund NJ USv. LongBeach Discriminatory $4mcompensatory L.A.MSA MortgageCo. Pricing fund (C.D.Cal. 1996) USv. Delta Discriminatory $7.25mcompensatory Kings& FundingCorp. Pricing &remediation Queens (E.D.N.Y.2000) fund Cntys.,NY USv. MidAmerica Redlining $11.25mremediation Chicago Bank(N.D.Ill. 2002) fund MSA USv. FirstAmerican Redlining $5.7mremediation Chicago& Bank(N.D.Ill. 2004) fund Kankakee, IL,MSAs USv. CentierBank Redlining $4.38mremediation Gary,IN, (N.D.Ind. 2006) fund MSA MayorofBaltimore Reverse $7.5mremediation Baltimore v. WellsFargo Redlining fund;Coordinated MSA (D.Md. 2008) withUSv. WellsFargo settlement USv. FirstUnited Redlining $660,000 Alabama SecurityBank remediation Market (S.D.Ala. 2009) fund Area CityofMemphisv. Reverse $7.5mremediation Shelby WellsFargo Redlining fund;Coordinated Cnty.,TN (W.D.Tenn. 2010) withUSv. WellsFargo settlement †USv. Citizens Redlining $3.63mcompensatory Detroit RepublicBankcorp., &remediationfund MSA Inc. (E.D.Mich. 2011) USv. Midwest Redlining $1.45mremediation St. Louis Bankcentre fund MSA (E.D.Mo. 2011) †USv. GFI Discriminatory $3.56mcompensatory NY&NJ MortgageBankers Pricing fund&civilpenalties (S.D.N.Y.2012) CityofMiamiv. Reverse Voluntarilydismissed Miami,FL BankofAmerica Redlining afterprotracted (S.D.Fla. 2013) appellatelitigation overstanding 20

Case Allegation Outcome Locationof Discrimination CityofL.A.v. Reverse MotionforSummary L.A.,CA WellsFargo Redlining Judgmentagainst (C.D.Cal. 2013) Citygranted CityofL.A.v. Reverse Outcomeunclear; L.A.,CA Citigroup,Inc. Redlining Initialmotionto (C.D.Cal. 2013) dismissdenied CityofL.A.v. Reverse Dismissedwith L.A.,CA JPMorganChase Redlining prejudice;Unclearif (C.D.Cal. 2014) therewasan accompanying settlement †USv. EagleBank& Redlining $0.98mremediation St. Louis TrustCo. ofMissouri fund MSA (E.D.Mo. 2015) †CFPBv. HudsonCity Redlining $32.75mremediation NewYork& SavingsBank fund Philadelphia (D.N.J.2015) MSAs CityofOaklandv. Reverse MotiontoDismiss Oakland,CA WellsFargo Redlining grantedafter (N.D.Cal. 2015) appeal CFPBv. BancorpSouth Redlining $10.8mcompensatory Memphis Bank &remediationfund MSA (N.D.Miss. 2016) USv. Union Redlining $9mremediation Cincinatti, SavingsBank fund Dayton,& (S.D.Ohio2016) Columbus,OH, &Indianapolis, IN,MSAs †USv. KleinBank Redlining $0.6mremediation Minneapolis (D.Minn. 2017) fund MSA USv. First Redlining $1.7mremediation Indianapolis MerchantsBank fund MSA (S.D.Ind. 2019) †USv. Trustmark Redlining $5.25mremediation Memphis Nat’lBank fund MSA (W.D.Tenn. 2021) †Indicates“pretreatment”caseswherebyapplicationstothedefendantbankareonlyincludedinthe4-yearprelitigationcontrolwindowbutnotthe4-yearpost-litigationtreatmentwindow. Jacksonv. Novastarisdesignated “pretreatment”becauseofabankruptcyfilingbythedefendantthatcoincidedwiththecase. USv. TrustmarkNat’l Bankisdesignated“pretreatment”becauseitwasfiledin2021,thusonlyleavinga2-yearpost-litigationtreatment window. Theremainingfive“pretreatment”casesaredesignatedassuchbecauseofM&A’sinitiatedeithertheyear thatlitigationwasfiledorwithinthe4-yearpost-litigationtreatmentwindow. 21

3.2 HMDA Data The HMDA dataset records the near-universe of mortgage applications every year. We downloaded the HMDA dataset for every year from 1991 to 2023, filtered to applications for conventionalmortgagesforowner-occupiedhomepurchaseswithindesignatedcore-basedstatisticalareas, droppedapplicationsformanufacturedhomes,andkeptobservationsinwhichthelistedapplication outcome was: approval without origination, approval with origination, or a denial. Given the differences in underwriting standards among conventional, FHA, USDA, and VA loans, our analysis looks exclusively at conventional mortgages to control for such differences. Given computational limitations associated with analyzing the full sample, as a general matter we take a random 10% sub-sample of all observations in HMDA. Nevertheless, several defendants in our sample cases are relatively small lenders, meaning that sub-sampling would deprive our analysis of the requisite statistical power. Accordingly, for all defendant banks except members of “the big five” (i.e., WellsFargo,Citi,BankofAmerica,Chase,andU.S.Bank),weretainallobservationsthatmeetthe foregoingcriteria. Fordefendantbanksthatarepartofthebigfive,wetakea10%sub-sampleofapplicationsbyWhiteborrowersbutretainallobservationsofapplicationsbynon-Whiteborrowers. To account for these cross-bank and cross-applicant differences in sub-sampling, our regressions includeweightsindicatingwhenobservationsare“over-sampled”byvirtueofusnottakinga10% sub-sample. We use datasets (3)-(5) to merge the litigation data with the HMDA dataset. Respondent IDs in HMDA are not the same as RSSD IDs, which are unique identifiers assigned to each financial institution by the Federal Reserve. Dataset (3) allows us to map each HMDA respondent ID onto a Federal Reserve RSSD ID. Given financial institution mergers, acquisitions, and internal re-organizations, RSSD IDs are not always constant from one year to the next. Dataset (4) tracks each change in an RSSD ID, what that RSSD ID transforms into, and whether the transformation resultsfromamergeroracquisition. Weusedataset(4)tomapeachRSSDIDontothemostrecent RSSDIDthatdoesnotreflectanRSSDIDchangeattributabletoamergeroracquisition. Wethen usedataset(5)to“rollup”eachRSSDIDtoitshighesttraceableparentfinancialinstitutionorbank 22

holdingcompany. Thesestepsallowustotreateveryloanoriginatorinaconglomerateoffinancial institutionsasoneunitaryentity.10 InTable3,weprovidesummarystatisticsfromourHMDAdataset. Thesamplecontainsmore than ten million mortgage records, with seven percent filed by Black applicants. Unconditional denialratesare14percentforWhiteapplicants,17percentacrossallnon-Whiteapplicantsand30 percentforspecificallyBlackapplicants. On the income statistics, which are presented in thousands of dollars, we see that the average applicantmade$85,800,withtheaverageWhiteapplicantmaking$84,300andtheaverageamong Black applicants at $68,200. As expected, the incomes of approved applicants are higher than the average and the incomes of denied applicants are much lower. We also see that the average loan amount of $196,400 is more than twice the average income of approved applicants. Finally, 57 percentofallloansweresecuritizedandsold,with30percentsoldtoGSEsand27percentsoldto non-GSEs. Table3: HMDASummaryStatistics White Black Non-White All TotalApplications 8,253,043 892,518 2,405,810 10,658,853 Pct. TotalApplications 88.5% 7.0% 11.5% 100.0% TotalDenied 1,019,801 169,375 154,973 1,174,774 Pct. Denied 14.2% 29.7% 16.7% 14.5% TotalOriginated 5,658,275 348,473 717,836 6,376,111 Pct. Originated 79.0% 61.2% 77.1% 78.8% Avg. Income(Thousands) 84.3 68.2 97.5 85.8 SDIncome(Thousands) 64.0 52.1 67.5 64.6 Avg. IncomeifDenied(Thousands) 62.9 54.9 74.6 64.5 Avg. IncomeifApproved(Thousands) 87.8 73.8 102.0 89.4 Avg. LoanAmount(Thousands) 196.4 181.6 273.8 205.1 SDLoanAmount(Thousands) 147.3 136.4 181.5 153.5 Pct. Securitized&Sold 56.7% 43.4% 56.4% 56.7% Pct. SoldtoGSE 30.0% 17.6% 31.3% 30.1% Pct. Soldtonon-GSE 26.6% 25.6% 24.9% 26.4% 10OurassumptionisthatifAlphaBankHoldingCompanyownsBankA,BankB,andMortgageLenderC,alawsuit againstBankBwillaffecttheloanoriginationpatternsofallentitiesheldbyAlpha. 23

4 Empirical Analysis Our main specification is an applicant level regression. The idea underlying our empirical approachisthat,aftercontrollingforotherrelevantvariables,significantdecreasesindiscrimination immediately following a fair lending lawsuit are consistent with the lawsuit having some causal role in the observed decrease.11 We use what amounts to a triple difference estimation, where the comparisonisbetweenBlackandWhiteapplicants,12 beforeandafterlitigation,atlitigatedbanks relative to non-litigated banks. Thus, for applicant i from Census tract c applying for a mortgage atbankbincityminyeartwerun: Y = β +β LitWindow XPostLit XBlack +β LitWindow XPostLit ibmt 0 1 mt mt i 2 mt mt +β LitWindow XBlack +β LitWindow 3 mt i 4 mt +β Black +β LnIncome +β LnLoanSize 5 i 6 i 7 i +β LMI +β Male +β SameSex +β JointApplicants 8 c 9 i 10 i 11 i +δ +δ +ε . (1) b mt ibmt The variable LitWindow turns on for a litigated banks in a window of [-4,+4] years around the litigation year, and in the geography of litigation if one is specified in the lawsuit. Applicants to a litigated bank in the year of litigation are dropped, as we do not observe the month that mortgage applicationsaresubmitted,andthuswhethertheyarrivebeforeorafterasuitisfiled. Thecoefficient β measurestheaveragedifferenceindenialratesforBlackapplicantsrelativetoWhiteapplicants 5 acrossthesample,conditionalonincomeandloanamountsought. Coefficientsβ andβ measure 3 4 the extent to which denial rates at litigated banks are higher, in the four years pre-litigation, for Black and White borrowers, respectively, relative to applicants at non-litigated banks in the same 11Inparticular,wemeasurediscriminationbasedonthedisparityintheprobabilitythataloanapplicationisdenied, approvedbutnotoriginated,ororiginated. If,aftercontrollingforapplicantincome,loansize,location,thespecific bank applied to, and year, that disparity narrows post-litigation, there is a plausible claim that the lawsuit played a causalroleintheobservednarrowing. 12OuranalysislooksexclusivelyatWhiteversusBlack applicants. Thisisbecausethemajorityofthelawsuitsin oursampleallegeracialdiscriminationagainstBlackborrowersinparticular. 24

yearandmetroarea. Thecoefficientβ measureswhetherthereisageneralchangeindenialrates 2 for applicants to litigated banks in the four years post-litigation. The focal parameter is β , which 1 capturesanychangeindenialratesforBlackapplicantsrelativetoWhiteapplicantspost-litigation. Because the decision to apply for a loan and to approve one is subject to local housing market conditions,weincludecity-by-yearfixedeffects. Tomakebefore-and-afterlitigationcomparisons within each lender, we also include lender fixed effects. The indicator for the window stretching from four years before to four years after each litigation year narrows the period over which comparisons pre- versus post-litigation comparisons are made. We cluster standard errors at the bankby-city-by-yearlevel,sinceapprovaldecisionsarelikelycorrelatedwithinbankduetopoliciesset atamanagementlevel. Wealsocontrolforwhetheranapplicantisfromalowtomoderateincome censustractwithLMI ,sincethatvariablemaycorrelatewithapplicantcreditworthiness. Inaddic tion,wecontrolforwhetheranapplicantismale(Male ),whetheranapplicationcomesfromjoint i applicants(JointApplicants ),andwhetherthoseapplicantsareofthesamesex(SameSex ). Ali i though mortgage discrimination because of sex, marital status, and LGBT status are illegal, we followBallewandPears(2023)incontrollingforthesevariablestoensurethatourfocalparameter exclusivelymeasuresdifferencesinrace-baseddiscrimination. Asrecentstudieshaveargued,staggereddifference-in-differencesanalysescanproducebiased estimates because of the staggered timing of treatment and dynamic treatment effects (Baker et al., 2022; Sun and Abraham, 2021). The problem is most serious in analyses with relatively small samplesizes,alargepercentofobservationsthataretreated,andsettingswithsubstantialtreatment effectheterogeneity(Bakeretal.,2022). Bycontrast,oursamplesizeisrelativelylargeandonlya smallshareofobservationsaretreated.13 Treatmenteffectheterogeneityisexpectedinoursetting, perhaps due to litigated banks operating in communities that vary substantially demographically or due to changes in how banks have responded over the three decades of our sample. Nevertheless, our use of LitWindow interacted with PostLit in estimating β is consistent with the mt 1 13Inouranalysis,only39outof13,075lenders(0.29%)aretreated. Andforourspilloveranalysis,only8,138out of254,703uniquelender-CBSApairings(0.01%ofoursample)aretreatedthroughexposuretoalawsuitfiledagainst alocalcompetitor. 25

suggestions in An et al. (2022), Baker et al. (2022), and Cengiz et al. (2019) in further mitigating theproblemsassociatedwithstaggereddifference-in-differencesresearchdesigns. Specifically,by narrowing our estimated treatment effects to a four-year post-treatment window—measured relativetoafour-yearpre-treatmentperiodamongthelitigatedbanks—ourmodelamountstoanevent studywithrelative-timeindicatorvariablesandacontrolgroupthatisnottaintedbythetreatment ofsomelendersbeforethetreatmentofothers(Bakeretal. 2022). One notable shortcoming in our approach is the absence of applicant-level credit score data. As Bhutta et al. (2022) show, a substantial contributor to observed racial differences in mortgage applicationdenialratesandinterestratesisaverageracialdifferencesinmetricsofcreditworthiness, suchascreditscores. Accordingly,ourcoefficientestimatesonvariablesandinteractionsinvolving racemaybebiasedtotheextentthatracemightserveasapartialproxyforcreditscore. Whileapotentialworry,severalfeaturesofourdesignmitigatethispotentiality. First,because β estimatesthechangeinBlackapplicantdenialratesattreatedlenders,frombeforetoafterlitiga- 1 tionrelativetothechangeamongWhiteapplicantsandchangesatnon-litigatedbanks,thisrelative estimate is unlikely to be biased by average racial differences in credit scores —at least in comparison to estimates of absolute differences in treatment by race that do not include a variable for credit score. This point is accentuated by the fact that there are no observed, meaningful changes in average racial differences in credit scores over time (Ballew and Pears, 2023). Second, to the extent that we find pre-treatment racial disparities at litigated banks above and beyond marketwidedisparities,thepre-treatmentdisparitiesaremorelikelytobetheproductofidiosyncrasies— idiosyncrasies that might include unlawful discrimination—at litigated banks rather than average racialdifferencesincreditscores. Third,ourfindingsofpre-treatmentracialdisparitiesareconsistentwiththeallegationsleveledinoursample’sfairlendingactions. Governmentagenciestendto haveinternalbankdata,includingafulllistofvariablesrelatedtoapplicantcreditworthiness,when conductingfairlendingexaminations,investigations,andenforcement. Andthecomplaintsinour samplelitigationactionstypicallyallegetheexistenceofracialdisparitiesevenaftercontrollingfor creditscore. Thus,ourfindingsofpre-treatmentdisparitiesconformwiththeconclusionsreached 26

by government investigations (which then lead to enforcement actions). Fourth, as discussed already, securitizing and selling loans to GSEs can serve as a proxy for applicant creditworthiness. Consistent with An et al. (2022), we make use of such securitization data to plausibly argue that ourfindingspertaintoraceratherthanlegitimatemeasuresofcreditworthiness. We also investigate potential spillover effects of litigation on local competitors of litigated banks. Thus,forbanksnotsubjecttolitigation,weestimatethebelowspecification: Y = γ +γ LitExposure XPostLit XBlack +γ LitExposure XBlack ibmt 0 1 mt mt i 2 mt i +γ Black +γ LnIncome +γ LnLoanSize 3 i 4 i 5 i +γ LMI +γ Male +γ SameSex +γ JointApplicants 6 c 7 i 8 i 9 i +δ +ε . (2) bmt ibmt LitExposureindicatesbanksoperatinginthesamecityasalitigatedbankwithinthe+/-fouryear windowofthelitigationdate. Thecoefficientγ measuresanydifferenceindisparityinthedenial 2 rate for Black applicants relative to White applicants versus non-exposed banks pre-litigation and γ tests for any spillover impact post-litigation. We include a bank by city by year fixed effect so 1 that any spillover effect is measured as a change in the denial rate disparity between White and Blackapplicantswithineachbankinagivencityandyear. 5 Results 5.1 Main Results Table4containstheresultsfromestimatingspecification(1)—lookingathowfairlendinglitigationactionsimpactindividuallevelmortgageapplicationdenialdecisions. Column(1)suggests that,onaverage,aBlackborrowerapplyingtoanon-litigatedbankwillbedeniedata9percentage point higher rate than a White borrower with the same income applying for the same sized loan to the same bank. However, if the same pair of borrowers applied to a litigated bank in the years 27

before litigation, the gap would be about 12 percentage points. This is roughly double the 14% overall denial rate in the sample. Interestingly, litigated banks also deny White applications at a slightlyhigherrate(0.7percentagepoint)thandotheircompetitorsinthesamecity. Followinglitigation,however,litigatedbankssubstantiallyreduceddenialratesforBlackborrowers, relative to White borrowers. A Black and a White applicant to a litigated bank, with the sameincomeandloansizesought,willhavethegapintheirdenialratesreducedby3.9percentage points post-litigation, relative to pre-litigation—in fact, by more than the initial denial rate disparity at litigated versus non-litigated banks (3 percentage points). The origination rate increases symmetrically with the decrease in the denial rate. However, at 3.2 percentage points, it does not increase as much as the denial rate decreases, which is explained by the fact that the rate at which mortgageapplicationsareacceptedbutnotoriginated(column(2))increasesafterlitigation. While it is not clear what drives this result, it may be a product of some litigated banks offering loans to Blackborrowersonunfavorableterms—whichapplicantsthendeclinetoaccept. Nevertheless,the second row of column (2) demonstrates that the rate at which mortgage applications are accepted butnotoriginatedforallapplicantsdecreasesby2.8percentagepoints,thuspartiallycounteracting the0.7percentagepointincreaseforBlackapplicants. Table 12 in Appendix 8.2 estimates specification (1) with an added control for the natural log of lender assets. As the Table demonstrates, our coefficients of interest do not meaningfully or qualitativelychange—eitherinsign,statisticalsignificance,ormagnitude—afteraddingthelender assetscontrol.14 InFigure1,welookathowtheBlack-Whitedenialratedisparityevolvesyearbyyearrelative to non-litigated banks. Four years before litigation is filed, the Black-White denial rate disparity is a statistically significant 2 percentage points higher than at non-litigated banks. In years three and two before litigation filing, there is no observed disparity and then a statistically significant 14This control comes from merging our data set with lender asset data from FR Y-9C reports (which report on domesticbankholdingcompanies),FFIECCallReports(banks),NCUACallReports(creditunions)andFRY-9CSP reports(smalldomesticholdingcompanies). Becauselenderassetdataareunavailableforsavingsandloanholding companiesregulatedbytheOfficeofThriftSupervisionbeforetheimplementationofDodd-Frank, aswellasother mortgagelendersthatarenotbanks,creditunions,orbankholdingcompanies,weloseaboutonequarterofoursample afteraddingthiscontrol. 28

Table4: FullLitigationSample-DirectEffectsonBlackApplicants (1) (2) (3) denied not_origin originate ∗∗∗ ∗ ∗∗∗ Post-LitXBlack -0.0387 0.0069 0.0318 (0.008) (0.004) (0.007) ∗ ∗∗∗ ∗∗∗ Post-Lit 0.0038 -0.0282 0.0243 (0.002) (0.002) (0.002) ∗∗∗ ∗∗∗ Pre-LitXBlack 0.0313 -0.0017 -0.0296 (0.006) (0.003) (0.005) ∗∗∗ ∗∗∗ ∗∗∗ Pre-Lit 0.0066 -0.0144 0.0078 (0.002) (0.001) (0.002) ∗∗∗ ∗∗∗ ∗∗∗ Black 0.0900 0.0024 -0.0924 (0.001) (0.000) (0.001) ∗∗∗ ∗∗∗ ∗∗∗ 1.male 0.0065 0.0009 -0.0074 (0.000) (0.000) (0.000) ∗∗∗ ∗∗∗ ∗∗∗ 1.lgbt 0.0331 -0.0060 -0.0271 (0.001) (0.001) (0.001) ∗∗∗ ∗∗∗ ∗∗∗ 1.joint -0.0135 -0.0049 0.0184 (0.000) (0.000) (0.000) ∗∗∗ ∗∗∗ ∗∗∗ 1.lmi 0.0318 0.0043 -0.0361 (0.000) (0.000) (0.001) ∗∗∗ ∗∗∗ ∗∗∗ ln_inc -0.0472 0.0105 0.0366 (0.000) (0.000) (0.000) ∗∗∗ ∗∗∗ ∗∗∗ ln_loan -0.0047 -0.0067 0.0113 (0.000) (0.000) (0.000) R2 0.204 0.074 0.242 N 8733267 8733267 8733267 Note: Pre-andpost-litigationeffectsestimatedwithinawindowof+/-fouryears. Theyearoflitigationisexcluded fromthesample. AllspecificationsincludebankandMSA-yearfixedeffects. Standarderrorsclusteredatthe bank-MSA-yearlevelinparentheses. ***p<0.01,**p<0.05,*p<0.1. smaller disparity (by 2 percentage points) at litigated banks compared with non-litigated banks. Theyearbeforelitigationfiling,thedisparityrisestoastatisticallysignificant4percentagepoints. Thisdisparitythenbeginsdecliningtheyearlitigationisfiled,andcontinuestodosothroughyear 29

plus four. In year two post-litigation, at litigated banks the Black-White denial rate disparity is a statistically significant 4 percentage points below the disparity at non-litigated banks. This result persists through years 3 and 4 post-litigation. Figure 1 plots analogous by-year results, expressed relativetothepre-litigationdisparityatlitigatedbanks. Figure1: Black-WhiteDenialDisparityOverTime 10 8 6 4 2 0 -2 -4 -6 -8 -10 ).p.p( ytirapsiD etaR laineD -4 -3 -2 -1 0 1 2 3 4 Year Relative to Litigation Filing Denial Rate Disparity vs Competitors 95% CI Note: EstimateddenialratedisparitiesbetweenBlackandWhiteapplicantsovertimeatbankssubjecttolitigation. Estimateddenialdisparitiesonlyincludeby-yeardisparitiesatsuchbanks,withtheyearthatlitigationwasfiled definedas0. Thefiguredoesnotincludeoverallracialdisparitiesatsuchbanks,nordoesitincludeindustry-wide denialdisparities. In Figure 3, we assess the individual impact of litigation on denial rate disparities for twentyfive separate litigation actions.15 Following twenty-one of these twenty-five litigation actions, the 15Wedonotincluderesultsforallthirty-eightactionsforseveralreasons. First,wecannotestimatepost-treatment effectsfortheseven“pretreatment”actions. SeeTable2. Second,fiveofoursamplelitigationsareallagainstWells Fargo(onebroughtbytheU.S.governmentandfourbymunicipalgovernments),andwecanonlyestimatetheindi- 30

Figure2: Black-WhiteDenialDisparityOverTime,RelativetoDisparityPre-Litigation 4 3 2 1 0 -1 -2 -3 -4 -5 -6 -7 -8 ).p.p( ytirapsiD etaR laineD 0 1 2 3 4 Year Relative to Litigation Filing Denial Rate Disparity vs Competitors relative to Preperiod 95% CI Note: EstimateddenialratedisparitiesbetweenBlackandWhiteapplicantsovertimeatbankssubjecttolitigation, relativetothepre-litigationdisparityatlitigatedbanks. Black-Whitedenialratedisparitydecreases,relativetobeforelitigation,consistentwithourresults pooling all litigation actions together. For these actions, the disparity decreases by a range of approximately2to22percentagepoints. Thedisparityincreasesfollowingfourofthesetwenty-fiveactions,butisstatisticallysignificant for only two of these four. For the actions that increase the denial rate disparity, the disparity increases by a range of 1 percentage point to approximately 58 percentage points in the case of US v. First United Security Bank (2009) (which we exclude from Figure 3 given how much of an vidualeffectsofonelawsuitagainstabank. Thus,weonlymeasuretheeffectsofUSv. WellsFargobutnottheeffects ofthemunicipalactionsagainstthebank. Third,intworemainingactions,thesamplesizeistoosmalltoestimatean individualeffect. 31

outlierresultitis). Itisunclearwhyexactlythedisparityincreasesafteranyoftheseactions. One possibleexplanationisthat,post-litigation,creditworthyBlackborrowerslookelsewhereforcredit, meaningthatonlythosemostlikelytobedeniedanywhereapplytosuchbanks,therebydrivingup theestimateddenialratedisparity. IntheredliningcaseofUSv. FirstUnitedSecurityBank(2009), for example, there are notably only 17 applications from Black applicants in HMDA from the regionwherethebankwasallegedlyredliningwithinthe4-yearpost-litigationperiod. Itisentirely possiblethatthissmallsampleofapplicantshaveunfavorableobjectivemetricsofcreditworthiness, thereforeexplainingtheincreaseindenialratesatthisinstitutionfollowinglitigation. Furtherstudyiswarrantedtoassessspecificreasonswhysomelitigationactionsdecreaselendingdiscriminationmorethanothersdo. Onepossibilityisthatdiscriminationdecreasesmorewhen themagnitudeofpre-litigationdiscriminationwasgreater. Asecondpossibility,whichcouldwork intandemwiththefirst,isthatthelitigationstrategiesemployed,orremedieswon,affectthemagnitudeofthedecreaseindiscrimination. 32

Figure3: Post-litigationChangeinBlackApplicantDenialRateAcrossLitigatedBanks .1 0 -.1 -.2 -.3 -.4 Case specific estimates 95% CIs Combined estimate Note: TheYaxisreportsthechangeintheBlack-Whitedisparityindenialrates(measuredinpercentagepoints), relativetopre-litigation. Markerssizedbylender’spost-litigationloanvolume. Notpictured: thepost-litigationtrend atFirstUnitedSecurityBankfollowingUSv. FirstUnitedSecurityBank(2009). Inthatlitigation,whichweexclude heregivenhowmuchofanoutlierresultitis,theBlack-Whitedenialdisparityincreasedbyastatisticallysignificant 58percentagepointsafterlitigation. 33

5.2 Allegation Type Next, we examine how the effects of litigation change based on the alleged type of discrimination. Tables 5, 6, and 7 present results of regressions estimating how the Black-White disparity changes after litigation challenging alleged discriminatory pricing and origination, redlining, and reverseredlining,respectively. Column (1) of Table 5 indicates that the Black-White denial rate disparity decreases after litigationallegingdiscriminatorypricingandorigination,consistentwithouroverallresults. Thestatisticallysignificantdecreaseeliminatesthepre-litigationdisparity,relativetonon-litigatedbanks. Pre-litigation, banks sued for discriminatory origination and pricing deny Black applicants at a 3.2 percentage point higher rate than they deny White applicants, relative to the disparity at nonlitigated banks. Post-litigation, however, such banks reduce the disparity by a statistically significant3.7percentagepoints. Bysymmetry,column(3)indicatesthattheoriginationrateincreases,althoughnotasmuchas the denial rate decreases, a product of the slight (and not statistically significant) uptick in Black applicationsthatareacceptedbutnotoriginated,asindicatedincolumn(2). AsdiscussedinParts 2.2 and 5.1, to the extent that the rate at which mortgage applications are accepted but not originated might serve as a proxy for the favorability of terms offered to borrowers,16 we would have expected a negative coefficient on this variable for discriminatory pricing litigation. While that doesnotoccur,thesecondrowofcolumn(2)demonstratesthatforallapplicants,therateatwhich mortgage applications are accepted but not originated decreases by 3 percentage points, thus partiallycounteractingthe0.5percentagepointincreaseforBlackapplicants. Table6showsthatpre-litigation,bankssuedforredliningactuallydenyapplicationsfromBlack applicants at a weakly significant 1.5 percentage point lower Black-White applicant disparity relativetothedisparityatnon-litigatedbanks. Thisresultissurprising,althoughitmaybeconsistent 16Oneavenuebywhichtheimpactoflitigationoncombatingdiscriminationcouldbeunderminedwouldinvolve a litigated bank approving more Black loan applications but offering unfavorable interest rates to such applicants, relativetowhattheircompetitorsoffer. Ifabankwantstodecreaseobserveddisparitiesindenialrates,butnonetheless does not want to originate loans for Black borrowers, it could offer loans with uncompetitive interest rates, leading applicantstooptagainstloanorigination. 34

with pre-litigation Black applicants to redlining banks being from majority-White neighborhoods. Post-litigation,theBlack-Whitedenialratedisparitydecreasesby1percentagepointandtheoriginationrateincreasesby0.8percentagepoint,althoughneitherisstatisticallysignificant. Inturn,Table7presentsresultsfromreverseredlininglitigation. Severalofthereverseredlining actions we sampled did not culminate in favorable settlements for the plaintiff. Nevertheless, the results are broadly consistent with our hypotheses in Part 2.2. With reverse redlining, banks specifically target borrowers from minority neighborhoods, offering them credit at supracompetitiveinterestrates. Althoughnotstatisticallysignificant,thepost-litigationBlack-Whitedenialrate disparityincreasesby2.7percentagepointsandthedisparityamongapplicationsthatareaccepted but not originated decreases by 1.3 percentage points, consistent with lenders becoming more discerningaboutapplicantcreditworthiness,ceasingtotargetBlackborrowerswithhigh-pricedmortgages, but nevertheless offering approved applicants more favorable loan terms. The disparity in origination rates widens by 1.4 percentage points (not statistically significant), as expected. Notably,amongallapplicantstherateatwhichapplicationsareacceptedbutnotoriginateddecreases by a statistically significant 3.5 percentage points, explaining the entirety of the statistically significant 3.5 percentage point increase in origination rates. This is consistent with banks providing loans on better terms to all applicants after reverse-redlining litigation. Notably, lawsuits against one large lender make up the majority of our reverse redlining cases, and it was a defendant in several concurrent actions while being sued for reverse redlining. We must therefore be cautious thattheresultsofTable7arenotbiasedbythoseotheractions. Ourinclusionofbankfixedeffects andcity-yearfixedeffectsshouldneverthelesscontrolinpartforthispotentialconfoundingfactor. 35

Table5: DiscriminatoryPricingandOriginationLitigationEffects (1) (2) (3) denied not_origin originate Post-LitXBlack -0.0374*** 0.0054 0.0320*** (0.008) (0.004) (0.008) Post-Lit 0.0043* -0.0300*** 0.0257*** (0.002) (0.002) (0.002) Pre-LitXBlack 0.0320*** -0.0016 -0.0304*** (0.006) (0.003) (0.005) Pre-Lit 0.0068*** -0.0150*** 0.0082*** (0.002) (0.001) (0.002) Black 0.0899*** 0.0024*** -0.0924*** (0.001) (0.000) (0.001) 1.male 0.0065*** 0.0009*** -0.0074*** (0.000) (0.000) (0.000) 1.lgbt 0.0331*** -0.0060*** -0.0271*** (0.001) (0.001) (0.001) 1.joint -0.0135*** -0.0049*** 0.0184*** (0.000) (0.000) (0.000) 1.lmi 0.0318*** 0.0043*** -0.0361*** (0.000) (0.000) (0.001) ln_inc -0.0472*** 0.0106*** 0.0366*** (0.000) (0.000) (0.000) ln_loan -0.0047*** -0.0067*** 0.0114*** (0.000) (0.000) (0.000) R2 0.204 0.074 0.243 N 8627367 8627367 8627367 Note: Pre-andpost-litigationeffectsestimatedwithinawindowof+/-fouryears. Theyearoflitigationisexcluded fromthesample. AllspecificationsincludebankandMSA-yearfixedeffects. Standarderrorsclusteredatthe bank-MSA-yearlevelinparentheses. ***p<0.01,**p<0.05,*p<0.1. 36

Table6: RedliningLitigationEffects (1) (2) (3) denied not_origin originate Post-LitXBlack -0.0095 0.0017 0.0078 (0.018) (0.007) (0.020) Post-Lit -0.0010 -0.0070** 0.0079 (0.004) (0.003) (0.006) Pre-LitXBlack -0.0150* 0.0068 0.0082 (0.009) (0.005) (0.009) Pre-Lit -0.0076** 0.0082*** -0.0006 (0.003) (0.002) (0.005) Black -0.0216 0.0176 0.0040 (0.015) (0.013) (0.020) 1.w -0.0001 0.0046 -0.0045 (0.008) (0.006) (0.008) Black 0.0916*** 0.0032*** -0.0948*** (0.001) (0.000) (0.001) 1.male 0.0064*** 0.0009*** -0.0073*** (0.000) (0.000) (0.000) 1.lgbt 0.0336*** -0.0055*** -0.0281*** (0.001) (0.001) (0.001) 1.joint -0.0147*** -0.0052*** 0.0199*** (0.000) (0.000) (0.000) 1.lmi 0.0324*** 0.0046*** -0.0370*** (0.000) (0.000) (0.001) ln_inc -0.0455*** 0.0107*** 0.0348*** (0.000) (0.000) (0.000) ln_loan -0.0043*** -0.0069*** 0.0111*** (0.000) (0.000) (0.000) R2 0.196 0.071 0.232 N 8952436 8952436 8952436 Note: Pre-andpost-litigationeffectsestimatedwithinawindowof+/-fouryears. Theyearoflitigationisexcluded fromthesample. AllspecificationsincludebankandMSA-yearfixedeffects. Standarderrorsclusteredatthe bank-MSA-yearlevelinparentheses. ***p<0.01,**p<0.05,*p<0.1. 37

Table7: ReverseRedliningLitigationEffects (1) (2) (3) denied not_origin originate Post-LitXBlack 0.0271 -0.0130 -0.0141 (0.037) (0.021) (0.043) Post-Lit -0.0004 -0.0345*** 0.0349*** (0.011) (0.009) (0.012) Pre-LitXBlack -0.0127 0.0197 -0.0070 (0.015) (0.016) (0.023) Pre-Lit 0.0087 0.0037 -0.0124 (0.009) (0.007) (0.009) 1.w1.Black -0.0306* 0.0069 0.0237 (0.017) (0.006) (0.015) 1.w -0.0153*** 0.0098*** 0.0055 (0.005) (0.003) (0.005) Black 0.0916*** 0.0032*** -0.0948*** (0.001) (0.000) (0.001) 1.male 0.0064*** 0.0009*** -0.0073*** (0.000) (0.000) (0.000) 1.lgbt 0.0336*** -0.0055*** -0.0281*** (0.001) (0.001) (0.001) 1.joint -0.0147*** -0.0052*** 0.0199*** (0.000) (0.000) (0.000) 1.lmi 0.0324*** 0.0046*** -0.0370*** (0.000) (0.000) (0.001) ln_inc -0.0455*** 0.0107*** 0.0348*** (0.000) (0.000) (0.000) ln_loan -0.0043*** -0.0069*** 0.0111*** (0.000) (0.000) (0.000) R2 0.196 0.071 0.232 N 8952436 8952436 8952436 Note: Pre-andpost-litigationeffectsestimatedwithinawindowof+/-fouryears. Theyearoflitigationisexcluded fromthesample. AllspecificationsincludebankandMSA-yearfixedeffects. Standarderrorsclusteredatthe bank-MSA-yearlevelinparentheses. ***p<0.01,**p<0.05,*p<0.1. 38

5.3 Trends in Securitization Rates Table8estimatesspecification(1)withtheoutcomevariablesofwhether,afterbeingoriginated, loans are securitized and sold, and if so whether they are sold to a GSE or a private investor. Prelitigation, the Black-White disparity in securitizing and selling loans to Black borrowers among litigated banks is a 3.7 percentage point higher rate than pre-litigation. On net, this is the result of greatersalesratestoprivateinvestors(column(3))asopposedtoGSEs(column(2)). AsAnetal. (2022)argue,theseresultsareconsistentwithlitigatedbanksoriginatingloanstoBlackborrowers withrelativelylowermetricsofcreditworthiness. Crucially,thedisparityincreasesbyaninsignificant1.9percentagepointsafterlitigation. This resultcomesfroma10percentagepointincreaseinloanstoBlackborrowersthataresoldtoGSEs, and an 8 percentage point decrease in loans sold to private investors. Because the sale of a loan to GSEs can serve as a loose proxy for creditworthiness (An et al. (2022)), these results provide contextual evidence that overall, denial rates decrease and origination rates increase among Black borrowers, but that that result comes from more credit extended tocreditworthy borrowers. These results therefore also provide contextual evidence that, prior to litigation, litigated banks were not offeringloanstocreditworthyborrowers,eitherbecauseofovertbiasandprincipal-agentfailuresor because of statistical discrimination against Black borrowers. By contrast, post-litigation, lenders implementednewanti-biasmeasuresand/orfoundwaystobetterassessapplicantcreditworthiness without regard to race, explaining the combination of higher Black origination rates and higher (lower)ratesofsecuritizingandsellingloanstoGSEs(non-GSEs). 39

Table8: OverallDirectEffectsonSecuritizationandBlackBorrowers (1) (2) (3) secur secur_gse secur_non_gse Post-LitXBlack 0.0186 0.0983*** -0.0797*** (0.012) (0.013) (0.011) Post-Lit -0.0110** -0.1191*** 0.1081*** (0.005) (0.006) (0.007) Pre-LitXBlack 0.0368*** -0.0197*** 0.0566*** (0.006) (0.007) (0.009) Pre-Lit 0.0748*** 0.1164*** -0.0416*** (0.004) (0.004) (0.005) Black -0.0460*** -0.0502*** 0.0042*** (0.001) (0.001) (0.001) 1.male -0.0013*** -0.0004 -0.0009*** (0.000) (0.000) (0.000) 1.lgbt -0.0085*** 0.0044*** -0.0129*** (0.001) (0.001) (0.001) 1.joint 0.0201*** 0.0261*** -0.0060*** (0.000) (0.000) (0.000) 1.lmi -0.0247*** -0.0245*** -0.0002 (0.001) (0.001) (0.001) ln_inc -0.0547*** -0.0626*** 0.0078*** (0.001) (0.001) (0.001) ln_loan 0.0274*** 0.0101*** 0.0172*** (0.001) (0.001) (0.001) R2 0.287 0.343 0.378 N 7476455 7476455 7476455 Note: Pre-andpost-litigationeffectsestimatedwithinawindowof+/-fouryears. Theyearoflitigationisexcluded fromthesample. AllspecificationsincludebankandMSA-yearfixedeffects. Standarderrorsclusteredatthe bank-MSA-yearlevelinparentheses. ***p<0.01,**p<0.05,*p<0.1. 40

5.4 Spillovers In Table 9 we show results from estimation of specification 2, looking for spillover effects of litigation. The idea is that a lawsuit against a bank’scompetitor in a certain location may raise the salience to the bank of the risk of facing liability from practices deemed discriminatory. We test this hypothesis by examining how banks in particular MSAs respond after a competitor in their MSAistargetedwithMSA-specificlitigation. Thereisevidencethatlitigationhasspillovereffectsonnon-litigatedbanks. Aftergeographicspecific litigation of their local competitors, non-litigated banks reduced denial rates for Black applicantsby1percentagepointsrelativetoWhiteapplicantsatthesamebankata10percentlevel of statistical significance. Banks also increased relative origination rates for Black borrowers by 1.5 percentage points at a 5 percent level of statistical significance. This suggests that raising the salienceofthethreatoflitigationhasanon-trivialimpactonlenderbehavior. 41

Table9: SpilloverEffectsofLitigationonExposedBanks (1) (2) (3) denied not_origin originate ∗ ∗∗ Post-LitXExposedXBlack -0.0104 -0.0042 0.0146 (0.006) (0.004) (0.007) ∗∗ ∗∗∗ ExposedXBlack -0.0096 0.0109 -0.0013 (0.004) (0.003) (0.005) ∗∗∗ ∗∗∗ 1.Black 0.0850 0.0006 -0.0857 (0.001) (0.000) (0.001) ∗∗∗ ∗∗∗ ∗∗∗ 1.male 0.0067 0.0011 -0.0078 (0.000) (0.000) (0.000) ∗∗∗ ∗∗∗ ∗∗∗ 1.lgbt 0.0319 -0.0056 -0.0263 (0.001) (0.001) (0.001) ∗∗∗ ∗∗∗ ∗∗∗ 1.joint -0.0125 -0.0048 0.0173 (0.000) (0.000) (0.000) ∗∗∗ ∗∗∗ ∗∗∗ 1.lmi 0.0306 0.0045 -0.0351 (0.000) (0.000) (0.001) ∗∗∗ ∗∗∗ ∗∗∗ ln_inc -0.0451 0.0112 0.0339 (0.000) (0.000) (0.000) ∗∗∗ ∗∗∗ ln_loan 0.0033 -0.0040 0.0006 (0.000) (0.000) (0.001) R2 0.291 0.161 0.325 N 8383114 8383114 8383114 Note: Pre-andpost-litigationeffectsestimatedwithinawindowof+/-fouryears. Theyearoflitigationisexcluded fromthesample. Allspecificationsincludebank-MSA-yearfixedeffects. Standarderrorsclusteredatthe bank-MSA-yearlevelinparentheses. ***p<0.01,**p<0.05,*p<0.1. 42

5.5 Robustness Checks Our results demonstrate a consistent, strong link between fair lending litigation and reduced disparitiesinoutcomesforBlackversusWhiteapplicants. Weneverthelessconsidertwopotential alternative explanations—other than fair lending litigation—for the observed reductions in disparities: (1) Community Reinvestment Act agreements (“CRA agreements”) and (2) enforcement decisionsandorders(“EDOs”). CRAagreements(sometimesreferredtoas“CommunityBenefits Agreements”) are bank commitments to extend a certain amount of credit to minority borrowers and communities; Bostic and Robinson (2003) find a significant, positive association between the number of newly–initiated CRA agreements in a county and increased CRA, minority, and overall conventional mortgage lending. Given this association, our results could be biased upward if enoughofoursamplelitigationoccursaroundthesametimeasdefendantsenterCRAagreements. In turn, EDOs are public enforcement actions issued by regulatory supervisors requiring banks to cease an activity or promptly remedy a deficiency.17 Similar to Bostic and Robinson (2003)’s results on the effects of CRAs, An et al. (2022) find a significant association between EDOs and increased lending to minority borrowers. Accordingly, our results could also be biased upward if enoughlitigationinoursamplecoincideswithEDOissuances. We control for these alternative causal pathways by identifying CRA agreements entered and EDOs issued within the +/- 4 year window of the filing of litigation and then re-running specification (1)—first excluding defendant banks that enter CRA agreements within the time window, andsecondexcludingdefendantbanksthatfaceEDOswithinthewindow. Wereporttheresultsin tablesthat,apartfromtheseexclusions,areidenticaltoTable4. To conduct these analyses, we collect data on CRA agreements and EDOs that occur during our sample period. Our CRA data come from the National Community Reinvestment Coalition 17Unlike litigation in court, which is the subject of our study, EDOs are administrative actions commenced by financial regulators. While not always the case, regulators frequently issue EDOs in conjunction with enforcement actionsbroughtincourtbytheDOJorCFPB;insuchcases,theEDOandlawsuitpertaintothesameunderlyingset ofdeficiencies. ThereareneverthelessmanyotherinstanceswherealitigationdoesnothaveanaccompanyingEDO oranEDOdoesnothaveanassociatedlitigation. 43

(“NCRC”),whichhasrecordsofCRAagreementsfrombeforethe1990sthrough2025.18 Wehand matchCRAagreementsenteredbybanksinourdatabaseoflitigationactionsiftheagreementoccurswithinthe+/-4yearobservationwindowaroundthelitigationdate. WecollectdataonEDOs fromthethreefederalbankingregulatorsandhandmatchthemtoourdatabaseiftheyoccurwithin the +/- 4 year window around the litigation date (Prior to its closure in 2011 the Office of Thrift Supervision (OTS) was the primary federal regulator for some lenders and their EDO records are maintained by the OCC).19 We only hand match an EDO to a litigation if the EDO concerns institutional compliance with bank regulations (e.g., EDOs pertaining to deficiencies in safety and soundness,riskmanagement,mortgageloanservicingandforeclosures,consumerprotectionpractices,andfairlending)—weexcludeEDOsthatpertaintominor,one-offissuessuchasmisconduct by an individual employee. Appendix 8.3 outlines banks in our litigation sample that enter CRA agreements within the litigation window, as well as banks that are subject to EDOs within the litigationwindow. Table10reportsourresultsafterexcludinglitigationagainstbanksthatenterCRAagreements within the litigation window. As with our main specification in Table 4, there is a negative, statistically significant coefficient on the “Post-Lit X Black” variable, meaning that the Black-White disparityindenialratesnarrowsby3.8percentagepointsfollowinglitigation. Likewise,theBlack- Whitedisparityinoriginationratesnarrowsby3.2percentagepointsafterlitigation. AswithTable 4, the disparity in the rate that applications are accepted but not originated increases—here, by 0.6 percentage points, although that number is not statistically significant. The coefficients on the “Post-Lit”variablealsoalignwiththeresultsinTable4: thereisapositive,statisticallysignificant (at the 10% level) coefficient of 0.38 percentage points when application denial is the outcome variable, a negative, statistically significant (at the 1% level) coefficient of 3.0 percentage points for applications that are accepted but not originated, and a positive, statistically significant (at the 18Bostic and Robinson (2003) and Bostic and Robinson (2004) use data from the NCRC in their analyses of the linkbetweenCRAagreementsandminorityaccesstocredit. 19The EDOs from each respective agency are available at https://orders.fdic.gov/s/ (FDIC), https://www.occ.treas.gov/topics/laws-and-regulations/enforcement-actions/index-enforcement-actions.html (OCC), https://www.federalreserve.gov/supervisionreg/enforcementactions.htm (FRS), and https://www.occ.treas.gov/newsevents/newsroom/news-issuances-by-year/ots-issuances/index-ots-issuances.html(OTS). 44

1%level)coefficientofpercentagepointsfororiginationrates. Table 11 reports our results after excluding banks in our litigation sample that are subject to EDOswithinthelitigationwindow. Theresultsofthisrobustnesscheckmateriallydifferfromthe results in Tables 4 and 10. After excluding banks that are also subject to EDOs, the Black-White disparity in denial rates no longer narrows post-litigation and in fact increases by 0.8 percentage points, although this figure is not statistically significant. By contrast, as with Table 4, the gap in origination rates narrows by 1 percentage point (although this figure is also not statistically significant). In turn, the rate that applications from Black borrowers are accepted but not originated decreases, with the Black-White disparity narrowing by 1.8 percentage points. This figure is statisticallysignificantatthe1%level. Whilewenolongerseeapost-litigationnarrowingintheBlack-Whitedisparityindenialrates after excluding banks subject to EDOs, fair lending litigation appears to effectively reduce denial rates and increase origination rates for all applicants in this subsample. As Column 1 of Row 2 of Table 11 indicates, relative to before litigation, after being sued banks subject to litigation but not EDOs reduce denial rates for all applicants by 5.1 percentage points. Likewise, banks in this subsample increase origination rates by 4 percentage points post-litigation, relative to before litigation. Both figures are statistically significant at the 1% level. Notably, there is an overall increase in the rate that applications are accepted but not originated: 1.1 percentage points, which also is statistically significant at the 1% level. Because the Black-White disparity for this figure narrows by 1.8 percentage points, however, accepted-but-not-originated rates decrease on net for Blackapplicants(byabout0.7percentagepoint). It is not straightforwardly clear why the Black-White disparity in denial and origination rates ceases to decrease when we remove banks subject to EDOs. One possibility is selection bias. It is possiblethat(1)regulatorsonlyissueEDOsassociatedwithlitigationwhenthepre-litigationracial disparity is particularly pronounced, and (2) a larger disparity has more room to be narrowed— whether because of litigation, an EDO, or the combination of the two. If these conditions obtain, thenthelitigationweexcludedhappenstohavethehighestlikelihoodofeffectivelyreducingdenial 45

rates and increasing origination rates. Our results offer some evidence that is consistent with this hypothesis. Unlikewithouroverallsample,wheretheBlack-Whitedenialratedisparityatlitigated banks is a statistically significant 3 percentage points higher (relative to non-litigated banks) prelitigation20,inthissubsamplethepre-litigationdisparityisastatisticallysignificant2.8percentage points lower than at non-litigated banks. The same goes for origination rates. In the overall sample, the pre-litigation Black-White denial rate disparity is a statistically significant 3.0 percentage pointshigheratlitigatedbanks,relativetonon-litigatedbanks.21 Bycontrast,inthissubsamplethe pre-litigation disparity is 0.3 percentage point narrower at litigated banks, relative to non-litigated banks. ExcludinglitigationthathasaccompanyingEDOsmaysimplyexcludeoursamplelitigation thatisbestpositionedtoreducediscriminationalongthemetricswestudy. A second possibility is that litigation is most effective when paired with EDOs. While we do not rule out this hypothesis, exploring the nuanced intersection of litigation and EDOs is outside thescopeofthisstudy. AthirdpossibilityisthateliminatinglitigationwithassociatedEDOscompromisesstatisticalpowerbyremovingsufficientlymanytreatmentobservationsfromoursample. AsrecordedinTable14inAppendix8.3,14banksinoursampleweresubjecttoEDOswithinthe litigationwindow,meaningthatweloseasubstantialshareofourlitigationsampleafterexcluding theseenforcementactions. (Bycomparison,onlysixactionsinourlitigationsampleinvolvedbanks thatenteredCRAswithinthelitigationwindow,meaningthatfarfeweractionswereexcludedwhen conductingtherobustnesscheckforCRAs.) 20SeeColumn1,Row3ofTable4. 21SeeColumn3,Row3ofTable4. 46

Table10: CRARobustnessCheck (1) (2) (3) denied not_origin originate Post-LitXBlack -0.0379*** 0.0061 0.0318*** (0.008) (0.004) (0.008) Post-Lit 0.0038* -0.0295*** 0.0256*** (0.002) (0.002) (0.002) Pre-LitXBlack 0.0320*** -0.0017 -0.0303*** (0.006) (0.003) (0.005) Pre-Lit 0.0074*** -0.0149*** 0.0075*** (0.002) (0.001) (0.002) Black 0.0899*** 0.0024*** -0.0924*** (0.001) (0.000) (0.001) 1.male 0.0065*** 0.0009*** -0.0074*** (0.000) (0.000) (0.000) 1.lgbt 0.0331*** -0.0060*** -0.0271*** (0.001) (0.001) (0.001) 1.joint -0.0135*** -0.0049*** 0.0184*** (0.000) (0.000) (0.000) 1.lmi 0.0318*** 0.0043*** -0.0361*** (0.000) (0.000) (0.001) ln_inc -0.0472*** 0.0106*** 0.0367*** (0.000) (0.000) (0.000) ln_loan -0.0047*** -0.0067*** 0.0114*** (0.000) (0.000) (0.000) R2 0.204 0.074 0.242 N 8647555 8647555 8647555 Note: Pre-andpost-litigationeffectsestimatedwithinawindowof+/-fouryears. Theyearoflitigationisexcluded fromthesample. Allspecificationsincludebankfixedeffectsandcounty-yearfixedeffects. Standarderrors clusteredatthebank-MSA-yearlevelinparentheses. ***p<0.01,**p<0.05,*p<0.1. 47

Table11: EDORobustnessCheck (1) (2) (3) denied not_origin originate Post-LitXBlack 0.0079 -0.0179*** 0.0100 (0.010) (0.004) (0.010) Post-Lit -0.0507*** 0.0107*** 0.0400*** (0.004) (0.002) (0.004) Pre-LitXBlack -0.0282*** 0.0254*** 0.0029 (0.009) (0.003) (0.009) Pre-Lit 0.0051 -0.0412*** 0.0360*** (0.003) (0.002) (0.003) Black 0.0900*** 0.0021*** -0.0921*** (0.001) (0.000) (0.001) 1.male 0.0064*** 0.0009*** -0.0073*** (0.000) (0.000) (0.000) 1.lgbt 0.0326*** -0.0062*** -0.0264*** (0.001) (0.001) (0.001) 1.joint -0.0129*** -0.0048*** 0.0178*** (0.000) (0.000) (0.000) 1.lmi 0.0312*** 0.0043*** -0.0355*** (0.000) (0.000) (0.001) ln_inc -0.0471*** 0.0105*** 0.0367*** (0.000) (0.000) (0.000) ln_loan -0.0055*** -0.0068*** 0.0123*** (0.000) (0.000) (0.000) R2 0.210 0.076 0.249 N 8469160 8469160 8469160 Note: Pre-andpost-litigationeffectsestimatedwithinawindowof+/-fouryears. Theyearoflitigationisexcluded fromthesample. Allspecificationsincludebankfixedeffectsandcounty-yearfixedeffects. Standarderrors clusteredatthebank-MSA-yearlevelinparentheses. ***p<0.01,**p<0.05,*p<0.1. 48

6 Conclusion Webeganthisprojectbyquestioningtheefficacyoffairlendingenforcementactions. Wefind that, in the wake of legal settlements for discrimination against Black borrowers, lenders significantly reduced denial rates for Black applicants. The reductions offset pre-litigation racial disparities in denial rates by litigated banks, relative to those banks’ competitors. Origination rates for Blackapplicantsalsoincreasedpost-litigation. Wefurtherobserveevidenceofaspillovereffecton theapprovaldecisionsofnon-litigatedbanksoperatinginthesamecityasalitigatedbank. Inconclusion, the evidence suggests that enforcement of fair lending laws is an effective tool to reduce racialdiscriminationincreditmarkets. 7 References Aaronson, Daniel, Daniel Hartley, and Bhashkar Mazumder. 2021. The Effects of the 1930s HOLC‘Redlining’Maps. AmericanEconomicJournal: EconomicPolicy. Aaronson, Daniel, Jacob Faber, Daniel Hartley, Bhashkar Mazumder, and Patrick Sharkey. 2021. The Long-Run Effects of the 1930s HOLC “Redlining” Maps on Place-Based Measures of EconomicOpportunityandSocioeconomicSuccess. RegionalScienceandUrbanEconomics. An,Byeongchan,RobertBushman,AnyaKleymenova,andRimmyE.Tomy. 2022. SocialExternalitiesofBankEnforcementActions: TheCaseofMinorityLending. FederalReserveFinance andEconomicsDiscussionSeries. Appel,IanandJordanNickerson. 2016. PocketsofPoverty: TheLong-TermEffectsofRedlining. SSRNWorkingPaper. Arrow, Kenneth. 1973. “The Theory of Discrimination.” Edited by Orley Ashenfelter and AlbertRees. DiscriminationinLaborMarkets(PrincetonUniversityPress). Baker, Andrew C., David F. Larcker, and Charles C.Y. Wang. 2022. How Much Should We TrustStaggeredDifference-in-DifferencesEstimates? JournalofFinancialEconomics. Ballew, Hailey and Geoffrey Pears. 2023. Banks’ Commitment to Social Responsibility in 49

LendingAmidReducedEnforcementofFairLendingLaws. SSRNWorkingPaper. Baradaran,Mehrsa. 2019. JimCrowCredit. UCIrvineLawReview. Barr, Michael. 2005. Credit Where It Counts: The Community Reinvestment Act and Its Critics. N.Y.U.LawReview. Becker,GaryS.1957. “TheEconomicsofDiscrimination.” UniversityofChicagoPress. Becker, Gary S. 1968. Crime and Punishment: An Economic Approach. The Economic DimensionsofCrime(Springer). Bhutta, Neil, Aurel Hizmo, and Daniel Ringo. 2022. “How Much Does Racial Bias Affect MortgageLending? EvidencefromHumanandAlgorithmicCreditDecisions?” FinanceandEconomicsDiscussionSeries(FEDS)(FederalReserveSystem). Bostic, Raphael W. and Breck L. Robinson. 2003. Do CRA Agreements Influence Lending Patterns? RealEstateEconomics,31,23-51. Bostic, Raphael W. and Breck L. Robinson. 2004. The Impact of CRA Agreements on CommunityBanks. JournalofBankingandFinance,28,3069-95. Butler, Alexander W., Erik J. Mayer, and James Weston. 2022. Racial Disparities in the Auto LoanMarket. SSRNWorkingPaper. Cengiz, D., Dube, A., Lindner, A., and Zipperer, B. 2019. The Effect of Minimum Wages on Low-WageJobs. TheQuarterlyJournalofEconomics,134,1405–1454. Chetty, Raj, Nathaniel Hendren, and Lawrence F. Katz. 2016. The Effects of Exposure to Better Neighborhoods on Children: New Evidence from the Moving to Opportunity Experiment. AmericanEconomicReview. Dobbie, Will, Andres Liberman, Daniel Paravisini, and Vikram Pathania. 2021. Measuring BiasinConsumerLending. TheReviewofEconomicStudies. Faber, Jacob William. 2018. Segregation and the Geography of Creditworthiness: Racial InequalityinaRecoveredMortgageMarket. HousingPolicyDebate. Horowitz, Ben. 2018. Fair Lending Laws and the CRA: Complementary Tools for Increasing EquitableAccesstoCredit. FederalReserveBankofMinneapolisBlogPost. 50

Li, Nicholas. 2022. Racial Sorting, Restricted Choices, and the Origins of Residential SegregationinU.S.Cities. WorkingPaper. Nash, Andrew. 2008. The Origins of Fair Lending Litigation. Washington U. School of Law WorkingPaper. Rohner,Ralph. 1978. EqualCreditOpportunityAct. BusinessLaw. Rothstein, Richard. 2017. The Color of Law: A Forgotten History of How Our Government SegregatedAmerica. Liveright. Rouse, Cecilia, Jared Bernstein, Helen Knudsen, and Jeffery Zhang. 2021. Exclusionary Zoning: ItsEffectonRacialDiscriminationintheHousingMarket. CEABlogPost. Stiglitz, Joseph E. and Andrew Weiss. 1981. “Credit Rationing in Markets with Imperfect Information.” AmericanEconomicReview. Sun, L., Abraham, S. 2021. Estimating Dynamic Treatment Effects in Event Studies with HeterogeneousTreatmentEffects. JournalofEconometrics. Zhang, David Hao and Paul S. Willen. 2021. Do Lenders Still Discriminate? A Robust ApproachforAssessingDifferencesinMenus. NBERWorkingPaper. 8 Appendix 8.1 Cases Reviewed But Excluded from Our Sample 51

Figure4: CasesReviewedButExcludedfromOurSample Case Reason for Exclusion US v. First National Bank of Gordon (1996) Discrimination in consumer loans; no observations in HMDA US v. AIG Federal Savings Bank (2010) Mortgage discrimination but low obs in HMDA US v. Pacific Mercantile Bank (2018) Mortgage discrimination but low obs in HMDA US v. Sage Bank (2015) Mortgage discrimination but low obs in HMDA US v. Community State Bank (2013) Mortgage discrimination but low obs in HMDA US v. Southport Bank (2013) Mortgage discrimination but low obs in HMDA US v. Luther Burbank Savings Bank (2012) Mortgage discrimination but low obs in HMDA US v. Albank Federal Savings Bank (1997) Mortgage discrimination but low obs in HMDA US v. KleinBank (2017) Mortgage discrimination but low obs in HMDA CFPB and US v. National City Bank (2013) Mortgage discrimination but acquired by PNC in 2008 US v. Texas Champion Bank (2013) Unsecured loans US v. Countrywide Financial Corporation (2011) Mortgage discrimination but acquired by BAML in 2008 US v. Chevy Chase Bank (2013) Mortgage discrimination but acquired by Capital One in 2009 Ramirez v. Greenpoint Mortgage Funding, Inc. Mortgage discrimination but acquired by Capital One 2 yrs before (2008) suit Zamora v. Wachovia (2007) Mortgage discrimination but class cert denied Mortgage discrimination but class cert denied after appellate Miller v. Countrywide (2007) litigation National Community Reinvestment Coalition v. Mortgage discrimination but defendant filed for bankruptcy, so no Novastar Financial (2007) obs in HMDA; also concurrent w Jackson v. Novastar JAT Inc. v. Nat'l City Bank of the Midwest (2006) Business loans Boykin v. Bank of America Corporation (2004) Mortgage discrimination but private non-class suit Hargraves v. Capital City Mortgage Corporation (1998) Mortgage discrimination but private non-class suit Edwards v. Flagstar Bank (1995) Mortgage discrimination but private non-class suit Doane v. National Westminster Bank (1995) Mortgage discrimination but private non-class suit Latimore v. Citibank Fed. Sav. Bank (1995) Mortgage discrimination but private non-class suit Barrett v. H&R Block, Inc. (2008) Mortgage discrimination but class cert denied on appeal US v. Deposit Guaranty National Bank Home improvement loans US v. Old Kent Financial Corporation (2004) Commercial loans; Bank Acquired by Fifth Third Bank, May 2001 US v. First National Bank of Dona Ana County (1997) Mobile home mortgages loans Lawsuit over disparate impact from activity in the secondary Adkins v. Morgan Stanley (2012) mortgage market City of Pittsburgh Commission on Human Relations v. Key Bank USA Home improvement loans Powell v. American General Financial, Inc. Mortgage discrimination but private non-class suit Cooley v. Sterling Bank Private non-class suit over unsecured lines of credit Dumas v. Sentinel Mortgage Corporation Private non-class suit that didn't center mortgage loans Hood v. Midwest Savings Bank Private non-class suit that didn't center mortgage loans 52

Case Reason for Exclusion Church of Zion Christian Center v. Southtrust Bank Private non-class suit that didn't center mortgage loans Housing Opportunities Made Equal a/k/a HOME v. Nationwide Insurance Insurance redlining, not mortgage redlining Sallion v. Suntrust Banks, Inc. Mortgage discrimination but class cert denied on appeal Milton v. Bancplus Mortgage Corporation Mortgage discrimination but class cert denied on appeal US v. Auto Fare (2014) Auto lending market, not mortgage US v. Synchrony Bank f/k/a GE Capital Retail Bank (2014) Credit card lending US v. Fifth Third Mortgage (2014) disability discrimination US v. First United Bank (2014) consumer loans, not mortgages US v. Sallie Mae (2014) pre service loans, not mortgages US v. Santander (2014) unlawful reposession of automobiles US v. American Honda Finance Corp (2015) Auto lending US v. Fifth Third Bank (2015) Auto Lending US v. Toyota Motor Credit Corp (2015) Auto lending US v. Evergreen Bank Group (2015) motorcycle lending US v. Evolve Bank & Trust (2015) disability discrimination US v. Charter Bank (2016) vehicle secured consumer loans US v. The Home Loan Auditors (2016) mortgage loan modification US v. First Federal Bank of Flordia (2016) discriminiation on basis of familial status(maternity) US v. Hatfield (2017) sexual harrassment US v. COPOCO Community Credit Union (2017) illegal auto reposession US v CitiFinancial Credit Co (2017) illegal auto reposession US v. Westlake Services (2017) illegal auto reposession US v. Wells Fargo Bank, N.A., d/b/a Wells Fargo Dealer Services (2017) illegal auto reposession US v. Northwest Trustee Services (2017) illegal home foreclosure US v. Advocate Law Groups of Flordia (2018) mortgage loan modification US v. BMW Financial Services (2018) Motor vehicle lease terminations US v. Northwest Trustee Services (2018) Home forclosures US v. Hudson Valley Federal Credit Union (2018) illegal auto reposession US v. California Auto Finance (2018) illegal auto reposession US v. Guaranteed Auto Sales (2019) race discrimination on auto loans US v. Nissan Motor Acceptance Corp (2019) Auto reposessions US v. PHH Mortgage Corp (2019) home forclosures US v. Bank of America (2020) discrimination on basis of disability US v. Conn Credit (2021) overcharged interest rates 53

Case Reason for Exclusion US v. New Jersey Higher Education Student Assistance Authority (2021) Student loan case US v. Ally Financial Inc and Ally Bank (2013) auto lending US v. Nara Bank (2013) auto lending US v. Bank of America (2012) discrimination on basis of disability US v. Union Auto Sales (2012) Discrimination in auto lending US et al., v. Bank of America Corp., et al. (2012) Not discrimination on basis of race US v. Capital One (2012) not a mortgage case US v. Nixon State Bank (2011) consumer loans US v. Mortgage Guaranty Insurance Corp (2011) sex based discrimination US v BAC Home Loans Servicing (2011) wrongful forclosure US v Saxon Mortgage Services (2011) wrongful forclosure US v. Nationwide Nevada (2008) refused to purchase automobile contracts based on ethnicity US v. Compass Bank (2007) Discrimination on basis of marital status US v. Springfield Ford (2007) Discrimination on car loan rates US v. Pacifico Ford (2007) Discrimination on car loan rates US v. First Nat'l Bank of Pontotoc (2007) sexual harrassment of female borrowers US v. Fifth Third Bank (2004) discrimination on basis of race for business loans US v. Fidelity Federal Bank (2002) credit card program discrimination US v. Associates National Bank (2001) credit card program discrimination US v. Deposit Garanty National Bank (1999) Home improvement loan discrimination Cason v. Nissan Motor Acceptance Corp (2000) higher finance charges for African Americans at Nissan Dealerships Louisiana ACORN Fair Housing v. LeBlanc (1998) rent discrimination Regional Economic Community Action Program, Inc. v. City of Middletown (1998) discrimination against alcoholics US v. Big D Enterprises (1998) Discrimination against African American renters City refused to permit development of affordable, owner occupied, US v. City of Lake Station (1998) single family homes US v. City of Toledo (1998) discrimination on basis of disability US v. Crawford (1998) sexual harrassment US v. Damron (1998) refused to rent to African Americans US v. Gardner (1998) discrimination against children US v. Inland Empire Builders (1998) discrimination on basis of disability US v. Krueger (1998) sexual and racial harassment of black tenants US v. Lexington Village Apartments and Hillcrest Village (1998) defendant not a mortgage originator US v. Nejam Properties (1998) defendant not a mortgage originator US v. Richmond (1998) defendant not a mortgage originator 54

Case Reason for Exclusion US v. Vernon (1998) defendant not a mortgage originator US v. Village of Addison (1998) defendant not a mortgage originator US v. Housing Authority of the Town of Milford (1997) defendant not a mortgage originator US v. Las Vegas Jaycees Senior Citizens Mobile Home Community (1997) defendant not a mortgage originator US v. Nationwide Mutual Insurance (1997) Discrimination in home insurance US v. Rock Springs Vista Development Corp (1997) defendant not a mortgage originator US v. Town of Cicero (1997) defendant not a mortgage originator US v. Williams (1997) defendant not a mortgage originator US v. Associates National Bank (1997) discrimination in credit card applications US v. Big D Enterprises (1997) defendant not a mortgage originator US v. Choice Property Consultants (1997) defendant not a mortgage originator US v. City of Milwaukee (1997) defendant not a mortgage originator US v. Hagadone (1997) defendant not a mortgage originator US v. Harlan (1997) defendant not a mortgage originator US v. JDL Management (1997) defendant not a mortgage originator US v. Cedar Builders (1996) defendant not a mortgage originator US v. City of Waukegan (1996) defendant not a mortgage originator US v. Village of Hatch (1996) defendant not a mortgage originator US v. American Family Mutual Insurance (1995) defendant not a mortgage originator US v. City of Pharma (1995) defendant not a mortgage originator US v. Pinewood Associates (1995) defendant not a mortgage originator US v. Secutiry State Bank (1995) discrimination in consumer loans US v. Veal (1994) defendant not a mortgage originator US v. Blackpipe State Bank (1994) not discrimination on mortgages US v. Chevy Chase Bank (1994) not discrimination on mortgages US v. First National Bank of Vicksburg (1994) Home improvement loan discrimination US v. Jacksonville Housing Authority and City of Jacksonville (1994) defendant not a mortgage originator US v. Nedialkov (1994) defendant not a mortgage originator US v. Flagstar Corporation and Denny's (1993) not discrimination on mortgages US v. Plaza Mobile Estates (1991) defendant not a mortgage originator 55

8.2 Total Asset Covariates Table12: OverallResultswithBankAssetControls (1) (2) (3) denied not_origin originate Post-LitXBlack -0.0432∗∗∗ 0.0080∗∗ 0.0352∗∗∗ (0.008) (0.004) (0.008) Post-Lit 0.0032 -0.0247∗∗∗ 0.0215∗∗∗ (0.002) (0.002) (0.002) Pre-LitXBlack 0.0192∗∗∗ -0.0057∗∗ -0.0135∗∗ (0.006) (0.003) (0.005) Pre-Lit 0.0056∗∗∗ -0.0149∗∗∗ 0.0092∗∗∗ (0.002) (0.001) (0.002) Black 0.1045∗∗∗ 0.0047∗∗∗ -0.1092∗∗∗ (0.001) (0.001) (0.001) 1.male 0.0077∗∗∗ 0.0006∗∗ -0.0083∗∗∗ (0.000) (0.000) (0.000) 1.lgbt 0.0389∗∗∗ -0.0053∗∗∗ -0.0335∗∗∗ (0.001) (0.001) (0.001) 1.joint -0.0132∗∗∗ -0.0046∗∗∗ 0.0178∗∗∗ (0.000) (0.000) (0.000) 1.lmi 0.0366∗∗∗ 0.0043∗∗∗ -0.0408∗∗∗ (0.001) (0.000) (0.001) ln_inc -0.0494∗∗∗ 0.0098∗∗∗ 0.0396∗∗∗ (0.000) (0.000) (0.000) ln_loan -0.0110∗∗∗ -0.0072∗∗∗ 0.0183∗∗∗ (0.000) (0.000) (0.001) ln_total_assets 0.0058∗∗∗ -0.0084∗∗∗ 0.0026∗∗∗ (0.000) (0.000) (0.000) R2 0.169 0.051 0.183 N 6135311 6135311 6135311 Standarderrorsinparentheses ∗p<0.10,∗∗p<0.05,∗∗∗p<0.01 Note: Pre-andpost-litigationeffectsestimatedwithinawindowof+/-fouryears. Theyearoflitigationisexcluded fromthesample. AllspecificationsincludebankandMSA-yearfixedeffects. Standarderrorsclusteredatthe bank-MSA-yearlevelinparentheses. ***p<0.01,**p<0.05,*p<0.1. 56

8.3 CRA and EDO Robustness Check Table13: BanksThatEnterCRAAgreementsWithinLitigationWindow Case Allegation Outcome Locationof Discrimination USv. Huntington Discriminatory $420,000 Cleveland MortgageCo. Pricing compensatory MSA (N.D.Ohio1995) fund USv. MidAmerica Redlining $11.25mremediation Chicago Bank(N.D.Ill. 2002) fund MSA USv. Fleet Discriminatory $4m Westbury, MortgageCorp. Pricing compensatory& NY& (E.D.N.Y.1996) remediation Woodbridge, fund NJ CityofL.A.v. Reverse Outcomeunclear; L.A.,CA Citigroup,Inc. Redlining Initialmotionto (C.D.Cal. 2013) dismissdenied USv. Union Redlining $9mremediation Cincinatti, SavingsBank fund Dayton,& (S.D.Ohio2016) Columbus,OH, &Indianapolis, IN,MSAs USv. First Redlining $1.7mremediation Indianapolis MerchantsBank fund MSA (S.D.Ind. 2019) 57

Table14: BanksSubjecttoEDOsWithinLitigationWindow Case Allegation Outcome Locationof Discrimination Payaresv. J.P.Morgan Discriminatory $300/classmember; National Chase(C.D.Cal. 2007) Pricing $1.965min (privateclassaction) attorneys’fees USv. J.P.Morgan Discriminatory $53mcompensatory National Chase(S.D.N.Y2017) Pricing fund USv. WellsFargo Discriminatory $175mcompensatory National (D.D.C.2012) Pricing fund USv. Suntrust Discriminatory $21mcompensatory National Mortgage Pricing fund (E.D.Va. 2012) Puellov. Discriminatory $200/classmember; National CitifinancialServices, Pricing $400,000in Inc. (D.Mass. 2008) attorneys’fees (privateclassaction) CityofMiamiv. Reverse Voluntarilydismissed Miami,FL BankofAmerica Redlining afterprotracted (S.D.Fla. 2013) appellatelitigation overstanding CFPBv. BancorpSouth Redlining $10.8mcompensatory Memphis Bank(N.D.Miss. 2016) &remediationfund MSA CityofOaklandv. Reverse MotiontoDismiss Oakland,CA WellsFargo Redlining grantedafter (N.D.Cal. 2015) appeal CityofL.A.v. Reverse MotionforSummary L.A.,CA WellsFargo Redlining Judgmentagainst (C.D.Cal. 2013) Citygranted CityofMemphisv. Reverse $7.5mremediation Shelby WellsFargo Redlining fund;Coordinated Cnty.,TN (W.D.Tenn. 2010) withUSv. WellsFargo settlement MayorofBaltimore Reverse $7.5mremediation Baltimore v. WellsFargo Redlining fund;Coordinated MSA (D.Md. 2008) withUSv. WellsFargo settlement USv. FirstAmerican Redlining $5.7mremediation Chicago& Bank(N.D.Ill. 2004) fund Kankakee, IL,MSAs CityofL.A.v. Reverse Outcomeunclear; L.A.,CA Citigroup,Inc. Redlining Initialmotionto (C.D.Cal. 2013) dismissdenied †USv. Citizens Redlining $3.63mcompensatory Detroit RepublicBankcorp., &remediationfund MSA Inc. (E.D.Mich. 2011) †Indicates“pretreatment”cases. 58