Foreclosing on Opportunity: State Laws and Mortgage Credit

Foreclosing on Opportunity: State Laws and Mortgage Credit KarenM.Pence BoardofGovernorsoftheFederalReserve System Karen.Pence@FRB.GOV (202)452-2342 May13,2003 Abstract Foreclosurelawsgovern therightsofborrowersandlenderswhenborrowers defaultonmortgages. Manystatesprotectborrowersbyimposingrestrictionson theforeclosureprocess; theserestrictions,inturn,imposelargecostsonlenders. Lendersmayrespondtothesehighercostsbyreducingloansupply;borrowersmay respondtotheprotectionsimbeddedintheselawsbydemandinglargermortgages. Iexamineempiricallytheeffectofthelawsonequilibriumloansize.Iexploit therichgeographic information availableinthe1994 and 1995 HomeMortgage Disclosure Act data to compare mortgage applications for properties located in censustractsthatbordereachother,yetarelocatedindifferentstates.Usingsemiparametricestimationmethods,Ifindthatdefaulter-friendlyforeclosurelawsare correlatedwithafourpercenttosixpercentdecreaseinloansize.Thisresultsuggeststhatdefaulter-friendlyforeclosurelawsimposecostsonborrowersatthetime ofloanorigination. IamgratefultoJohnKarlScholz,YuichiKitamura,andBobHavemanfortheirguidancewiththisprojectandtoDavidBrown,MaryDiCarlantonio,GregGrothe,Michele Rambo,KeikoSnell,andJamesTedrickforoutstandingresearchassistance. Forhelpfulcomments,IthankthefacultyandstudentsoftheUniversityofWisconsin,myFederalReserveBoardcolleagues,ChuckCapone,andseminarparticipantsattheUniversitiesofMaryland,California-Berkeley,Texas-Austin,TexasA&M,WesternOntario, and North Carolina-Greensboro, as well as the participants from Wharton, Freddie Mac,theSanFranciscoFed,theTreasuryDepartment,andtheCongressionalBudget Office. IreceivedgenerousfinancialsupportfromaU.S.DepartmentofHousingand Urban Development doctoral dissertation grant, the Christensen Award in Empirical Economics,andtheSocialScienceResearchCouncilPrograminAppliedEconomics, fundedby the MacArthurFoundation. This research does not necessarily reflect the views of the Board of Governorsof the Federal Reserve System, its members, or its staff; the U.S. Department of Housing and Urban Development; the Social Science ResearchCouncil;ortheMacArthurFoundation.

1 Introduction Whenaborrowerdefaultsonahomemortgage,thelendermayattempttorecoverits losses by repossessing and selling the property. However, estimated losses on these foreclosuresrangefrom30percentto60percentoftheoutstandingloanbalancesbecauseoflegalfees,foregoneinterest, andpropertyexpenses.1 Stateforeclosurelaws also affect these losses: foreclosures in some states are quick, low-cost procedures thatprovidescantprotectionsforborrowers,whilelawsinotherstatesconfersubstantial benefits on borrowersandcorrespondinglylargecosts on lenders. Laws in these “borrower-friendly”statesareintendedtoprotecthomeownersindistress. However,if lenderspassthehigherassociatedcosts ontoborrowers,thelawsmayhavetheunintendedconsequenceofreducingthesupplyofmortgagecredit. In this paper, I use semiparametric methods to examine the effect of foreclosure lawsonthesize ofapprovedmortgageloans. Thiseffectis, a priori,ambiguous. As mentionedabove,lendersmayrespondto thehigherexpensesassociatedwith costly foreclosurelawsbycharginghigherinterestrates,requiringlargerdownpayments,or both.Jones(1993),forexample,documentsthatlendersinAlberta,Canada,increased downpayment requirements after suffering large default rates linked to foreclosure laws. However,defaulter-friendlyforeclosurelawsprovideborrowerswithwealthinsuranceagainstfallinghouseprices.Ifrisk-averseborrowersvaluethisinsurancemore thanitscost,mortgagedemandmayincrease. It is difficult to identify the effects of foreclosure laws on the mortgage market becausebothlaws andreal estate marketsexhibitstrongregionalpatterns. One type ofpropertylaw governingforeclosuresis prevalentonthe East Coast, anotherin the Midwest,andyetanotherontheWestCoast. Realestatemarketsalsovarystrikingly acrosstheUnitedStates. In1990,forexample,realhousepricesincreased17percent in Seattle, Washington, and 15 percent in Aurora, Illinois, while they decreased 11 1SeeCapone(1996).ClauretieandHerzog(1990)andCiochetti(1997)alsofindlossratesinthisrange; theNationalHomeEquityMortgageAssociationestimatesforeclosurelossesat50centsonthedollar. 1

percentinNewHaven,Connecticut,andCharleston,WestVirginia.2 Inasimplecrosssectionalregression,aregionalshocktothehousingmarketcouldbemisinterpretedas aneffectoftheforeclosurelaws. I control for this identification problem by comparing approved home mortgage applicationsincensustractsthataregeographicallyneareachotherbutarelocatedin differentstates. The proximity of the houses suggests that they may take on similar valuesforunobservedcharacteristicsthat mightotherwisebias theestimation. However,differentforeclosurelawsgovernthemortgages. StudiessuchasHolmes(1998) andBlack(1999)haveusedasimilar“borders”identificationstrategytoestimatethe effect of state right-to-worklaws on the location of manufacturing and the effect of schoolqualityonhouseprices,respectively.Icarryoutthisstrategyusingthemillions ofgeocodedloanapplicationscollectedin1994and1995undertheHomeMortgage DisclosureAct. Tocapturelocalvariationinrealestatemarketsinasflexibleamanneraspossible,I implementasemiparametricestimatorthatallowsunobservedcharacteristicstotakeon adifferentvalueateachcensustract.Aslongasthesecharacteristicschangesmoothly overspace,whilethelawschangediscontinuouslyatthestateborder,theeffectofthe lawsisidentified. Althoughacensus-tractfixedeffectsmodelalsoallowsunobserved characteristics to vary by census tract, it fails in this application because the census tracts fixed effects are collinear with the state laws. This semiparametric estimation strategyfits withinthe regressiondiscontinuityframeworkdiscussed byHahn, Todd, andVanderKlaauw(2001)andPorter(2002). After controlling for these geographically varying factors, I find that loan sizes arefourtosixpercentsmallerinstateswithdefaulter-friendlyforeclosurelaws. This finding is robust to several changes in specification. In contrast, a specification that ignores these regional factors indicates that foreclosurelaws do not affect loan size, suggestingthatstudiesthatomitthesefactorsmayyieldmisleadingresults. 2SeePoterba(1991). 2

Thedecreaseinloansizesuggeststhatlendersrespondtocostlyforeclosurelaws by reducing loan supply. Smaller loan sizes may also reflect, in part, an effect of the laws on house prices: buyers may not be willing to pay as much for a house if theyhavedifficultyobtainingfinancing. AlthoughIcontrolforhousepriceswiththe semiparametricestimation techniqueandwith census tract housingcharacteristics, it ispossiblethatthecoefficientcapturestheeffectofthelawsonbothloansizeandon houseprices.3 These findings suggest that policymakersface a tradeoff. They can facilitate the availability of low-priced mortgage credit, or they can provide protections to homeownerswho experiencefinancialdifficulties, but theycannotdo both. While several recent papers have explored a similar tradeoff engendered by bankruptcylaw,4 very fewhaveconsideredtheroleofforeclosurelaw,despitetheprimaryrolethathousing playsinmosthouseholdportfolios.Thisresearchbeginstofillthatgap. 2 Costs and Benefits of Foreclosure Laws Previouspapershavefocusedonthreeareasofpropertylawthataffectforeclosures:judicialforeclosureprocesses,statutoryrightsofredemption,anddeficiencyjudgments. In this paper, I examine the effects of all three, although I argue below that a link betweenjuducialforeclosureprocessesandlendercostsismostplausible. 1. Twenty-onestates, as shown in figure 1, require a judicial foreclosure process inwhichthelendermustproceedthroughthecourtstoforecloseonaproperty. Thesestatesareconcentratedinthenortheasternandmidwesternregionsofthe United States, although Florida, South Carolina, Louisiana, and New Mexico, amongotherstates, alsorequiretheprocedure. Inallotherstates, lendershave theoptionofusingasimpler,quicker,andcheapernonjudicialprocedurecalled 3HMDAdoesnotincludethevalueofthepropertyunderlyingtheloanorotherloantermssuchasthe interestrate. 4See,forexample,Gropp,Scholz,andWhite(1997). 3

powerofsale,inwhichatrusteeoverseesthesaleoftheproperty, 2. Afterthecompletionoftheforeclosuresale,thehomeownercanstillregainthe propertyin thenine states that permita statutory right ofredemption. Upto a yearafterthesale,dependingonthestate,homeownerscanredeemtheirpropertyfortheforeclosuresalepriceplusforeclosureexpenses. Insomestates,the homeownerretainsownershipduringthisperiod.5 Statutoryrightsofredemption areprevalentinfarmingstates, wherecropsmayfail oneyearandsucceedthe next.Asshowninfigure2,moststatutoryright-of-redemptionstatesarelocated intheGreatPlainsregion. Statelegislaturesinstalledtheprovisioninresponse toborrowerdemandsformoreprotectionduring19th-centurydepressions.6 3. Moststatesallowcreditorstocollectadeficiencyjudgmentequaltothelender’s foreclosurelossesagainsttheborrower’sotherassets.Althoughdeficiencyjudgments are not often pursued, the threat of a deficiency judgment can be used to obtainconcessionsfromtheborrower. As shownin figure3, ninestates located in the western half of the United States forbid deficiency judgments for thetypicalhomemortgagedefaultcase.7 Theseprohibitionswerearesponseto perceivedlendingabusesduringtheGreatDepression,whenlenderswouldpurchaseforeclosedpropertiesatpricesfarbelowtheloanbalanceandthenobtain adeficiencyjudgmentagainsttheborrower’sotherassets.8 Foreclosurelawscanaffectcostsbyimposingtransactioncosts;byprolongingthe length of time in which lenders forgo interest on the loan and incur carrying costs 5Asdiscussedinthedataappendixintable1,somestatesallowastatutoryrightofredemptionifthe lenderfollowsajudicialforeclosureprocessbutnototherwise.Icodeastateasallowingastatutoryrightof redemptionifitisavailableundertheprocedurealendergenerallyfollows. 6SeeCapone (1996), p.126. Theprinciple underlying thestatutory rightofredemption iseven more ancient,datingbacktoancientHebrewLaw.SeeLev.25:25-31. 7Nostateforbidsdeficiencyjudgmentsinallcases. California, likemoststatesthatrestrictdeficiency judgments,prohibitsthemonlyforowner-occupied,one-tofour-familyhomes.Aswiththestatutoryright ofredemption,Icodeastateasforbiddingadeficiencyjudgmentifitisunavailableundertheforeclosure processalendergenerallyfollows. 8SeeCapone(1996),p.134. 4

such as taxes and maintenance; and by shifting the relative bargainingpower of the lenderandborrower.Evenwhenlenderspursuealternativestoforeclosure,foreclosure laws may affect the outcome. Lenders may be more willing to grant concessions to defaultersinstateswithcostlylawsbecausepursuingaforeclosureissoexpensive. Thelengthoftheforeclosureprocessappearstobeanespeciallykeyfactor. IllustrativecalculationsbyCapone(1996),forexample,suggestthatdelayingtheforeclosureprocessona$100,000loanbysixteenmonthsincreasedcostsbyover$13,500in 1996.9 Anotherindicatorofthevalueofalengthyforeclosureprocessisthepresence of“equityskimmers,”whobuypropertiesfromdefaultingborrowersandthenrentout the property while manipulating the legal system to extend the process as much as possible.10 Judicialproceduresaresubstantiallymoretimeconsumingthanpower-of-saleprocedures.Wood(1997)findsthatjudicialforeclosures,onaverage,take148dayslonger thannonjudicialforeclosures,whileFreddieMac’s guidelinesformortgageservicers indicatethatforeclosuresinthemosttime-consumingstate,Maine(ajudicialforeclosurestate),takealmost300dayslongerthaninthequickeststate,Texas(apower-ofsalestate).11 Judicialprocessesalsoimposemoretransactioncostsandarethoughtto introduce more uncertainty into the foreclosure proceedings than power-of-sale processes. Fromtheborrower’sperspective,thedelaysassociatedwithajudicialforeclosureprovidea lengthyperiodoffreerent. Inaddition,judicialforeclosureprocesses providesafeguardsagainstlenderexcesses.12 Thelinkbetweenajudicialforeclosureprocessandlendercostshasbeennotedin boththepopularpress(Fleishman(2002))andinmoreformalstudies. Asfarbackas theGreatDepression,judicialforeclosurescosttheHomeOwners’LoanCorporation 9Seep.43. 10SeePolk(1988). 11Estimates taken from Jankowski (1999), p. 2-11. These estimates assume no complications in the foreclosureprocess;filingforbankruptcy,forexample,canaddanadditionalsixmonths. 12In2000,inresponsetocomplaintsaboutpredatorylendingintheDistrictofColumbia,housingactivists pressedforchangesincitylawthatwouldallowborrowerstocontestpredatoryforeclosuresincourt. 5

twiceasmuchaspower-of-saleforeclosures.13 Morerecently,Touche,RossandCo. (1975), Clauretie (1989), Clauretie and Herzog (1990), Ciochetti (1997), and Wood (1997)havealso documenteda relationshipbetweenjudicial foreclosuresand lender costs. Statutoryrightsofredemptionanddeficiencyjudgmentsalsoaffectlenders’costs. Statutoryrightsofredemptionprolongtheforeclosureprocessandmaydepressprices at the foreclosure sale, since the new owner cannot obtain a clear title for the property. Deficiencyjudgmentsaffectrelativebargainingpower: thethreatofadeficiency judgmentcanbeusedtoextractconcessionsfromborrowers. However,inpracticeborrowersrarelyexercisestatutoryrightsofredemptionand lendersrarelypursuedeficiencyjudgments.Borrowersgenerallydefaultwhenthesize oftheloanexceedsthevalueoftheproperty;sincethisconditionis usuallystill true a yearafter the foreclosuresale, theyhave no incentiveto redeemthe property. Deficiency judgments are rarely profitable because most borrowers in foreclosure have veryfew resources. Capone (1996)notes that deficiency judgmentsare “rarely used in practice” and are generallypursuedonly against “investors, repeat defaulters, and nonhardshipcases.”14 By contrast, judicialforeclosuresaffect everyforeclosurein a statethatrequirestheprocess. Researchershavenotbeenabletodocumentaconsistentlinkbetweentheselaws and lender costs. Clauretie (1989) and Clauretie and Herzog (1990) find lower loss ratesinstatesthatprohibitstatutoryrightsofredemption.WhenClauretieandHerzog (1990)excludeCalifornia from the specification, they find higherloss rates in states that prohibitdeficiency judgments, but when they include California they find lower lossrates.15 Jones(1993)andAmbrose,Capone,andDeng(2001)findhigherdefault rates in areas that prohibit deficiency judgments. However, Wood (1997) finds that 13Author’scalculationfromBridewell(1938),p.555. 14Seep.135. 15ClauretieandHerzog(1990)attributetheseresultstotheriseinpropertyvaluesinCaliforniaoverthe sampleperiod,whichpresumablyreducedloanlosses. 6

FannieMae’slossesarehigherinstatesthatprohibitstatutoryrightsofredemptionand allowdeficiencyjudgments. Tocomplicatethesituationfurther,almosteverystate’slawisidiosyncraticregarding statutory rights of redemptionand deficiency judgments. It is easy to code state law with respect to judicial foreclosure processes, but not with respect to these two measures. Four papersin this literature code deficiencyjudgmentsdifferently,16 and evensomelawyersspecializinginforeclosurelawsareunawareofwhetherastatutory rightofredemptionoradeficiencyjudgmentispermittedintheirstate.17 Thecoding decisionsmadeinthispaperaredescribedinthedataappendixintable1. 3 The Effect of Foreclosure Law on Mortgage Supply and Demand Asdescribedinmoredetailbelow,ifbothlendersandborrowerstakeforeclosurelaws intoaccountwhenmakingtheirdecisions,theeffectofforeclosurelawsonequilibrium loansizeisambiguous. Lenders. Supposemanyidenticallenderscompetebyofferingmenusofloancontracts, indexedby loan size and interest rate. Borrowers pick the utility-maximizing contractfromthismenuandlendersmakezeroprofits. Defaulter-friendlyforeclosurelawsaffectthelender’szero-profitcurveintwoways: theyincreasethelender’slossesiftheborrowerdefaultsandtheyincreasetheprobabilityofdefaultitself.Toseethisinasimplemodel,assumethatiftheborrowerrepays theloan,thelenderreceivestheinterestrate (cid:0) (cid:1) (cid:2) (cid:0) (cid:3) (cid:4) (cid:1) timestheloanamount (cid:2) .18 Iftheborrowerdefaults,thelendertakesbackthehouse,valuedatthedownpayment 16SeeClauretie(1989),LinandWhite(2001),Wood(1997),andthispaper. 17Inanattempttoclarifysomestates’laws,Icalledlocalrealestatelawfirmsthatspecializedinforeclosurelaw. 18SeeLehnertandPence(2001)forageneralequilibriummodeloftheeffectofdefaulter-friendlyforeclosurelawsonthesupplyofmortgagecredit. SeeAmbrose,Buttimer, andCapone(1997)foraformal mortgagepricingmodelinwhichdefaulter-friendlylawsincreasetheprobabilityofdefault. 7

(cid:3) plustheloanamount (cid:2) timesanystochasticchange (cid:0) (cid:1) (cid:2) (cid:4) (cid:3) (cid:4) (cid:5) invaluesincepurchase. Thelendermustalsopaytheborroweraforeclosurebenefit (cid:6) thatislargerin defaulter-friendlystatesthaninlender-friendlystates.If (cid:7) istheexpectationsoperator, thelender’szeroprofitcurveis (cid:5) (cid:6) (cid:0) nodefault (cid:1) (cid:0) (cid:1) (cid:2) (cid:1) (cid:2) (cid:5) (cid:6) (cid:0) default (cid:1) (cid:0) (cid:7) (cid:5)(cid:7) (cid:0) (cid:3) (cid:2) (cid:2) (cid:3) (cid:8) (cid:0) (cid:6) (cid:1) (cid:0) (cid:8) (cid:2) (cid:4) (cid:9) Toseethattheloantermsaffecttheprobabilityofdefault,assumethattheborrower defaultsifthegainfromdefault, (cid:6) ,outweighsthegainfromrepayingtheloan, (cid:2) (cid:3) (cid:0) (cid:1) (cid:2) (cid:5) (cid:0) (cid:3) (cid:2) .19 Thisdecisionis drivenbythe stochasticchangeinhouseprices. At the trigger value (cid:5) (cid:0) , the borrower is indifferent between defaulting and not defaulting: (cid:5) (cid:0) (cid:0) (cid:3) (cid:2) (cid:2) (cid:3) (cid:0) (cid:1) (cid:2) (cid:4) (cid:6) or (cid:5) (cid:0) (cid:4) (cid:6) (cid:3) (cid:2) (cid:2) (cid:1) (cid:2) (cid:2) (cid:9) Highervaluesof (cid:5) (cid:0) implya higherprobabilityofdefault. Assumingafixedinitial housevalue, (cid:5) (cid:0) is increasingin (cid:6) , (cid:2) , and (cid:1) , indicatingthat borrowersaremore likelytodefaultiftheyliveinastatewithgenerousforeclosurebenefits. Inaddition, borrowerswithhigherloan-to-valueratiosaremorelikelytodefault. Torecoupthese expectedlosses,lenderswillchargemoreleveragedborrowershigherinterestrates,so thelender’szero-profitcurveslopesupwardin (cid:0) (cid:2) (cid:10) (cid:1) (cid:3) space. Insummary,thelender’szero-profitcurveshiftsinwardwiththegenerosityofthe foreclosure laws. Borrowers could experience this reduced credit supply as higher interestrates,higherdownpaymentrequirements,orboth. Borrowers. Ifborrowersareawareofforeclosurelawswhentheypurchasetheir homes,defaulter-friendlylawsmayincreasetheirdemandformortgages.Considering 19Obviously,thissimplemodelignorestransactioncostsandthevalueoftheoptiontodefaultinthefuture, amongotherfactors. 8

housingsolelyasanasset,aborrower’sexpectedreturntofinancingahouseis (cid:10) (cid:11) (cid:12) (cid:2) (cid:6) (cid:10) (cid:7) (cid:3) (cid:5) (cid:0) (cid:3) (cid:2) (cid:2) (cid:3) (cid:4) (cid:0) (cid:1) (cid:0) (cid:6) (cid:10) (cid:2) (cid:3) (cid:2) (cid:5) (cid:0) (cid:3) (cid:9) Asthegenerosityoftheforeclosurelaw (cid:6) increases,aborrowerreceivesgreaterinsuranceagainstfallinghouseprices.Generousforeclosurebenefitsalsoresultinhigher interestrates, therebydecreasingthewealthofa borrowerwhorepaysthemortgage. Supposetheincreaseininterestratesoffsetsthehigherforeclosurebenefits,sothatthe lender’sexpectedprofitsareconstantacrossstates. Evenso,arisk-averseborrower’s utility frominvestingin housingis higherin a state with defaulter-friendlylaws, becausetheborrowergainsgreaterbenefitsatatimewhenthemarginalutilityofwealth ishigh. It may seem implausible, however, that borrowers are aware of foreclosure laws when they purchase property. In a study of the effect of state bankruptcy laws on mortgageapplications, for example, Berkowitz and Hynes (1999)assumed that loan demanddoesnotrespondtobankruptcylaws.20 Inthiscase,thetheoryoffersaclearcut prediction: unchanged mortgage demand and reduced mortgage supply imply a decreaseintheequilibriumloansize. InstitutionalFactors.Manymortgagemarketparticipantsareskepticalthatmortgagetermsvarywithforeclosurelaws.Lendersassertthattheydonottakeforeclosure costsintoaccountwhensettingmortgageterms. Inaddition,theyarguethattheinstitutionsthatoriginatemortgagesdonotbearthecostsofforeclosure. Mostmortgages aresoldtoFannieMae,FreddieMac,orothersecondarymarketinstitutionssoonafterorigination,andmortgageswiththegreatestriskofdefaultarerequiredtoacquire privatemortgageinsurance(PMI). Nonetheless,mortgagemarketparticipantsareclearlyawarethattheyfacedifferent foreclosurecostsacrossstates.FannieMae’svicepresidentofcreditportfoliomanage- 20[W]ethinkthatitisunlikelythatmostdebtorsarecognizantof,orestimate,[bankruptcy]exemptionsat thetimeofborrowing.”p.812 9

mentnotedinaninterviewthat“Itisinefficienttohaveexactlythesameforeclosurecostinsurancecoverageforeveryloanacrossthenation...becausethesecostsvaryfrom statetostate.”21 Furthermore,mortgagemarketparticipantsdonothavetoknowwhy theirlossesarehigherinsomeareasofthecountrytoadjusttheirloanterms. Secondarymortgagemarketinstitutionsandprivatemortgageinsurershavelargely thesameinformationsetasmortgageoriginators. Privatemortgageinsurersmustapproveanapplicationbeforetheoriginatoracceptsit,andinsomecircumstancesFannie MaeandFreddieMaccangivebackloanstolendersonwhichtheysufferdisproportionatelosses. EvenwithPMI,FreddieMacandFannieMaesufferlossesonforeclosures. Wood(1997)suggeststhataftercollectingPMI,FannieMae’slossesonmortgageswithLTVsinthe80to85percentrange(whichcarryPMI)were65percentof theirlossesonmortgageswithLTVsinthe75to80percentrange(whichdonotcarry PMI).22 In addition, not all loans are sold to secondarymarket institutions; in 1995, only57percentofsingle-familymortgagesweresecuritized.23 Lendersappeartohaverespondedtoforeclosurelawsinthepastbyadjustingloan terms.Asmentionedearlier,Jones(1993)documentedthatlendersrespondedtocostly foreclosurelawsinCanadabyincreasingdownpaymentrequirements.Meador(1982) and Wood (1997) found higher interest rates in states with defaulter-friendly laws, whileAlston(1984)foundhigherinterestratesandadecreasedsupplyofmortgagesin statesthatprohibitedforeclosuresduringtheGreatDepression. Thus,ifbothborrowersandlendersincorporateforeclosurelawsintotheirbehavior,simpletheorymodelsdonotpredicthowloansizevarieswiththelaws. Mortgage market institutions also send conflicting signals about whether the laws affect their behavior. In the next sections, I examine empirically whether loan size varies with stateforeclosurelawsinthecontemporarymortgagemarketwhenregionaleffectsare controlledforfully. 21SeeHochstein(2000). 22SeeTable4.5. 23SeeFreddieMac(1999),p.44TableA6. 10

4 Estimation Asmentionedpreviously,bothforeclosurelawsandrealestatemarketsexhibitregional patterns,therebyraisingconcernsthataregionalshockwillbemisinterpretedasaneffectofthelaws.Iaddressthisidentificationobstaclebyselectingmortgageapplications ingroupsofcountiesthatbordereachother,yetarelocatedindifferentstates. These mortgageapplicationsaresubjecttodifferentforeclosurelawsbutmaytakeonsimilar valuesforunobservedvariablesthatwouldotherwisebiastheanalysis.Holmes(1998) andBlack(1999)useasimilar“borders”identificationstrategytoestimatetheeffect of state right-to-worklaws on the locationof manufacturingand the effectof school qualityonhouseprices. BecauserealestatemarketsvaryacrosstheUnitedStates,relevantvariablesunobservedbytheeconometricianmayalsovarysubstantiallywithlocation.Tocapturethis unobservedvariationas flexiblyas possible, I implementa semiparametricestimator thatallowsunobservedvariablesto takeona differentvalueat eachcensustract, yet still identifies the effect of the laws. As an added virtue, each mortgageon the data setiscomparedonlywithneighborswithinafewmileradius,furtherincorporatingthe localnatureofrealestatemarkets. Intheirstudiesoftheeffectsofbankruptcylawonthemortgagemarket,Berkowitz andHynes(1999)andLinandWhite(2001)controlforregionalfactorswithstatefixed effectsmodelsthatexploitchangesinbankruptcylawovertime. Thisapproachisnot well suited to studies of foreclosure laws, however, because the laws rarely change. Whenforeclosurelawchangesdooccur,theyarerarelyrandomevents. Rather,they areusuallyprecipitatedbyeventssuchastheGreatDepression,raisingconcernsofa correlationbetweenthechangesinthelawandthemortgagemarketoutcomesunder study. In this context, a “borders” identification strategy is more appealing than a changes-over-timestrategy. To motivate the importance of my method, I first estimate two models that use 11

standardparametrictools. First, I runasimple ordinaryleast squaresregressionthat acknowledgesregionaleffectsonlythroughitsstandarderrors,whichallowforcorrelationacrossobservationswithinthesameMSA.ThenIrunafixedeffectsmodelthat assignsaseparatefixedeffecttoeachgroupofcounties.Becausethecorrelationacross observationsmaynotbeconstantwithinanMSA,thestandarderrorsinthismodelalso allowforwithin-MSAcorrelations. Thefixedeffectsmodel,however,maskssubstantialheterogeneitybothacrossand withincountygroups. Somecountygroups,suchasthetwocountiesintheareacoveredbyFargo,NorthDakota,andMoorhead,Minnesota,arepartofcompact,homogeneousmetropolitanareas,whileothers,suchasthecountiesinthevastNewYorkCity area,aresprawlingandheterogeneous.ItmaybeundesirabletotreatFargo-Moorhead andNewYorkCityequivalentlyortouseasinglefixedeffecttodescribethebehaviorofallthewidelyvaryingcomponentsoftheNewYorkCityarea. Anothersource of heterogeneity is county size: counties in the western United States are generally muchlargerthantheireasterncounterparts. Thuswesterncountygroupsmayextend fartherfromtheborderandencompassmorevariation.24 Finally,eventhemostcareful attempts to group similar counties togetherwill lead inevitablyto some arbitrary classifications. These inflexibleand arbitraryassumptionsembeddedin the fixed effectsmodelmayobscurethetruerelationshipbetweenforeclosurelawsandmortgage marketoutcomes. Tocontrolforregionalfactorsinamoreflexiblemanner,Iestimatethefollowing partiallinearmodel,inwhich (cid:11) istheloansize, (cid:12) isanunknownfunctionthattakeson adifferentvalueateachcensustract,thematrix (cid:13) includestheforeclosurelawsand othercontrolvariables,and (cid:14) isanidiosyncraticerrorterm: (cid:11) (cid:4) (cid:12) (cid:0) (cid:15) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) (cid:2) (cid:13) (cid:22) (cid:2) (cid:14) (cid:9) 24Arizona’s114,006squaremilesaredividedinto15counties,forexample,whileGeorgia’s59,441miles aredividedinto159counties. 12

Identificationof (cid:22) inthismodelrequirestwoassumptions. Thefirstisthat (cid:12) isa smoothfunction. Underthis assumption,theforeclosurelawvariablesareidentified, yetunobservedvariablesareallowedtotakeonadifferentvalueateachcensustract. Amodelwithaseparatefixedeffectforeachcensustractisnotidentified,becausethe fixedeffects wouldbe collinearwith the foreclosurelaw variables.25 Yet a modelin which a smooth function takes on a different value at each census tract is identified becausethefunctionchangessmoothlyatthestateborder,whiletheforeclosurelaws change discontinuously. This identification strategy fits into the regression discontinuity framework discussed by Hahn, Todd, and Van der Klaauw (2001) and Porter (2002). TobecertainthatIamcapturingtheeffectsofthelaws,Imustcontrolforallother factors that changeat state bordersand may be correlated with foreclosurelaws and themortgagemarketoutcomesunderstudy. Statedmoreformally,thesecondidentificationassumptionisthefamiliar (cid:7) (cid:0) (cid:14) (cid:13) (cid:1) (cid:10) (cid:15) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) (cid:4) (cid:9) . Asdiscussedinthenext section, I control for a rich assortment of state laws, county policies, and neighborhoodcharacteristicsthatchangediscontinuouslyatthestateborder.Ialsoexaminethe sensitivityofmyresultstoavarietyofrobustnessteststhatincluderemovingselected borders,MSAs,andstatesfromthesample. Figure4graphsthe (cid:12) functionoverthenortheasternUnitedStates. Thegraphsuggeststhatthesmoothnessidentificationassumptionisreasonable: (cid:12) doesnotchange invalue–orequivalently,inshade–atalmostanyofthebordersdepicted. Thegraph also indicatesthat the (cid:12) functioncorrespondsreasonablywell to idiosyncraticvariation in the real estate market. Specifically, areas that experienced greater growth in housepricesoverthe1990-94periodtakeonhighervalues.26 Asdiscussedinthedata section,thedependentvariableinthespecificationisthelogoftheloansizein1994or 1995,whileoneoftheindependentvariableisthelogof1990houseprices.Becauseof 25Censustractsaredefinedsoasnottospanstatelines. 26Thechange inhouseprices wascalculated with theOffice ofFederal HousingEnterprise Oversight HousePriceIndex,availableathttp://www.ofheo.gov/house/. 13

thelogtransformation,anypercentagechangeinhousepricesoverthe1990-94period thataffectsloansizesshouldbeabsorbedintheintercept (cid:12) function. Estimating the partial linear model. I estimate this model using the method outlinedinRobinson(1988),whosuggeststakingtheexpectationsofalltermsinthe modelwithrespectto(inthisapplication)thegeographiclocationofthecensustract (cid:7) (cid:0) (cid:11) (cid:15) (cid:1) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) (cid:4) (cid:7) (cid:0) (cid:12) (cid:0) (cid:15) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) (cid:15) (cid:1) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) (cid:2) (cid:7) (cid:0) (cid:13) (cid:15) (cid:1) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) (cid:22) (cid:9) Thenthepartiallinearmodelcanberewrittenas (cid:11) (cid:0) (cid:7) (cid:0) (cid:11) (cid:15) (cid:1) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) (cid:4) (cid:13)(cid:7) (cid:0) (cid:7) (cid:0) (cid:13) (cid:15) (cid:1) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) (cid:22)(cid:8) (cid:2) (cid:14) (cid:9) Theterm (cid:12) (cid:0) (cid:15) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) dropsoutoftheequationbecauseitequalsitsexpectation. FollowingRobinson(1988),Iestimatetheexpectedvalueofthedependentvariable (cid:11) and independent variables in (cid:13) , conditional on the census tract, with Nadaraya- Watson kernel regression. I subtract these expected values from the actual values, creatingtheresiduals (cid:11) (cid:0) (cid:7) (cid:0) (cid:11) (cid:15) (cid:1) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) and (cid:13) (cid:0) (cid:7) (cid:0) (cid:13) (cid:15) (cid:1) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) . Running ordinaryleastsquaresontheseresidualsyieldsanestimateof (cid:22) . Withtheestimateof (cid:22) inhand, (cid:12) canbecalculatedas (cid:7) (cid:0) (cid:11) (cid:15) (cid:1) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) (cid:0) (cid:7) (cid:0) (cid:13) (cid:15) (cid:1) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:3) (cid:22) . As calculated by kernel regression, the expected value of a variable in a given censustractissimplytheweightedaverageofthevariablesinsurroundingtracts. The weightingscheme,alsoknownasthekernel,placesgreaterweightongeographically closecensustractsthanonthosefartheraway.Thestatisticsliteraturesuggeststhatthe choiceofkernelhasonlyasmalleffectontheestimates;IuseboththeEpanechnikov kernelandthetriangularkernel. The bandwidth governs which census tracts are included in the weighted aver- 14

age. For example,with asmall bandwidth,onlynearbycensustracts are usedin the weightedaverage. Thebandwidthusedinthis paperis somewhatunusualbecauseit ismeasuredinmiles. Whiletheusualkernelestimatorofthedensity (cid:23) atapoint (cid:24) is written (cid:13)(cid:23) (cid:0) (cid:24) (cid:3) (cid:4) (cid:17) (cid:1) (cid:25) (cid:0) (cid:0)(cid:1)(cid:6) (cid:1) (cid:26) (cid:0) (cid:24) (cid:1) (cid:0) (cid:25) (cid:24) (cid:3) (cid:10) where (cid:26) isthekernel, (cid:17) isthesamplesize,and (cid:25) isthebandwidth,thispaper’skernel takestheform (cid:13)(cid:23) (cid:0) (cid:24) (cid:3) (cid:4) (cid:17) (cid:1) (cid:25) (cid:0) (cid:0)(cid:1)(cid:6) (cid:1) (cid:26) (cid:0) (cid:27) (cid:28) (cid:18) (cid:20) (cid:21) (cid:17) (cid:15) (cid:16) (cid:29) (cid:16) (cid:20) (cid:30) (cid:16) (cid:16) (cid:17) (cid:15) (cid:16) (cid:17) (cid:18) (cid:19) (cid:18) (cid:25) (cid:20) (cid:0) (cid:21) (cid:15) (cid:20) (cid:24) (cid:1) (cid:21) (cid:17) (cid:27) (cid:24) (cid:28) (cid:17) (cid:31) (cid:28) (cid:16) (cid:18) (cid:3) (cid:9) Distancesarecalculatedbetweenthegeographiccentersofthecensustractsusingthe Haversineformula.27 Kernelregressionestimatescanbesensitivetothechoiceofbandwidth.Iestimate anoptimalbandwidthforeachvariableusingcross-validationandalsoexperimentwith bandwidthsslightlylargerandsmallerthantheseidealbandwidths.28 Anotheralternative, though, is that the distance measure should be adjusted for population density. Unobserved characteristics may change more rapidly in dense urban neighborhoods thaninsparselysettledareas. ThusIalsousevariantsofthefollowingbandwidth,in which (cid:21) (cid:0) (cid:16) (cid:21) istheareainsquaremilesofthecensustract: (cid:27) (cid:4) (cid:0) (cid:10) (cid:14)(cid:15) (cid:0) (cid:16) (cid:2) areaoffirsttract (cid:10) (cid:17) (cid:9) (cid:3) (cid:2) (cid:10) (cid:14)(cid:15) (cid:0) (cid:16) (cid:2) areaofsecondtract (cid:10) (cid:17) (cid:9) (cid:1) 2 (cid:3) (cid:9) Sincecensustractsaredesignedtohavearoughlyconstantnumberofpeople,populationdensityisapproximatelyinverselyproportionaltocensustractarea.Takingthe 27TheHaversineformulatakesthecurvature oftheearthintoaccount. Seehttp://www.census.gov/cgibin/geo/gisfaq?Q5.1. 28Specifically, Iperformcross-validation byminimizingtheestimatedprediction error, asdiscussedin PaganandUllah(1999),p.119. 15

averageensuresthattheestimatorissymmetrical,inthesensethatiftract ! iswithin themaximumdistancefromtract " ,tract " iswithinamaximumdistanceoftract ! , andbothdistancesreceivethesameweight. Thevalue“20”guardsagainstoutliersby imposinganupperboundonthemaximumdistance. Themeanofthisbandwithis6.5 milesandthemedianis4.3. Under the method outlined by Robinson (1988), (cid:22) converges at a (cid:2) (cid:17) rate even though (cid:12) convergesatthesemiparametricrate.Standarderrorscanbeestimatedusing the conventionalordinaryleast squares formula. Porter (2002)shows, however, that inaregression-discontuitycontext, (cid:22) convergesataslowerrateandtheconventional standarderrorformulacannotbereliedon. Toaddresstheseconcerns,Ibootstrapthe standarderrorsusingadesignmatrixorpairwisebootstrap. 5 Data Geography. As a first step to implementingthe bordersestimation strategy, I select countiesintheUnitedStates thatarepartofametropolitanstatistical area(MSA)as defined by the Census Bureau; that lie along state borders; and that border another metropolitan county. I impose this “metropolitan” requirement because the Home MortgageDisclosureAct,mydatasource,doesnotrequirelenderstoreportmortgage applicationsfornonmetropolitancounties. Thisselectioncriteriayields181counties, whichIsortinto55groupsofcohesiveandgeographicallycontiguouscounties,listed intable2. Someofthesegroups,suchasthethreecountiesintheCharlotte-Gastonia- RockHillMSAthatliealongtheNorthCarolina-SouthCarolinaborder,arefromthe same MSA; others, such as the westernmost county of the Pensacola, Florida, MSA andtheeasternmostcountyoftheMobile,Alabama,MSA,arefromdifferentMSAs. Although counties from the same MSA form a more natural comparison group, the partial linear model compares properties within a couple miles of each other; these propertiesshouldtakeonsimilarvaluesofunobservedvariablesregardlessofwhether 16

theyaretechnicallyinthesameMSA.29 Thissamplewillyieldestimatesoftheeffectsofthelawsonlyifasizeablenumber of borderingcounties have different legal structures. Despite the regional pattern of foreclosure law, the maps in figures 1 through 3, which superimpose the 55 county groups on the foreclosure law maps, suggest that there is still substantial variation on which to base the analysis. Turning first to the judicial foreclosure requirement, 28 county groups, shown in black in figure 1 and listed in table 3, contain at least onestatethatrequiresajudicialforeclosureandonestatethatallowsapower-of-sale procedure. These county groups are spread across the United States; thus, they are unlikelytobedominatedbyasingleregionaleffect. Furthermore,theyarelocatedin denselypopulatedareasoftheUnitedStates,sothepartiallinearregressionestimates arebasedonalargenumberofmortgageapplicationsneartheborder. Intotal,these countiescontain776,588mortgageapplications. Estimation of the effects of statutory rights of redemption and deficiency judgments,however,appearsmoreproblematic. Asshowninfigures2and3andintables 4and5,ninecountygroupscontainonestatethatrequiresastatutoryrightofredemption and one state that does not; five county groups contain one state that prohibits deficiency judgments and one that does not. These counties are concentrated in the Midwest – four of the five “deficiency judgment” borders include either Minnesota or Iowa– and in less populatedareas of the country,with 87,679applicationsin the statutory right-of-redemptiongroups and 75,154 in the deficiency judgment groups. EarlierIarguedthattherewasastrongerlinkbetweenlendercostsandjudicialforeclosureprocessesthanbetweenthesecostsandeitherstatutoryrightsofredemptionor deficiencyjudgments. Theseestimationissuessuggestthatevenifsucharelationship exists,itmaybehardtodiscernusingthisapproach. Tables3,4,and5alsolistwhetherthecountygroup’sborderisariver.Sincerivers 29Nonetheless,asshownlaterinthepaper,Iperformarobustnesstestinwhichthesampleincludesonly countygroupsfromthesameMSA;thisrestrictiondoesnotchangetheresults. 17

are a barrierto interaction, geographicallyclose census tracts may be less similar in theirunobservedcharacteristicswhentheyareseparatedbyariver.30 Fourteenofthe judicial borders are separated by a river, ten are not, and four have partial river borders.Forthesefourpairs,Icodethecountiesindividuallytoindicatewhethertheyare separatedbyariver. Discardingbordersthatcoincidewithriversreducesthestatutory right-of-redemptionsampletofourbordersandthedeficiencyjudgmentsampletoone border. Loan application data. Under federal law, mortgage lending institutions with assetsaboveacertainthresholdarerequiredtoreportbasicinformationoneverymortgageapplicationthattheyreceiveinametropolitanstatisticalarea.31 Thefederalgovernmentusesthisinformationtoassesswhethermortgagelendersareservingthehousingcreditneedsoftheircommunities. TheHMDAdataincludecharacteristicsofthe borrower(sex,race,income,presenceofco-applicant)andtheloan(amount,type,purpose,ifthepropertyisowner-occupied).Thedataalsolistthecensustractinwhichthe propertyislocated,whethertheapplicationwasapproved,andthenameofthelending institution.HMDAdataareavailablefromtheFederalFinancialInstitutionsExaminationCouncilfortheyears1990through2001. Thedatasets areenormous: the1994 HMDAdata,forexample,containover12millionloanapplications. Iusethedatafrom1994and1995formyestimation. Theseyearsstrikeabalance between competing goals: earlier years provide a better match to the 1990 Census, whereas later years reflect the most recent trends in the mortgage market.32 I limit thesampletoacceptedapplicationsforthepurchaseofowner-occupied,one-tofourfamilyhomes. I excludeall loansoriginatedbymanufacturedhousinglenders,since 30Hoxby(2000)andCutlerandGlaeser(1997)useriversasinstrumentsinstudiesoftheeffectsofschool choiceandsegregation, respectively. Theyfindthatthenumberofriversispositively correlated withthe numberofschooldistrictsandthedegreeofsegregation;thesefindingssuggestthatriversaffectinteractions. 31Thislimitwas$10millionuntil1996,$28millionin1997,and$29millionin1998and1999. The 1998HMDAdataincludedanestimated75%ofallhomepurchaseloans(Canner,Passmore,andLaderman (1999)). 32Practicalreasonsalsodictateuseof1994and1995.HMDAdidnotinclude1990censustractidentifiers until1992,andtheregulationsgoverningwhichinstitutionsarerequiredtoreportunderHMDAchangedin 1993and1996.Inaddition,thequalityofthedataisbelievedtohaveincreasedovertime. 18

theunderwritingstandardsforsomeoftheseloansareclosertoautoloansthanhome mortgages. I also delete loans for less than $10,000; loans with missing or invalid censustractidentifiers;andloansinwhichtheborrowerclaimedtohavezeroincome. Afterimposingtheserestrictions,Iamleftwithasampleof1,252,562loanapplications inmetropolitancountiesalongstateborders. Neighborhoodcharacteristics. Iobtainadditionalinformationaboutborrowers’ neighborhoodsbymergingvariablesfromthe1990CensusontotheHMDAdataforthe tractinwhichthepropertyislocated. Neighborhoodcharacteristicsaffectthedemand forhousingandtheriskinessoftheloan.Inaddition,ifpeoplesortintoneighborhoods in whichthe residentsare like them, the neighborhoodcharacteristicsmayproxyfor unobservedapplicantcharacteristicsthatarecorrelatedwithdefault. As describedinmoredetailinthedataappendixintable1,Iincludecharacteristicsofthetractresidents,includingtheirage,educationlevel,andemploymentstatus. These variables affect a borrower’sdemand for housingand ability to repay a mortgage. Ialsoincludecharacteristicsofthehousingstock,suchasmedianhousevalue, medianyearbuilt, andpercentowner-occupied. Thequalityofthe housingstockaffectsdemandforhousingaswellasthevolatilityofneighborhoodhouseprices,which in turn increases default risk. For each tract, I also obtain its predicted crime rates, its geographicarea (measured in square miles), and the latitude and longitude of its geographiccenter. Stateandcountylawsandpolicies. Bankruptcylaws,likeforeclosurelaws,vary dramaticallyacrossstates. Aborrowerwhofilesforachapter7bankruptcyretainsequityinthehomeuptothestate-specifichomesteadexemptionandretainsotherassets up to the personal propertyexemption. All other assets are forfeitedto repay creditors. In1994,homesteadexemptionsvariedfrom$0tounlimitedacrossstates,while personalpropertyexemptionsvariedfrom$2,000to$40,500.33 33AndreasLehnertgenerouslyprovidedthisdata. Thedataappendixintable1providesdetailsonthe codingofthesevariables. 19

Bankruptcylawsappeartoaffectthemortgagemarket,butthedirectionoftheeffectisunclear. Aborrowercandeclarebankruptcywithoutdefaultingonamortgage. Solongashomeequityislessthanthehomesteadexemption,bankruptcyproceedings do not necessarily affect the mortgage lender. Indeed, Berkowitz and Hynes (1999) and Agarwal, Chomsisengphet, and Elul (2002) argue that generous bankruptcy exemptions decrease the probabilityof mortgagedefault because borrowershave more fundswithwhichtorepaytheirmortgages. However,LinandWhite(2001)notethat generousexemptionsincreasetheprobabilitythataborrowerwillfile forbankruptcy andthatforeclosureproceedingsaresubstantiallymoreexpensivewhenbankruptcyis involved. Adding more confusion, Berkowitz and Hynes (1999) and Lin and White (2001)findoppositeresultswhentheyexamine,usingtheHMDAdata,whetherborrowersaremorelikelytobedeniedmortgagecreditinstateswithgenerousbankruptcy exemptions. Iprovideanindependenttestoftheeffectofbankruptcylawsbyincludingthehomesteadandpersonalpropertyexemptionsinthespecification.Thispaperis, tomyknowledge,amongthefirsttoincludethefullsetofbankruptyandforeclosure laws.34 Other state and local policies that may affect housing demand are state income taxes, proxied here by the maximum income tax rate in the state; county per-pupil spending;andcountypropertytaxespercapita.Propertytaxesmayaffecthouseprices bysignalingthelocallevelofpublicgoodsprovisionorbyindicatingahigherfinancial burdenforhomeowners.Ialsoincludethenumberofhomebuildingpermitspercapita andthenumberofbankspercapitaineachcounty,apossibleindicatorofthedegree ofcompetitionamonglocalmortgagelenders. Table 6 contains sample means and standard deviations for the variables in the specification. The table indicates that 62percentof the mortgageapplicationsin the sampleareinstatesthatrequireajudicialforeclosureprocess; 7percentareinstates that require a statutory right of redemption; and 6 percent are in states that forbid 34Agarwal,Chomsisengphet,andElul(2002)alsoincludesthefullsetoflegalvariables. 20

deficiencyjudgments.Theaverageforeclosureprocesstook185days. SpecificationImplications. ThevariablesontheHMDAdataposetwoissuesfor the specification. First, the data do not include the value of the property associated with the loan. I proxy for this house value with census tract housing characteristics suchasthemedianhousevalueinthetractandwiththe (cid:12) function,which,asshown earlier,appearstosoakupsomeofthelocalvariationinhouseprices.Ifthesevariables capturethehousepricevariationfully,thecoefficientontheforeclosurelawvariables willreflecttheeffectofthelawsontheloan-to-valueratio. Ifnot,thecoefficientmay alsoexpresstheeffectofthelawsonthevalueoftheproperty.Ifborrowershaveamore difficulttimeobtainingmortgagecredit,theymaynotbewillingtopayasmuchfora house. LehnertandPence(2001)modelthisdynamicexplicitlyandpresentevidence thatforeclosurelaws arecapitalizedintohouseprices. Inthepresentpaper,I cannot ruleoutthisexplanationcompletely. Second, authors such as Berkowitz and Hynes (1999)and Lin and White (2001) haveusedtheHMDAdatatoexaminewhetherstatelawsaffecttheprobabilitythata mortgageapplicationisrejected.Manyborrowers,though,consultwithlendersbefore formallyfilingamortgageapplication. Lendersmaytellborrowershowtoshapetheir applicationstomaximizetheirchancesofapproval;theymightsuggest,forexample, thatalargerdownpaymentwouldincreasetheirprobabilityofacceptance.Ifborrowers adapt their applications accordingly, denial rates may not differ across states even if loan terms do. In addition, denial rates may be higher in states that are less risky for lenders, because lenders in these states may solicit and encouragemarginal loan applications more aggressively.35 Instead of focusing on denial rates, I focus on the loansizesofacceptedapplicationsinthispaper. 35Canner,Passmore,andLaderman(1999),forexample,notethatsubprimelendersgenerallyhavehigher denialratesthanprimelenders,partlybecauseoftheiractivesolicitionofmarginalloanapplications. 21

6 Results Topreviewtheresults,Ifindthatloansizesaresmallerinstateswithdefaulter-friendly foreclosurelawsandthatunobservedregionalfactorscanplayaninfluentialroleinestimatingthisrelationship.Table7presentsresultsfortheforeclosurelawparametersfor theordinaryleastsquares,fixedeffects,andpartiallinearmodels.Allspecificationsuse thelogoftheloansizeasthedependentvariable.Theordinaryleastsquaresandfixed effectsmodelsemployEicker-WhitestandarderrorsthatarerobusttoheteroskedasticityandthatallowforarbitrarycorrelationsacrossobservationswithinthesameMSA. The partial linear model uses a triangular kernel and a bandwidth of ten miles for every variable, slightly larger than the nine mile bandwidth that the cross-validation algorithmindicatedwasappropriateforthejudicialforeclosureandstatutoryright-ofredemptionvariables.Standarderrorsinthepartiallinearmodelarebootstrappedwith 200replications. Table7providesdramaticevidenceofaconnectionbetweenloansizeandajudical foreclosurerequirementonceregionalfactorsareaccountedforproperly.Theordinary least squares specification, shown in the first column, suggests that there is no relationshipbetween a judicial foreclosurerequirementand loan size. The coefficient is essentiallyzeroinbothsizeandstatistical significance,asomewhatastonishingfindinginaspecificationwith1,252,562observations.Accountingforregionalfactorswith countygroupfixedeffects,however,suggeststhatloansizesareastatisticallysignificant3.5percentsmallerinstateswithjudicialforeclosurerequirements. Controlling forgeographyinan evenmoreflexiblemannervia thepartiallinear modelyields an even strongerresult: loan sizes are 6.8 percent smaller in judicial foreclosurestates. Thisestimateisalso statisticallysignificant. Theseresultssuggestthereducedmortgagesupplyinstates with a judicialforeclosurerequirementdominatesanyeffectof thelawsonmortgagedemand. Unlike the judicial foreclosure requirement, even after introducing regional con- 22

trols loan sizes appear to be unaffectedby statutory rights of redemption. This provisionis associated with a statistically insignificant 1.5percentdecrease in loan size intheordinaryleastsquaresspecificationand0.7percentincreaseinthefixedeffects specification,aswellaswithaborderline-significant1.7increaseinthepartiallinear specification. Sinceveryfewhouseholdsexerciseastatutoryrightofredemption,itis notsurprisingto see noconnectionbetweenthis provisionand themortgagemarket. Inaddition,thisresultisbasedonarelativelysmallsampleofborders. Thedeficiencyjudgmentresultsvaryconsiderablywiththespecification,presumably because of the small sample of loan applications on which these estimates are based. However, as noted earlier, other researchers have also had difficulty robustly estimating the effects of deficiencyjudgments. In the ordinaryleast squares specification,loansizesareanimplausible13percenthigherinstatesthatforbiddeficiency judgments. Inthefixedeffectsspecification,thedeficiencyjudgmentcoefficientisessentiallyzero,whilethepartiallinearregressionestimatessuggestthatloansizesare5 percenthigherinstatesthatprohibitdeficiencyjudgments.Whiledeficiencyjudgment prohibitionsmay affect the mortgagemarket, this paper’s approachmay not reliably estimatetheireffect. Table 7 also contains the estimates and standard errors for the other parameters in the model. The coefficients suggest that the bankruptcy homestead exemption is negativelycorrelatedwithloansize,whilethebankruptcypersonalpropertyexemption ispositivelycorrelatedwithloansize. However,neitherexemptionappearstohavea substantialeffectonthemortgagemarket: theestimate suggeststhata 1,000percent increase in the homestead exemptionis associated with a 4 percent decrease in loan size.Althoughthepersonalpropertycoefficientisthelargerofthetwo,thisdifference probablyreflectsthesizeoftheexemptions:thehomesteadexemptionrangesfrom$0 to$500,000acrossstates, whilethepersonalpropertyexemptionrangesfrom$2,000 to$40,500.36 36Itop-codethehomesteadexemptionat$500,000forstateswithunlimitedhomesteadexemptionsand 23

Theotherstateandcountyvariableshavesimilarlysmalleffects.A$1,000increase in per-pupil spending is associated with a 0.8 percent (fixed effects) or 1.6 percent (partial linear) increase in loan size; a 1 percentage point increase in the maximum state income tax is associated with a 0.003 percent (fixed effects) or 0.002 percent (partiallinear)increaseinloansize;anda$100increaseinpropertytaxespercapitais associatedwitha0.003percent(fixedeffects)or0.002percent(partiallinear)decrease inloansize. Theloanapplicationandcensustractvariablesgenerallyfollowsensible patterns–loansizeincreaseswithincome,forexample.37 Borrowerswhoaretypically downpaymentconstrained–minorityborrowersandthosewhotakeoutFHAandVA loans–havelargerloansizes,controllingforcensustracthousingcharacteristics. Fixedeffectsrobustnesstests. Turningbacktothefixedeffectsmodel,wemight worrythatsomeofthecountygroupsrepresentinappropriatecomparisongroups. In somecountygroups,theareasoneithersideoftheborderrepresentdifferentmetropolitanstatisticalareas(MSAs).Bydefinition,MSAsareareaswithhighdegreesofsocial andeconomicintegration;differentMSAs,eveniftheyaregeographicallycontiguous, maytakeondifferentvaluesofunobservedvariables.Inothercountygroups,thestate border coincides with a river. Due to this barrier to interaction, the areas on either sideoftherivermayhavedevelopeddifferentlyandthushavedissimilarvaluesofthe unobservedvariables.38 Toaddresstheseconcerns,table8showsfixedeffectsestimatesfortwosubgroups: countygroupscomposedofonlyoneMSAandcountygroupsnotseparatedbyariver. The estimates suggest that loan sizes are 5.4 percent smaller when the sample is restrictedtosame-MSAcountiesandare5percentsmallerwhenthesampleisrestricted add“one”toeachstate’svalue beforetaking thelogofthehomestead exemption. Iusethelogarithmic transformationoftheexemptionsbecausetheirdistributionsareskewed. Enteringtheexemptionslinearly intothespecificationorusingdummyvariablesforexemptionquantilesdoesnotchangethesignorstatistical significanceofthebankruptcyexemptioncoefficients. 37Ininterpretingthecensustractestimates,notethatthe“percent”variablesarestoredaswholenumbers. Forexample,20percentis20not0.2. 38Seetable2foralistofsame-MSAcountygroupsandtables3,4,and5foralistofcountygroupsin whichriversarenotaborder. 24

tono-riverbordercounties; bothestimatesarestatisticallysignificant. Thus, restrictingthesampletomorehomogeneousgroupsofcountiesappearstoincreasesomewhat theassociationbetweenloansizeandajudicialforeclosurerequirement.Thestatutory right-of-redemptioncoefficientbehavesfairlyconsistently,hoveringnearzeroinboth robustnesstests.However,thedeficiencyjudgmentcoefficientrangesfrom-0.11inthe “norivers”testto0.04inthe“sameMSA”test. Inanotherrobustnesstest,Imeasurethecostlinessofthelawswiththeanticipated lengthoftheforeclosureprocess,insteadofwithdummyvariablesforjudicialforeclosure,statutoryrightofredemption,anddeficiencyjudgment.Thismeasurecomesfrom FreddieMac’sguidelinesandcapturessubtledifferencesbetweenstateswithroughly similarlaws.39 KentuckyandWisconsinbothhavejudicialforeclosureprocesses,for example,buttheanticipatedforeclosurelengthis138daysinKentuckyand300days inWisconsin.40 Asnotedearlier,stateswithlengthyforeclosureprocessestendtobe states that require a judicial foreclosure. On average, judicial foreclosures take 148 dayslongerthanpower-of-saleforeclosures;41thedifferencebetweenthequickestand mosttime-consumingstates,accordingtoFreddieMac’sguidelines,is289days. The results from this days specification are consistent with the foreclosure law dummyresults. A 100-dayincrease in the length of the foreclosureprocess is associatedwitha 1.8percentdecreaseinloansize; a 200-dayincreaseis associatedwith a 3.6 percent decrease in loan size, almost identical to the judicial parameter in the earlierspecification. Next,everycountygrouppotentiallycontainssomeidiosyncraticfactorthatmight makeit an inappropriatecomparisongroup. To ensurethat the results arenotdriven byanysuchidiosyncraticfactor,Irunthefixed-effectspecification55times,eachtime 39SeeJankowski(1999),pp.2-11. 40Judicialforeclosureprocessesmandateawaitingperiodbetweenthejudge’sordertoselltheproperty andtheactualsale. Theborrowercanstoptheforeclosureatanypointduringthisperiodbypayingoffthe entireloanbalance. ThisperiodextendsforonetotwomonthsinKentuckyandsixtotwelvemonthsin Wisconsin.Thesekindsofdifferencesdrivethevaryinglengthsoftheforeclosureprocesses. 41SeeWood(1997). 25

excludingadifferentcountygroupfromtheestimation. Ialsorunaparallelexercise inwhichIexcludeeachstatefromthedatasetinturn. Table9showsthesmallestand largestparametervaluesresultingfromthesespecifications,alongwiththenameofthe correspondingexcludedMSAorstate. TheresultssuggestthatnooneMSAorstateexertsanundueinfluenceonthejudicialforeclosureparameterestimate.Theestimatesrangefrom-0.046to-0.03forthe MSAspecificationsandfrom-0.047to-0.025forthestatespecifications,neatlybracketingtheoriginalpointestimate,-0.035.Thestatutoryright-of-redemptioncoefficients spanzerobutremainfairlysmallinmagnitude.Thedeficiencyjudgmentcoefficientis morevolatile,rangingfrom-0.027to0.049intheMSAspecificationsandfrom-0.035 to0.05inthestatespecifications. In a final robustness test, I run the model separately over 1994 and 1995 out of aconcernthattimetrendsin themortgagemarketmaybedistortingtheresults. The judicialcoefficientisalmostidenticalinbothyears:-0.037in1994and-0.034in1995. Theseresultsarenotshowninthetablesbutareavailableuponrequest. Partiallinearrobustnesstests.Thepartiallinearmodelresultscanbesensitiveto thechoiceofbandwidthandkernel.Toexaminetherobustnessoftheestimatestothese choices, tables 10 and11 show results from sevenbandwidth- kernel combinations. Fourspecificationsusethesamebandwidthforeverybandwidthandeverycensustract. Thesebandwidths–eightandtenmiles–brackettheoptimalbandwidthforthejudicial foreclosure parameter (nine miles), as determined by the cross-validation algorithm. Twootherspecificationsalsoimposethesamebandwidthoneveryvariablebutallow thisbandwidthtovarywithpopulationdensity.Thefinalspecificationusestheoptimal bandwidth,asdeterminedbycross-validation,foreveryvariable. The coefficient on the judicial foreclosure parameter is remarkably stable across specifications, ranging from -0.053 to -0.072. Regardless of the choice of kernel or the form of the bandwidth, the coefficient indicates a negative, statistically signifi- 26

cantrelationshipbetweenajudicialforeclosurerequirementandloansize. Decreasing thebandwidthincreasesthestandarderrors,afindingthatisconsistentwiththebiasvariance tradeoffin kernel estimation. The Epanechnikovspecifications also exhibit smaller standard errors, consistent with its “optimal kernel” reputation.42 However, the point estimates do not differ across kernels, therebyreinforcingthe conventional wisdomthatkernelchoiceisunimportant.Therobustnessoftheparameterestimateto smallchangesinbandwidthsuggeststhatthebandwidthchoicesarereasonable. Thecoefficientonthestatutoryrightofredemptionislessstable,switchingsignsas Imovefromafixed-milebandwidthtoabandwidththatvarieswithpopulationdensity. Thisinstabilitydoesnotresultfromasub-optimalbandwidthchoice,sincethecrossvalidation algorithm suggests that nine miles is also the optimal bandwidth for this parameter.Thedeficiencyjudgmentcoefficientsarealsovariable,rangingfrom0.05to 0.10,althoughtheirsignandstatisticalsignificanceareconstantacrossspecifications. Aswiththeearlierfixedeffectsresults,theseresultsdonotappearrobusttochangesin thespecification. 7 Conclusion In this paper I establish a robust inverse relationship between a judicial foreclosure requirementand mortgageloansize. This relationshipholdsin botha simple model withMSAfixedeffectsandina moreflexiblesemiparametricmodelinwhichunobservedvariablestakeonadifferentvalueateverycensustract. Therelationshipdoes notholdwhenIignoretheregionalpatternofforeclosurelaws;thisresultunderscores theimportanceofcontrollingforthesefactors. Statutory rights of redemption, in contrast, do not appear to affect the mortgage marketsubstantively.Sincefewborrowersexercisethisright,thisresultisnotsurprising. Theeffectofa prohibitionondeficiencyjudgmentsvariesacrossspecifications. 42SeePaganandUllah(1999),pp.27-28. 27

Bothcoefficientsarebasedonalimitednumberofbordersandarenot,ingeneral,measuredprecisely;somepreviousauthorshavealsohaddifficultyestimatingtheireffects robustly. Thispaperalsohighlightsawayofapplyingpartiallinearregressionthatisuseful for estimating the effects of state laws and policies. The method comparesobservationstoneighborswithinafewmiles,therebyincorporatingregionaleffectsinaflexiblemanner. Moreover,itcontrolsforanyunobservedfactorsthatvarysmoothlywith distance—a plausibleassumptionfor manyresearch questionswith a spatial component. Overall,theresultsfromthisestimationsuggestthatborrowersinstateswithdefaulterfriendlylawsfaceareducedsupplyofmortgagecredit. Ofcourse,evenatthisprice, borrowersinthesestates maystill valuetheinsuranceprovidedbygenerousforeclosurelaws. Nonetheless,theresultshighlightalargelyunexaminedtradeoff: defaulterfriendly foreclosure laws may assist homeowners experiencing hard times, but they also impose costs on a much larger pool of borrowers at the time of loan origination. Althougha decade-longrun-upin houseprices has precludedmuchdiscussion offoreclosures,recentincreasesinforeclosureratesmaybringrenewedinterestinthis tradeoff. 28

References Agarwal,S.,S.Chomsisengphet,andR.Elul(2002).Personalbankruptcyexemptionsandmortgagedefault.Unpublishedmanuscript,BrownUniversity. Alston,L.(1984).Farmforeclosuremoratoriumlegislation:Alessonfromthepast. AmericanEconomicReview74,445–457. Ambrose, B. W., R. J. Buttimer, Jr., and C. A. Capone (1997). Pricing mortgage defaultandforeclosuredelay.JournalofMoney,CreditandBanking29(3),314– 25. Ambrose,B.W.,C.A.Capone,Jr.,andY.Deng(2001).Optimalputexercise: An empiricalexaminationofconditionsformortgageforeclosure.JournalofReal EstateFinanceandEconomics23(2),213–234. Barta, P. (2001).Pinched homeowners are finding shelter in modified loans. Wall StreetJournalOctober30,A1. Berkowitz,J.andR.Hynes(1999).Bankruptcyexemptionsandthemarketformortgageloans.JournalofLawandEconomics42(2),809–830. Black,S.(1999,May).Dobetterschoolsmatter? Parentalvaluationofelementary education.QuarterlyJournalofEconomics114(2),557–599. Bridewell, D. (1938, January).The effect of defectivemortgagelaws on homefinancing.LawandContemporaryProblems5,545–563. Canner,G.B.,W.Passmore,andE.Laderman(1999,November).Theroleofspecialized lenders in extending mortgages to lower-income and minority homebuyers.FederalReserveBulletin85(11),709–723. Capone,Jr., C. A.(1996).ProvidingAlternativestoMortgageForeclosure: A Report to Congress. Washington, D.C.: U.S. Department of Housing and Urban Development. Ciochetti,B.A.(1997,Spring).Losscharacteristicsofcommercialmortgageforeclosure.RealEstateFinance,53–69. Clauretie,T.M.(1989).Stateforeclosurelaws,riskshifting,andthePMIindustry. JournalofRiskandInsurance56(3),544–554. Clauretie, T. M. and T. Herzog (1990). The effect of state foreclosure laws on loanlosses: Evidencefromthemortgageinsuranceindustry.JournalofMoney, CreditandBanking22(2),221–233. Cutler,D.M.andE.L.Glaeser(1997,August).Areghettosgoodorbad?Quarterly JournalofEconomics112(3),827–872. Fleishman,S.(2002).Facingforeclosure.WashingtonPostFebruary2,H1. FreddieMac(1999,April).SecondaryMortgageMarkets16(1). Gropp, R., J. K. Scholz, and M. J. White (1997). Personal bankruptcyand credit supplyanddemand.QuarterlyJournalofEconomics112(1),217–51. 29

Hahn, J., P. Todd, and W. Van der Klaauw (2001, January).Identificationand estimationoftreatmenteffectswitharegression-discontinuitydesign.Econometrica69(1),201–9. Hochstein,M.(2000,October23).FannieMaeouttoeatprivatemortgageinsurers’ lunchagain?AmericanBanker165(203),3A. Holmes, T. (1998). The effect of state policies on the location of manufacturing: Evidencefromstateborders.JournalofPoliticalEconomy106(4),667–705. Hoxby, C. M. (2000, December).Does competitionamongpublic schools benefit studentsandtaxpayers?AmericanEconomicReview90(5),1209–1238. Jankowski, D. (1999).National Mortgage Servicers Reference Directory (16ed.). PerryHall,MD:VernonEnterprises. Jones, L. D. (1993, April). Deficiency judgments and the exercise of the default optioninhomemortgageloans.JournalofLawandEconomics36(1),115–38. Keyles,S.A.(1995).Foreclosurelawandrelatedremedies:Astate-by-statedigest. Chicago,Ill.: SectionofRealProperty,ProbateandTrust Law, AmericanBar Association. Lehnert,A.andD.M.Maki(2002).Consumption,debt,andportfoliochoice:Testing the effects of bankruptcy law. Finance and Economics Discussion Series 2002-14,FederalReserveBoard,Washington,D.C. Lehnert,A.andK.M.Pence(2001).Thepriceofprotection:Foreclosurelaw,mortgagesandhouseprices.Unpublishedmanuscript,FederalReserveBoard,WashingtonD.C. Lin, E. Y. and M. J. White (2001). Bankruptcy and the market for mortgage and homeimprovementloans.JournalofUrbanEconomics50(1),138–62. Meador,M.(1982).Theeffectofmortgagelawsonhomemortgagerates.Journal ofEconomicsandBusiness34,143–148. Pagan, A. and A. Ullah (1999). Nonparametric Econometrics. Cambridge, UK: CambridgeUniversityPress. Polk (1988, February). Stop the attack of the equity skimmers. Mortgage Banking48(5),82–4. Porter,J.(2002).Asymptoticbiasandoptimalconvergenceratesforsemiparametric kernelestimatorsintheregressiondiscontinuitymodel.Workingpaper,Harvard University. Poterba,J.M.(1991).Housepricedynamics: Theroleoftaxpolicyanddemography.BrookingsPapersonEconomicActivity1991(2),143–83. Robinson, P. M. (1988). Root-N-semiparametric regression. Econometrica 56(4), 931–954. Touche,RossandCo.(1975).TheCostofMortgageLoanForeclosures:CaseStudiesofSixSavingsandLoanAssociations. Wood,C.(1997).TheImpactofMortgageForeclosureLawsonSecondaryMarket LoanLosses.Ph.D.thesis,CornellUniversity. 30

TABLE 1: DataAppendix–SourcesandVariableDefinitions DataType Source Variables ForeclosureLaws Jankowski(1999),cross-checkedwithKeyles(1995)andwithapplicablecaselawifnecessary. Judicialforeclosureprocedure: Nineteenstatesallowonlyjudicialprocedures. IalsocodeOklahomaandHawaiiasjudicialforeclosurestatesbecausenolendersusetheprohibitivelycostly nonjudicial procedures inthese states. The NationalMortgage Servicers Reference Directory statesthatOklahoma’snonjudicialprocedure is“rarelyexercised”(p. 7-159) becauseallborrowerscanelectthejudicialforeclosureprocessinstead. InHawaii,lendersdonotpursuenonjudicialforeclosuresbecausenotitlecompanywillinsurethesemortgages(Jankowski(1999), p.7-59). In1999,fiveyearsaftermyHMDAdata,Hawaiichangeditslawtomakeiteasierfor lenderstoobtainnonjudicialforeclosures. Nonjudicialproceduresareusedwidelyinallother statesthatofferthem. Statutoryrightofredemption: Icodeastateaspermittingastatutoryrightofredemptionifthe borrower always has thatoption under theforeclosure process (judicialor power-of-sale)that isstandard inthatstate. NewJersey allows aten-day statutory rightof redemption under the standardpower-of-saleforeclosureprocedure.Sincethisperiodissubstantiallylessthanthesixmonthtoone-yearperiodallowedinotherstates,IcodeNewJerseyasnotallowingastatutory rightofredemption.Underthisdefinition,ninestatesallowastatutoryrightofredemption. Deficiencyjudgment:Icodeastateasforbiddingdeficiencyjudgmentsiftheyarenotpermitted undertheforeclosureprocess(judicialorpowerofsale)thatisstandardinthatstate. ConsumerBankruptcyLaw LehnertandMaki(2002)AppendixA 1994 and 1995 homestead and personal property exemptions, by state, for unmarried households. Iusethenaturallogoftheexemptionsbecausetheirdistributionshavelongrighttails. I add“1”tothehomesteadexemptionbeforetakingthelogbecauseitequalszeroinsomestates. BorrowerandLoanCharacteristics 1994and1995HomeMortgageDisclosureAct Loansize;income;censustract;dummyvariablesforconventional,FHA,VA,andFmHAloans; forblack,Hispanic,andotherrace,andsinglefemaleborrowers;andforapplicationswithacoapplicant. continuedonnextpage 31

Table1(continuedfrompreviouspage) DataType Source Variables NeighborhoodPopulationandHousingCharacteristics 1990Census;asextractedbyCensusCD+Maps Age: Percentofresidentswhoare0-17,18-21,22-29,30-39,40-49,50-59,60-69,or70years andolder. Education:Percentofresidents25yearsorolderwhodidnotfinishhighschool,finishedhigh school,attendedsomecollege,graduatedfromcollege,orattendedsomegraduateschool. Laborforceparticipation:Percentofresidents16yearsorolderwhoareemployed,unemployed, oroutoftheworkforce. Housingcharacteristics: Percentofhomesthatareowner-occupied, rented,vacant, ormobile; medianhomeage,medianrent,medianhousevalue. Censustracthomicideandrobberyindexscores CAPIndex,Inc.:www.capindex.com. Indexesforhomicidesandrobberiespercensustract,aspredictedbytheCAPIndexmodel.The indexrunsfrom0to2000;100isthenationalaverage. Becauseofthelongrighttail,Itakethe naturallogofbothvariables. Lendercharacteristics U.S.DepartmentofHousingandUrbanDevelopment DummyvariablesindicatingwhetherthelenderassociatedwiththeHMDAmortgageapplicationspecializesinsubprimeormanufacturedhomelending. StateTaxRates NBERTaxsimmodel:www.nber.org/taxsim/state-rates/ Maximumstatetaxratefor1994or1995,asapplicable.Themaximumrateisthoughttobeless tarnishedwithendogeneity. Countybuildingpermitspercapita,1994 BureauoftheCensusextractedviaCensusCD+Maps Building permits per capita for new, private, one-unit housing structures, 1994. Available by county. Countypropertytaxespercapita,1992 1992CensusofStateandLocalGovernmentsextractedviaCensusCD+Maps Commercialandsavingbankofficespercapita,1994 FederalDepositInsuranceCorporationextractedviaCensusCD+Maps Availablebycounty. continuedonnextpage 32

Table1(continuedfrompreviouspage) DataType Source Variables Countyper-pupilspending FiscalYear(FY)1995PublicElementary-SecondaryEducationFinancesdatafromtheCensus Department’sAnnualSurveyofGovernmentFinances FY1995coverstheSeptember1994-June1995schoolyear.SincetheFY1994dataisasurvey thatexcludes someschool districts,whilethe FY1995 dataisacensus that coversallschool districts,IusetheFY1995 dataforbothyearsofHMDAdata. Tocalculatecountyperpupil spending,Isumallcurrentspending(TCURELSC)forallschoolsinthecountyanddivideitby thesumofallstudentsinthecounty(V33).Onlypublicschoolsareincludedinthiscalculation. Sincecapitalexpendituresareerraticintheirtiming,excludingcapitalspendingfromper-pupil spendingisstandardintheeducationliterature. 33

TABLE 2: METROPOLITAN AREASAND COUNTIESINTHESAMPLE CountyGroupName CountyGroupName CountyName Obs CountyName Obs MobileAL-PensacolaFL LasCrucesNM-ElPasoTX BaldwinCounty,Alabama 4029 DonaAnaCounty,NewMexico 3743 EscambiaCounty,Florida 7657 ElPasoCounty,Texas 12738 MobileAL-PascagoulaMS WilmingtonNC-MyrtleBeachSC JacksonCounty,Mississippi 3358 BrunswickCounty,NorthCarolina 1880 MobileCounty,Alabama 9313 HorryCounty,SouthCarolina 6010 NewOrleansLA-GulfportMS Charlotte-Gastonia-RockHillNC-SC (cid:0) HancockCounty,Mississippi 1028 GastonCounty,NorthCarolina 4004 St.TammanyParish,Louisiana 6723 MecklenburgCounty,NorthCarolina 24805 ColumbusGA-AL (cid:0) YorkCounty,SouthCarolina 3697 MuscogeeCounty,Georgia 3558 JohnsonCity-Kingsport-BristolTN (cid:0) RussellCounty,Alabama 785 HawkinsCounty,Tennessee 886 MemphisTN-AR-MS (cid:0) SullivanCounty,Tennessee 3781 CrittendenCounty,Arkansas 1362 WashingtonCounty,Virginia 1075 DeSotoCounty,Mississippi 3797 EriePA-ClevelandOH ShelbyCounty,Tennessee 26989 AshtabulaCounty,Ohio 2241 TiptonCounty,Tennessee 1271 ChautauquaCounty,NewYork 2358 TexarkanaTX-TexarkanaAR (cid:0) ErieCounty,Pennsylvania 6820 BowieCounty,Texas 2148 WheelingWV-OH (cid:0) MillerCounty,Arkansas 891 BelmontOhio 1256 FortSmithAR-OK (cid:0) MarshallWestVirginia 562 CrawfordCounty,Arkansas 1426 OhioCounty,WestVirginia 918 SebastianCounty,Arkansas 3288 PittsburghPA-SteubenvilleOH SequoyahCounty,Oklahoma 631 BeaverCounty,Pennsylvania 3294 RenoNV HancockCounty,WestVirginia 534 PlacerCounty,California 7508 JeffersonCounty,Ohio 1155 WashoeCounty,Nevada 10984 WashingtonCounty,Pennsylvania 3808 CheyenneWY-FortCollinsCO Youngstown-WarrenOHSharonPA LaramieCounty,Wyoming 3024 MahoningCounty,Ohio 6455 LarimerCounty,Colorado 8187 MercerCounty,Pennsylvania 2513 WeldCounty,Colorado 4610 TrumbullCounty,Ohio 4947 ChattanoogaTN-GA (cid:0) Parkersburg-MariettaWV-OH (cid:0) CatoosaCounty,Georgia 1193 WashingtonCounty,Ohio 1629 HamiltonCounty,Tennessee 7512 WoodCounty,WestVirginia 2204 WalkerCounty,Georgia 1189 Chicago-GaryIL-IN (cid:0) Minneapolis-St.PaulMN-WI (cid:0) CookCounty,Illinois 123836 DakotaCounty,Minnesota 12460 KankakeeCounty,Illinois 2885 St.CroixCounty,Wisconsin 1803 LakeCounty,Indiana 11082 WashingtonCounty,Minnesota 6982 WillCounty,Illinois 19428 continuedonnextpage 34

Table2(continuedfrompreviouspage) CountyGroupName CountyGroupName CountyName N CountyName N Augusta-AikenGA-SC (cid:0) Baltimore,MD-York/Lancaster,PA AikenCounty,SouthCarolina 3087 BaltimoreCounty,Maryland 18138 ColumbiaCounty,Georgia 3268 CarrollCounty,Maryland 4463 RichmondCounty,Georgia 4138 HarfordCounty,Maryland 6845 St.LouisMO-IL (cid:0) LancasterCounty,Pennsylvania 11121 JeffersonCounty,Missouri 5992 YorkCounty,Pennsylvania 10395 MadisonCounty,Illinois 7550 Hartford,CT-SpringfieldMA St.CharlesCounty,Missouri 11472 HampdenCounty,Massachusetts 8590 St.ClairCounty,Illinois 6254 HartfordCounty,Connecticut 19013 St.LouisCounty,Missouri 31476 TollandCounty,Connecticut 3397 St.LouisCityMissouri 5939 Poughkeepsie,NY-Danbury,CT (cid:0) Davenport-Moline-IA-IL (cid:0) DutchessCounty,NewYork 5142 RockIslandCounty,Illinois 3956 FairfieldCounty,Connecticut 5826 ScottCounty,Iowa 4716 PutnamCounty,NewYork 2310 RockfordIL-BeloitWI Lawrence,MA (cid:0) BooneCounty,Illinois 1410 EssexCounty,Massachusetts 16798 RockCounty,Wisconsin 4823 RockinghamCounty,NewHampshire 4304 WinnebagoCounty,Illinois 9462 Lowell-FitchburgMANashuaNH (cid:0) Chicago-KenoshaIL-WI (cid:0) HillsboroughCounty,NewHampshire 7360 KenoshaCounty,Wisconsin 3939 MiddlesexCounty,Massachusetts 32570 LakeCounty,Illinois 23201 WorcesterCounty,Massachusetts 2440 McHenryCounty,Illinois 11214 ProvidenceRI,Attleboro-WorcesterMA LouisvilleKY-IN (cid:0) BristolCounty,Massachusetts 10802 ClarkCounty,Indiana 2838 BristolCounty,RhodeIsland 1044 FloydCounty,Indiana 2176 HampdenCounty,Massachusetts 80 HarrisonCounty,Indiana 20 NorfolkCounty,Massachusetts 16720 JeffersonCounty,Kentucky 20009 ProvidenceCounty,RhodeIsland 10284 OldhamCounty,Kentucky 1470 WindhamCounty,Connecticut 96 Evansville-HendersonIN-KY (cid:0) WorcesterCounty,Massachusetts 12904 HendersonCounty,Kentucky 981 NewLondon,CT (cid:0) VanderburghCounty,Indiana 4721 NewLondonCounty,Connecticut 6098 WarrickCounty,Indiana 1905 WashingtonCounty,RhodeIsland 2836 Clarksville-HopkinsvilleTN-KY (cid:0) Philadelphia-Wilmington-AtlanticCity (cid:0) ChristianCounty,Kentucky 1737 BurlingtonCounty,NewJersey 12777 MontgomeryCounty,Tennessee 5682 CamdenCounty,NewJersey 11367 ShreveportLA-MarshallTX CecilCounty,Maryland 2120 CaddoParish,Louisiana 5748 ChesterCounty,Pennsylvania 11001 HarrisonCounty,Texas 1237 DelawareCounty,Pennsylvania 10930 LakeCharlesLA-BeaumontTX GloucesterCounty,NewJersey 6266 CalcasieuParish,Louisiana 3971 NewCastleCounty,Delaware 14425 OrangeCounty,Texas 1624 PhiladelphiaCounty,Pennsylvania 24026 continuedonnextpage 35

Table2(continuedfrompreviouspage) CountyGroupName CountyGroupName CountyName N CountyName N Cincinnati-HamiltonOH-KY-IN (cid:0) Portland-SalemOR-WA (cid:0) BooneCounty,Kentucky 2764 ClarkCounty,Washington 12402 ButlerCounty,Ohio 9808 MultnomahCounty,Oregon 19925 CampbellCounty,Kentucky 2569 Portsmouth-RochesterNH (cid:0) ClermontCounty,Ohio 5335 RockinghamCounty,NewHampshire 2341 DearbornCounty,Indiana 1614 StraffordCounty,NewHampshire 1776 HamiltonCounty,Ohio 21283 YorkCounty,Maine 1432 KentonCounty,Kentucky 4046 NewYorkCity (cid:0) OmahaNE-IA (cid:0) BergenCounty,NewJersey 18172 DouglasCounty,Nebraska 10745 BronxCounty,NewYork 4758 PottawattamieCounty,Iowa 1527 FairfieldCounty,Connecticut 18471 SarpyCounty,Nebraska 3109 HudsonNewJersey 5784 KansasCityMO-KS (cid:0) MiddlesexCounty,NewJersey 16664 CassCounty,Missouri 2121 NewYorkNewYork 9485 ClayCounty,Missouri 5418 PassaicCounty,NewJersey 7924 JacksonCounty,Missouri 15622 RichmondCounty,NewYork 7183 JohnsonCounty,Kansas 16605 RocklandCounty,NewYork 5470 LeavenworthCounty,Kansas 1845 UnionCounty,NewJersey 9420 PlatteCounty,Missouri 2347 WestchesterCounty,NewYork 17724 WyandotteCounty,Kansas 2540 Sussex,NJ-PortJervisNY (cid:0) Huntington-AshlandWV-KY-OH (cid:0) OrangeCounty,NewYork 5954 BoydCounty,Kentucky 1141 SussexCounty,NewJersey 4101 CabellCounty,WestVirginia 1812 Trenton-HunterdonNJ,Philadelphia GreenupCounty,Kentucky 752 BucksCounty,Pennsylvania 16583 LawrenceCounty,Ohio 1347 HunterdonCounty,NewJersey 4324 WayneCounty,WestVirginia 618 MercerCounty,NewJersey 8540 DetroitMI-ToledoOH Easton,PA-Washington,NJ FultonCounty,Ohio 1389 NorthamptonCounty,Pennsylvania 5910 LenaweeCounty,Michigan 2733 WarrenCounty,NewJersey 2759 LucasCounty,Ohio 12803 Washington,DC (cid:0) MonroeCounty,Michigan 4156 ArlingtonCounty,Virginia 4305 Fargo-MoorheadND-MN (cid:0) DistrictofColumbia 9639 CassCounty,NorthDakota 3538 FairfaxCounty,Virginia 32154 ClayCounty,Minnesota 1298 FrederickCounty,Maryland 5745 SouthBendIN-BentonHarborMI LoudounCounty,Virginia 7941 BerrienCounty,Michigan 4454 MontgomeryCounty,Maryland 24910 St.JosephCounty,Indiana 7179 PrinceGeorge’sCounty,Maryland 22333 GrandForksND-MN (cid:0) Duluth-SuperiorMN-WI (cid:0) GrandForksCounty,NorthDakota 1696 DouglasCounty,Wisconsin 1063 PolkCounty,Minnesota 573 St.LouisCounty,Minnesota 4006 (cid:0) Indicatesthatallcountieswithinacountygroupbelongtothesamemetropolitan statisticalareaasdefinedbytheCensusBureau. 36

TABLE3: JUDICIAL BORDERS CountyCross-BorderGroup River Noriver FortSmithAR-OK none SouthBendIN-BentonHarborMI none Clarksville-HopkinsvilleTN-KY none ShreveportLA-MarshallTX none DetroitMI-ToledoOH none WilmingtonNC-MyrtleBeachSC none Charlotte-Gastonia-RockHillNC-SC none ProvidenceRI,Attleboro-Worcester none NewLondon,CT none Hartford,CT-SpringfieldMA none Borderispartiallyariver PittsburghPA-SteubenvilleOH Ohio/none Portsmouth-RochesterNH Piscataqua/none KansasCityMO-KS Missouri/none Washington,DC Potomac/none Borderisentirelyariver MobileAL-PensacolaFL Perdido NewOrleansLA-GulfportMS Pearl Augusta-AikenGA-SC Savannah St. LouisMO-IL Mississippi Davenport-Moline-RockIslandIA-IL Mississippi Huntington-AshlandWV-KY-OH Ohio LakeCharlesLA-BeaumontTX Sabine Fargo-MoorheadND-MN RedRiver Minneapolis-St.PaulMN-WI St. CroixandMississippi GrandForksND-MN RedRiver Duluth-SuperiorMN-WI Superior LasCrucesNM-ElPasoTX RioGrande WheelingWV-OH Ohio Parkersburg-MariettaWV-OH Ohio 37

TABLE 4: STATUTORYRIGHT-OF-REDEMPTION BORDERS CountyCross-BorderGroup River MobileAL-PascagoulaMS none SouthBendIN-BentonHarborMI none DetroitMI-ToledoOH none KansasCityMO-KS none MobileAL-PensacolaFL Perdido ColumbusGA-AL Chattahoochee Minneapolis-St.PaulMN-WI St. CroixandMississippi Duluth-SuperiorMN-WI Superior LasCrucesNM-ElPasoTX RioGrande TABLE 5: DEFICIENCY JUDGMENT BORDERS CountyCross-BorderGroup River RenoNV none Davenport-Moline-RockIslandIA-IL Mississippi OmahaNE-IA Missouri Minneapolis-St.PaulMN-WI St. CroixandMississippi Duluth-SuperiorMN-WI Superior 38

TABLE 6: SAMPLE STATISTICS Mean Std.Dev. STATELAWS Judicialforeclosure 0.62 0.49 Statutoryrightofredemptionrequired 0.07 0.25 Deficiencyjudgmentprohibited 0.06 0.23 Lengthofforeclosureprocess 185.21 93.2 Loghomesteadexemption 8.79 2.64 Logpersonalpropertyexemption 9.07 0.65 Maximumstateincometax 5.10 2.37 LOANAPPLICATIONDATA Logloanamount 11.45 0.63 Logincome 10.83 0.63 Black 0.09 0.28 Hispanic 0.05 0.21 Otherrace 0.05 0.23 Co-applicant 0.67 0.47 Singlefemale 0.15 0.36 FHAmortgage 0.16 0.37 VAmortgage 0.06 0.23 FmHAmortgage 0.00 0.04 CENSUSTRACTDATA Pcthouseholdsmovedin1960–69 9.69 5.78 Pcthouseholdsmovedin1970–79 18.75 6.70 Pcthouseholdsmovedin1980–84 13.81 3.95 Pcthouseholdsmovedin1985–88 30.31 8.57 Pcthouseholdsmovedin1989–90 18.37 8.69 PercentHSgraduates 29.03 9.44 Percentsomecollege 25.55 6.32 Percentcollegegraduate 17.20 9.49 Percentgraduatedegree 9.88 8.22 Percentbelowpovertyline 6.99 7.23 Percentblackresidents 7.88 16.7 Log1989percapitaincome 9.74 0.39 Pctunemployed 4.89 3.20 PercentAge18-21 5.15 2.98 PercentAge22-29 12.69 4.61 PercentAge30-39 18.10 3.65 PercentAge40-49 14.13 3.42 PercentAge50-59 9.19 2.60 PercentAge60-69 8.09 3.35 PercentAge70+ 7.58 4.63 Percentvacant 5.89 5.53 Percentmobilehomes 3.41 7.84 continuedonnextpage 39

Table6(continuedfrompreviouspage) Mean Std.Dev. Percentrented 25.58 16.6 Pct2or3bedrooms 66.45 14.9 Pct4or5bedrooms 22.03 15.6 Medianyearbuilt 1964.08 13.9 Logmedianhousevalue 11.59 0.60 Medianrent 567.05 196 Logofhomicidescore 4.03 0.69 Logofrobberyscore 2.98 1.15 COUNTYDATA Per-pupilspending(thousands) 6.35 1.60 Homebuildingpermitspercapita 0.00 0.00 Bankspercapita 0.00 0.00 Propertytaxespercapita 797.13 393 Year 1994.46 0.50 FHA = Federal Housing Administration; VA = Department of Veterans Affairs; FmHA=FarmersHomeAdministration. 40

TABLE7: PARAMETER ESTIMATES Parameter OLS FixedEffects PartialLinear FORECLOSURELAWVARIABLES Judicialforeclosure -0.006 -0.035 -0.068 (0.013) (0.010) (0.005) Statutoryrightofredemptionrequired -0.015 0.007 0.017 (0.033) (0.023) (0.009) Deficiencyjudgmentprohibited 0.128 -0.003 0.052 (0.055) (0.038) (0.008) OTHERPOLICYVARIABLES Loghomesteadexemption -0.010 -0.007 -0.004 (0.002) (0.002) (0.001) Logpersonalpropertyexemption -0.009 0.024 0.024 (0.013) (0.015) (0.002) Per-pupilspending 0.005 0.008 0.016 (0.005) (0.008) (0.001) Maximumstateincometax -0.001 0.003 0.002 (0.003) (0.002) (0.001) Propertytaxespercapita 0.000 0.000 -0.000 (0.000) (0.000) (0.000) LOANAPPLICATIONVARIABLES LogIncome 0.454 0.450 0.444 (0.016) (0.015) (0.001) Black 0.023 0.023 0.021 (0.007) (0.007) (0.002) Hispanic 0.040 0.035 0.017 (0.019) (0.017) (0.002) Otherrace 0.048 0.042 0.041 (0.010) (0.010) (0.002) Co-applicant 0.075 0.077 0.075 (0.006) (0.006) (0.001) Singlefemale -0.018 -0.019 -0.020 (0.006) (0.006) (0.001) FHAmortgage 0.056 0.056 0.057 (0.014) (0.015) (0.001) VAmortgage 0.172 0.174 0.172 (0.011) (0.011) (0.001) FHMAmortgage 0.105 0.092 0.097 (0.020) (0.021) (0.006) CENSUSTRACTANDCOUNTYVARIABLES Pcthouseholdsmovedin1960–69 0.001 0.001 0.001 (0.001) (0.001) (0.0001) Pcthouseholdsmovedin1970–79 -0.001 -0.002 -0.002 (0.001) (0.001) (0.0001) continuedonnextpage 41

Table7(continuedfrompreviouspage) Parameter OLS FixedEffects PartialLinear Pcthouseholdsmovedin1980–84 -0.000 -0.001 -0.002 (0.001) (0.001) (0.0001) Pcthouseholdsmovedin1985–88 -0.000 -0.002 -0.002 (0.001) (0.001) (0.0001) Pcthouseholdsmovedin1989–90 -0.000 -0.002 -0.002 (0.001) (0.001) (0.0001) PercentHSgraduates -0.005 -0.002 -0.002 (0.001) (0.001) (0.0001) Percentsomecollege -0.001 -0.001 -0.000 (0.001) (0.001) (0.0001) Percentcollegegraduate -0.002 0.001 0.001 (0.001) (0.001) (0.0001) Percentgraduatedegree -0.005 -0.004 -0.003 (0.001) (0.001) (0.0002) Percentbelowpovertyline -0.002 -0.001 -0.000 (0.001) (0.001) (0.0001) Percentblackresidents 0.001 0.000 -0.001 (0.001) (0.001) (0.0001) Log1989percapitaincome 0.010 -0.005 0.047 (0.034) (0.033) (0.003) Pctunemployed -0.007 -0.005 -0.004 (0.002) (0.001) (0.0002) Percentvacant -0.001 0.000 0.001 (0.001) (0.001) (0.0002) Percentmobilehomes -0.002 -0.001 -0.001 (0.001) (0.000) (0.0001) Percentrented 0.003 0.003 0.003 (0.001) (0.001) (0.0001) Pct2or3bedrooms 0.004 0.004 0.003 (0.001) (0.001) (0.0001) Pct4or5bedrooms 0.006 0.005 0.005 (0.001) (0.001) (0.0001) Medianyearbuilt -0.001 -0.000 0.001 (0.0004) (0.000) (0.000) Logmedianhousevalue 0.381 0.360 0.269 (0.022) (0.023) (0.003) Medianrent 0.000 -0.000 -0.000 (0.000) (0.000) (0.000) PercentAge18-21 0.001 0.000 -0.001 (0.001) (0.001) (0.0002) PercentAge22-29 -0.003 -0.003 -0.004 (0.002) (0.002) (0.0002) continuedonnextpage 42

Table7(continuedfrompreviouspage) Parameter OLS FixedEffects PartialLinear PercentAge30-39 0.003 0.001 -0.001 (0.002) (0.001) (0.0002) PercentAge40-49 0.001 -0.002 -0.003 (0.002) (0.001) (0.0002) PercentAge50-59 -0.002 -0.003 -0.005 (0.001) (0.001) (0.0002) PercentAge60-69 -0.003 -0.006 -0.006 (0.001) (0.001) (0.0002) PercentAge70+ -0.001 -0.001 -0.002 (0.001) (0.001) (0.0002) Logofhomicidescore 0.016 0.040 0.039 (0.012) (0.009) (0.001) Logofrobberyscore -0.019 -0.042 -0.041 (0.011) (0.009) (0.001) Homebuildingpermitspercapita 7.721 1.848 5.343 (2.451) (1.436) (0.316) Bankspercapita -256.72 -71.210 -178.07 (91.46) (61.10) (20.44) Year -0.024 -0.023 -0.023 (0.004) (0.004) (0.0007) N 1,252,562 R (cid:0) 0.58 0.59 n.a. NOTE. The dependentvariable in the regression is (cid:18)(cid:19) (cid:20) (loan size). Eicker-White standard errors that allow for within-MSA correlations are shown in parentheses for the OLS and fixed effects specifications. Bootstrappedstandard errorsbased on 200 replicatesareshownforthepartiallinearregressionspecification.Partiallinearregressionestimatesbasedonabandwidthoftenmilesandatriangularkernel. 43

TABLE 8: FIXED EFFECTS ROBUSTNESS TESTS Parameter BaseSpec SameMSA NoRiver Judicialforeclosure -0.035 -0.054 -0.050 (0.010) (0.016) (0.015) Statutoryrightofredemptionrequired 0.007 0.008 -0.017 (0.023) (0.035) (0.031) Deficiencyjudgmentprohibited -0.003 0.041 -0.11 (0.038) (0.036) (0.028) Days -0.00018 n.a. n.a. (0.00006) n.a. n.a. N 1,252,562 944,966 818,297 R (cid:2) 0.58 0.59 0.59 0.59 (daysspecification) NOTE. Thedependentvariableintheregressionis (cid:18)(cid:19) (cid:20) (loansize). Eicker-Whitestandarderrorsthatallowforwithin-MSAcorrelationsareshowninparentheses. MSA= metropolitanstatisticalarea. TABLE9: “LEAVE-ONE-OUT”FIXED EFFECTSROBUSTNESS TESTS MSA State Parameter MinValue MaxValue MinValue MaxValue Judicialforeclosure -0.046 -0.030 -0.047 -0.025 (0.012) (0.010) (0.012) (0.009) WashingtonDC NewYorkCity Maryland Connecticut Statutoryrightof -0.006 0.021 -0.014 0.027 redemptionrequired (0.025) (0.022) (0.026) (0.027) LasCrucesNM NewYorkCity Wisconsin Connecticut Deficiencyjudgment -0.027 0.049 -0.035 0.050 prohibited (0.034) (0.036) (0.043) (0.036) DavenportIA RenoNV Wisconsin California NOTE. Thedependentvariableintheregressionis (cid:18)(cid:19) (cid:20) (loansize). Eicker-Whitestandarderrorsthatallowforwithin-MSAcorrelationsareshowninparentheses. MSA= metropolitanstatisticalarea. 44

TABLE10: PARTIAL LINEAR REGRESSION ROBUSTNESS TESTS Kernel: Triangular Epanechnikov Bandwidth: 10mi 8mi 10mi 8mi Judicialforeclosure -0.068 -0.071 -0.067 -0.072 (0.004) (0.005) (0.004) (0.005) Statutoryrightofredemption 0.017 0.022 0.015 0.019 (0.008) (0.010) (0.008) (0.008) Deficiencyjudgmentprohibited 0.052 0.055 0.050 0.054 (0.009) (0.010) (0.008) (0.009) NOTE. Thedependentvariableinallregressionsis (cid:18)(cid:19) (cid:20) (loansize). Bootstrappedstandarderrorsbasedon200replicatesareshowninparentheses. TABLE11: MOREPARTIAL LINEAR REGRESSION ROBUSTNESS TESTS Kernel: Triangular Bandwidth: 10mi (cid:10) (cid:14)(cid:15) (cid:0) (cid:16) (cid:2) area (cid:10) (cid:17) (cid:9) (cid:1) (cid:10) (cid:14)(cid:15) (cid:0) (cid:21) (cid:2) area (cid:10) (cid:17) (cid:9) (cid:1) CVAL Judicialforeclosure -0.068 -0.053 -0.054 -0.085 (0.004) (0.005) (0.006) (0.005) Statutoryrightofredemption 0.017 -0.038 -0.040 0.038 (0.008) (0.011) (0.014) (0.009) Deficiencyjudgmentprohibited 0.052 0.10 0.082 0.049 (0.009) (0.017) (0.021) (0.009) NOTE. Thedependentvariableinallregressionsis (cid:18)(cid:19) (cid:20) (loansize). Bootstrappedstandarderrorsbasedon200replicatesareshowninparentheses. CVAL=optimalcrossvalidationbandwidthusedforeveryvariableintheregression. 45

ssecorP erusolceroF laiciduJ a gniriuqeR setatS - 1 erugiF (cid:0) 045 seliM dnegeL ecneP neraK redroB etatS swaL ’setatS spuorG ASM .9991 ,sinneD ,ikswoknaJ :ecruoS egagtroM lanoitaN ehT seirtnuoC rehtO ssecorp laiciduj eriuqeR puorg nihtiw reffid swaL .yrotceriD ecnerefeR s’recivreS .aera lacitsitats natiloportem=ASM ssecorp laiciduj eriuqer ton oD puorg nihtiw eerga swaL

noitpmedeR fo thgiR yrotutatS a gniriuqeR setatS - 2 erugiF (cid:0) 017 seliM dnegeL ecneP neraK redroB etatS swaL ’setatS spuorG ASM .9991 ,sinneD ,ikswoknaJ :ecruoS egagtroM lanoitaN ehT seirtnuoC rehtO stsixE noitpmedeR fo thgiR puorg nihtiw eerga swaL .yrotceriD ecnerefeR s’recivreS .aera lacitsitats natiloportem = ASM noitpmedeR fo thgiR oN puorg nihtiw eergasid swaL

stnemgduJ ycneicifeD gnitibihorP setatS - 3 erugiF (cid:0) 007 seliM dnegeL ecneP neraK redroB etatS swaL ’setatS spuorG ASM .9991 ,sinneD ,ikswoknaJ :ecruoS egagtroM lanoitaN ehT seirtnuoC rehtO tnemgduJ ycneicifeD tibihorP puorg nihtiw eergasid swaL .yrotceriD ecnerefeR s’recivreS .aera lacitsitats natiloportem = ASM tnemgduJ ycneicifeD wollA puorg nihtiw eerga swaL

Figure 4 - Alpha Function from the Partial Linear Model, Northeastern United States PPoorrttssmmoouutthh --66..99%% FFiittcchhbbuurrgg --77..4444%% BBoossttoonn HHaarrttffoorrdd --11%% --1122..55%% SSttaammffoorrdd 22..33%% LLaannccaasstteerr 66..44%% DDCC 11%% Change in the OFHEO Legend House Price Index, 1990:Q4 - 1994:Q4 90 (cid:0) Miles State Border -2 to -1 Std. Dev. Alpha -1 to 0 Std. Dev. Karen Pence Less than -3 Std. Dev. Source: Home Mortgage Disclosure Act data, 0 to 1 Std. Dev. Census. OFHEO=Office of Federal Housing -3 to -2 Std. Dev. 1 to 2 Std. Dev. Enterprise Oversight