Preventive Servicing Is Good for Business and Affordable ...
Transcript of Preventive Servicing Is Good for Business and Affordable ...
243
HOUSING POLICY DEBATE VOLUME 18 ISSUE 2© 2007 METROPOLITAN INSTITUTE AT VIRGINIA TECH. ALL RIGHTS RESERVED.
Preventive Servicing Is Good for Business and Affordable Homeownership Policy Michael A. StegmanJohn D. and Catherine T. MacArthur Foundation
Roberto G. Quercia, Janneke Ratcliffe, and Lei DingUniversity of North Carolina at Chapel Hill
Walter R. DavisStatistics New Zealand
AbstractThis article documents the growing importance of preventive servic-
ing—businesspracticesthatemphasizeearlyinterventionindelinquencyanddefaultmanagementpracticesthatalsohelpfinanciallytroubledborrowersavoidforeclosure.Wesuggestthattheloanservicingsideoftheaffordablehousingdeliverysystemmaybeunderappreciatedandundercapitalized.
Weuseadatabaseofmorethan28,000affordablehousingloanstotestseveralpreventiveservicing–relatedpropositionsandfindthatafterwecon-trolforloanandborrowercharacteristics,thelikelihoodthatadelinquentmortgagor within this universe will ultimately default varies significantlyacrossservicers.Thissuggeststhatloanservicingisanimportantfactorindeterminingwhetherlow-andmoderate-incomeborrowerswhofallbehindin their mortgage payments will end up losing their homes through fore-closure.Italsosuggestsaneedforpolicymakerstoincorporatepreventiveservicingintoaffordablehomeownershipprograms.
Keywords:Affordability;Defaults;Mortgageservicing
IntroductionFavorabledemographics,unparalleledeconomicgrowth,andlowinter-
estratescombinedwithanaggressivelysupportivepublicpolicypropelledhomeownershipratesupwardduringthe1990stoahistorichighof69per-cent in2004,with“householdsofallages, incomes, racesandethnicities
HOUSING POLICY DEBATE
244 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
joininginthehomebuyingboom”(JointCenterforHousingStudies2005,1). Home Mortgage Disclosure Act records verify an impressive increasein mortgages to low- and moderate-income families—38 percent between1993and1997,comparedwithanincreaseof27percentforloanstohigher-incomehouseholds(Quercia,McCarthy,andWachter2003).Thisgrowthinmortgagestolow-andmoderate-incomehouseholdswassupportedbythewidespreadintroductionof“affordable”mortgageproductsfeaturingflex-ibleunderwriting—including lowdownpayments,higherdebt ratios,andreducedcashreserves—combinedwith theuseofnontraditionalmeansofverifyingcreditworthiness(Querciaetal.2002).
Whatarobusteconomyandliberalizedunderwritingmakepossible,aneconomyindeclinecandiminish.Asthe2001recessiontookholdandwasfollowedbya“job-loss”recovery,thenumberoffinanciallystressedfamiliesgrew,andthenumberofhomeownersspending50percentormoreoftheirincomeformortgage,insurance,andutilitiesincreasedby1.2million(78per-cent)between1997and2003(Lipman2005).Suchhighhousingcostsinevi-tablyleadtomorevolatilepaymentpatternsforlow-andmoderate-incomehomeowners,whosehigherlevelofdebttoincomemakesthemmoresensi-tivethanhomeownersingeneraltoevenmodestdeclinesinthedemandforlabor.Thispartlyexplainswhydefaultratesformoreflexiblyunderwritten,higherloan-to-valueratio(LTV)government-backedloanstendtobesignifi-cantlyhigherthantheyareforconventionalmortgagesasawhole.1Areviewoftheliteratureonthewell-establishedrelationshipbetweenLTVsandpro-pensitytodefaultfoundthat“defaultdependsclearlyonloancharacteristics,mainlyhomeequity”(QuerciaandStegman1992,376).
Efforts to increase the homeownership rate by increasing the numberof new mortgagors can be successful only to the extent that these newhomeownerscanmaintaintheirtenure.AccordingtoarecentU.S.Depart-mentofHousingandUrbanDevelopment(HUD)study,“[P]oliciesthatpro-moteonlytemporaryspellsofhomeownershipwillhavelittleimpactonthenationalhomeownershiprate.Whatisimportantispromotingnewowner-shipspellsthataresustainable”(2004,v).
Ifthe1990swerethedecadeofaffordablehomeownership,thencurrenteconomicrealitiessuggestthatthe“newfrontierinsuccessfullendinginlow-andmoderate-incomecommunities justmaybe inpost-purchaseservices”(Meyer1997,4).Theseservicesincludecollectionsanddefaultmanagement
1Forexample,from1993to2000,therateofforeclosuresforFederalHousingAdminis-tration(FHA)loanswasatleasttwicethatofconventionalloansandthe90-daypast-dueratewasthreetosixtimeshigherforFHAloansthanforconventionalones(HUD2003).
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 245
strategiesdesignedtokeepdelinquenthomeownersfromlosingtheirhomesthroughforeclosure.Thisarticlesuggeststhatwiththeproliferationofafford-ablelendingproducts,whichlayermortgagedefaultrisks,theservicingsideoftheaffordablehousingsystemhasbecomemorecriticalthanever.
Afterdiscussingwhatwemeanbypreventiveandsmart servicing,wewillreviewtheeconomicsoftheloanservicingbusiness.Then,usingadata-baseofmorethan28,000affordablehomeloansthatwearecollectingaspartofamultiyearevaluationofasecondarymortgagemarketdemonstra-tionundertakenbySelf-Help,a leadingcommunitydevelopmentfinancialinstitution,weempiricallytestseveralservicing-relatedpropositions:
1.That affordable mortgages can have a high likelihood of recoveryfromearly-stagedelinquency
2.Thatcertainborrower,loan,property,andpaymenthistoryattributesaffectcurerates(curemeansthattheborrowermakesupthemissedpayments)
3.Thatloanservicingplaysaroleindeterminingtheultimateoutcomeofdelinquencyspells
Preventive and smart servicing Inthisarticle,weusetheterm“preventiveservicing”todescribedelin-
quencymanagementpracticesthatemphasizeearlyinterventionanddefaultmanagementpracticesthathelpfinanciallytroubledborrowersavoidfore-closure(Oliveretal.2001).2“Smartservicingtools”meanstechnologiesthatenablepreventiveservicing,including“modelingsoftwareandscriptedmenuoptionstoengagecollectorandborrower”(Fields2001,27). Intheeffec-tivemanagementofnonperformingloans,collectionseffortsaresupportedbysmartsystemsthatprioritizecollectionscallsandidentifylossmitigationpathwaysappropriatetoeachcase.
Thekeytoeffectiveandefficientpreventiveservicingisinknowingwhichborrowerstocontactatwhichtime,andsuccessinavoidingforeclosureoftendepends on early intervention. For example, in a 2001 review of FederalHousingAdministration(FHA) loans,oneservicerhadasuccessrate thatexceeded45percentforworkoutsprocessedwithinthefirsttwomonthsofdelinquency,butonlya10percentsuccessrateiftheworkoutwasprocessed
2Forafulldiscussionofalternativestoforeclosure,seeCuttsandGreen(2005).
HOUSING POLICY DEBATE
246 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
aftersevenmonths(HUD2002).Deloitte,theconsultingfirm,estimatesthatorganizationsthatadoptwhatitreferstoaspredictiveanalyticsandotherpreventiveservicingpractices“canenjoysubstantialbenefitsincludinga20percentimprovementinthedollarscollected,reductionsofupto30percentinoveralloperatingcosts,anddeclinesofupto20percentinrollrates(oraccounts rolling to the next stage of delinquency)” (Caldwell and Jordan2006,2).However,engagingadelinquentmortgagorquicklyissometimeseasiersaidthandone.Asacaseinpoint,whenCountrywideMortgagetriedtohelp300seriouslydelinquentChicago-areahomeownersavoid foreclo-sure,only38responded(Sichelman2001).Frustratedbysuchlowresponserates,thecompanysenta“teamoflossmitigationspecialists”tomeetwith61borrowers,and“despitethefactthat32ofthesecustomerswereinthelatestagesofforeclosure,Countrywidewasabletocreaterepaymentpro-gramsfor59oftheresidents”(Mozilo2000,18),dramaticallyunderscoringthepowerofpreventiveservicing.
Affordable mortgages, smart servicing, and mortgage economics Lendingprogramsdesignedtoincreasethehomeownershiprateforlow-
andmoderate-incomeborrowersthroughliberalizedunderwritingguidelinesalsoincreasethelikelihoodthatthosehomeownerswilldefault.Beginninginthelate1980sandcontinuingthroughthe1990s,lenders,mortgageinsur-ers,andsecondarymarketingagenciesrespondedtopolicygoalsandregula-torypressuresbyrelaxingtheguidelinesthatsometimesbarredlower-incomehouseholdsfromqualifyingforamortgage.Theseguidelinesevolvedtomin-imizetherequireddownpayment,reducetheborrower’sassetandcontribu-tionrequirements,raisetheallowablemonthlydebtpaymentsatparticularincomelevels,andloosencreditworthinessstandards.Whilethesefeaturesallowmorehouseholds tobuyhomes, theycan increase the riskofdelin-quencyanddefaultby40to400percent(Steinbach1995).Becauseoftheincreasedincidenceofdefaultandbecauselow-andmoderate-incomebor-rowerstypicallyhavelittlefinancialcushiontofallbackonintheeventofaneconomicsetback(BakuandSmith1998),callsto“implementhomebuyersupportstructuresdesignedtoensurelong-termhomeownership”pointthewaytopreventiveservicingmeasuresforthesehomeowners(Steinbach1995,39).
The smart servicing tools now in use canbe particularly effective fortheseborrowers.Thecombinationofearlyintervention,thetargetingofbor-rowersmostatriskofdefault,andworkoutoptionsthatallowborrowerstorecoverfromfinancialsetbacksallhelpservicersfocusresourcesonthose
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 247
loansthatcanbenefitthemostfromintervention.Ineffect,flexibleguidelinesandpreventiveservicinghavethesameultimateaim:tobolstertheranksofhomeowners.
Becauseofthehighcostofdefaults,smartandpreventiveservicinggen-erallymakesgoodbusinesssenseforanymortgage.Forexample,theFHA-expectedlossperdefaultedloanrangesfrom30percentto40percentoftheloanamount,3 thussuggestingthatavoidingforeclosurecansaveasmuchas$60,000to$80,000onthemaximumFHAloaninstandard-costareas.Thoughmostconventionaloraffordablemortgagedefaultsdonotend inforeclosure,thosethatdocanbequitecostly.Forexample,CuttsandGreenciteFocardi’sfindingsthatforasampleofloansthatwentthroughthefullformalforeclosureprocess,thetotalcost,includinglostinterestduringdelin-quency, foreclosure costs, and disposition of the foreclosed property, ran$58,759andtookanaverageof18monthstoresolve,whilevoluntarytitletransferalternatives toforeclosurehadaveragecosts inexcessof$44,000andtooknearlyoneyeartoconclude(2005,352).
Ina1996article,AmbroseandCaponeillustratedhowinrisingmarkets“savingstothemortgageinvestorsandinsurersfrompreventingoneforeclo-sure(asuccessfulworkout)arelargeenoughtopaytheaddedcostsoffourworkouts”andthateveninsoftermarketswherepricesaredecliningbyasmuchas5percentayear,onesuccessfulworkoutcouldoffset thecostofmorethantwoadditionalworkouts(citedinCaponeandMetz2003,10).Sinceaffordablehousingmortgageswiththeirtypicallysmallerequitycush-ionsaremorelikelytodefault4andtoresultingreaterlossseverityfortheinvestororinsurer,thiscostbenefitequationshouldbeevenmorecompellingfortheaffordablesegment.
Thecostsofdefaultarelargelybornebytheholdersofcreditordefaultrisk (e.g., Fannie Mae, Freddie Mac, FHA, and mortgage insurers); thus,theystandtobenefitthemostfromtheuseofsmartandpreventiveservicing.Becausetheservicerhastherelationshipwiththeborrower,however,inves-torsandinsurersmustlooktothatentitytoimplementtheseloss-minimiz-ingpractices,whichareoftenlaborintensiveorrequireinvestmentsinnewtechnologiesandtraining—orboth.
3Forexample, the2005analysisofFHA’sMutualMortgage InsuranceFund indicatesthattheaveragelossseverityrateexperiencedbyFHAfrom2000to2004rangedfrom30.59percentonstreamlinedrefinanceadjustable-ratemortgagesto48.76percenton15-yearnon-streamlinedrefinancefixed-ratemortgages.From2002to2003,theaveragelossrateperclaimwas35percent(HUD2005).
4Forexample,amortgageinsurancecompanyfoundthatloanswithdownpaymentsof3percentoftheborrower’sownfundshadtwicethedefaultrateofloanswitha5percentdownpayment(Steinbach1995).
HOUSING POLICY DEBATE
248 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
Servicer economics reliesonmaximizingefficiency,withconsolidationdrivenbytechnologyandeconomiesofscale.Between1989and2005,theshareofthemarketheldbythe5largestservicerswentfrom7percentto42percent,withthe25largestholding69percentofthemarket(InsideMort-gageFinance2006). Inthedrivetotakeadvantageofeconomiesofscale,productivitygainedsignificantly,fromaround700loansservicedperdirectfull-timeemployeein1992to1,229in2000,anincreaseofabout60percentinlessthan10years.Directservicingcostsperloanperyear,whichwere$80to$90inthelate1980s,wereshavedtolessthan$50in2000(Oliveretal.2001).
Ontherevenuesideoftheequation,thelargestsinglesourceofrevenueisservicingfees.Thestandardfeeforconventionalconforminghomeloansisonequarterof1percent(25basispoints)oftheoutstandingbalanceoftheloanannually.Ongovernment-backedloans,thefeeisgenerally44basispointsannually(Muolo2000),areflectionofloweraverageloanbalance,morereportingrequirements,andhighercostsassociatedwithservicingFHAandU.S.DepartmentofVeteransAffairs(VA)loans.
Some previous research indicates that servicing methods can make adifferenceinoutcomesforsubprimeborrowers.Alargeservicerexamined23,000 delinquency transitions over nearly four years—from current todelinquent,fromdelinquenttoforeclosureortocure,andfromforeclosuretocure—insubprime loansfromseveraldifferentservicers (Sjaastadetal.2005).Theanalysisvalidates thepredictivenessof such factorsas currentFICO(Fair,Isaac&Company)score,LTV,loanage,andpaymenthistoryforeachtransitiontype,yetaftercontrollingforthesefactors,theauthorscon-cludethat“servicerscanhavespecific,strongimpactsontransitionratesanddelinquencyduration times” in subprimemortgages (Sjaastadetal.2005,16).Thisservicereffectseemsreasonable,foroncealoanisclosed,theser-vicerhasanongoingrelationshipwiththeborrowerandisinapositiontoinfluenceoutcomes.
Toillustrate,whenaloanreaches60or90daysdelinquent,oneservicerpromptlyreferstheloantoaforeclosureattorney,asstatelawallows,andinsists that the borrower pay full arrearages and fees to halt foreclosure,somethingafinanciallytroubledborrowerusuallycannotdo.Anotherser-vicercontactstheborrower,exercisesforbearance,acceptspartialpayments,andactivelyhelpsformulateanalternativetoforeclosure.Inthefirstcase,thedelinquentloanquicklytransitionstoforeclosure,whileinthesecond,theborrowerremainsintheearly-delinquencystageuntileithercatchingupordefaulting severalmonths later.Delinquent loan servicingpractices aredifferentiatedbysuchfactorsasallocationofresources(bothstaffandtech-
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 249
nology),theprocessinplacefornonperformingloans,investorpreference,andorganizationalcommitmenttolossmitigation,manyofwhicharecon-sideredinthisarticle.
Our literaturereviewfoundfewempiricalstudiesof theeconomicsofservicing affordable mortgage portfolios—those characterized by below-average loan balances and disproportionately large numbers of low- andmoderate-incomeborrowers—but there are scattered indications from theresearchandourexaminationoftheSelf-Helpportfolioofthespecialchal-lengesposedbythismarketsegment.
Over 30 years ago, the relationship between average loan size anddiscounted, net servicing revenue was found to be positive and nonlinear(McConnell1976). In2001, theaverageconventionalmortgage loanbal-ancewas$120,393(HMDA2001).TheaverageloanbalanceinSelf-Help’sprogram is substantially lower: $79,800. Thus, the annual servicing feesgeneratedontheaverageSelf-Helploanare34percentlowerthanthefeesfortheaverageconventionalloan($200versus$301).However,latefeesgener-atedby the increased incidenceofdelinquencymayoffset thisdifferentialsomewhat.5Morerecently,LindaSimmons,aStratmorGrouppartner,ana-lyzedalargepoolofmortgagedatafrom1999to2002andfound“astrongrelationship between portfolio composition and servicing costs” (DeZube2003,48).TheStratmorGroupfoundthatconventionalconformingloanscostanaverageof$48ayearforcoreservicingoverthestudyperiod,“whilegovernment-backedloanscameinat$98”(DeZube2003,48).Borrowerswho have affordable conventional loans have loan sizes and risk profilesmore like thoseofgovernment-backedborrowersandmay thereforehavesimilareconomics.
Onthepositiveside,affordablemortgagesmayexhibitslowerandlessvolatileprepaymentpatterns.Intheiranalysisoftheprepaymentratesasso-ciatedwithpoolsofFHA-insuredloans,DengandGabriel(2002)foundthatborrowerswithlowercreditscoresandhigherdefaultriskswerelesslikelytoprepaythanborrowerswithbettercredit.BytrackingFreddieMacloanspurchasedfrom1993to1997throughtheendof1999,VanOrderandZorn(2004) found that low-income and minority borrowers prepay mortgageloans less rapidlywhenconditionsare favorable forrefinancing. If indeedlow-andmoderate-incomemortgagesprepaymoreslowlythanothertypesofconventionalmortgages,theirservicingrevenuestreamwillbelongerandpotentiallymorevaluable.
5Latefees, incurredforpaymentsreceivedafterthe15thofthemonth,aretypically3percent to5percentof thepayment (sources includeCaliforniaRealEstateCenter.com2006andMortgageNewsDaily.com2005).
HOUSING POLICY DEBATE
250 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
Tosummarize,thebusinesschallengesinservicingmortgagesoriginatedlargelyforCommunityReinvestmentAct(CRA)creditarecompoundedbytheirsmallersizeandhigherdelinquencyanddefaultrates.Althoughthesechallengesmaybeoffsettosomeextentbyslowerprepaymentpatterns,wearguethatthemortgageservicingindustryneedstoimplementthemosteffi-cientdefaultmanagement systemspossible tokeep costsdownwhile stillmeetinginvestorrequirementstominimizelosses.Webelievethatconsistentwith thepublicpolicygoalof sustaininghomeownership, investor/insurerandservicergoalsmightbealignedinthedevelopmentofsmartservicingtools.
The evolution of smart servicing tools6
Advancedinformationtechnologieswerefirstappliedtocollectionman-agementinthecreditcardindustryduringtheearly1990sasthepercent-ageofthosemakingminimumornomonthlypaymentsswelledalongwithaggregatedebtlevels.Thissectorwasafertiletestinggroundfordecisioningsystems that analyze thebehaviorpatternsof delinquent cardholders anddeterminewhois least likelytopaywithoutearlyintervention(Waggoner2002).
In a parallel fashion, first-generation automated mortgage servicingmodels,introducedinthemid-1990s,identifydelinquentmortgageaccountsthatarelikelytobenefitmostfromearlyintervention.Thesetools,suchasFreddieMac’sEarlyIndicatorSM(EI)andFannieMae’sRiskProfiler®(RP),useacombinationofborrowerinformationcontainedinthemortgageappli-cation, loan-specificpaymentpatterns,andupdatedcredit informationontheborrower(FannieMae2005a;FreddieMacn.d.).Personalcontactwiththeborrowerisnotrequiredtoproduceapropensityscore(Stanton2001).Thisapproach,whichoptimizesservicers’collectionbudgetsbyprioritizingcallschedules,isonlythefirstpartofsmartservicingsystems.
The second phase applies a combination of scripting systems such asFreddieMac’sEarlyResolutiontoaidservicerswhentheycontactborrowersandanalytic toolssuchasGE’smortgage insurancecompany’sLossMiti-gationOptimizer, FreddieMac’sWorkoutProspector®, andFannieMae’sWorkoutProfilerTMtoassesstheviabilityofforeclosurealternativesineachcase.Thesetoolsessentiallyapplyconsistentunderwritingmethodologytokeepborrowersintheirhomesbyassessingwhatoneindustryexpertreferstoasthe“threeCs”—capacity,collateral,andcommitment(Kehr2005).
6ThissectionisbasedlargelyonadraftpreparedforthisprojectbyStevenHornburg.
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 251
Because investorsand insurersstandtogain themost fromaggressivelossmitigation,theyoftenservedascatalystsfortheinitialintroductionofutilitiesandtechnologyinthisareaandinvestedheavilyinthem.Today,suchtoolsareincreasinglybeingdevelopedorenhancedbythird-partysoftwarecompaniesandarebeingintegratedmoredirectlyintotheservicers’systems(Kehr2005).Whilemanyofthesetoolsgohandinhand,theycanbeusedindependentlyandwithmortgagesofmanydifferenttypes.
Havinglauncheditsownlossmitigationprogramin1996,FHAiscom-mittedtopreventiveservicingaswell.By2005,FHAofficialsaffirmedtheirbeliefthatthelossmitigationprogram“isbecomingthedominantapproachtoaddressadefault”(“FHACites90,000LossMitCases”2005,91).Thecost-effectivenessofpreventiveservicingisevident.In2002,FHApaidout$5.5 billion in claims on 64,000 foreclosures, while paying just $98 mil-liontohelpkeep73,000financiallytroubledborrowersintheirhomeswithrecastloans(Harney2002).By2005,thenumberoflossmitigationclaimspaid rose to90,000,withFHAavoidinganestimated$2billion in losses(“FHACites90,000LossMitCases”2005).
Instead of developing its own tools, FHA relies on servicers to applycommerciallyavailable smart servicing tools to its loans. Italso institutedasystemofscoringservicersandawardsfinancial incentives—$27millionin2005—foroutstandinglossmitigationperformance(“FHACites90,000LossMitCases”2005.FHAisnottheonlyinvestortoscoreorpayservicersforusingalternativestoforeclosure.FreddieMac’sServicerPerformancePro-filesusesuchindicatorsascurerates,workouts,andtimelinessofforeclosureto benchmark servicers (Melchiorre 1999). One servicer reported earning$25,000permonthfromlossmitigationincentives(O’Connor2003).Finan-cialincentiveshaveledsomeservicerstoseetheir“lossmitigationdepart-mentsmovefromcostcenterstoprofitcenters”(HUD2002,3).
Evidently, smartandpreventiveservicinghasbeenfinanciallyeffectivefor investors and insurers and even for servicers.Buthoweffectivehas itbeeninsustaininghomeownership?Throughaggressiveoutreachfeaturingacombinationofpreventiveandsmartservicing,FannieMaeanditsloanservicingpartnershelpednearly34,000financiallystrappedborrowersavoidforeclosurein2004alone;accordingtoKehr(2005),morethan92percentreceivedworkoutsthatallowedthemtostayintheirhomes.7Thesuccessoflossmitigationeffortsisoftengaugedbytheworkoutratio,whichgenerally
7Notallworkoutsresultinretainingthehome;otherstrategiesusedtoavoidforeclosureinvolvetheborrower’sgivinguporsellingthehomevoluntarily,suchasinthecaseofdeeds-in-lieuandpreforeclosuresales(Cornwell2003).
HOUSING POLICY DEBATE
252 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
comparesthenumberofloansthatreceivesuccessfulworkoutswiththetotalnumberofloansthateithergetworkoutsorbecomerealestateowned(whenthetitleistransferredtothelenderthroughforeclosureorinlieuofforeclo-sure).Detailsofthismeasurecanvary;forexample,FreddieMacexcludesrepaymentplansintallyingworkouts,butFannieMaeincludesthem.
From1996to2000,FannieMae’sworkoutratiorosefrom32percentto53percent,whileFreddieMac’srosefrom26percentto38percent(Cordell2001;Cornwell2003).FHA’sworkoutrationearlydoubledfromlessthan15percentinlate1997toalmost30percentattheendof1999(Herbert,Gruenstein, and Burnett 2000) and over 50 percent by 2002 (Cutts andGreen2005).Moststrikingly,thenumberofFHAproblemloansthatwereresolvedusingahomeretentionworkoutincreasedsome88times,from770in1997to68,003in2003.AfterintroducingitsLossMitigationOptimizer,GE’smortgage insurancecompanyreportedalmostdoublingitscureratio—theshareofallloansforwhichthecompanyreceivesnotificationofdelin-quencybutthatlaterbecomecurrent(Venetis2001).8Afterimplementingacollectionscoringtool,oneservicerreporteda50percentincreaseinFHAandVAworkouts (Abraham1999); another claims thatworkouts tripledin the initialperiodafter a“BackInTheBlack” lossmitigation systemwasimplemented(O’Connor2003).From2000to2004,145,000FreddieMacborrowers,or“130familieseverybusinessday,”werekeptintheirhomeswithloanmodifications,repaymentplans,andforbearance(“Now,‘Afford-ableServicing’”2005,1).
PerhapstheclearestevidencecomesfromCuttsandGreen’s2005exami-nationofFreddieMac–ownedloansthatwentfrom60to120daysdelin-quentbetweenJanuaryandSeptember2001.Followingtheseloansfor18months,CuttsandGreen found thatusinga repaymentplan lowered theprobabilityoffailureby80percentforborrowerswithhigherincomesandby68percentforlow-andmoderate-incomeborrowers.
Clearly,then,thesetechnologiesandtoolshavehadanimpressiveimpactonthewaynonperformingmortgagesareservicedandonservicers’abilitytocost-effectivelykeepmoreborrowersintheirhomes.
Servicing issues and Self-Help data Inthissection,weuseauniquedatasetofmorethan28,000mortgages
madetolow-andmoderate-incomeborrowerstoexploretheimpactofvari-
8Notificationtypicallycomesat90days,however,itcouldbesooner,forexample,inthecaseofmonthlypaypolicies.
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 253
ousfactorsondelinquencytransitions(fromdelinquenttoseriouslydelin-quentortocure)andtoisolatetheeffectsofservicing.
TheSelf-Helpsecondarymarketdemonstrationprogramisamultibil-lion-dollar initiativedesignedtoexpandhomeownershipopportunities forcreditworthy, low-income, low-wealth individuals who are not effectivelyservedbytheconventionalmarket.TheprogramisapartnershipamongtheFordFoundation;theCenterforCommunitySelf-Help,aNorthCarolina-basedcommunitydevelopmentorganization;andFannieMae,asecondarymarketentity.Thegoaloftheprogramistoprovidetangibleevidencethatlow-wealthborrowersare“bankable”andthatsecondarymarketactorscansignificantlyexpandtheirpurchaseofaffordablehousingloanswithoutcom-promisingeithertheirbalancesheetortheirsafetyandsoundness.
WithaFordFoundationgranttounderwriteasignificantportionofthecreditrisk,Self-Helppurchasesaffordablemortgages(CRAloans)thatcouldnototherwisebereadilysoldinthesecondarymarketbecauseoftheirper-ceivedhigherrisk.Self-HelpthensellstheloanstoFannieMaewhileretain-ingfullrecourse,meaningthatSelf-HelpisobligedtocoveranylossesFannieMaemayincurasaresultofdefaults.Self-Helpcontractswithparticipatingmortgagelenderstooriginateandsubservicetheloans.
Self-Help mortgage products feature flexible underwriting and typi-cally includeoneormoreof the followingfeatures:a lowdownpaymentornoneatall,lowerreserves,higherdebt-to-incomeratios,borrowerswithspottycreditrecordsornoestablishedcredit,andwaiverofprivatemortgageinsurance.
Dataforouranalysiscomefrom28,131loansthatweremadetolow-andmoderate-incomemortgagorsandpurchasedbySelf-HelpbeforeJanu-ary1,2003;theaverageoriginalbalancewas$79,800.Foreachloan,thedataincludeafullsetofloancharacteristics,someborrowercharacteristics,andthecompletepaymenthistoryfromthetimeofpurchasebySelf-Help.9
ThedatabaseisbeingcompiledaspartofourongoingmultiyearevaluationofSelf-Help’sprogramwiththeaimof
1.Measuring loan performance by tracking delinquency, default, andforeclosureratesoverthecriticalfirstfiveyearsofthemortgageterm
9Asmallnumberofmonthsaremissingfromthepaymenthistoryfile.Wehavedataon580,276loan-months,with532loan-monthsmissing.Weuseanalgorithmtofillinthelikelyvaluesforthemissingmonths.Forexample,ifmonth1waspaidontimeandmonth3waspaidontime,weassumethatmonth2wasalsopaidontime.Similarly,ifmonth1waspaidontimebutmonth3showsa60-daydelinquency,weassumethatmonth2waslate.Inthefewinstanceswhereaprecisedeterminationwasnotpossible,weimputedmonthssuchthatanydelinquencyspellwouldbeasshortaspossible.
HOUSING POLICY DEBATE
254 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
2.Documentingthesocialandeconomicimpactsofhomeownershiponborrowers
3.Assessing to the extent practicable the impacts of the program onneighborhoodconditionsandhousingmarketdynamics
Aspartoftheevaluation,wearebuildingapanelofover2,700hom-eowners(asof2005)whoarebeinginterviewedinsixannualwavesrunningthrough2008.Someofthedescriptivedatacitedlaterinthearticlearefromourbaselinephoneinterviews,whichwereconductedbetweenlate2001and2003.Thisbaselinesurveygatheredinformationonthehomepurchaseandmortgageprocess,alongwithdemographicandeconomicdata.Coreques-tionsarerepeatedeveryyear,whiledifferentmodules(socialcapital,wealthandassets,financialbehaviorsandattitudes,senseofcommunity,andmoversurvey)areusedindifferentyears.
Who are these homeowners?
DemographicandothercharacteristicsoftheborrowersAlmost40percentofSelf-Helpborrowersareminoritiesand44percent
arewomen.Anestimated16percentaresingleparents.10Itisimportanttonotethathalfhadincomesbelow60percentofthelocalmedianhouseholdincomewhentheyclosedontheirloans(table1).Moreover,alittleover40percentofallSelf-HelpfamilieshadFICOscoresoflessthan660ornocreditscoreatall.ComparedwithborrowerswhoseconformingloanswereboughtbyFannieMaein1999,Self-Helphomeownersareabouttwiceaslikelytobeminorities,almostfivetimesaslikelytohaveincomesbelow60percentoftheareamedianincome,asdefinedbyHUD,andtwoandahalftimesaslikelytobefemale(CenterforCommunitySelf-Helpn.d.).
Datafromourbaselinesurveyof3,727Self-Helphomeowners,whichoccurredgenerallyaround12to24monthsafterhomepurchase,allowustolookatthefamilyandemploymentsituationsoftheseborrowersingreaterdetail.Asarule,Self-Helpfamilieshaveastrongcommitmenttotheworkforce; more than three-quarters of all sampled households with marriedcouplesorpartnershavetwowageearners.Moreover,16percentofallbor-rowersand10percentofallspousesorpartnerswhoworkhavemorethanonejob,includingpart-time,weekend,andeveningwork,and96percentofoursurveyedborrowersworkatleastfull-time(35ormorehoursperweek).Almost30percentaverageatleast50hoursofpaidworkperweek.
10ThisfigureisanestimatefortheentirepopulationofSelf-Helpborrowersbasedonthebaselinepanelsurvey.
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 255
Table 1. Self-HelpLoans:DescriptiveStatisticsforLoansPurchasedbeforeJanuary1,2003(N=28,131)
Variable Percentage Median
Creditscore Nocreditscore 3.5 FICO620orlower 17.2 FICO621to660 21.0 FICO661to720 30.4 FICOabove720 28.0
Loancharacteristics LTV(%) 97.0 Back-endratio(%) 36.7 Ageofloanatpurchase(months) 11
Borrowercharacteristics Femaleborrower 44.0 Blackborrower 21.3 Hispanicborrower 16.6 Singleparenta 15.9 First-timehomebuyer 43.8 Incomeatorigination $29,472 IncomeasapercentageofAMI 59.7
Postpurchaseemploymenthistorya Currentunemployment(borrower) 3.1 Currentunemployment(spouse) 8.3 Currentlymorethanonejob(borrower) 16.1 Currentlymorethanonejob(spouse) 10.4 Anyunemploymentsincepurchase(borrower) 16.8 Anyunemploymentsincepurchase(spouse) 59.3 Currentlyworksfull-time(borrower) 95.9 Currentlyworksmorethan50hoursperweek(borrower) 27.8 Two-wage-earnerhouseholds 75.9 (amongmarried/partneredhouseholds)
Geography Rural 20.5 NorthCarolinaborrower 40.1 Californiaborrower 12.9 Oklahomaborrower 7.0 SouthCarolinaborrower 6.0 Ohioborrower 4.3 Virginiaborrower 3.9 Georgiaborrower 2.7 Texasborrower 2.6 Floridaborrower 2.5 Illinoisborrower 2.3 Otherstates 15.6
Source:Self-Helpdatabase,panelsurvey,andauthors’calculations.aThisestimateisbasedondatacollectedaspartofthebaselinepanelsurvey,afive-yearstudyofover3,700Self-Helpborrowers.AMI=areamedianincome(medianincomeofthemetropolitanstatisticalarea[MSA]orthestatefornon-MSAareas).
HOUSING POLICY DEBATE
256 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
Despitethefactthatourinterviewswereconductedinthemidstoftheeconomicslowdown,wefoundlowunemploymentratesamongallborrow-ers(3.1percent),althoughmuchlargernumbersofborrowers(16.8percent)andtheirspousesorpartners(59.3percent)hadexperiencedatleastonespellofjoblessnesslastingaweekormoresinceclosingontheirloans(table1).11
Thephenomenonofhighlabormarketvolatilityandjobchurningamonglow-andmoderate-incomeborrowersduringperiodsofeconomicdecline,characterizedbycutbacksinovertimeandthelossofsecondjobsandpaidspousalwork,contributestohigherratesoflateandmissedmortgagepay-mentsamongalllow-andmoderate-incomefamilies.Thosewithsinglewageearnersareparticularlyaffected.Becausethesefamiliesaremorelikelythanotherstoloseincomeintermittentlyfromperiodicjobcutbacksandlayoffs,theyarealsomorelikelythanhigher-incomeborrowerstofallbehindintheirloanpaymentsandaremorelikelytohaveahardertimecatchingup.Thisputsevenmorepressureonservicerswhoseportfolioscontainlargenumbersofaffordablemortgagesandexplainswhytechnologythatenablesservicerstoidentifywhichborrowersaremostlikelytofallfartherbehindoncetheymissasinglepaymentcanbeavaluabledefaultmanagementtool.
ThedelinquencyexperiencesofSelf-HelpmortgagesAsofMarch2005,about2percentofallloansintheSelf-Helpportfo-
liohad reached foreclosure. In fact, comparedwith industrybenchmarks,Self-Help loans perform quite well. As of the first quarter of 2005, 3.03percentofSelf-Helploansweremorethan90daysdelinquentorinforeclo-sure,comparedwith1.89percentforallloans,5.15percentforFHAloans,and5.23percentforsubprimeloans(MortgageBankersAssociation2005).TakenincombinationwithSelf-Help’shistoricallossseverityrateofjust26percentoftheoriginal loanbalancecomparedwith37percentforsimilarFHAloans(HUD2005),thisperformancesuggeststhatSelf-Help’slossmiti-gationeffortsareeffective.
Consistent with preventive servicing’s emphasis on early intervention,only10percentofloansthatbecameatleast30daysdelinquenteventuallyreached foreclosure, comparedwithalmost30percent thatbecame60ormoredaysdelinquent.Forthosegoingover90days,morethan40percentreachedforeclosure.
Becausethisarticledealswithservicinganddefaultmanagementratherthanwithoriginationsandtheassessmentofinitialcreditrisk,wearemore
11Postpurchasejoblessnessamongspousesincludesthosewhoarenotinthelaborforce.
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 257
interestedinexaminingtheoutcomesofSelf-Helploansoncetheybecome30dayspastduethanwearewithdeterminingthefactorsthatcauseloanstobecomedelinquentinthefirstplace.(TheanalysisofinitialdelinquencyandoverallSelf-HelploanperformanceisthesubjectofotherpapersproducedbytheCenterforCommunityCapitalism[Querciaetal.2002,forexample].)SincethefailurerateforaSelf-Helploanthatisover90daysdelinquentismorethanfourtimeshigherthanitisforonethatis30daysdelinquent,wefocusedonthetransitionswithinthisperiod.Fromthepaymenthistory,wewereabletodeterminewhenaloanbecamedelinquent,generatinga“delin-quencyspell.”BetweenSeptember1998andDecember2004,wetrackedtheperformanceof28,131Self-Help loansoriginatedbeforeDecember2003.Only21.2percent—or5,969loans—everexperiencedadelinquencyspell:15.7 percent experienced only relatively mild delinquency (a delinquencyspellof30or60days),and6.7percentexperiencedatleastonespelllastingatleast90days.
Theservicingchallengeassociatedwithaffordablelendingcanbemoregraphicallyillustratedbyexaminingthenumberofdelinquencyspellsgen-eratedbyeachloan.Astable2shows,alargeshareoftheever-delinquentloans(2,564outof5,969)generatedonlyonedelinquencyspelleach,last-inganaverageof2.5months,beforecuringordefaulting.Theremainderof these loans experienced two or more separate delinquency spells. Thisincludes1,403repeatedlydelinquentloanswithfourormorespells.Whileaccountingforabout5percentofallloans,thisgroupgeneratedoverhalfofalldelinquencyspells.
Adelinquencyspellcanbecuredorprogress toa90-daydelinquency(hereinafterreferredtoas“default”).Sincemostdelinquencyspellsresolve
Table 2.Numberof30-DayDelinquencies
Numberof AverageLength Loans Percentage ofDelinquency
Neverdelinquent 22,162 78.8
Onlyonedelinquency 2,564 9.1 2.50months
Twodelinquencies 1,246 4.4 2.66months
Threedelinquencies 756 2.7 2.57months
Fourdelinquencies 477 1.7 2.73months
Fiveormoredelinquencies 926 3.3 2.46months
Total 28,131 100 2.55months
Source:Self-Helpdatabaseandauthors’calculations.
HOUSING POLICY DEBATE
258 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
12SincesomeloanswereoriginatedafterSeptember1998,theaveragelengthofobserva-tionforallloansisabout53months.
13Becausetheendingdateoftheobservationperiodwasarbitrarilydetermined,thestatusof“still30or60daysdelinquent”wasnotconsideredasanindependentoutcome.Besides,thereareveryfewspellsinthiscategory(2.6percent).
14Tomakesurethatalargenumberofearly-stagedelinquencieswerenotdrivenbypend-ing refinancing plans, we examined the number of delinquencies that ended in repayment.However,only234,orlessthan2percent,ofdelinquencyspellsendedinpayoff.
15Weuse90daysasaproxyfordefaultbecausecompleteddefaultsusuallylagsubstan-tiallyandthedataprovidemoreoccurrencesofdelinquenciesexceeding90days.Themeasure,then,istowhatextentloansmovetothemoreseverecategoryof“seriousdelinquency”ormorethan90dayslate.
16Some90-daydelinquentloanscuredandthensubsequentlybecamedelinquentagain.
onewayortheotherwithinsixmonthsandourobservationperiodlastedmorethansixyears,12nearlyallspells(97.4percent)achievedanoutcomeofeithercureordefaultbeforetheendoftheobservationperiod.Loansthatwerestilldelinquentattheendoftheperiodwereexcludedfromoursam-ple.13Oncealoanwascured,anysubsequentdelinquencygeneratedanewspell.Anobservationwasconsideredcurediftheloanwaspaidoffwhilestill30to60daysdelinquent.14
Asaresult,wehaveasampleof15,038delinquencyspellsthatweregen-eratedby5,886borrowersandendedineithercureordefaultbyDecember2004,foranaverageof2.6delinquencyspellsperloan.Some85percentofthesedelinquencyspellswerecured,while15percentendedindefault.15 The2,200defaultsweregeneratedby1,870loans,16representingjust7percentofallloanstrackedbut31percentofallever-delinquentloans(table2).Theother69percentofever-delinquentloansneverproceededbeyond60dayspastdue.Delinquencyspellsthatcuredordefaultedlastedanaverageof2.1monthsor4.1months,respectively.
Astable3shows,13.4percentoffirst-timedelinquenciesendedindefault,andthisrategenerallyincreasedwiththenumberofpriordelinquencies,to17.6percentofthoseexperiencingafourthdelinquencyspell.Yetsomewhatsurprisingly, the default rate for borrowers experiencing a fifth or higherdelinquency spell dropped to 14.2 percent, nearly identical to the defaultrateamongloansexperiencingafirstspell.Thisfindingsuggestsaninflexionpointabovewhichthenumberofpreviousdelinquenciesexperiencedbyseri-allydelinquentborrowersdoesnotincreasethelikelihoodthatthecurrentspellwillendindefault.Infact,smartservicingtechnologiesseektoidentifythosefrequentlydelinquentborrowerswhoareunlikelytoactuallydefault,savingcollectionsresourcestofocusonother,higher-riskdelinquencies.
We want to determine the extent to which delinquency outcomes areaffectedbytheloanservicer.Theaveragenumberofdelinquencyspellsper
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 259
servicerinthepopulationis537.Eightservicershadinexcessofthisaverage;weclassifiedtheseasthe“mainservicers.”Fouroftheseaccountedformorethan1,000ofthedelinquencyspellseach,whiletheother4servicedbetween537and1,000.Theremaining20servicerseachmanagedlessthantheaver-agenumberofspells;atthefarendofthespectrumwereseveralwhohandledlessthan50.Table4showsthenumberandoutcomesofdelinquencyspellsfortheeightmainSelf-Helpservicers,whichtogethermanaged83percentofthedelinquencyspellswetracked(and76percentofthe5,886loans),andforallotherscombined.Themain8servicerswerenotnecessarilylargerthantheother20;infact,somewerequitesmall.TheysimplyhadalargershareoftheSelf-Helploansthatexperienceddelinquencyspells.
Table4 shows thatServicers2,6,and7exhibit the lowestcure rates(below 80 percent) while Servicers 1, 5, and the “others” category havethehighest (just around90percent).Thesearenaturallyaccompaniedbyhigherandlowerdefaultrates,respectively.InbetweenareServicers3and4,whichfallclosertothelattergroup,andServicer8,whichfallsclosertotheformer.
Mostoftheloanswereservicedbythesameentitythatoriginatedthem.However,thereweretwocaseswhereanoriginatorusedmorethanoneser-vicingoperation.Wefoundthatinoneofthesecases,therewasameasur-abledifferenceincurerates(9.6percentversus12percent)betweenthetwoservicersused.Intheothercase,whereoneofthesubservicershandledone-twentiethasmany loansas theother,bothhadnearly identicalcurerates
Table 3.Percentageof30-DayDelinquenciesThatEndinDefault(byDelinquencyOrder)
Numberof Numberof
Spells Defaults Percentage
Firstdelinquency 5,886 786 13.4
Seconddelinquency 3,327 520 15.6
Thirddelinquency 2,097 319 15.2
Fourthdelinquency 1,345 236 17.6
Fifthorhigherdelinquency 2,383 339 14.2
Total 15,038 2,200 14.6
Source:Self-Helpdatabaseandauthors’calculations.
HOUSING POLICY DEBATE
260 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
Table 4.Outcomesof30-DayDelinquenciesbyServicer
Percent90 Numberof Numberof Percent Days Loans Delinquencies Cured Delinquent
Servicer1 293 656 88.87 11.13
Servicer2 594 1,461 79.19 20.81
Servicer3 1,133 3,474 86.70 13.30
Servicer4 824 2,594 87.24 12.76
Servicer5 377 861 90.36 9.64
Servicer6 595 1,688 77.84 22.16
Servicer7 261 599 79.63 20.37
Servicer8 405 864 83.91 16.09
Others 1,404 2,841 89.02 10.98
Total 5,886 15,038 85.37 14.63
Source:Self-Helpdatabaseandauthors’calculations.Note:Eightmajorservicers(handlingmorethantheaveragenumberofdelinquencies)werechosenforinclusioninouranalysis.
(12.2percentand12.5percent).Inbothofthesecaseswithoneoriginatorandmultipleservicers,therewasonlyoneservicerwithalargeenoughnum-berofdelinquencyspellstobecountedamongthemainones;therestwereincludedintheotherscategory.Inallothercases,asingle,uniqueservicerwasmanagingaportfolioofloansoriginatedbythesamelender.
LogitregressionmodelTobetterisolatetheimpactsofdifferentservicingmethodsandotherrisk
factors,ourgoalinmodelingtheoutcomesofthe15,038delinquencyspellsistoidentifyfactorsthatpredictthelikelihoodthatagivendelinquencywillcureorworsenandgointodefault.Wemodelthebinaryoutcomesusingalogitregressionapproachwherethedependentvariableistheoddsofcureandthereferencecategoryisdefault.
(1)=
+=−
k
jijj
i
i Xpp
Log11
βα
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 261
Here,Pi is theprobabilityof transition froma30-daydelinquency tocure,Χiisacolumnvectorofthecovariatemeasuresfordelinquencyspelli,andβisarowvectorofcoefficientsfortheoutcomeofcuring.Thismodelassumesthatoutcomesatanyonepointintimeareindependentofoutcomesinanypreviouspointintime.Tocontrolforthepotentialstatisticalprob-lemsassociatedwithrepeatedevents,themodelwasestimatedusingStata’slogitprocedurewithanadjustmenttothestandarderrorsforclusteringbyloan.17
Inadditiontoinvestigatingtheoutcomesforalldelinquencyspells,itisalsoimportanttolookattheoutcomesofthefirstdelinquencyspellstoelimi-natetheeffectofpossibledependenceamongdifferentspells.Itispossiblethatsomeservicerscouldtreatfirstdelinquenciesdifferentlyfromserialones.Tocheckthedifferentimpactofservicersontheoutcomeoffirstdelinquen-cies,weusedthesamemodeltorunaseparateregressiononfirstdelinquencyspells.
Werecognize thatdifferences incureanddefault ratesmaybedue todifferencesinthecompositionofeachportfolio.Table5profilestheloanshandledbythemaineightservicersandallothers.Thedifferencesarenota-ble:Servicer6’sportfoliohasthegreatestshareoflowcreditscoresandthelowest-incomeborrowers.LoanshandledbyServicer8haveespeciallyhighLTVs, while loans handled by Servicer 5 experienced the highest housingappreciationratesduringthestudyperiod.Astotheloansservicedby“oth-ers,”theonlyobviousdifferenceintermsofriskfactorsisintheirsomewhatbetterappreciationrates(7percentversus4percentforallbutServicer5).Borrowersservicedbytheotherscategoryalsohavehigherincomethanbor-rowershandledbymostofthemaineight.
Consequently,whenweestimatethelogitregressionmodel,wecontrolforvariousloanandborrowercharacteristicsthatmightaffectdelinquencyoutcomes.Controlvariablesinthemodelincludethefollowing:
1. Loan characteristics. Housing equity is a key factor in determiningdefault.Delinquentborrowerswithsubstantialaccumulatedequityhavemoreresourcestoavoidforeclosure,whilethosewithlittleornoequityhave less incentiveand lessability to retain theirhome.We includeadummyvariableforloanswithanLTVof95percentorhigher.Wealsoincludeloanageinmonthsatthetimeofdelinquencyandthenumber
17Theestimationusedthe“cluster”option,whichspecifiesthatobservationsareindepen-dentacrossgroups(loans)butnotnecessarilywithingroups(StataCorporation2003).Thisoptionaffectstheestimatedstandarderrorsandvariance-covariancematrixoftheestimator,butnottheestimatedcoefficients.
HOUSING POLICY DEBATE
262 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
Table 5. LoanandBorrowerCharacteristicsbyServicer
MainServicers Variable All 1 2 3 4 5 6 7 8 Others Creditscore
Nocreditscore(%) 3.5 0.3 4.4 0.6 1.2 2.27 7.7 7.1 1.3 5.78
FICO620orlower(%) 17.2 7.8 23.1 25.1 23.8 9.42 46.5 26.4 15.5 11.43
FICO621to660(%) 21.0 17.6 26.5 20.4 23.7 20.61 21.3 35.1 20.2 19.41
FICO661to720(%) 30.4 33.7 29.1 26.8 29.9 37.22 15.2 22.4 36.5 30.88
FICOabove720(%) 28.0 40.6 17.0 27.1 21.5 30.49 9.3 9.2 26.6 32.50
Loancharacteristics
MedianLTV(%) 97.0 97.0 97.0 97.0 99.0 97.0 97.0 97.0 102.3 97.0
Medianback-endratio 36.7 37.7 37.5 35.4 36.6 37.3 34.9 37.1 37.2 37.0 (%)
Ageofloanat 11 13 3 23 18 9 5 2 9 12 purchase(months)
Estimated 4.3 4.2 3.3 3.4 3.6 13.4 4.1 4.0 4.4 7.0 appreciationrate(%)a
Borrowercharacteristics
Femaleborrower(%) 44.0 43.3 48.7 51.1 46.03 36.71 50.1 37.5 43.2 37.14
Blackborrower(%) 21.3 15.2 33.9 34.7 37.66 9.88 60.2 12.5 6.8 13.50
Hispanicborrower(%) 16.6 4.8 4.2 2.6 2.89 42.98 2.0 7.9 15.1 26.56
First-timehomebuyer 43.8 56.0 18.0 64.8 88.61 32.09 82.4 18.0 48.3 25.39 (%)
Incomeat 29,472 31,074 30,000 23,640 24,870 35,634 24,504 30,276 27,996 33,516origination($)(median)Incomeasa 59.7 61.3 64.3 57.0 58.1 61.2 48.2 52.8 59.8 61.8percentageofAMI(median)
Totalloansb 28,131 1,962 1,852 4,661 2,318 2,378 1,055 850 2,464 10,591
Source:Self-Helpdatabaseandauthors’calculations.Note:Totalsmaynotequal100percentbecauseofrounding.aTheestimatedannualhouseappreciationrateisbasedonthehousepriceatoriginationandtheupdatedhousepricesestimatedinDecember2005.bSelf-HelploanspurchasedbeforeJanuary1,2003.AMI=areamedianincome(medianincomeofthemetropolitanstatisticalarea[MSA]orthestatefornon-MSAareas).
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 263
ofpreviousdelinquencies(ofcourse,thisvariableisnotincludedinthemodelfocusingonthetransitionoffirstdelinquencies).
2. Borrower characteristics. To control for underwriting and borrowerdifferences,we includeadummyforback-endratiosof38percentorhigher,18 incomeasapercentageofareamedian income,dummies forgenderandrace/ethnicity,andadummyforfirst-timehome-buyerstatus.Sincetheimpactofcreditscoreonloanperformancemightnotbelinear,wefollowindustrypracticeandpreviousliteraturebyalsoincludingasetofcreditscoredummyvariablesforborrowerswithnocreditscore,borrowerswithlowscores(below620),borrowerswithmoderatescores(620to659),andborrowerswithrelativelyhighscores (660to719).Thecategorywiththehighestscores(720orabove)issetasareferencegroup.
3. Economicconditions.Becausechangesinpropertyvaluesafterorigina-tioncanaffecthousingequity,weuseavariablefortheestimatedannualrate of appreciation, determined using the change between the valueofthehouseatoriginationandupdatedvaluesforeachpropertyasofDecember2005.Tohelp separate lender/servicereffects fromregionaleconomiceffects,wealsoincludedummyvariablesforthestateshavingthemostloansintheSelf-Helppopulationstudied:NorthCarolinaandCalifornia.
Totestfordifferencesamongservicers,weincorporatedummyvariablesforthemaineightSelf-Helpservicersintoourlogisticmodel.Wetreatdelin-quenciesservedby“others”asthereferencegroup.Theoddsratioestimatedfromthemodelistheoddsofcure(versusworsening)ofadelinquencyspellforoneservicerrelativetotheoddsofcurebyservicersintheotherscategory.Table4indicatedthatthereferencegroup(“others”)hadaveryhighcurerate (89.02percent);onlyServicer5hadahigher rate than this. Inotherwords,aftercontrollingforothercharacteristicsinourlogitmodel,wearecomparingthepracticesofmajorservicerswithsomeofthebestperformers(asthesimplefrequencytablesuggests).
Someoftheexplanatoryvariablesarehighlycorrelated.Generally,bor-rowerswithhighLTVshavelowercreditscores.BlackandHispanicborrow-ersalsohavelowercreditscoresandhigherLTVsthannon-Hispanicwhites.
18According to theSellingGuide createdbyFannieMae, its suggestedbenchmark fortotaldebt-to-incomeratiois36percentforstandardloansand38percentforcommunitylend-ingproducts(Allregs.com2006).NearlyallSelf-Helplendersallowahighermaximumratio,rangingfrom40percentto48percent.
HOUSING POLICY DEBATE
264 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
PropertiesinCaliforniahavehigherhouseappreciationrates.Further,thereisasignificantcorrelationbetweentheloanagevariableandthenumberofpreviousdelinquencies.However,thesevariableshavebeenwidelyusedintheliterature,andacheckofvarianceinflationfactorsindicatesthatmulti-collinearityisnotparticularlyseriousamongindependentvariables.19Asaresult,thefinalmodelofthisanalysisincludesalltheseimportantvariablessuchascreditscores,LTV,race,gender,andstatedummies.
Tables6and7presenttheresultsofthelogitregressionmodel.Table6isbasedonalldelinquencyspells,whiletable7focusesonthetransitionoffirstdelinquencyspellsonly.20Apositivecoefficientmeansthattheoddsofcureincreaseastheindependentvariableincreases;anegativecoefficientmeansthattheoddsdecrease.Anoddsratiorepresentstheoddsofcurewhenanindependentvariableisequalto1relativetotheoddsofcurefortherefer-encecategory(forcategoricalvariables).Themodelfitisstrong.21Further,theeightservicervariableswerejointlysignificantatthe0.001level,whichsuggests that the introduction of these variables appreciably increases theexplanatorypowerofthemodel.
Ofprimary interest is the fact thataftercontrolling for loanandbor-rower characteristics, house appreciation rate, and regional dummies, wefindsignificantdifferencesintheoddsofcureamongservicers.Specifically,delinquenciesforServicers2,3,6,and7aresignificantlylesslikelytocurethantheyarefortheservicersintheotherscategory,whiletheoddsofcureforServicers1,4,5,and8arenotsignificantlydifferentfromthose“others.”Inotherwords,30-daydelinquenciesforServicers2,3,6,and7aresignifi-cantlymorelikelytoendindefaultthantheyareforservicersintheotherscategory.Theoddsofcure forServicers2,6,and7areabout40percentlowerthantheyareforservicersintheotherscategory,whiletheoddsareabout20percentlowerforServicer3.
19Varianceinflationfactorsareameasureofthemulticollinearityinaregressiondesignmatrix.Forthisstudysample,thevarianceinflationindicatorsformostvariablesarelessthan3(generallyconsideredacceptable).Theonlyexceptionsareseveralcategoricalcreditscorevariables.
20Thefinalstudysamplefortable6isreducedfrom15,308to13,878becauseofmissingdata.Thestudysamplefortable7is5,443.
21Toassessthepredictivequalitywiththemodelfocusingonallthespells,wefindthattheshareofcorrectlypredictedobservationswas87percentforcureand23percentfordefaultifwespecifythecutoffas>0.8(anestimatedprobabilityof>0.8forcurewouldbeassignedtobeapositiveoutcome).Theoverallshareofcorrectlyclassifiedobservationsis78percent.Ofcourse,ahighercutoffwouldleadtobetterspecificationfordefaultandpoorersensitivityforcure.Forexample,ifthecutoffissetas0.85,theshareofobservationsthatwerecorrectlypredicted is67percent for cureand47percent fordefault.Theoverall correctly classifiedshareis64percent.
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 265
Table 6. LogisticRegressionResultsfortheOutcomesof30-DayDelinquencies:ProbabilityofCure
Variable Coefficient StandardError OddsRatio
Constant 1.423** 0.193
Nocreditscore –0.349* 0.102 0.705
FICObelow620 –0.046 0.119 0.955
FICO620to659 0.031 0.131 1.032
FICO660to719 0.036 0.136 1.036
LTV95orabove –0.087 0.069 0.917
Femaleborrower 0.128* 0.066 1.136
Blackborrower 0.076 0.071 1.079
Hispanic 0.301** 0.163 1.351
IncomeasapercentageofAMI 0.006** 0.002 1.006
Ageofloan(months) 0.004** 0.001 1.004
Numberofpreviousdelinquencies –0.028 0.014 0.973
Appreciationratea 0.028** 0.010 1.028
NorthCarolinaborrower 0.119 0.104 1.126
Californiaborrower –0.101 0.176 0.904
Servicer1 0.121 0.193 1.129
Servicer2 –0.610** 0.072 0.543
Servicer3 –0.222* 0.088 0.801
Servicer4 –0.213 0.099 0.808
Servicer5 0.088 0.157 1.093
Servicer6 –0.545** 0.068 0.580
Servicer7 –0.527** 0.086 0.590
Servicer8 –0.235 0.109 0.790
Modelfit
Wald_2 209.2**
df 22 Source:Self-Helpdatabaseandauthors’calculations.Note:Thestudysampleis13,878sinceanumberofobservationsweredroppedbecauseofmissingdata.aTheestimatedannualhouseappreciationrateisbasedonthehousepriceatoriginationandupdatedhousepricesestimatedinDecember2005.AMI=areamedianincome(medianincomeofthemetropolitanstatisticalarea[MSA]orthestatefornon-MSAareas).*p≤0.05.**p ≤0.01.
HOUSING POLICY DEBATE
266 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
Table 7.LogisticRegressionResultsfortheOutcomesoftheFirst30-DayDelinquencies:ProbabilityofCure
Variable Coefficient StandardError OddsRatio
Constant 1.553** 0.26
Nocreditscore –0.375 0.164 0.687
FICOatorbelow620 –0.057 0.162 0.945
FICO620to659 0.005 0.175 1.005
FICO660to719 0.070 0.191 1.073
LTV95orabove –0.099 0.105 0.906
Femaleborrower 0.342** 0.125 1.408
Blackborrower 0.143 0.116 1.154
Hispanic 0.237 0.209 1.267
IncomeasapercentageofAMI 0.008** 0.003 1.008
Ageofloan(months) –0.003 0.002 0.997
Appreciationratea 0.026** 0.010 1.027
NorthCarolinaborrower 0.070 0.154 1.073
Californiaborrower 0.072 0.298 1.074
Servicer1 –0.199 0.184 0.820
Servicer2 –0.547** 0.118 0.578
Servicer3 –0.135 0.152 0.874
Servicer4 –0.085 0.174 0.919
Servicer5 0.216 0.292 1.241
Servicer6 –0.949** 0.066 0.387
Servicer7 –0.894** 0.081 0.409
Servicer8 –0.472** 0.115 0.624
Modelfit
Wald_2 161.4**
df 21
Source:Self-Helpdatabaseandauthors’calculations.Note:Thestudysampleis5,443sinceanumberofobservationsweredroppedbecauseofmissingdata.aTheestimatedannualhouseappreciationrateisbasedonhousepriceatoriginationandupdatedhousepricesestimatedinDecember2005.AMI=areamedianincome(medianincomeofthemetropolitanstatisticalarea[MSA]orthestatefornon-MSAareas).*p ≤0.05.**p≤0.01.
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 267
Astothecontrolvariables,it isparticularlynoteworthythatvariablesthatarekeyindicatorsofdefaultriskatthetimeofunderwriting(basedoncreditscoreandLTV)generallydonotappeartobesignificantinpredictingwhetheranalreadydelinquentloanwilldefault.Theonlyexceptionisthatthereissomeevidencethatborrowerswithoutcreditscoresarelesslikelytobe cured.But someborrower characteristicsdomatter.Female,Hispanic,andhigher-incomeborrowersaremore likely tocure. Inparticular,delin-quentborrowerswhowereHispanichad39percentbetteroddsofcuring(versusworsening)thandelinquentborrowerswhowerewhite.Theevidencealsoindicatesthatolderloansaremorelikelytocure,suggestingthatexperi-encedborrowershandledelinquenciesbetterthannewonesdo.Fromamac-roeconomicstandpoint,locationineitherNorthCarolinaorCaliforniadidnothaveasignificanteffectonoutcomes,butthereisconsiderableevidencethatthehigherthehouseappreciationrate,themorelikelythatdelinquentloanswouldbecuredandthelesslikelythatborrowerswouldletthedelin-quencyproceed.
Whenwefocusonthefirstdelinquencyspell(table7),theresultsaregen-erallyconsistentwiththeresultsbasedonalldelinquencyspells.Firstdelin-quenciesforServicers2,6,and7arestilllesslikelytobecured,whilethecoefficientsforServicer3becomeinsignificant,likethoseofServicers4and5.ButthecoefficientforServicer8issignificantandnegativeinthismodel.AmongServicers2,6,7,and8,theoddsofcureareparticularlylow—about60percentlower—forServicers6and7.Thusitseemsthattherelativeoddsofcurefortheseservicers,whileworseforalldelinquencies,lagevenmoreforfirstdelinquencies.
Table 8 shows predicted cure and default rates by servicer. Using theresultsfromtable6(allspells),weestimatethatServicer2hasthe lowestpredictedcurerate(andthehighestpredicteddefaultrate),whileServicer1hasthehighestpredictedcurerate(andthelowestpredicteddefaultrate).Inotherwords,theprobabilitythatadelinquentloanservicedbyServicer2willdefaultisnearlydoublethatforServicer1(table8).
Thus there appear to be roughly two groups of servicers in terms ofoutcomes.DelinquencieshandledbyServicers2,3,6,and7havealowerprobabilityofcureandahigherprobabilityofdefault.Servicers1,4,5,and8,alongwiththeservicersintheotherscategory,havehighercureratesandalowerriskofdefault.22
22AlthoughServicer8hasalowercurerateforthefirst-30delinquencyspells,itsoverallcurerateisnotstatisticallydifferentfromthatofthe“others.”
HOUSING POLICY DEBATE
268 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
From telephone interviews with the main servicers and supplementalinformation provided by Self-Help’s Servicing Manager, we learned thattherearesignificantdifferencesinthewaypersonnel,systems,andprocessesareappliedtononperformingloansbetweenthe30thand90thdayofdelin-quency—theperiod reflectedby theanalysis (Russell2005).Forexample,oneservicerhasonecollectorforevery15,000loansitservicesandusesnorisk-scoringsystematall.Thisisinstarkcontrasttotheservicersthathaveatleastonecollectorforevery5,000loansandusebothcollectionsscoringandforeclosurealternativeassessmentmodels,supportedbyscriptingsoftware.
Servicersuseanumberofdifferentcollectionsandlossmitigationtools.CollectionsscoringtoolsincludeFreddieMac’sEIandFannieMae’sRP.EIfeaturesbothascorethatprioritizescallstoborrowersintheearlystageofdelinquencyandamodulethatpredictstheshareofmoreadvanceddelin-quencies thatcouldresult ina loss (FreddieMacn.d.).RPusesborrower,loan,andpropertycharacteristicstosegmentloans(whethercurrentlydelin-quentornot)bylikelihoodofgoingmorethan60daysdelinquentwithinsixmonths. Investorshave laidout specific servicingproceduresbasedonscore—forexample,recommendingearliercontactforhigh-riskratherthanforlow-riskloans(FannieMae2005a,2005b).Toolstoassessalternativesto foreclosure includeFreddieMac’sWorkout Prospector®,which applies
Table 8. PredictedOutcomesofaTypical30-DayDelinquencybyServicer
Percentin Default(90 Percent Days Cured Delinquent)
Servicer1 90.41 9.59
Servicer2 81.94 18.06
Servicer3 87.00 13.00
Servicer4 87.10 12.90
Servicer5 90.13 9.87
Servicer6 82.89 17.11
Servicer7 83.13 16.87
Servicer8 86.84 13.16
Others 89.31 10.69 Source:Self-Helpdatabaseandauthors’calculations.Note:Basedonthelogitregressionmodelintable6.Additionalborrowerandloancharacteristicshavebeencontrolledfor.
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 269
23Thisprogramwasdevelopedonathird-partyplatform.FreddieMacsoldittothethirdparty,ComputerSciencesCorporation,in2004(ComputerSciencesCorporation2004).
aconsistentapproachtoassess theviabilityof foreclosurealternatives foreach case (Cutts and Green 2005), and Workout ProfilerTM, part of Fan-nieMae’sHomeSaverSolutionNetworkTM(HSSN),aWeb-based,interac-tivemodelthatincorporatessuchdataasfinancialsituation,propertyvalue,andpaymenthistory(O’Connor2003).AnexampleofascriptingsystemisEarlyResolution®,23aprogramtotrainandguidepersonnelastheyinteractwithborrowerstodeterminethebestmethodtohelpthemkeeptheirhomes(O’Connor2003).
Servicersalsofollowdifferentstepstoaddressdelinquencies.Self-Helpsuggeststhatitssubservicersfollowaspecifictimelinethroughthedelinquencyprocess(figure1).ButwhileservicersgenerallyfollowedSelf-Help’smanual,thereweresomemarkeddeviationsinactualdelinquencymanagement.
Anotherareaofdifferentiationderives fromaSelf-Help-drivendelin-quencyinterventionprogram.Inthetraditionalprocess,theborrowermusttakesomeinitiativetogetintoalossmitigationplan,eitherbycontactingtheservicerorahousingcounselororbyfillingoutaninformationpacketthatcomes in themail fromtheservicer. In2002,Self-HelpbegantestingamoreproactiveapproachbyengagingBalance,anaffiliateofConsumerCreditCounselingofSanFrancisco,tocontactborrowersbytelephoneonthe45thdayofdelinquency.TheprogramwasactivatedwithServicers1,2,3,and6between2002andlate2003,aswellaswithoneotherthatisnotprofiledinthisarticle(Holder2005).BecauseSelf-Helpdeliberatelyphasedinservicersaccordingtowhereinterventionmighthavethemostimpact,itisnotsurprisingthatmostoftheparticipatingservicersfallinthelowercureratecategory.
ServicerprofilesProfilesofthedifferentstaffing,systems,andprocessesusedbythemain
Self-Helpservicersaredescribedinthefollowingparagraphs.Theseservicersand subservicers ranged in size from less than50,000 loans serviced to2million.Theyarelocatedinvariouspartsofthecountry,andtheirportfo-liosincludeloansfrommanydifferentstates.Theydonotfocusstrictlyonconventional,affordablemortgages,butratherservicearangeofmortgagetypes.Sowhilenotstatisticallyrepresentativeofallservicers,thesebriefcasestudiescharacterizeawiderangeofinstitutions,practices,andforeclosureprevention strategies.Whether oneof these strategies ismore effective or
HOUSING POLICY DEBATE
270 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
Sour
ce:C
ente
rfo
rC
omm
unit
ySe
lf-H
elp
(200
3).
Fig
ure
1.S
elf-
Hel
pD
elin
quen
cyM
anag
emen
tT
imel
ine
inD
ays
Sche
dule
ear
lyde
linqu
ency
coun
selin
g
3060
90
7–10
16
20
45
50
10
5
Colle
ctio
nca
ll/pa
ymen
tre
min
der
notic
e
Follo
w-u
ple
tter
Late
pay
men
t no
tice
(a la
tefe
e is
usu
ally
asse
ssed
)
HUD
lette
r (su
gges
ts
coun
selin
g)
Loss
miti
gatio
nal
tern
ativ
esle
tter
Bre
ach/
dem
and
lette
r
Refe
r to
fore
clos
ure
(con
tinue
loss
m
itiga
tion)
Refe
r to
atto
rney
(fore
clos
ure
times
var
yby
sta
te)
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 271
efficientthananothercannotbedeterminedfromthisanalysis,buttheextentofthedifferenceshintsatwhyoutcomesvary.
Servicers1,4,5,and8(highercurerates).Servicer1isamidsizeorgani-zation(between100,000and300,000loansservicedinall)thatusesEIfordelinquencyscoring.Thereisastrongcollectionsstaff,withonecollectorforevery4,000loansintheportfolio.ThisenablesspecializationsuchthattwocollectorsareassignedtoSelf-Helploans.TheyfollowSelf-Help’stimelineclosely.Intermsoflossmitigation,however,Servicer1hasonlyonespecial-isttoaddressitsportfolioofmorethan100,000loansanddoesnotuseaworkoutsystem.Self-Helprolledoutitsdelinquencyinterventionpilotwiththisservicer15monthsbeforetheendofthe53-monthobservationperiod,soitwouldhaveasomewhatlimitedimpactonourresults.
Servicer4isamega-servicer(morethan1millionloanshandled).Itaver-ages20,000 loansper collector; also, itoutsources certain calls for early-term,low-riskdelinquenciesandusesbothRPandEI.Affordablemortgagesarenotidentifiedperse,butinvestorprocessrequirementsarefollowedforall loans. Servicer 4 also averages about 20,000 to 25,000 loans per lossmitigationemployeeandusesaproprietaryworkoutsystembutnoscriptingtechnology.Thisisoneofonlytwoservicersthatreportedpayinganincen-tivetoitslossmitigationstaff,inthiscaseforworkoutsthatareapprovedorclosed.
Servicer5,alsoamega-servicer,usesbothRPandEIandhasaloans-to-collectorratiothatfallsatthehigherendoftherangeamongSelf-Help’sservicers. Staffmembers specialize by investor, not product type. For lossmitigation, the servicer uses the HSSN and supplements decision makingwithinternallydevelopedspreadsheets.Thisservicer’sprocessissomewhatunusualinthatitstartscollectionscallingonday1andsendsthelossmiti-gationalternatives letterveryearly—onthe30thday—eveninadvanceoftheHUDletter. (TheHUDlettergivestheborroweratoll-freenumbertocalltoobtainthename,number,andaddressofthenearestHUD-approvedcounselingagency.Thelossmitigationletterinformstheborrowerthatheor she may qualify for alternatives to foreclosure. The letter lists options(repaymentplan, loanmodification,deed in lieu, andpreforeclosure sale)andprovidesadescriptionofwhateachonemeans.)Thislikelyencouragesworkoutsearlyinthedelinquencytimeline.
Servicer 8 is small (less than 100,000 loans) and uses RP. Overall, itstaffsonecollectionspersonper6,000loans,withonecollectorassignedtoits2,000affordableloans(includingSelf-Help).Thelossmitigationemploy-ees—aboutoneforevery20,000loansintheportfolio—donotspecialize;WorkoutProfilerTMisusedforlossmitigation(Russell2005).Fromaprocess
HOUSING POLICY DEBATE
272 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
standpoint,Servicer8takeslittleactionbetweentheinitialdelinquencyletterandthe60thday,whentheHUDletter,lossmitigationletter,andbreachlet-terareallsentwithinafewdaysofeachother(Holder2005).
Servicers2,3,6,and7(lowercurerates).Servicer2issmallandhasonecollectionspersonforevery15,000loansandnoriskscoringsystem.Thereisonlyonelossmitigationcounselor(forabout50,000loans),andtheHSSNisusedforworkouts.Staffmembersdonotspecializebyinvestororproducttype.Servicer2sendsouttheHUDlettertwoweekslaterthansuggested,butotherwisegenerallyfollowsSelf-Help’stimeline.Becauseofthelargenumberofdelinquenciesanditslimitedresources,Servicer2wasselectedasthefirstparticipantintheinterventionprogram.
Servicer3islarge(300,000to1millionloans)andusesbothRPandEI.Itemploysapproximatelyonecollectionspersonper10,000loans.Collec-torsdonotspecialize.Onthelossmitigationside,thereisonespecialistfor80,000loans,andtheHSSNisusedforworkouts.Servicer3basicallyfollowsSelf-Help’sinvestorrequirementsandbeganparticipatingintheinterventionprograminAugust2002,aboutmidwaythroughtheobservationperiod.
Servicer6isanotherlargeorganization.Theratioofloansservicedpercollector is between 5,000 and 10,000, while the ratio of loss mitigationspecialiststoloansisaboutoneper20,000.Servicer6usesRPbutnosmartworkoutsystem.HUDlettersarenotsent,butabreachletter issentveryearly(onthe35thday),andloansarereferredtoforeclosureearlieraswell.ThisservicerhasparticipatedinthedelinquencyinterventionprogramsinceJuly2003(thefinal18monthsoftheobservationperiod).
Servicer7fallsinthemega-servicercategory,averaging12,000loansperfull-timecollectionsemployee.Thisservicerreportshavingadedicatedteamforhigher-riskproducts;makingintroductorycallswhennewloansinthiscategory are added to theportfolio; paying staff incentives forworkouts;andusingRP,EI,multiplelossmitigationsystems,andascriptingsystem.Thisservicer’sactualprocessisdistinctinthatneitheraHUDletternoralossmitigationletterisautomaticallysent,andthebreachletterdoesnotgooutuntilthe90thday,sothereoftenappearstobelittlecommunicationwithdelinquentborrowersbetweenthe30thand90thdays.
Evidently, there is substantialvariation in the servicers’approaches toforeclosureprevention—somuch,infact,thatit isdifficulttodiscernbestpractices.WebelievethatthisvariationisnotconfinedtotheSelf-Helpser-vicers,butspeculatethatitiswidespread,particularlyassales,mergers,andconsolidationsresultintherealignmentofservicingplatformsandcultures.Further research into successful strategies for affordable mortgages could
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 273
helpservicersdevelopmoreeffectivecombinationsofpeople,processes,andtechnology thatwouldhelpagreaternumberof troubledborrowerskeeptheirhomes.
ConclusionsTheunprecedentedgrowthinhomeownershipamonglow-andmoderate-
incomefamiliesduringthe1990smeantthatthebackendofthehome-buyingtransaction—loanservicingandpostpurchasedefaultmanagement—becamemuchmoreimportant.Althoughdelinquencyanddefaultratesincreasedastheeconomyfellintorecession,manymorefinanciallysqueezedhomeown-erswouldpossiblyhavelosttheirhomesifnotfortheinnovationsinloanservicingdiscussedinthisarticle.Atthiswriting,theloomingpossibilityofsofteninghousingmarketsandrisingforeclosures,andtheproliferationofaggressivemortgageproductswithadjustableratesandnegativeamortiza-tionsuggestthatthesetoolsmaybecomeevenmoreimportantinthenearfuture.Forexample,interest-onlyloans,onwhichnoprincipalpaymentsaremadetobringdownmortgagedebt,madeupnearlyone-thirdofmortgageoriginationsduring2004and2005(FishbeinandWoodall2006).Accordingtooneestimate,ofthe7.7millionhouseholdsthattookoutadjustable-ratemortgagestobuyorrefinancein2004and2005,“upto1millioncouldlosetheirhomesthroughforeclosureoverthenextfiveyearsbecausetheywon’tbeabletoaffordtheirmortgagepayments,andtheirhomeswillbeworthlessthantheyowe”(Knox2006,A1).
As mortgage products featuring more flexible underwriting standardsgaininpopularityandasmorelow-andmoderate-incomefamilies,immi-grants,andminoritiesjointhefirst-timehome-buyingmarket,theeconomicfortunesofmortgageinstitutionswillbeincreasinglytiedtotheeffectivenessoftheirloanservicingoperations.Clearly,themoreaggressivetheunderwrit-ing,themoreimportantpreventiveservicingbecomes.Whileclosecoordina-tionbetweencreditunderwritingpoliciesandservicingstrategiesisimportantfortheentireindustry(CaldwellandJordan2006),itisevenmoreimportantforaffordablehousingportfolios.
Althoughitisnotpossibletogeneralizeforallaffordablelendingpro-grams,earlyfindingsfromourevaluationoftheSelf-Helpsecondarymarketdemonstration program are nevertheless suggestive. A significant share ofSelf-Helpborrowersandcoborrowerswhobought theirhomesbefore theonsetofthe2000-01recessionhaveexperiencedmultiplespellsofjobless-nessandmissedhousingpaymentssincetakingouttheirloans.YetthevastmajorityofSelf-Helpborrowershavenevermissedapayment,andasmall
HOUSING POLICY DEBATE
274 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
fraction account for a disproportionately large share of total delinquencyspellsanddefaults.
Againstthisoverallbackdrop,wefindthatevenaftercontrollingforloanand borrower characteristics and regional economic conditions, the oddsthata late-payingborrowerwillmanagetocatchuponpayments insteadofsinkingintoseriousdelinquencycanvaryasmuchas60percentbetweenservicers.Thissuggeststhatservicingstrategiesdomatter.
Weshouldalsonotethat,assuggestedbyindustrydataaswellasourowninterviews,defaultmanagementcostsdriveservicingprofits,andloanservic-ingremainsa labor-intensiveprocess,smart technologiesnotwithstanding.Someservicersweinterviewedstillvaluepersonnelovertechnology.Whilethismaybeawisechoiceintheshortrun,withoutthedecision-makingtech-nology thatwillbecomethe industrystandard,affordable lenderswill lagbehindindustrybenchmarksoverthelongterm,andmorelow-andmoder-ate-incomeborrowersthannecessarywilllosetheirhomes.
Withtherateofappreciationforhousingpricesslowing,therangeofcost-effectiveworkoutoptionsthatwouldenabledefaultedborrowerstoremainintheirhomesmaybenarrowing,sothatitbecomesevenmoreimportanttodiffusesmartservicingtechnologiesaswidelyaspossibleamongaffordablehousinglendersandservicers.Theevidencethatpreventiveservicingreallyworksshouldspurfoundationandgovernmentsupportforthedisseminationofbestpractices.Forexample,foundationandgovernmentgrantsforaccessandtrainingcouldacceleratetherateatwhichthesesmartservicingsystemsareadoptedbysmallerlendersandservicersofaffordablemortgages.Andthis,inturn,couldmaketheirpreventiveservicingeffortssmarterandmorecost-effective.
Finally,policyaimedataffordablehomeownershipshouldbemoresym-metrical;mortgageinnovationsthatlowerthecostofentrymustbematchedbysupportformoresophisticatedservicingsystems.Intheresearchrealm,thereisanurgentneedforsystematic,firm-levelstudiesofservicingpracticesforaffordablemortgages.Asthepresentstudysuggests,withoutbeingableto“unbundle”servicingactivitiesandobservehowfirm-levelpracticesareactuallyimplemented,definingbestpracticeswillbeimpossible.
AuthorsMichaelA.StegmanisthedirectorofPolicyfortheProgramonHuman
and Community Development at the John D. and Catherine T. MacAr-thur Foundation. Roberto G. Quercia is an associate professor of CityandRegionalPlanningattheUniversityofNorthCarolinaatChapelHill.
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 275
Janneke Ratcliffe is the associate director of the Center for CommunityCapitalismattheUniversityofNorthCarolinaatChapelHill.LeiDingisaseniorresearchassociateattheCenterforCommunityCapitalismattheUniversityofNorthCarolinaatChapelHill.WalterR.DavisistheprincipalmethodologistatStatisticsNewZealand.
TheauthorsthankPhilipE.Comeau,AmyCrewsCutts,DebbieKehr,andMarquitaWebbfortheirhelpfulcommentsandsuggestionsandgrate-fullyacknowledgetheFordFoundationforitssupportofthisresearch.
References
Abraham,Jesse.1999.ALifePreserverforLoans.MortgageBanking59(9):56–65.
AllRegs.com.2006.FannieMaeSingle-FamilyGuides:2002SellingGuide.WorldWideWebpage<http://www.allregs.com>(accessedJuly10,2006).
Baku, Esmail, and Marc Smith. 1998. Loan Delinquency in Community LendingOrganizations:CaseStudiesofNeighborWorksOrganizations.HousingPolicyDebate9(1):151–75.
Caldwell, J.H., and Elizabeth Jordan, 2006. Unleashing the Power. DrivingProfits through Risk-Based Mortgage Collections. Deloitte. World Wide Web page<http://www.deloitte.com/dtt/cda/doc/content/US_FSI_UnleashingthePower_MBStudy.pdf>(accessedJuly11,2007).
CaliforniaRealEstateCenter.com. 2006. Mortgage Payment Behind? World Wide Webpage <http://www.Californiarealestatecenter.com/mortgage-payment-behind.htm>(accessedJuly11,2007).
Capone,CharlesA.Jr.,andAlbertMetz.2003.MortgageDefaultandDefaultResolu-tions:TheirImpactonCommunities.PaperpresentedattheFederalReserveBankofChi-cagoConferenceonSustainableCommunityDevelopment,Washington,DC.March27.
Center for Community Self-Help. 2003. Self-Help Servicing Manual, rev. January 1.Durham,NC.
CenterforCommunitySelf-Help.n.d.ComparisonofSelf-HelpBorrowerswithThoseforConformingLoansPurchasedbyGSEs.Unpublishedpaper.Durham,NC.
Computer SciencesCorporation. 2004.CSCAcquiresMortgage IndustryUtility fromFreddieMac:CompanyPlans toExpandDefaultManagement Services. Press release,February19.WorldWideWebpage<http://www.csc.com/solutions/managementconsult-ing/news/2599.shtml>(accessedJuly11,2007).
Cordell, Larry. 2001. Innovative Servicing Technology: Smart Enough to SustainHomeownershipGains?Paperpresentedatthe2001ConferenceonHousingOpportu-nity,sponsoredbytheResearchInstituteforHousingAmerica,Washington,DC.April26.
HOUSING POLICY DEBATE
276 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
Cornwell,Ted.2003.FannieMaeSeesaContinuedRiseinWorkoutRatiosduring2003.NationalMortgageServicingNews27(25):7.
Cutts,AmyCrews,andRichardK.Green.2005.InnovativeServicingTechnology:SmartEnoughtoKeepPeopleinTheirHouses?InBuildingAssets,BuildingCredit:CreatingWealthinLow-IncomeCommunities,ed.NicolasP.RetsinasandEricS.Belsky,348–78.Washington,DC:BrookingsInstitutionPressandJointCenterforHousingStudies.
Deng,Yongheng,andStuartA.Gabriel.2002.EnhancingMortgageCreditAvailabilityamongUnderservedandHigherCredit-RiskPopulations:AnAssessmentofDefaultandPrepaymentOptionExerciseamongFHA-InsuredBorrowers.WorkingPaperNo.02–10.UniversityofSouthernCalifornia,MarshallSchoolofBusiness.
DeZube,Dona.2003.RunawayChurn.MortgageBanking63(5):46–52.
FannieMae.2005a.FannieMaeRiskProfiler®4.0UserGuide.Washington,DC.
Fannie Mae. 2005b. Servicing Changes to Risk Profiler®. Announcement No. 05–01.Washington,DC.
FHACites90,000LossMitCases.2005.NationalMortgageNews30(5):91.
Fields,RuthG.2001.RescuingLatepayers.MortgageBanking61(5):24–28.
Fishbein,AllenJ.,andPatrickWoodall.2006.ExoticorToxic?AnExaminationoftheNon-TraditionalMortgageMarketforConsumersandLenders.Washington,DC:Con-sumerFederationofAmerica.
Freddie Mac. n.d. EarlyIndicatorSM Version 5.0 Overview. World Wide Web page<http://www.freddiemac.com/service/earlyind/pdf/eiv5overview.pdf> (accessed February9,2006).
Harney,Kenneth.2002.It’sBetterNottoForeclose,Home-LoanGiantsFind.Washing-tonPost,November9.
Herbert,Christopher,DebbieGruenstein,andKimberlyBurnett.2000.AnAssessmentofFHA’sSingle-FamilyMortgageInsuranceLossMitigationProgram.CambridgeMA:AbtAssociatesInc.
Holder,Mary.2005.PersonalinterviewwiththeLossMitigationandServicingManageratSelf-Help,September24.
Home Mortgage Disclosure Act. 2001. Public Loan Origination Records. From com-puter-generatedtablesprovidedtoauthorsbyHaroldL.Bunce,deputyassistantsecretaryforEconomicAffairs,U.S.DepartmentofHousingandUrbanDevelopment,onJune12,2003.
InsideMortgageFinance.2006.The2006MortgageMarketStatisticalAnnual.Vol. I.ThePrimaryMarket.Bethesda,MD.
JointCenterforHousingStudies.2005.TheStateoftheNation’sHousing2005.Cam-bridge,MA.
HOUSING POLICY DEBATE
PreventiveServicingandAffordableHomeownershipPolicy 277
Kehr,Debbie.2005.Telephone interviewwith theDirectorofServicingOperationsatFannieMae,August30.
Knox,Noelle.2006.SomeHomeownersStruggletoKeepUpwithAdjustableRates.USAToday,April3,p.A1.
Lipman,BarbaraJ.2005.TheHousingLandscapeforAmerica’sWorkingFamilies2005.NewCenturyHousing5(1):1–56.
McConnell,JohnJ.1976.ValuationofaMortgageCompany’sServicingPortfolio.Jour-nalofFinancialandQuantitativeAnalysis11(3):433–53.
Melchiorre, Camillo T. 1999. Hitting Home Runs in Servicing. Mortgage Banking59(5):32–39.
Meyer, Laurence H. 1997. Models for Neighborhood Development and Reinvest-ment.SpeechattheAnnualMeetingonNeighborhoodHousingServicesofNewYorkCity, June 18. World Wide Web page <http://www.federalreserve.gov/BoardDocs/Speeches/1997/199706182.htm>(accessedJuly15,2007).
Mortgage Bankers Association. 2005. National Delinquency Survey: Second Quarter2005.Washington,DC.
MortgageNewsDaily.com.2005.ForeclosureHappens,ButThereAreSolutions.WorldWide Web page <http://www.mortgagenewsdaily.com/822005_Default_Mortgage.asp>(accessedJuly11,2007).
Mozilo,AngeloR.2000.LoanWorkoutPrograms:ABestBusinessPractice.MortgageBanking60(10):17–18.
Muolo,Paul.2000.TheRushtoBuyReceivables.U.S.Banker110(5):56–58.
Now,“AffordableServicing.”2005.NationalMortgageNews29(23):1,26.
O’Connor, Robert. 2003. Rescuing Borrowers at Risk. Mortgage Banking Magazine64(3):26–31.
Oliver,GeoffreyA.,TimothyDavis,BernadetteKogler,andJeffreyKlein.2001.ProfitableServicersintheNewMillennium.MortgageBanking61(9):32–41.
Quercia,RobertoG.,GeorgeW.McCarthy,andSusanM.Wachter.2003.TheImpactsofAffordableLendingEffortsonHomeownershipRates.JournalofHousingEconomics12(1):29–59.
Quercia,RobertoG.,andMichaelA.Stegman,1992.ResidentialMortgageDefault:AReviewoftheLiterature.JournalofHousingResearch3(2):341–79.
Quercia,RobertoG.,MichaelA.Stegman,WalterR.Davis,andEricStein.2002.Per-formanceofCommunityReinvestmentLoans:ImplicationsforSecondaryMarketPur-chases.InLow-IncomeHomeownership:ExaminingtheUnexaminedGoal,ed.NicolasP.RetsinasandEricS.Belsky,348–74.Washington,DC:BrookingsInstitutionPressandJointCenter forHousingStudies.Alsoavailableat<http://www.kenan-flagler.unc.edu/assets/documents/CC_reinvestment.pdf>.
HOUSING POLICY DEBATE
278 M.A.Stegman,R.G.Quercia,J.Ratcliffe,L.Ding,andW.R.Davis
Russell, Julie. 2005. An Exploration of Successful Default Management Strategies forAffordableHomeLoans.Master’sthesis.UniversityofNorthCarolinaatChapelHill.
Sichelman,Lew.2001.CompassionateServicing.MortgageBanking61(5):16–23.
Sjaastad, John E., Jason Vinar, Philip Jones, and Jeremy Prahm. 2005. DelinquencyDynamics:ServicerEffects.MarketPulse1:1–24.
Stanton,ThomasH.2001.Opportunities forReducingDelinquencies andDefaults inFederalHousingCreditPrograms:AReviewofNewTechnologiesandPromisingPrac-tices.JournalofPublicBudgeting,Accounting,&FinancialManagement13(2):157–92.
StataCorporation.2003.StataProgrammingReferenceManual,Release8.CollegeSta-tion,TX:StataPressPublications.
Steinbach,GordonH.1995.ReadytoMaketheGrade.MortgageBanking55(9):36–42.
U.S. Department of Housing and Urban Development. 2002. Follow-up NationwideReview:DepartmentofHousingandUrbanDevelopment’sLossMitigationProgram.AuditReport:DistrictInspectorGeneralforAudit,RockyMountainDistrict.No.2002–DE–001.Denver,CO.
U.S.DepartmentofHousingandUrbanDevelopment.2003.U.S.HousingMarketCon-ditions,1stQuarter2003:Table18.WorldWideWebpage<http://www.huduser.org/periodicals/ushmc/spring03/index.html>(accessedOctober10,2005).
U.S.DepartmentofHousingandUrbanDevelopment.2004.TheSustainabilityofHom-eownership:FactorsAffectingtheDurationofHomeownershipandRentalSpells.ReportpreparedbyAbtAssociatesInc.Washington,DC.
U.S.DepartmentofHousingandUrbanDevelopment.2005.ActuarialReviewof theMutual Mortgage Insurance Fund FY 2005. World Wide Web page <http://www.hud.gov/offices/hsg/comp/rpts/actr/2005actr.cfm>(accessedJuly10,2006).
VanOrder,Robert,andPeterZorn.2004.ThePerformanceofLow-IncomeandMinor-ityMortgages:ATaleofTwoOptions.Workingpaper.UniversityofPennsylvaniaandFreddieMac.
Venetis,Kyriaki.2001.GELaunchesE–LMOFastTrackasPartofItsMIServicerPlat-form.CreditandCollectionsWorld.com.WorldWideWebpage<http://www.creditcollec-tionsworld.com/news/083001_1.htm>(accessedOctober10,2005).
Waggoner,DarrenJ.2002.DebtSales:ReadytoErupt?Collections&CreditRisk7(6):44.WorldWideWebpage<http://www.collectionsworld.com/06cov02.htm>(accessedJune14).