TRMswww.vtigu.com

FederalReserveBankofChicagoComplexMortgagesGeneAmromin,JenniferHuang,ClemensSialm,andEdwardZhongNovember24,2010WP2010-17ComplexMortgagesGeneAmrominFederalReserveBankofChicagoJenniferHuangUniversityofTexasatAustinClemensSialmUniversityofTexasAustinandNBERandEdwardZhongUniversityofWisconsin-MadisonNovember24,2010WethankEthanCohen-Cole,SerdarDinc,PeteKyle,JayHartzell,JeongminLee,RobertMcDonald,LauraStarks,SheridanTitman,MichelleWhiteandseminarparticipantsatthe2010FinancialEconomicsandAccountingConference,theFederalReserveBankofChicago,theUniversityofLausanne,theUniversityofTexasatAustin,andtheUniversityofZurichforhelpfulcommentsandsuggestions.
GeneAmrominisattheFederalReserveBankofChicago,230SouthLaSalleStreet,Chicago,IL60604.
Email:gamromin@frbchi.
org;JenniferHuangisattheMcCombsSchoolofBusiness,UniversityofTexasatAustin,Austin,TX78712.
Email:jennifer.
huang@mccombs.
utexas.
edu;ClemensSialmisattheMcCombsSchoolofBusiness,Uni-versityofTexasatAustin,Austin,TX78712.
Email:clemens.
sialm@mccombs.
utexas.
edu;andEdwardZhongisattheDepartmentofEconomics,UniversityofWisconsin-Madison,Madison,WI53715.
Email:edzhong@gmail.
com.
1ComplexMortgagesAbstractComplexmortgagesbecameapopularborrowinginstrumentduringthebullishhous-ingmarketoftheearly2000sbutvanishedrapidlyduringthesubsequentdownturn.
Thesenon-traditionalloans(interestonly,negativeamortization,andteasermortgages)enablehouseholdstopostponeloanrepaymentcomparedtotraditionalmortgagesandhencerelaxborrowingconstraints.
Atthesametime,theyincreasehouseholdleverageandheightendependenceonmortgagerenancingtoescapechangesincontractterms.
Wedocumentthatcomplexmortgageswerechosenbyprimeborrowerswithhighincomelevelsseekingtopurchaseexpensivehousesrelativetotheirincomes.
Borrowerswithcomplexmortgagesexperiencesubstantiallyhigherexpostdefaultratesthanborrowerswithtraditionalmortgageswithsimilarcharacteristics.
"Theavailabilityofthesealternativemortgageproductsprovedtobequiteimportant,and,asmanyhaverecognized,islikelyakeyexplanationofthehousingbubble.
"–BenS.
Bernanke1IntroductionOverthelastdecade,theresidentialmortgagemarkethasexperiencedasignicantincreaseinproductcomplexity,followedbyarapidreversionbacktosimpleproducts.
Inthispaper,westudythemortgagecontractchoiceofindividualhouseholdsandtheirsubsequentdefaultbehavior.
ThemenuofhouseholdmortgagechoicesintheUnitedStateswasdominatedfordecadesbyfully-amortizinglong-termxedratemortgages(FRM)and,toalesserextent,byadjustableratemortgages(ARM)thatlockedintheinitialinterestratefortherstvetosevenyearsofthecontract.
Fromthevantagepointoftheborrower,FRMcontractspreservecontracttermsestablishedatoriginationforthelifetimeoftheloan.
Forpracticalpurposes,thesamecanbesaidoftheprevailingARMcontracts,giventheaverageborrowertenureataparticularhouseofaboutsevenyears.
Knowingthemonthlyservicingcostsandamortizationschedulessimpliesthehouseholdbudgetingproblem.
Themortgagemarkethasexperiencedasignicantincreaseinproductcomplexityintheearly2000s.
Theproductsthatgainedprominenceduringtheperiodofrapidhousepriceappreciationfeaturedzeroornegativeamortization,shortinterestrateresetperiods,andverylowintroductoryteaserinterestrates.
Wetermthese"complexmortgages"(CM).
Figure1showstheproportionofxedrate,adjustablerate,andcomplexmortgageproductsoriginatedovertheperiodbetween1995and2009,asreportedbyLPSAppliedAnalytics(ourprimarydatasourcedescribedindetailbelow).
TheshareofcomplexproductsintheU.
S.
remained1below2%untilthesecondhalfof2003beforejumpingtoabout30%ofmortgageoriginationsjusttwoyearslater.
Insomegeographicareascomplexmortgagesaccountedformorethan50%ofmortgageoriginations.
Thecomplexproductsfadedalmostasquickly,decliningtolessthan2%oforiginationsin2008.
Complexmortgagesappeartobeatthecoreoftherecentriseanddeclineinhousingprices.
Toobtainanimpressionoftherelationbetweenrisklevelsandmortgagecomplexity,weaggregatetheloan-leveldatainto366MetropolitanStatisticalAreas(MSAs)andthensortallMSAsintoquintilesaccordingtotheproportionofcomplexmortgageloansin2004–therstyearofsubstantialoriginationsofcomplexloans.
Figure2summarizestheaveragequarterlychangesinhousepricesforthebottom,themiddle,andthetopMSAquintiles.
WeobservethatMSAsinthetopcomplexityquintileexperiencestrongerhouseappreciationbefore2006andfasterdepreciationafter2006.
ThisresultprovidesanindicationthathousepricechangesweremorepronouncedinMSAswithhighproportionofcomplexloans.
ItalsosuggeststheimportanceofunderstandingthereasonsforCMusageandthedriversoftheireventualperformance.
Thedeningfeatureofcomplexmortgagesisthedeferralofprincipalrepayment.
Asaresult,complexmortgagesarecharacterizedbylowmortgagepaymentsduringtherstfewyearsofthecontract,whichrelaxeshouseholdliquidityandborrowingconstraintsandenableshouseholdstotakelargeexposuresinhousingassets.
Thelackofmortgageamortizationinevitablyproducestwoeects:ahigherloan-to-value(LTV)ratioforanygivenpathofhousepricesandagreaterrelianceonrenancingtoescapeincreasesinpaymentsonceacontractenterstheamortizationphase.
Complexmortgagescanbeoptimalborrowinginstrumentsifhouseholdsexpecttheirin-comelevelsorhousingpricestoincreaseovertime,asdiscussedbyPiskorskiandTchistyi(2010).
Theycanalsobeoptimalinstrumentsforlendersconcernedwiththeirexposurein2anassetbubbleenvironment(BarlevyandFisher(2010)).
Inaddition,complexmortgagesmightalsoberationallychosenbyhouseholdsthatexhibitrelativelyhighlaborincomeriskandliveinareaswithvolatilehouseprices.
Thesehouseholdshaveanincentivetominimizetheinitialmortgagepaymentsandtokeepthemortgagebalancerelativelyhighbecausetheyhavetheoptiontodefaultincaseofadverseincomeandhousepriceshocks.
Theseincentivestorationallydefaultshouldbeparticularlypronouncedinnon-recoursestates,wherelendersdonothaveaccesstothenon-collateralizedassetsofhouseholdsincaseofdelinquency.
Inthiscase,complexmortgagesshouldbeahallmarkofsophisticatedborrowerskeenlyawareofthevalueofthedefaultoption.
Ontheotherhand,thelowinitialpaymentsofcomplexmortgagesmightobfuscatethelong-termborrowingcostsforhouseholds,assuggestedbyCarlin(2009)andCarlinandManso(2010).
Lendersmighthaveanincentivetointroducecomplexproductstohidetheactualfeesembeddedinnancialproducts.
Whereasitisrelativelyeasyforahouseholdtocomparethecostsofplain-vanillaxedratemortgagesacrossdierentlenders,itismorediculttocomparecomplexloansthatoftenincludeintricateresetschedules,prepaymentpenalties,andshort-livedteaserinterestrates.
Lendersmightbeparticularlyeagertooertheseproductsiftheyarecondentthattheycansecuritizetheseloans.
Inthiscase,weshouldobservethatcomplexmortgagesaretakenoutprimarilybyunsophisticatedinvestorsthatdonotunderstandthespecicfeaturesoftheirmortgagecontracts.
Tostudythemortgagechoicesofhouseholdsandthedefaultexperiences,wemakeexten-siveuseoftheLPSAnalyticsdata.
Thedatabase,describedindetailinSection2,containsloanlevelinformationforalargesampleofmortgagesintheUnitedStates.
Ofparticularrelevanceforouranalysisistheabilitytoidentifyprecisecontractterms,bothatthetimeoforiginationandoverthelifetimeoftheloan.
Ourmainresultindicatesthatcomplexmortgagesaretakenoutbywell-educatedhouse-3holdswithrelativelyhighincomelevelsandwithprimecreditscores.
Wendthathouse-holdsborrowingcomplexmortgagesearnsignicantlyhigherannualincomes($141,998)thanhouseholdsborrowingxedratemortgages($88,642)oradjustableratesmortgages($101,005).
Furthermore,only7%ofborrowersusingcomplexloanshavecreditscoresbelow620(com-monlyconsideredsubprimecreditscores),whereas10%ofxedrateborrowersand23%ofadjustablerateborrowersfallintothissubprimecategory.
Wealsondthatahigherpro-portionofthepopulationinneighborhoodswithahighpropensityofcomplexloanstendtohaveacollegedegree.
Finally,complexloansaremoreprevalentinnon-recoursestates,wherenon-collateralizedassetsofthehouseholdsareprotected.
Thus,theseresultsindicatethatcomplexloansarenotprimarilyoriginatedtonaivehouseholdsthatarefooledbylendersintoinappropriatemortgagecontracts.
Nonetheless,thesehouseholdsarestretchingtopurchasemoreexpensivehousesrelativetotheirincomes,asindicatedbytheirhighervalue-to-income(VTI)ratios.
HigherVTIratiosareassociatedwithgreaterpropensitytousecomplexcontractsevenaftercontrollingforMSA-levelincomeandVTImeasures.
Thissuggeststhatatleastapartoftherelationshipisduetohouseholdsusingcomplexmortgagestogetmoreexpensivehouseswithinhighhousingpriceareas.
Wealsondthatareaswithhigherpasthousepriceappreciationandhigherpopulationgrowthhavemorecomplexmortgages,whereasareasthatexperiencedsustainedhousepricedecreasesinthepasttenyearshavefewercomplexmortgages.
Thisevidencesuggeststhattheexpectationofcontinuedhousepriceappreciationisalikelydrivingforcebehindthepopularityofcomplexmortgages.
Next,westudythedefaultbehaviorofborrowersofcomplexmortgages.
Thefocusoninitialloanaordabilitymightmotivatehouseholdstoborrowtooextensivelyandtounderes-timaterenancingrisk,whichisexacerbatedbyhistoricallyshortresetperiodsandrecastingofnegativeamortizationloans.
Aftercontrollingforobservablecharacteristicsthatinclude4theFICOcreditscoreandincome,wendthathouseholdswithcomplexmortgagesaremorelikelytodefault.
Thisholdstrueafterthesetofcontrolsisexpandedtoincludetime-varyingloan-to-valueratios,whichsuggeststhathigherCMdefaultsarenotdueexclusivelytohigherexpostleverage.
Sincecomplexmortgagestypicallyhavelowermonthlypaymentsrelativetotheirxedratecounterpartsduringtherstyearsafterorigination,thehigherdefaultratessuggestthateitherCMhouseholdsaremorelikelytodefaultstrategically,orthatthesehouseholdshavemorevolatileincomestreams.
Overall,ourndingssuggestthatinadditiontothewell-documentedimpactofsubprimemortgages,householdswithcomplexmortgagesmightbeasignicantdrivingforcebehindthemountingdefaultsduringtherecentcrisis.
Whiletheextensionofcredittosubprimeborrowersandmortgagesecuritizationhasre-ceivedmuchattentionfollowingthenancialcrisisof2007-2009,thechoiceandimpactofmortgagecomplexityremainslargelyunexplored.
MianandSu(2009)showthatthesharpincreaseinmortgagedefaultsin2007issignicantlyampliedingeographicareaswithahighdensityofsubprimeloansthatexperiencedanunprecedentedgrowthinmortgagecreditpriorto2007.
Keys,Mukherjee,Seru,andVig(2010)focusontheroleofmortgagesecuritizationprocess,ndingthatsecuritizationloweredthescreeningincentivesofloanoriginatorsfortheirsubprimeborrowers.
Jiang,Nelson,andVytlacil(2010b)studytherelationbetweenmortgagesecuritizationandloanperformanceandndthatthelenderapplieslowerscreeningeortsonloansthathavehigherexanteprobabilitiesofbeingsecuritized.
1Ourpapercontributestothisliteraturebysuggestinganadditionalandimportantchannellinkingmortgagemarket1AdditionalpapersonsecuritizationandtheexpansionofcredittosubprimeborrowersincludeAdelino,Gerardi,andWillen(2009),Bond,Musto,andYilmaz(2009),Keys,Mukherjee,Seru,andVig(2009),Lout-skinaandStrahan(2009),Mayer,Pence,andSherlund(2009),StantonandWallace(2009),Agarwal,Ambrose,Chomsisengphet,andSanders(2010),Bajari,Chu,andPark(2010),BarlevyandFisher(2010),Berndt,Hol-lield,andSandas(2010),Campbell,Giglio,andPathak(2010),CorbaeandQuintin(2010),DemyanykandHemert(2010),Garmaise(2010),Gerardi,Rosen,andWillen(2010),Glaeser,Gottleb,andGyourko(2010),Goetzmann,Peng,andYen(2010),Jiang,Nelson,andVytlacil(2010a),Piskorski,Seru,andVig(2010),Purnanandam(2010),Rajan,Seru,andVig(2010),StantonandWallace(2010),andWoodwardandHall(2010).
5innovationstothenancialcrisisof2007-2009.
Afewrecentpapershaveinvestigatedtheroleofnon-traditionalmortgagecontractsintherecentcrisis.
PiskorskiandTchistyi(2010)studyoptimalmortgagedesigninanenvironmentwithriskyprivatelyobservableincomeandcostlyforeclosureandshowthatthefeaturesoftheoptimalmortgagecontractareconsistentwithanoptionadjustableratemortgagecontract.
CorbaeandQuintin(2010)presentamodelwhereheterogeneoushouseholdsselectfromasetofmortgagecontractsandhaveachoiceofdefaultingontheirpayments.
Usingtheirmodel,theyndthatthepresenceofsubprimemortgageswithlowdownpaymentssubstantiallyampliesforeclosureratesinthepresenceofalargeexogenousshocktohouseprices.
Inacontemporaneouspaper,BarlevyandFisher(2010)describearationalexpectationsmodelinwhichbothspeculatorsandtheirlendersuseinterest-onlymortgageswhenthereisabubbleinhouseprices.
Theyprovideevidencethatinterestonlymortgageswereusedextensivelyincitieswhereinelastichousingsupplyenablespronouncedboom-bustcycles.
Ourpaperstudiesempiricallythecharacteristicsandthedefaultexperiencesofborrowersofcomplexloans.
Theremainderofthispaperisstructuredasfollows.
Section2describesourdatasourcesandreportssummarystatistics.
InSection3westudythemortgagechoiceofhouseholdsanddescribethemainfeaturesofmortgagecontracts.
InSection4westudythedelinquencyofdierentcontracttypes.
2DataSourcesandSummaryStatisticsOurstudyreliesonseveralcomplementarydatasourcesthatcovervariousaspectsofthehous-ingmarketduringtheperiodbetween2003and2007.
Inparticular,themicrolevelanalysisofmortgagecontractchoiceandperformancereliesheavilyontheproprietarymortgage-leveldatabaseoeredbyLenderProcessingServices(LPS)AppliedAnalytics(formerlyknownasMcDashAnalytics).
LPScollectsdatafromsomeofthenation'slargestmortgageservicers6thatreportcontractandborrowerdetailsatthetimeofloanorigination,aswellasmonthlyinformationonmortgageperformance.
TheLPSdatacoveragehasgrownsteadilyovertime,with9outof10largestservicersreportingtothedatabaseby2003.
Ourdatabasecoversabout10millionmortgageswithatotalloanvalueofmorethan$2trillionbetween2003and2007.
Forthepurposesofourstudy,theavailabilityofgranularinformationonmortgagecontracttermsisofparticularimportance.
Foreachoftheloans,LPSprovidesinformationontheloaninterestrate,theamortizationschedule,andthesecuritizationstatus.
Foradjustableratemortgages(ARMs),weknowtherateatorigination,thefrequencyofresets,thereferencerate,andtheassociatedcontractualspread.
Forloansthatdonotamortizesteadilyovertheirterm,weknowthehorizonoftheinterest-onlyperiod,whethernegativeamortizationisallowedandifso,towhatextentandoverwhatperiodoftime.
Thisinformationallowsustopreciselycategorizeloancontracts.
TheLPSdataalsocontainskeyinformationonborrowerandpropertycharacteristicsatthetimeoforigination.
Theseincludetheappraisedpropertyvalue,theloan-to-valueratio(LTV),propertytype(singlefamilyorcondominium),whetherthepropertywastobeoccupiedbytheborrower,andtheborrower'screditworthinessasmeasuredbytheirFICO(FairIsaacCorporation)creditscore.
2AnimportantfeatureoftheLPSdatabaseisthatunlikesomeotherdatasources,itisnotlimitedtoaparticularsubsetoftheloanuniverse.
TheLPSdatacoverprime,subprime,andAlt-Aloans,3andincludeloansthatareprivatelysecuritized,thosethataresoldto2AsBajari,Chu,andPark(2010)emphasize,animportantfeatureoftheFICOscoreisthatitmeasuresaborrower'screditworthinesspriortotakingoutthemortgage.
FICOscoresrangebetween300and850Typically,aFICOscoreabove800isconsideredverygood,whileascorebelow620isconsideredpoor.
AsreportedontheFairIsaacCorporationwebsite(www.
myco.
com),borrowerswithFICOscoresabove760areabletotakeout30-yearxedratemortgagesatinterestratesthatare160basispointslower,onaverage,thanthoseavailableforborrowerswithscoresinthe620-639range.
3Alt-Aloansareamiddlecategoryofloans,moreriskythanprimeandlessriskythansubprime.
They7GovernmentSponsoredEnterprises(GSEs),andloansthatheldonbanks'balancesheets.
Althoughthisallowsforabroadsetofmortgagecontracts,thecoverageissomewhatskewedinfavorofsecuritizedloansthataremorelikelytobeservicedbylargecorporationsreportingtoLPS.
Therelativescarcityofportfolioloansisrelevanttoussincesomeofthecontractsofinterest,suchasoptionARMs,arecommonlyheldinlenders'portfolios.
Still,thelargeoverallsizeofthedataensuresthatwehaveamplecoverageofallcontracttypes.
WecomplementborrowerinformationinLPSwithhouseholdincomedatacollectedundertheHomeMortgageDisclosureAct(HMDA).
Doingsoallowsustocomputesomeofthekeymeasuresofloanaordability,suchastheratioofhousevaluetoincome(VTI).
Wefurtheraugmenttheloan-leveldatawithinformationontrendsinlocalhomeprices.
Quarterlydataonhomepricesisavailablebymetropolitanstatisticalarea(MSA)fromtheFederalHousingFinanceAgency(FHFA)-anindependentfederalagencythatisthesuccessortotheOceofFederalHousingEnterpriseOversight(OFHEO)andothergovernmententities.
4WeusetheFHFAHousePriceIndex(HPI)includingalltransactionsthatisbasedonrepeatsalesinformation.
Weuseahousepriceindextoconstructborrower-specicvariablesoncumulativegrowthinlocalhouseprices.
Atthemoreaggregatelevel,weutilizezipcodelevelinformationfromthe2000U.
S.
Censustocontrolforbroaddemographiccharacteristics,suchaseducationlevels.
WealsomakeuseoftheannualpercapitaincomeandunemploymentratedataattheMSAlevelfromtheBureauofEconomicAnalysis(BEA).
aregenerallymadetoborrowerswithgoodcreditscores,buttheloanshavecharacteristicsthatmakethemineligibletobesoldtotheGSEs-forexample,limiteddocumentationoftheincomeorassetsoftheborrowerorhigherloan-to-valueratiosthanthosespeciedbyGSElimits.
4AspartoftheHousingandEconomicRecoveryActof2008(HERA),theFederalHousingFinanceRegulatoryReformActof2008establishedasingleregulator,theFHFA,forGSEsinvolvedinthehomemortgagemarket,namely,FannieMae,FreddieMac,andthe12FederalHomeLoanBanks.
TheFHFAwasformedbyamergeroftheOceofFederalHousingEnterpriseOversight(OFHEO),theFederalHousingFinanceBoard(FHFB),andtheU.
S.
DepartmentofHousingandUrbanDevelopment'sgovernment-sponsoredenterprisemissionteam(seewww.
fhfa.
govforadditionaldetails).
8TodeterminewhetherlenderrecoursehasanimpactonmortgagechoicesandmortgagedefaultswefollowGhentandKudlyak(2010)andclassifyU.
S.
statesasrecourseornon-recoursestates.
Innon-recoursestates,recourseinresidentialmortgagesislimitedtothevalueofthecollateralsecuringtheloan.
Ontheotherhand,inrecoursestatesthelendermaybeabletocollectondebtnotcoveredbytheproceedingsfromaforeclosuresalebyobtainingadeciencyjudgment.
5ThesummarystatisticsonthesevariablesarepresentedinTable1andwewilldiscussdierencesinthesevariablesacrossmortgagetypesinmoredetailinSection3.
2.
AllofthevariablesdiscussedabovearesummarizedinTable9.
3MortgageChoiceThissectiondescribesindetailthedierencesincharacteristicsofthemainmortgagecontractsoeredintheU.
S.
duringthelastdecadeandthedeterminantsofthemortgagechoice.
3.
1MortgageContractDesignInthissectionweillustratethedierentpaymentpatternsofsomepopularU.
S.
mortgagecontracts.
Weclassifyallmortgageproductsintothreegroups:(1)FixedRateMortgages(FRM);(2)AdjustableRateMortgages(ARM);and(3)ComplexMortgages(CM).
6Fixedratemortgagesarelevel-paymentfully-amortizingloanswithmaturitiesthatgener-allylastfor15or30years.
Forexample,ahouseholdborrowing$500,000ona30-yearxedratemortgagewitha5%interestratewillberequiredtomakeequalmonthlypaymentsof$2,684for360months.
After30yearsthemortgagewillbepaidocompletely.
Borrowers5GhentandKudlyak(2010)classifythefollowingstatesasnon-recourse:Alaska,Arizona,California,Iowa,Minnesota,Montana,NorthDakota,Oregon,Washington,andWisconsin.
6AdditionalinformationonvariousmortgagecontractscanbeobtainedfromthewebsiteofJackM.
Gut-tentagathttp://www.
mtgprofessor.
com.
9generallyhavetheoptiontoprepaythemortgageiftheysellthepropertyoriftheyrenancetheirloanduetoadecreaseinmortgageinterestrates.
Adjustableratemortgagesarefully-amortizingloanswheretheinterestratechangesafteraninitialperiodaccordingtoapreselectedinterestrateindex.
Theinitialperiodwithaxedinterestratetypicallylastsbetweentwoandsevenyears.
Themortgagesexhibitcapsandoorsthatpreventtheinterestratesfromchangingtoomuchoverthelifetimeoftheloan.
InterestratesonARMsgenerallyarelowerthanthoseonFRMsduetotheincreasingtermstructureofinterestratesandtheavailabilityoftheprepaymentoptioninFRMs.
7Forexample,a5/1ARMwitha30-yearmaturity,a$500,000initialbalance,anda4.
5%initialinterestratewillhaveinitialmortgagepaymentsof$2,533permonthfortherst60months.
Subsequently,thepaymentscanincreaseordecreasedependingonthelevelofinterestrates.
Iftheinterestrateincreasesto7%,thenthemonthlypaymentinthesixthyearwillincreaseto$3,221.
8Complexmortgagesincludeavarietyofback-loadedmortgagecontracts.
Mostcomplexmortgagesareadjustableratemortgagesandexhibittime-varyingpayments.
Themostpop-ularcontractisanInterestOnly(IO)mortgage.
IOborrowersonlyneedtopaythemortgageinterestforaninitialtimeperiodthattypicallylastsbetweenveandtenyears.
Subsequently,themortgagebecomesafully-amortizingloan.
Forexample,a5-yearIOadjustablerateloanwitha30-yearmaturity,a$500,000initialbalance,anda4.
5%initialinterestratewillhaveinitialmortgagepaymentsof$1,875permonthfortherst60months.
Subsequently,thepaymentsresetaccordingtothefutureinterestrates.
Iftheinterestrateincreasesto7%,7Fixedratemortgagescanberenancedwheninterestratesdecrease,whichisaveryvaluableoptionthatispricedintheinitialinterestrate.
Therearenumerouspapersonprepayments.
Seeforexample,DunnandMcConnell(1981),SchwartzandTorous(1989),Stanton(1995),DunnandSpatt(1999),Longsta(2005),Campbell(2006),Amromin,Huang,andSialm(2007),Gabaix,Krishnamurthy,andVigneron(2007),andSchwartz(2007).
8SeveralpapersstudythetradeobetweenFRMsandARMs(e.
g.
,CampbellandCocco(2003),Vickery(2007),andKoijen,VanHemert,andVanNieuwerburgh(2009)).
10thenthemonthlypaymentinthesixthyearwillalmostdoubleto$3,534.
Evenifinterestratesremainat4.
5%,themortgagepaymentwillincreaseto$2,779permonthattheendoftheinitialinterest-onlyperiod.
Thepaymentsincreaseevenmoreformortgageswithlongerinterest-onlyperiods.
AsecondtypeofacomplexmortgageisaNegativeAmortizationMortgage(NEGAM),suchasanOptionARM.
Thesemortgagesgiveborrowerstheoptiontoinitiallypayevenlessthantheinterestdue.
Thedierencebetweentheinterestdueandtheactualmortgagepaymentisaddedtotheloanbalance.
Thesemortgagescarrytheriskoflargerincreasesinmortgagepayments,whenthemortgageisrecasttobecomeafullyamortizingloanafter5-10yearsorwhentheloanbalanceexceedstheinitialbalanceatoriginationbymorethanacertainamount(typically10-25%).
Finally,athirdtypeofacomplexmortgageisaTeaserRateMortgage(TRM).
ForTRMs,theinitialinterestrateissignicantlybelowthefullyindexedrate.
Teaserrateloanstypicallychargeinvestorsinterestratesofbetween1-2%duringtherst1-12months.
Mostteaserratemortgagesalsofeaturenegativeamortization.
Insum,complexmortgagesareback-loadedproductswithlimitedamortizationduringtherstyearsafterorigination.
Asmentionedintheintroduction,complexmortgagescanbeoptimalifhouseholdsexpecttheirincomelevelsorhousingpricestoincreaseovertime(PiskorskiandTchistyi(2010)).
However,thelowinitialpaymentsofcomplexmortgagesalsocarryanumberofrisks,fromobfuscatingthelong-termborrowingcostsofhouseholds(Carlin(2009),CarlinandManso(2010))togreaterrelianceonrenancingtoavoidincreasesinpayments.
Thisobfuscationmightbeparticularlypronouncedforteaserratemortgages,whoselowpaymentsonlyapplyforarelativelyshortinitialperiod.
113.
2SummaryStatisticsbyMortgageTypeTable2reportsstatisticsforourbroadmortgagecategories–fully-amortizingxedrate(FRM),fully-amortizingadjustablerate(ARM)andcomplex(CM)mortgagetypes.
Ourdatacontaininexcessof10millionloancontractsoriginatedbetween2003and2007.
Inoursample,69percentofmortgagesarexedratemortgages,12percentareadjustableratemortgages,andtheremaining19percentarecomplexmortgages.
Complexmortgages,onaverage,areassociatedwithhigherloanamountsrelativetothetraditionalARMandFRMmortgages,andareusedtonancemoreexpensivehouses.
Forexample,theaveragehomevalueforcomplexloansis$513,728,whereastheaveragehomevaluesforFRMsandARMsare$264,878and$309,465,respectively.
Countertosomeofthecommonlymadeassertionsaboutcomplexmortgages,theyareextendedtoborrowerswithhighincomelevels.
Indeed,themeanincomeofacomplexmortgageborrowerisabout60%higherthanthatofaborrowerwithatraditionalplain-vanillaxedratemortgage.
Nevertheless,theaverageratioofhousevaluetoincome(VTI)–ameasureofaordability–isconsiderablyhigherincomplexmortgagecontracts,suggestingthatcomplexmortgageborrowersarepurchasingmoreexpensivehousesrelativetotheirincome.
Yet,thishigherspendingonhousesisnotreectedintheloan-to-value(LTV)ratio,asallmortgagetypeshavesimilarrstlienLTVvalues.
9PanelAofFigure3depictsthecumulativedistributionfunctionoftheVTIratioforborrowerswithdierentmortgagecontracts.
ThegureindicatesthatCMborrowerstendtohavesubstantiallyhigherVTIratiosthanbothARMandFRMborrowers.
MedianhouseholdsusingFRMs,ARMs,andCMshavevalue-to-incomeratiosof3.
0,3.
1,and3.
7,respectively.
Putdierently,foragivenlevelofincomeCMborrowers9LPSdataiscollectedattheloanandnotpropertylevel,whichlimitsone'sabilitytoconstructanaccurateestimateofthetotaldebtsecuredbythehouse.
Inparticular,weareunabletoaccountforsecond-lienmortgagesloans(theso-called"piggybackloans")usedtonancethehouse.
Primarilyforthisreason,wedonotemphasizetheimportanceofLTVinourempiricalanalysisandinsteadfocusontheVTIratio.
12purchasedhousesvaluedatabout20%more.
Thelowerinitialpaymentsoncomplexmortgagesthusappeartoenablehouseholdstopurchaseexpensivehomesrelativetotheirincomelevels.
WealsondthatborrowersofcomplexmortgageshavebettercreditscoresthanARMborrowersandsimilarcreditscoresasFRMborrowers.
Whereas23%ofARMborrowershaveFICOcreditscoresbelow620,thesamecanbesaidofonly10%ofFRMand7%ofCMborrowers.
PanelBofFigure3summarizestheentiredistributionofFICOscoresfordierentmortgagecontracts.
WhereasmanyborrowersusingARMstendtohavesub-primecreditscores,thecreditqualityofborrowersusingCMsisfairlysimilartothatoftheFRMborrowers.
Theseresultsemphasizethattheclienteleforcomplexmortgagesdierssignicantlyfromthatforsubprimeloans.
Severalotherloancharacteristicsaredierentforcomplexmortgages.
CMborrowersaremorelikelytoliveinacondominiumandareslightlymorelikelytousethepropertytheyarenancingforinvestmentpurposes.
Wealsondsignicantdierencesinthefrequencyofprepaymentpenaltiesacrossmortgagetypes.
UnlikeFRMs,asignicantfractionofARMsandCMsfacepenaltiesiftheloansareprepaidwithinthersttwoorthreeyears.
Around40%ofthemortgagesinoursamplearefromrenancingtransactions,whereastheremainingproportionisfromnewhomepurchases.
Complexmortgageshaveaslightlyhighershareofrenancingscomparedtonewpurchases.
Sincecomplexloansareparticularlypopularforexpensivehomes,theyarealsomorelikelytoexceedtheconformingloanlimit(i.
ebejumboloans).
Hence,although79%ofFRMsaresecuritizedbygovernment-sponsoredenterprises(GSEs,suchasFannieMae,FreddieMac,amdGinnieMae),only24%ofCMsgothroughtheGSEs.
PrivatesecuritizationpartiallyosetsthelackofGSEinvolvementintheARMandCMmarkets.
ComplexmortgageborrowersreceivesignicantlylowerinitialinterestratesthanFRMorARMborrowers.
Themeaninitialinterestrateoncomplexmortgagesof5.
12%issignicantly13lowerthantheratesonFRMs(6.
16%)andARMs(5.
97%).
Thisresultisprimarilycausedbyteaserratemortgagesthatcharge,onaverage,aninitialinterestrateofonly1.
30%.
ForeachARMandCMloanweimputetheratesuchborrowersmighthavereceivedhadtheychosenaconventional30-yearxedratemortgageinstead.
Wedenesuchhypotheticalrateastheaverageinterestrateonall30yearFRMsoriginatedinthesamemonth,state,withsimilarloansize(whetherornotabovetheconforminglimit),LTVratio,andFICOscore.
ThehypotheticalFRMinterestrateissimilaracrossthevariouscontracts.
Whereasthevariablesaboveareavailableattheloanlevel,wealsoreportsomeadditionalvariablesobservedattheMSAorthestatelevel.
WendthatCMborrowerstendtoliveincitieswithhigherincomelevelsandwithhigherVTIratios.
Thus,someofthevariationinincomelevelsandVTIratiosisdrivenbydierencesinthesecharacteristicsacrosscities.
Fromaspatialstandpoint,complexmortgagesaremorecommoningeographicareasthatexperiencedhighhousepriceappreciation.
Theaverage3-yearcumulativepriceappreciationamongcomplexborrowersamountedtoastaggering44%,ascomparedwith30%amongtraditionalFRMborrowers.
Wealsodocumentthatonly12%ofcomplexmortgageswereoriginatedinareasthathadexperiencedfourquartersofdeclinesinhousepricesoverthepreceding10years,asopposedto13%ofFRMsand16%ofARMs.
Unfortunately,wedonotobservetheeducationlevelofborrowersdirectly.
However,wecancomputetheproportionofpeopleinzipcodeswithacollegeeducation.
Householdsusingcomplexmortgagestendtoliveinareaswithahigherproportionofcollegegraduates.
Finally,thepopulationgrowthrateandtheunemploymentrate,whichcapturemacroeconomicconditionsattheMSAlevel,aresimilarinareaswithdierentmortgagecompositions.
Complexmortgagesweresubstantiallymorepopularinnon-recoursestates,wherethelendercannotaccessassetsofthedefaultinghouseholdsbeyondthevalueofthecollateralsecuringtheloan.
Whereasonly22%ofFRMsareinnon-recoursestates,44%ofCMsare14originatedinsuchstates.
Table3breaksoutthekeysummarycharacteristicsamongdierentcomplexmortgagetypes.
Teaserloans,onaverage,appeartobeusedtonancemoreexpensivehomesandareassociatedwithhigherloanvalues.
TheyalsodisplaythehighestVTIratios.
Itisworthnotingthatfewoftheteasercontractsareoeredtosubprimeborrowers.
Asexpected,teaserloanscommonlycarryprepaymentpenalties.
Finally,evenamongcomplexproducts,teaserloansaretakenoutinareaswithhigherhousepriceappreciation,oftentorenanceanexistingmortgageobligation.
Finally,IOcontractsappeartohavebeensubjecttostricterunderwritingcriteria.
Whereasonly11%ofIOswereunderwrittenonthebasisoflessthanfulldocumentation,morethan40%ofNEGAMandTRMloanswereissuedinthismanner.
3.
3GeographicDistributionofMortgagesFigure4showstheconcentrationofcomplexmortgagesindierentcountiesacrosstheUnitedStatesin2002,2005,and2008.
ConsistentwithFigure1,wendthatcomplexmortgageswerefairlyuncommonin2002.
Thedistributionofcomplexmortgageslooksdramaticallydierentin2005,whenmultiplecountiesinCalifornia,Colorado,Florida,andNevadahadCMsharesinexcessof40%.
Insomezipcodesinthesestatesmorethanhalfofmortgageoriginationswerecomplexloans.
Whilethispatternlookssuggestive,numerousareaswithhighhousepriceappreciationhadfewcomplexmortgagesevenatthepeakofthehousingboom.
Forexample,CMcontractsaccountedforonlyabout5%ofloansintheAlbany,NYmetropolitanareawherehousepricesrosebymorethan80%between2001and2007.
Incontrast,CMsprovedtobeverypopularintheDetroitMSA,wherenominalhousepricesremainedatduringthisperiod.
Itisalsoworthnotingthatinsomeareasrapidpriceincreasespreceded15thesurgeinCMcontracts,whereasotherareashadtheoppositerelationship.
103.
4AordabilityofDierentMortgageContractsComplexmortgageproductshaverelativelylowpaymentsduringtheirrstyearsandtherebyenablehouseholdstopurchasemoreexpensivehomes.
Figure5depictstheratiobetweenthemonthlypaymentsofARMsandCMsrelativetofully-amortizingFRMsoriginatedinthesamemonthforborrowerswithsimilarcharacteristics(i.
e.
,loansoriginatedinthesamestateswithsimilarFICOscoresandloan-to-valueratios).
Weobservethat64.
5%ofARMsand85.
6%ofCMshavepaymentsthatarelessthanthoseofcomparableFRMsduringtherstyear.
Furthermore,9.
0%ofARMsand49.
8%ofCMshavepaymentsthataremorethan20%lower.
PanelsBandCshowthatthepaymentsonthevastmajorityofCMsremainlowerthanthoseonFRMseventhreeorveyearsaftertheorigination.
Forexample,wendthatveyearsafteroriginationthepaymentratioislessthanonefor87.
6%ofCMs,andlessthan0.
8for62.
5%ofCMs.
Thus,arelativelysmallfractionofcomplexmortgageshavesubstantialresetsofmortgagepaymentsduringtherstveyearsthatcannotbemanagedbyrenancingintoanewcontract.
11ThisresultindicatesthatCMborrowerscontinuedtohaverelativelylowpaymentsthroughoutthemortgagecrisisof2007-2009.
Mortgagedefaultsduringthecrisiswouldlikelyhavebeensignicantlyhigherifcomplexmortgageshadresettheirminimumpaymentsafterashorterintroductorytimeperiod.
ThendingthatARMsandCMspaymentswerelowerthanthoseforcomparableFRMsforanextendedperiodoftimecanbeexplainedbyseveralfactors.
First,short-terminterestrateshavedecreasedoveroursampleperiodtherebyreducingthepaymentsonARMsandCMs,10GrangercausalitytestscarriedoutattheMSAlevelpresentmixedevidenceoftherelationshipbetweenchangesinhousepricesandCMshares.
Theresultsarealsohighlysensitivetothechoiceofevaluationperiod.
ThissubjectisdiscussedingreaterdetailinaconcurrentpaperbyBarlevyandFisher(2010).
11Unfortunatelywedonothavesucientlylongtimeseriesavailabletostudytheresetsinmoredetailsincemostofthecomplexmortgagesinoursampleareoriginatedbetween2004and2006.
16whicharegenerallytiedtosuchrates.
Second,Figure5onlyshowsthepaymentsofmortgagesthatsurvivedandwerenotpreviouslyrenanced.
Householdsthatobtainmortgageswithlowerinterestratesandlowertotalpaymentsarelesslikelytorenancealoan,resultinginatendencyoftheactualpaymentsonsurvivingARMsandCMstodecreaseovertimerelativetotheFRMs.
Byvirtueoftheiramortizationstructure,complexloanslargelymaintainahighleverageratioovertime.
Figure6depictsthedistributionoftheremainingmortgagebalanceone,three,andveyearsaftermortgageoriginationrelativetotheoriginalbalanceforthethreemortgagecontracttypes.
Evenveyearsafterorigination(PanelC)around51%ofcomplexmortgagesarewithin2.
5%oftheirinitialloanbalanceandaround16%ofborrowersincreasedtheirloanbalancebymorethan2.
5%.
ThiscreatesasharpcontrastwithFRMandARMborrowerswhograduallypaydowntheirmortgages.
Thus,CMborrowerstendtokeepsubstantiallyhigherdebtlevelsthanhouseholdswithmoretraditionalmortgageproducts.
ThismakesCMborrowersmoresusceptibletoeconomicshocks.
Thisdynamicdeteriorationinrelativeleverageratiosbecomesparticularlydramaticintheeventofslowerhousepriceappreciation,asexperiencedduringthehousingcrisisof2007-2009.
123.
5DeterminantsofMortgageChoiceInthissectionweanalyzethedeterminantsofmortgagechoicemoresystematically.
Inpar-ticular,weestimatethelikelihoodofselectionofaparticularmortgagecontracttype(ARMorCM)relativetoabaselinecontract,whichwetaketobeanFRM.
Theserelativelikeli-hoodsareestimatedasafunctionofloan-andborrower-levelcovariates,aswellasMSA-level12Thehigherlong-termloan-to-valueratiosofcomplexloansmayhavecontributedtoafurtherdeteriorationinhousingmarkets,assuggestedbytheleverageeectofStein(1995)andLamontandStein(1999).
Additionalpapersthatstudythemacro-economicaspectsofhousingpricesincludeLustigandVanNieuwerburgh(2005),Ortalo-MagneandRady(2006),Piazzesi,Schneider,andTuzel(2007),BrunnermeierandJulliard(2008),Favilukis,Ludvigson,andVanNieuwerburgh(2010),Landvoigt,Piazzesi,andSchneider(2010),andVanNieuwerburghandWeill(2010).
17aggregates.
Formally,weusemaximumlikelihoodtoestimatethefollowingmultinomiallogitregressions:Prob(Yi=m)Prob(Yi=FRM)=eβmXi+FEStatei+FEYeari+i,(1)whereProb(Yi=m)/Prob(Yi=FRM)isprobabilityofobtaininganARMorCMrelativetoaFRM,Xisavectorofmortgage-speciccovariates,FEYearareindicatorvariablesfortheoriginationyears,andFEStatearegeographicindicatorvariables.
Table4reportstheestimatedcoecients.
ThersttwocolumnsuseonlyindividualhouseholdlevelcharacteristicstoexplainthemortgagechoiceandthelasttwocolumnsincludeMSAlevelaggregatesandstatexedeects.
AllregressionsincludetimexedeectsandthestandarderrorsareclusteredbyMSA.
SincesomeoftheMSAlevelvariablesarenotavailableforthefullsample,thecorrespondingspecicationsincludefewerobservationsthantheoverallsamplesummarizedinTable2.
WendthathouseholdswithhigherincomelevelsaresignicantlymorelikelytoobtainacomplexmortgagethantotakeoutamoretraditionalFRMloan.
Despitetheirhigherincome,thesehouseholdsarestretchingtopurchasemoreexpensivehomes,asindicatedbytheirhigherestimatedcoecientsonvalue-to-income(VTI)ratios.
AlthoughARMloansarealsomorelikelyinhigherVTItransactions,theeconomiceectofVTIisstrongerforCMcontracts.
HouseholdswithlowerFICOscoresaresignicantlymorelikelytochooseanARMoraCMcontract,althoughthecoecientestimateissubstantiallysmallerforCMsthanforARMs.
Thethemeofcomplexmortgagesas"aordabilityproducts"forhouseholdswithpref-erencesforrelativelyexpensivehomesrelativetotheirincomesisreectedinseveralothercoecients.
Forinstance,wendthatCMcontractsaremuchmoreprevalentformortgages18abovetheGSEconformingloanlimit.
Suchmortgagesaresubjecttotheso-calledjumbospread,whichincreasestherelativeappealofpayment-shrinkingCMproducts.
Moststrik-ingly,however,CMborrowersaremuchmorelikelytoprovideincompletedocumentationfortheirloans.
ThegreaterrelianceofCMcontractsonlow-documentationunderwritingiscon-sistentwithborrowereorttoinatetheirincometoqualifyforahigherloanamountneededforanexpensivehouse.
Overall,thereislittleevidencethatatypicalcomplexmortgageistakenoutbyarelativelypoorandnaivehousehold.
Wendthatthetypeofpropertyhasanimpactonmortgagecontractchoice.
MortgagesusedtonancecondominiumsandinvestmentpropertiesaremorelikelytobeARMsorCMs.
Complexmortgagesmightbeparticularlyattractiveforsuchtypesofproperties,sinceownersofcondominiumsandinvestmentpropertieshavepotentiallyfewerindirectcostsofstrategicallydefaultingontheirproperties.
Theymightthereforehaveanincentivetopaydowntheirmortgagebalancerelativelyslowlytoincreasetheoptionvalueofstrategicdefault.
Wealsondthathouseholdsinnon-recoursestatesaresignicantlymorelikelytoobtainacomplexmortgagethanhouseholdsinrecoursestates.
Thismightalsobecausedbythehigheroptionvalueofdefaultinginnon-recoursestates.
Householdsinsuchstateshavesmallerincentivestopaydowntheirmortgagesastheycansimplywalkawayincaseofdefaultwithoutworryingaboutthelenderaccessingtheirotherassets.
However,itisinterestingthatlendersdidnotcurtailtoamoresignicantdegreetheprevalenceofcomplexloansinnon-recoursestates.
ItispossiblethatthepositiveassociationbetweenCMcontractchoiceandbothVTIandincomereectsthepropensityofCMstobeconcentratedinhighincomeandhighhousepriceMSAs.
However,specicationsthatincorporateMSA-levelcontrolsandstatexedeectspreservetheserelationships.
Althoughsomeofthecoecientsareattenuatedinthosespec-ications,theyremainhighlysignicant.
Thissuggeststhatwithinindividualgeographies,19complexmortgagechoiceisfavoredbytherelativelywell-othatarestretchingtheirbudgetowconstraintstoaordmoreexpensivehouses.
Complexmortgagesarebackloadedcontractsinwhichreducedmortgagepaymentsarefollowedbyhigherpaymentsneededtocatchuponthedelayedprincipalrepayment.
Thereareseveralexplanationsjustifyingthispreferenceforanincreasingpaymentpath.
First,individualhouseholdsmightanticipatefutureincomegrowth,dueeithertofavorablelocaleconomicconditionsortotheirpersonalwageprole,especiallyforyoungerhouseholds.
Forthesehouseholdsitmakessensetopurchaseexpensivehomesrelativetotheirincomesunderthepermanentincomehypothesis(Gerardi,Rosen,andWillen(2010)andCocco(2010)).
Second,householdsmightexpecthousepricestoappreciateinthefuture,whichenablesthemtorenancetheirloanstomeetthehigherfuturepayments(BarlevyandFisher(2010)).
Third,thepopularityofthesebackloadedproductsmightbeanoutcomeoflaxlendingstandardsduetoagencyissues,inwhichlenderscareonlyaboutthefeesgeneratedfromoriginatingtheloansandnotaboutfuturedefaultswhentheyselltheloansviasecuritization(Carlin(2009),Keys,Mukherjee,Seru,andVig(2010)andJiang,Nelson,andVytlacil(2010a)).
Wecannotperfectlyseparatethesethreeexplanations.
However,resultsinTable4shedsomelightontheirrelativeimportanceinthechoiceofmortgagecontracts.
Sincewecannotobservehouseholdexpectationsfortheirincomeandhousepricegrowth,weusethepriorthreeyears'housepriceappreciationandanindicatorvariableforwhethertheareaexperiencedanannualdeclineoverthepriortenyearsasproxiesforexpectedincomeandhousepricegrowthrates.
Thesetwovariablescapturetheextenttowhichhouseholdsextrapolatepastlocalexperiencestobuildtheirexpectationsaboutfuturehousepricedynamics.
Borrowersandlendersinareaswhichexperiencedarecentdeclineinhousepricesmighthavebeenmorecautiousinchoosinginstrumentsthatexhibitloworevennegativeamortization.
Ontheotherhand,borrowersandlendersingeographicareaswhereappreciationwassubstantial20mighthavebeenmorewillingtoacceptnon-amortizingloansiftheyexpectedtheappreciationtocontinueinthefuture.
Inaddition,weincludethepriorone-yearpopulationgrowthrateintheMSAasaproxyforexpectedincomeandhousepricegrowth.
Geographicareaswithsignicantpopulationgrowthmightbeareaswherehouseholdsexpectsignicanthousepriceandincomegrowth.
Wendthatthepricedeclineindicatorvariableandthepopulationgrowthratesigni-cantlyaectthechoiceofCM.
Inparticular,CMcontractsaremorepopularinareasthatdidnotexperienceanannualhousepricedeclineoverthepriortenyearsandinareaswithhighpopulationgrowth.
Thisevidencesuggeststhattheexpectationsofcontinuedhousepriceandincomegrowtharelikelyadrivingforcebehindthepopularityofcomplexmortgages.
Finally,ifcomplexmortgagesareaectedbyagencyconictsandarepushedtonaivehouseholdstomaximizethecommissionsforloanocers,thenwemightexpecttheseloanstobeconcentratedinlowincomeareaswithpoorlyeducatedhouseholds.
Wedonotndsupportforthishypothesis.
Citieswithlowerproportionsofcollegeeducatedhouseholdsandwithlowermedianincomesdonotexhibithigherproportionsofcomplexloans.
Table5reportsthecoecientsofmultinomiallogitregressionsthatfurtherdierentiatebetweenvarioustypesofcomplexcontracts.
TheestimatesareconsistentwiththeunivariateresultsinTable2.
Inparticular,weseethatNEGAMandespeciallyTRMcontractswereusedbyhigh-incomeborrowerstorenancetheirhigh-pricedprimaryresidences,oftenonthebasisofonlylimitedincomeandassetdocumentation.
Itislikelythatsuchrenancingswereserialinnature,whichwouldfurtherunderscorethefragilityofsuchcontractsinenvironmentswheretherenancingmarketsfreezeup.
214MortgageDelinquenciesInthissectionwestudythedelinquencyofdierenttypesofmortgages.
Amortgageisdelinquentiftheborrowerisatleast60dayslateinmakingthemortgagepayments.
4.
1ReasonsforMortgageDelinquenciesDelinquenciesmightdieracrossmortgagetypesforvariousreasons.
First,ARMsandCMsaregenerallyadjustedaccordingtoshort-terminterestratesandmighthavehigherdelin-quencyratesbecausetheirmortgagepaymentsincreaseinarisinginterestrateenvironment.
Overoursampleperiodtheinterestrateshavenotrisensubstantially,suggestingthatthischannelislikelynotofsignicantimportance.
Second,CMsgenerallyexhibitanincreasingpaymenttrendoverthelifeoftheloansincetheinitialpaymentsarenotfullyamortizingasdescribedpreviously.
Mortgagedelinquenciesmightbecomemorelikelyafterthevariousresetswhenthepaymentssuddenlyincrease.
Ontheotherhand,CMsmightexhibitlowerdelinquencyratesduringtheinitialperiodwhenmortgagepaymentsarerelativelylow.
Somecomplexmortgagecontracts(e.
g.
,OptionARMs)giveborrowerstheexibilitytoadjusttheirmortgagepaymentsastheirincomelevelsuctuate,whichmightreducetheprobabilityofdefaults.
AsweobserveinFigure5,mostcomplexmortgageshavelowermortgagepaymentsthancorrespondingFRMsorARMsovertherstveyearssinceorigination.
Third,CMspaydowntheirmortgagebalanceataslowerratethanFRMsandARMsassummarizedinFigure6.
Therefore,borrowersofcomplexloanshaveabiggerincentivetodefaultontheirloansincaseofcashowdicultiesorforstrategicreasons.
Whereasaborrowerwithacomplexmortgagemightjustwalkawayfromtheirmortgagecontractiftheyexperiencenancialdiculties,aborrowerwithaFRMoranARMmightbemorelikelyto22selltheirhomesincetheembeddedequityishigherforfullyamortizingmortgagecontracts.
Fourth,borrowersthatareattractedtoARMsandCMsmightdierintheirpreferences.
Borrowersthatarewillingtobearinterest-rateriskmightbemorerisk-tolerantasshownbyCampbellandCocco(2003).
Finally,borrowersusingtraditionalmortgageproductsmightbemoreinuencedbyethicalnormsthatmotivatethemtopaybacktheirdebtevenifitwouldbemoreeconomicaltodefaultonamortgagecontract,asdiscussedbyGuiso,Sapienza,andZingales(2009).
4.
2SummaryofMortgageDelinquencyPanelAofTable6reportstheproportionofmortgagesthataredelinquentafterone,three,andveyearsbymortgagetype.
WeobservethatFRMshavethelowestdelinquencyratesatallhorizons,CMshavelowerdelinquencyratesthanARMsataoneyearhorizonbuthigherdelinquencyratesatlongerhorizons.
Forexample,22.
75%ofCMs,18.
48%ofARMs,and11.
95%ofFRMsaredelinquentata5-yearhorizon.
Thus,atlongerhorizonstheprobabilityofdelinquencyincreasesforCMs.
Figure7showstheproportionofmortgagedelinquenciesforFRMs,ARMs,andCMsfortherstveyearsafterorigination.
Ineachmonthwedepicttheproportionofremainingmortgagesthatbecomedelinquentforthersttime.
Weobservethatcomplexmortgageshavestrictlyhigherdelinquencyratesthanxedratemortgagesatallhorizons.
Mortgagedelinquenciesofcomplexloansreachpeaksof1.
3%and1.
2%ofsurvivingloansafter27and39monthssinceorigination.
Thesepeaksoccurthreemonthsaftercommonresetintervals,sincedelinquencybeginswhenamortgagepaymentisatleast60dayslate.
WeobserveasimilarpeakforARMsafterahorizonof27months.
WhereasARMshaveslightlyhigherratesofdelinquencyatshorthorizons,CMshavesubstantiallyhigherratesatlongerhorizons.
Itmustbekeptinmindthatborrowersof23complexloanshaverelativelyhighdelinquencypropensitiesdespitehavingsignicantlyhighercreditscoresthanARMborrowers,assummarizedinTable2.
Itisalsoinsightfulthatthedelinquencyrateincreasessubstantiallyevenbeforetheminimumloanpaymentsareresetaftertwoorthreeyears,indicatingthatsomeborrowersofcomplexloansdonotevenmaketherelativelylowinitialmortgagepayments.
4.
3HazardRateModelToinvestigatethedeterminantsofmortgagedelinquencies,werunthefollowingCoxpropor-tionalhazardmodel:h(i,t)=h0(t)eβXi,t+FEYeari+FEYeart+FEStatei+,(2)wherethehazardrateh(t)istheestimatedprobabilityofrsttime60daydelinquencyattimetconditionalonsurvivingtotimet,h0(t)isthebaselinehazardrate,Xisavectorofhousehold-speciccovariates,andFEYeariandFEYeartaretwoindicatorvariablesfortheoriginationyearandcalendaryearstocontrolfordierentvintageeectsandmacroeconomicconditions.
Insomespecications,wealsoincludeFEStatetocontrolforstatexedeects.
Theloansampleisexpandedtoaloan-yearlevelsothattime-varyingcovariatescanbein-cluded.
Also,timeisscaledsothattherstobservationdateisthecalendaryearoforigination(time0),andsubsequentcalendaryearsaremeasuredrelativetotheyearoforigination.
Im-plicitly,loansofdierentvintagesarecomparedwitheachother,sothatthebaselinehazardrepresentstheprobabilityofdelinquencyforaborrowerwithcovariatesof0attyearsafterorigination.
Insomespecicationwesplitupcomplexmortgagesintothethreesub-types(IO,NEGAM,andTRM).
Table7reportstheestimatedcoecientsofthepropensityofrsttime60daydelinquency,24sothatthechangeinprobabilityofdelinquencycanbereadasoddsratios.
Forexample,incolumn1,thecoecientof0.
792forCMmeansthattheratiooftheprobabilityofdelinquencyforaborrowerwithacomplexmortgageandtheprobabilityofdelinquencyforaborrowerwithsimilarcharacteristicsbutaxedratemortgageise1*0.
792/e0*0.
792=2.
2;orthecomplexborrowerisabout2.
2timesmorelikelytobedelinquent.
Inthersttwocolumns,weuseonlyborrowercharacteristicsatthetimeofloanorigi-nationtoestimatethedelinquencyprobability.
Inlasttwocolumns,weincludetime-varyingcharacteristicsandstatexedeects.
ThecurrentLTVratioisdenedasthemortgageloanamountattheendofthepriorperioddividedbythecurrenthomevalue.
ThecurrenthomevalueisestimatedbyadjustingthehomevalueatoriginationbythehousepriceappreciationattheMSAlevelsincetheorigination.
Householdswithcomplexloanswillpaydowntheirmortgagesataslowerpace(asillustratedinFigure6)andwillhavehighercurrentLTVratios.
Inaddition,areaswithhousepricedeclineswillhavehighercurrentLTVratios.
HouseholdswithLTVratiosexceeding100%willhavehigherincentivestodefaultontheirloanssincetheydonothaveanyhomeequityatstake.
Thus,includingthecurrentLTVratiointhehazardmodelcontrolsfordynamicleveragelevels,whichdieracrossmortgagetypes.
Finally,theunemploymentlevelcapturestheproportionofunemployedinanMSAandtheincomegrowthisdenedasthegrowthrateofincomeattheMSAlevelsincethemortgagewasoriginated.
WendthatCMshavesignicantlyhigherdelinquencyratesthanFRMsinallspecica-tions.
Delinquencyratesareparticularlyhighforteaserratemortgages,whicharepresumablytheleasttransparentmortgagecontractweanalyze.
HouseholdsthatborrowusingARMsalsohavesignicantlyhigherpropensitiestobedelinquent,althoughthecoecientestimateissub-stantiallysmallerthanthecoecientoncomplexmortgages.
Thepropensitytobedelinquentdecreaseswiththeincomelevelatorigination.
Furthermore,borrowerswithlowercreditscores,subprimeborrowers,loansoriginatedwithlowornodocumentation,loansabovethe25conforminglimit,andinvestmentpropertiesaresignicantlymorelikelytobedelinquent.
ThelasttwocolumnsconsidertheimpactoftheadditionalMSAlevelvariablesandstatexedeects.
Wendthathouseholdsinareaswithhighunemploymentanddepressedin-comegrowthsincetheoriginationoftheloanaremorelikelytobedelinquent,suggestingthatthedicultytomeetcashowpaymentiscertainlyadriverofmortgagedelinquency.
However,localincomeshocksarelikelytoaectborrowersofdierentmortgagessimilarly.
Toinvestigatetheimpactofhousepriceappreciationanddierentamortizationschedules,weincludethecurrentLTVratio.
WendthathouseholdswithhighercurrentLTVratiosaresignicantlymorelikelytodefault,suggestingthatstrategicdefaultislikelyacontribu-tortomortgagedelinquencyaswell.
ThissourceofdelinquencyisalsolikelytoexplainthesignicantlyhigherdelinquencyrateforCMsovertime,sincetheLTVforCMsincreasessig-nicantlyovertimerelativetoARMsorFRMsduetotheloworevennegativeamortizationintherstfewyears.
ItisalsoremarkablethatthecoecientsonCMsremainhighlystatisticallysignicantevenaftercontrollingforthecurrentLTVratio,thelocalincomegrowthrate,thelocalunem-ploymentrate,andstatexedeects,suggestingthatCMborrowersmightbefundamentallydierentfromFRMborrowers.
Theymightbemoreriskseekingingeneral,asrevealedbytheirchoicesforCMcontracts.
Theymighthaveriskierincomeormightbemorereceptivetotheideaofstrategicdefault.
Additionalworkisneededtofullydisentanglethevarioussourcesofdelinquency.
TheseresultsareconsistentwiththestructuralmodelofCorbaeandQuintin(2010),whondthatthepresenceofnontraditionalmortgagesampliedtheforeclosurecrisisbetweentherstquarterof2007andtherstquarterof2009.
264.
4BankruptcyThedecisiontodefaultonamortgageisrelatedtothedecisiontodeclarebankruptcy.
Con-trastingthedeterminantsofpersonalbankruptcywiththedeterminantsofmortgagedelin-quencygivesusimportantinsightsaboutthemotivationofthedelinquencybehavior.
Itisnotnecessarythathouseholdsthatdefaultontheirmortgagesarealsodeclaringbankruptcy.
Norisitnecessarythathouseholdsthatdeclarebankruptcydefaultontheirmortgages.
Forexample,inoursampleonly13%ofhouseholdsthataredelinquentontheirmortgagealsode-clarebankruptcy.
Furthermore,only29%ofhouseholdsthatdeclarebankruptcyalsodefaultontheirmortgageloans.
13Bankruptcyissignicantlylesscommonthanmortgagedefaults.
Inoursample,13%ofmortgagesbecomedelinquentatanytimeduringtheirlife,whereasonly2%ofmortgageborrowersalsodeclarebankruptcy.
PanelBofTable6showstheproportionofhouseholdswithdierentmortgagetypesthatdeclarebankruptcy.
WeobservethatFRMshavethelowestbankruptcyrateatallhorizons.
HouseholdsborrowingusingCMshavehigherbankruptcyratesthanARMsataveyearhorizon.
Forexample,3.
18%ofCMs,2.
94%ofARMs,and2.
15%ofFRMshouseholdsdeclarebankruptcywithinaveyearhorizonaftertheyoriginateamortgage.
Table8reportsthepropensityofhouseholdstodeclarepersonalbankruptcyandcontrastsitwiththosethataredelinquentontheirmortgage.
Notsurprisingly,mostcoecientshavethesamesignsinbothregressions.
Forexample,higherincomeandhigherFICOscoresreducethepropensitiesofbothdelinquencyandbankruptcy.
Itisinterestingthatsomevariablesshowupwithdierentsignsinthetworegressions.
Forexample,althoughinvestmentpropertieshavehighermortgagedelinquencyrates,house-holdswithinvestmentpropertiesarelesslikelytoleforpersonalbankruptcy.
Thisevidence13SeeLi,White,andZhu(2010)foradiscussionoftherelationshipbetweenbankruptcylawsandmortgagedefaults.
27suggeststhatownersofinvestmentpropertiesaremorelikelytowalkawayfromthepropertywhenitiseconomicaltodoso,eveniftheycanaordtocontinuethemortgagepayment.
Moreover,loanswithlowdocumentationaremorelikelytobedelinquent,butthatvariabledoesnotpredictpersonalbankruptcy,suggestingthatthesehouseholdsmightbemorelikelytostrategicallydefault.
4.
5PrepaymentAnotherreasonthathouseholdsgointodelinquencyisthattheycannotrenancetheirpre-viousmortgagewhentheyhaveahighLTVratioorexperienceabadincomeshock.
PanelCofTable6summarizestheproportionofmortgagesthatareprepaid.
Mortgagesareprepaidiftheborrowerspay-otheirloanbeforematurityeitherbyrenancingtheloanorbypayingothemortgageusingtheproceedsfromsellingthehouseorthroughothermeans.
WendthatARMsaremorelikelytobeprepaidthanFRMs,whileCMshaveintermediatelevelsofprepayments.
Unfortunately,wedonotobservewhetherhouseholdsprepaytheirmortgagestorenancetheirloanorwhethertheyprepaytheirmortgagesbecausetheysoldtheirhomes.
ThelastcolumnofTable8reportsthepropensityofhouseholdstoprepay.
Manyvariableshavetheoppositesignforthedelinquencyandtheprepaymentregressions,sincevariablesthatincreasetheprobabilityofprepaymentlikelywilldecreasetheprobabilityofdelinquency.
Forexample,loanswithhighcurrentLTVarelesslikelytobeprepaidandmorelikelytogointodelinquency.
However,therearesomeexceptions.
Forexample,CMsandARMsarebothmorelikelytobeprepaidandmorelikelytogointodelinquency.
Loansthatwereusedtorenanceanotherloanarebothlesslikelytobeprepaidandlesslikelybedelinquent.
285ConclusionsTherecenthousingcrisisbroughttheextensionofcredittosubprimeborrowersandagencyproblemsinherentinmortgagesecuritizationtotheforefrontofacademicresearch.
Thispaperfocusesonadierentaspectofcreditmarketsduringthistime–namely,theproliferationofnon-amortizingmortgages.
Inadditiontovariableinterestrates,suchmortgagesalsofeaturedchangesinamortizationschedulessetobyavarietyoftriggers.
Thesecomplexmortgagecontractsbecameextremelypopularduringthemid2000sandvanishedalmostcompletelyafterthehousingcrisisof2007-2009.
Wendthatcomplexmortgageswerethecontractofchoiceforrelativelyhighcreditqualityandhigh-incomehouseholdsseekingtopurchasehousesthatwereexpensiverelativetotheirincomes.
WefurtherndthatCMcontractswerenotsimplyaninevitableoutcomeofhighhouseprices.
Evenwithinhighhousepriceareasthesecontractsareassociatedwithhouseholdsstretchingtoaordmoreexpensivehouses,oftenonthebasisofstatedincomealone.
Wedocumentthatcomplexmortgagesexperiencedsubstantiallyhigherdefaults,controllingforavarietyofborrowerandloancharacteristics,aswellasmacroeconomicshocks.
Higherdefaultratescannotbeattributedsolelytogreaterleverageofcomplexmortgagesandtheonsetofamortizationresetsbroughtaboutbyinabilitytorenancecomplexloans.
ThatcomplexloansweremorelikelytobeunderwrittenusingstatedincomemayalsoindicategreaterinherentearningsvariabilityofCMborrowers,whichwouldmakethemmoresusceptibletoeconomicshocks.
29ReferencesAdelino,M.
,K.
Gerardi,andP.
Willen(2009).
Whydon'tlendersrenegotiatemorehomemortgagesRedefaults,self-cures,andsecuritization.
NBERWorkingPaper15159.
Agarwal,S.
,B.
W.
Ambrose,S.
Chomsisengphet,andA.
B.
Sanders(2010).
Thyneighbor'smortgage:Doeslivinginasubprimeneighborhoodaectone'sprobabilityofdefaultForthcoming:RealEstateEconomics.
Amromin,G.
,J.
Huang,andC.
Sialm(2007).
Thetradeobetweenmortgageprepaymentsandtax-deferredsavings.
JournalofPublicEconomics91,2014–2040.
Bajari,P.
,C.
S.
Chu,andM.
Park(2010).
Anempiricalmodelofsubprimemortgagedefaultfrom2000to2007.
UniversityofMinnesotaandFederalReserveBoard.
Barlevy,G.
andJ.
Fisher(2010).
Backloadedmortgagesandhousepricespeculation.
FederalReserveBankofChicago.
Berndt,A.
,B.
Hollield,andP.
Sandas(2010).
Theroleofmortgagebrokersinthesubprimecrisis.
CarnegieMellonUniversity.
Bond,P.
,D.
K.
Musto,andB.
Yilmaz(2009).
Predatorymortgagelending.
JournalofFinancialEconomics94,412–427.
Brunnermeier,M.
K.
andC.
Julliard(2008).
Moneyillusionandhousingfrenzies.
ReviewofFinancialStudies21,135–180.
Campbell,J.
Y.
(2006).
Householdnance.
JournalofFinance61,1553–1604.
Campbell,J.
Y.
andJ.
F.
Cocco(2003).
Householdriskmanagementadnoptimalmortgagechoice.
QuarterlyJournalofEconomics118,1449–1494.
Campbell,J.
Y.
,S.
Giglio,andP.
Pathak(2010).
Forcedsalesandhouseprices.
Forthcom-ing:AmericanEconomicReview.
Carlin,B.
I.
(2009).
Strategicpricecomplexityinretailnancialmarkets.
JournalofFi-nancialEconomics91,278–287.
Carlin,B.
I.
andG.
Manso(2010).
Obfuscation,learning,andtheevolutionofinvestorsophistication.
UniversityofCaliforniaLosAngelesandMIT.
Cocco,J.
F.
(2010).
Understandingthetrade-osofalternativemortgageproducts.
LondonBusinessSchool.
Corbae,D.
andE.
Quintin(2010).
Mortgageinnnovationandtheforeclosureboom.
Uni-versityofTexasandUniversityofWisconsin.
Demyanyk,Y.
andO.
V.
Hemert(2010).
Understandingthesubprimemortgagecrisis.
Forthcoming:ReviewofFinancialStudies.
Dunn,K.
B.
andJ.
J.
McConnell(1981).
Valuationofmortgage-backedsecurities.
JournalofFinance36,599–617.
Dunn,K.
B.
andC.
S.
Spatt(1999).
Calloptions,points,anddominancerestrictionsondebtcontracts.
JournalofFinance54,2317–2337.
Favilukis,J.
,S.
C.
Ludvigson,andS.
VanNieuwerburgh(2010).
Themacroeconomicef-fectsofhousingwealth,housingnance,andlimitedrisk-sharingingeneralequilibrium.
LondonSchoolofEconomicsandNewYorkUniversity.
Gabaix,X.
,A.
Krishnamurthy,andO.
Vigneron(2007).
Limitsofarbitrage:Theoryandevidencefromthemortgagebackedsecuritiesmarket.
JournalofFinance62,557–596.
30Garmaise,M.
(2010).
Afterthehoneymoon:Relationshipdynamicsbetweenmortgagebrokersandbanks.
UniversityofCaliforniaatLosAngeles.
Gerardi,K.
S.
,H.
S.
Rosen,andP.
S.
Willen(2010).
Theimpactofderegulationandnan-cialinnovationonconsumers:Thecaseofthemortgagemarket.
JournalofFinance65,333–360.
Ghent,A.
C.
andM.
Kudlyak(2010).
Recourseandresidentialmortgagedefault:TheoryandevidencefromU.
S.
states.
BaruchCollegeandFederalReserveBankofRichmond.
Glaeser,E.
L.
,J.
Gottleb,andJ.
Gyourko(2010).
CancheapcreditexplainthehousingboomHarvardUniversity.
Goetzmann,W.
N.
,L.
Peng,andJ.
Yen(2010).
Thesubprimecrisisandhousepriceappreciation.
YaleUniversityandUniversityofColorado.
Guiso,L.
,P.
Sapienza,andL.
Zingales(2009).
Moralandsocialconstraintstostrate-gicdefaultonmortgages.
EuropeanUniversityInstitute,NorthwesternUniversity,andUniversityofChicago.
Jiang,W.
,A.
A.
Nelson,andE.
Vytlacil(2010a).
Liar'sloanEectsoforiginationchannelandinformationfalsicationonmortgagedelinquency.
ColumbiaUniversity.
Jiang,W.
,A.
A.
Nelson,andE.
Vytlacil(2010b).
Securitizationandloanperformance:Acontrastofexanteandexpostrelationsinthemortgagemarket.
ColumbiaUniversity.
Keys,B.
J.
,T.
Mukherjee,A.
Seru,andV.
Vig(2009).
Financialregulationandsecuriti-zation:Evidencefromsubprimeloans.
JournalofMonetaryEconomics56,700–720.
Keys,B.
J.
,T.
Mukherjee,A.
Seru,andV.
Vig(2010).
Didsecuritizationleadtolaxscreeingevidencefromsubprimeloans.
QuarterlyJournalofEconomics125,307–362.
Koijen,R.
S.
J.
,O.
VanHemert,andS.
VanNieuwerburgh(2009).
Mortgagetiming.
JournalofFinancialEconomics93,292–324.
Lamont,O.
andJ.
C.
Stein(1999).
Leverageandhouse-pricedynamicsinU.
S.
cities.
RANDJournalofEconomics30,498–514.
Landvoigt,T.
,M.
Piazzesi,andM.
Schneider(2010).
Thehousingmarket(s)ofsandiego.
StanfordUniversity.
Li,W.
,M.
J.
White,andN.
Zhu(2010).
DidbankruptcyreformcausemortgagedefaulttoriseFederalReserveBankofPhiladelphia,UniversityofCaliforniaatSanDiego,andUniversityofCaliforniaatDavis.
Longsta,F.
A.
(2005).
Borrowercreditandthevaluationofmortgage-backedsecurities.
RealEstateEconomics33,619–661.
Loutskina,E.
andP.
E.
Strahan(2009).
Securitizationandthedecliningimpactofbanknancialconditiononloansupply:Evidencefrommortgageoriginations.
JournalofFinance64,861–922.
Lustig,H.
andS.
VanNieuwerburgh(2005).
Housingcollateral,consumptioninsuranceandriskpremia:Anempiricalperspective.
JournalofFinance60,1167–1219.
Mayer,C.
,K.
Pence,andS.
Sherlund(2009).
Theriseinmortgagedefaults.
JournalofEconomicPerspectives23,23–50.
Mian,A.
andA.
Su(2009).
Theconsequencesofmortgagecreditexpansion:EvidencefromtheU.
S.
mortgagedefaultcrisis.
QuarterlyJournalofEconomics124,1449–1496.
Ortalo-Magne,F.
andS.
Rady(2006).
Housingmarketdynamics:Onthecontributionofincomeshocksandcreditconstraints.
ReviewofEconomicStudies73,459–485.
31Piazzesi,M.
,M.
Schneider,andS.
Tuzel(2007).
Housing,consumption,andassetpricing.
JournalofFinancialEconomics83,531–569.
Piskorski,T.
,A.
Seru,andV.
Vig(2010).
Securitizationanddistressedloanrenegotia-tion:Evidencefromthesubprimemortgagecrisis.
Forthcoming:JournalofFinancialEconomics.
Piskorski,T.
andA.
Tchistyi(2010).
Optimalmortgagedesign.
ReviewofFinancialStud-ies23,3098–3140.
Purnanandam,A.
(2010).
Originate-to-distributemodelandsubprimemortgagecrisis.
Uni-versityofMichigan.
Rajan,U.
,A.
Seru,andV.
Vig(2010).
Thefailureofmodelsthatpredictfailure:Distance,incentivesanddefaults.
UniversityofMichigan,UniversityofChicago,andLondonBusinessSchool.
Schwartz,A.
(2007).
Householdrenancingbehaviorinxedratemortgages.
HarvardUni-versity.
Schwartz,E.
S.
andW.
N.
Torous(1989).
Prepaymentandthevaluationofmortgage-backedsecurities.
JournalofFinance44,375–392.
Stanton,R.
(1995).
Rationalprepaymentandthevaluationofmortgage-backedsecurities.
ReviewofFinancialStudies8,677–708.
Stanton,R.
andN.
Wallace(2009).
CMBSsubordination,ratingsination,andthecrisisof2007-2009.
UniversityofCaliforniaatBerkeley.
Stanton,R.
andN.
Wallace(2010).
Thebear'slair:Indexedcreditdefaultswapsandthesubprimemortgagecrisis.
UniversityofCaliforniaatBerkeley.
Stein,J.
C.
(1995).
Pricesandtradingvolumeinthehousingmarket:Amodelwithdown-paymenteects.
QuarterlyJournalofEconomics110,379–406.
VanNieuwerburgh,S.
andP.
-O.
Weill(2010).
WhyhashousepricedispersiongoneupReviewofEconomicStudies77,1567–1606.
Vickery,J.
(2007).
Interestratesandconsumerchoiceintheresidentialmortgagemarket.
FederalReserveBankofNewYork.
Woodward,S.
E.
andR.
E.
Hall(2010).
Diagnosingconsumerconfusionandsub-optimalshoppingeort:Theoryandmortgage-marketevidence.
NBERWorkingPaper16007.
32Table1:SummaryStatisticsThistablereportsmeans,standarddeviations,medians,andrstandthirdquartilesforourdatasample.
MeanStd.
Dev.
1stQuart.
Median3rdQuart.
LoanAmount218,065181,464108,300168,000268,918HouseValue317,294297,950145,000234,000388,000Income100,21188,25150,00075,000117,000VTI3.
541.
942.
223.
184.
41FirstLienLTV0.
750.
180.
670.
790.
86FICO70767662715762FICOlessthan6200.
110.
31000Subprime0.
070.
26000LowDocumentation0.
140.
34000Condo0.
130.
34000InvestmentProperty0.
100.
30000Renance0.
410.
49001WithPrepaymentPenalty0.
130.
34000PrepaymentPenaltyTerm(inMonths)30.
1713.
4824.
0036.
0036.
00AboveConformingLimit0.
110.
31000GovernmentSecuritized0.
640.
48011PrivateSecuritized0.
250.
43001InitialInterestRate(in%)5.
941.
445.
506.
006.
50HypotheticalFRMInterestRate(in%)6.
190.
455.
886.
136.
50MSAlevelvariablesMedianIncome77,64120,68962,00074,00088,000MedianVTI3.
280.
822.
603.
153.
80HousePriceChangePrior3Years0.
330.
210.
140.
290.
46DecreaseinHousePricesPrior10Years0.
130.
34000CollegeorMore0.
350.
160.
220.
320.
45PopulationGrowth(in%)1.
101.
430.
290.
821.
74UnemploymentRate(in%)5.
031.
404.
104.
805.
70Non-RecourseState0.
270.
44001NumberofObservations10,208,52233Table2:SummaryStatisticsbyMortgageTypeThistablereportssummarystatisticsforFixedRateMortgages(FRM),AdjustableRateMortgages(ARM),andComplexMortgages(CM).
AllFRMARMCMLoanAmount218,065179,415220,374357,887HouseValue317,294264,878309,465513,728Income100,21188,642101,005141,998VTI3.
543.
403.
464.
07FirstLienLTV74.
1773.
8877.
0173.
45FICOCreditScore707710684710FICOlessthan6200.
110.
100.
230.
07Subprime0.
070.
030.
240.
10LowDocumentation0.
140.
110.
090.
26Condo0.
130.
110.
170.
18InvestmentProperty0.
100.
090.
110.
11PrepaymentPenalty0.
130.
060.
250.
33PrepaymentPenaltyTerm(inMonths)30.
1737.
3927.
5727.
85Renance0.
410.
410.
340.
45AboveConforminglimit0.
110.
050.
130.
33GovernmentSecuritized0.
640.
790.
430.
24PrivateSecuritized0.
250.
150.
410.
54InitialInterestRate(in%)5.
946.
165.
975.
12HypotheticalFRMInterestRate(in%)6.
196.
176.
206.
23MSAlevelvariablesMedianIncome77,64174,10576,53091,254MedianVTI3.
283.
133.
283.
84HousePriceChangePrior3Years0.
330.
300.
320.
44DecreaseinHousePricesPrior10Years0.
130.
130.
160.
12CollegeorMore0.
350.
330.
360.
39PopulationGrowth(in%)1.
101.
111.
121.
08UnemploymentRate(in%)5.
035.
045.
214.
87Non-RecourseState0.
270.
220.
260.
44NumberofObservations10,208,5227,071,3171,202,3831,934,82234Table3:SummaryStatisticsofComplexLoansbyMortgageTypeThistablereportssummarystatisticsfordierenttypesofcomplexmortgagesincludingInterest-OnlyMortgages(IO),NegativeAmortizationMortgages(NEGAM),andTeaserRateMortgages(TRM).
AllCMIONEGAMTRMLoanAmount357,887352,757343,059393,023HouseValue513,728501,394497,894571,770Income141,998141,348135,024153,249VTI4.
074.
034.
024.
27FirstLienLTV73.
4574.
0574.
1870.
67FICOCreditScore710720689710FICOlessthan6200.
070.
050.
160.
03Subprime0.
100.
080.
230.
00LowDocumentation0.
260.
110.
420.
49Condo0.
180.
200.
170.
15InvestmentProperty0.
110.
140.
060.
08PrepaymentPenalty0.
330.
140.
390.
83PrepaymentPenaltyTerm(inMonths)27.
8528.
0128.
2827.
38Renance0.
450.
340.
540.
64AboveConforminglimit0.
330.
320.
290.
42GovernmentSecuritized0.
240.
310.
220.
06PrivateSecuritized0.
540.
530.
510.
57InitialInterestRate(in%)5.
125.
996.
031.
30HypotheticalFRMInterestRate(in%)6.
236.
246.
316.
10MSAlevelvariablesMedianIncome91,25489,39092,52595,133MedianVTI3.
843.
753.
864.
07HousePriceChangePrior3Years0.
440.
430.
430.
49DecreaseinHousePricesPrior10Years0.
120.
110.
110.
16CollegeorMore0.
390.
400.
360.
39PopulationGrowth(in%)1.
081.
180.
980.
93UnemploymentRate(in%)4.
874.
725.
035.
08Non-RecourseState0.
440.
390.
490.
55NumberofObservations1,934,8221,087,058484,574363,19035Table4:MortgageChoiceMultinomialLogitRegressionsThistablereportsthecoecientsofmultinomiallogitregressionsformortgagechoice.
Thecoe-cientsaremeasuredrelativetoFRM.
Thesignicancelevelsareabbreviatedwithasterisks:One,two,andthreeasterisksdenotesignicanceatthe10,5,and1%level,respectively.
Individual-levelCovariatesStateFixedEectsARMCMARMCMLog(Income)0.
4400.
7730.
2740.
507(0.
024)(0.
034)(0.
015)(0.
028)Value-to-Income0.
0800.
1260.
0280.
041(0.
013)(0.
016)(0.
006)(0.
009)FICO/1000.
3790.
0540.
4050.
053(0.
013)(0.
020)(0.
014)(0.
022)Subprime2.
3041.
4812.
3061.
448(0.
040)(0.
069)(0.
040)(0.
077)LowDocumentation0.
0060.
8920.
0360.
914(0.
037)(0.
047)(0.
031)(0.
043)AboveLoanLimit0.
7181.
3820.
7071.
306(0.
053)(0.
064)(0.
041)(0.
044)Condo0.
6640.
7420.
4830.
453(0.
054)(0.
049)(0.
037)(0.
027)InvestmentProperty0.
2830.
1100.
3460.
072(0.
025)(0.
040)(0.
017)(0.
029)Renance0.
5350.
0210.
5600.
116(0.
022)(0.
043)(0.
020)(0.
050)Non-RecourseStates0.
1530.
720(0.
078)(0.
090)CollegeorMore0.
8710.
110(0.
058)(0.
086)HousePriceChange0.
1520.
317(0.
152)(0.
194)DecreaseinHousePrices0.
0720.
212(0.
036)(0.
036)MSAMedianIncome0.
2761.
006(0.
120)(0.
161)MSAMedianVTI0.
2640.
248(0.
050)(0.
058)MSAPopulationGrowth2.
6734.
398(1.
650)(1.
852)OriginationYearDummiesYesYesStateDummiesNoYesObservations10,166,5828,944,87236Table5:MortgageChoiceMultinomialLogitRegressionsforDetailedClassica-tionThistablereportsthecoecientsofmultinomiallogitregressionsformortgagechoice.
Thecoe-cientsaremeasuredrelativetoFRM.
Thesignicancelevelsareabbreviatedwithasterisks:One,two,andthreeasterisksdenotesignicanceatthe10,5,and1%level,respectively.
StateFixedEectsARMIONEGAMTRMLog(Income)0.
2770.
4330.
5200.
725(0.
015)(0.
029)(0.
022)(0.
038)Value-to-Income0.
0290.
0560.
0070.
016(0.
006)(0.
010)(0.
006)(0.
007)FICO/1000.
4100.
1130.
3030.
304(0.
014)(0.
020)(0.
018)(0.
035)Subprime2.
2911.
3162.
2132.
771(0.
041)(0.
052)(0.
097)(0.
204)LowDocumentation0.
0820.
0671.
7421.
901(0.
032)(0.
041)(0.
060)(0.
049)AboveLoanLimit0.
6991.
3621.
1141.
376(0.
042)(0.
047)(0.
040)(0.
053)Condo0.
4810.
4730.
4720.
306(0.
036)(0.
032)(0.
031)(0.
048)InvestmentProperty0.
3400.
2430.
3300.
120(0.
017)(0.
032)(0.
038)(0.
048)Renance0.
5480.
4350.
1660.
685(0.
020)(0.
057)(0.
042)(0.
076)CollegeorMore0.
8530.
4390.
3420.
640(0.
057)(0.
098)(0.
076)(0.
097)HousePriceChange0.
1920.
3310.
3000.
712(0.
153)(0.
183)(0.
256)(0.
285)DecreaseinHousePrices0.
0720.
1850.
2920.
298(0.
036)(0.
051)(0.
049)(0.
046)MSAMedianIncome0.
3120.
7611.
6991.
677(0.
120)(0.
176)(0.
209)(0.
247)MSAMedianVTI0.
2530.
2770.
0820.
133(0.
050)(0.
058)(0.
099)(0.
076)MSAPopulationGrowth2.
5604.
8821.
8074.
677(1.
610)(2.
031)(1.
708)(2.
729)OriginationYearDummiesYesStateDummiesYesObservations8,944,87337Table6:MortgageDelinquencies,HouseholdBankruptcies,andPrepaymentDe-cisionsThistablereportstheproportionofmortgagesthatareatleast60daysdelinquent,theproportionofhouseholdswithmortgagesthatdeclarebankruptcy,andtheproportionofmortgagesthatareprepaidafterone,three,andveyears.
Mortgagesareprepaidifaborrowerrenancestheloanorpaysbacktheloancompletelybeforematurity.
PanelA:ProportionofMortgagesthatareDelinquentFRMARMCM1Year2.
656.
434.
023Years9.
3115.
6317.
565Years11.
9518.
4822.
75NumberofLoans6,895,0471,174,3281,917,719PanelB:ProportionofHouseholdsDeclaringBankruptcyFRMARMCM1Year0.
250.
520.
263Years1.
512.
282.
205Years2.
152.
943.
18NumberofLoans6,895,0471,174,3281,917,719PanelC:ProportionofMortgagesthatarePrepaidFRMARMCM1Year7.
6615.
1012.
053Years28.
3247.
1238.
335Years37.
2959.
9845.
34NumberofLoans6,895,0471,174,3281,917,71938Table7:HazardModelofMortgageDelinquencyThistablereportsthehazardrateformortgagedelinquency.
Thesignicancelevelsareabbrevi-atedwithasterisks:One,two,andthreeasterisksdenotesignicanceatthe10,5,and1%level,respectively.
Individual-levelCovariatesStateFixedEectsCM0.
7920.
689(0.
020)(0.
014)IO0.
7610.
664(0.
026)(0.
019)NEGAM0.
7740.
687(0.
021)(0.
014)TRM0.
9640.
800(0.
027)(0.
022)ARM0.
3460.
3430.
3260.
324(0.
013)(0.
013)(0.
012)(0.
012)LogIncome0.
2490.
2500.
1640.
165(0.
018)(0.
018)(0.
017)(0.
016)ValuetoIncome(VTI)0.
0300.
0300.
0140.
014(0.
008)(0.
008)(0.
008)(0.
008)FICO/1001.
1081.
1061.
0581.
057(0.
016)(0.
016)(0.
018)(0.
018)Subprime0.
4080.
4220.
4210.
430(0.
016)(0.
016)(0.
011)(0.
011)LowDocumentation0.
0520.
0390.
0530.
043(0.
015)(0.
013)(0.
012)(0.
010)AboveLoanLimit0.
4030.
3950.
4420.
438(0.
038)(0.
038)(0.
026)(0.
026)Condo0.
0860.
0840.
0640.
063(0.
041)(0.
041)(0.
026)(0.
026)InvestmentProperty0.
2890.
2900.
2830.
284(0.
033)(0.
033)(0.
030)(0.
030)Renance0.
1520.
1600.
1640.
170(0.
009)(0.
009)(0.
013)(0.
013)Non-RecourseState0.
1120.
108(0.
061)(0.
061)CollegeorMore1.
4151.
411(0.
061)(0.
062)CurrentLTV0.
7620.
761(0.
066)(0.
066)UnemploymentLevel0.
0370.
037(0.
008)(0.
008)IncomeGrowthsinceOrigination0.
0400.
040(0.
004)(0.
004)CalendarDummiesYesYesYesYesOrig.
YearDummiesYesYesYesYesStateDummiesNoNoYesYesObservations32,960,51332,960,51326,019,61626,019,61639Table8:HazardModelsofMortgageDelinquency,PersonalBankruptcy,andMort-gagePrepaymentThistablereportsthehazardrateformortgagedelinquency,personalbankruptcy,andprepaymentdecisions.
Thesignicancelevelsareabbreviatedwithasterisks:One,two,andthreeasterisksdenotesignicanceatthe10,5,and1%level,respectively.
DelinquencyBankruptcyPrepaymentCM0.
6890.
6310.
372(0.
014)(0.
017)(0.
019)ARM0.
3260.
2080.
545(0.
012)(0.
013)(0.
011)LogIncome0.
1640.
3580.
079(0.
017)(0.
024)(0.
012)ValuetoIncome(VTI)0.
0140.
1710.
001(0.
008)(0.
011)(0.
002)FICO/1001.
0580.
7630.
091(0.
018)(0.
012)(0.
013)Subprime0.
4210.
0750.
289(0.
011)(0.
022)(0.
017)LowDocumentation0.
0530.
0060.
008(0.
012)(0.
011)(0.
008)AboveLoanLimit0.
4420.
4080.
099(0.
026)(0.
040)(0.
020)Condo0.
0640.
1930.
051(0.
026)(0.
030)(0.
011)InvestmentProperty0.
2830.
2000.
270(0.
030)(0.
023)(0.
011)Renance0.
1640.
2320.
116(0.
013)(0.
015)(0.
010)CollegeorMore1.
4151.
3730.
123(0.
061)(0.
070)(0.
045)CurrentLTV0.
7620.
7070.
634(0.
066)(0.
062)(0.
063)UnemploymentLevel0.
0370.
0460.
037(0.
008)(0.
010)(0.
008)IncomeGrowthfromOrigination0.
0400.
0320.
012(0.
004)(0.
004)(0.
004)CalendarandOrig.
YearDummiesYesYesYesStateDummiesYesYesYesObservations26,019,61625,851,51925,989,41740Table9:VariableDescriptionsThistablereportsthedescriptionofthevariablesusedandthecorrespondingdatasources.
VariableDataSourceAggregationDescriptionLoanAmountLPSIndividualLoanamountHomeValueLPSIndividualAppraisedhomevalueatoriginationIncomeHMDAIndividualReportedIncomefromloanapplicationFICOLPSIndividualFICOatoriginationVTILPSIndividualAppraisalvaluedividedbyincomefromloanapplicationFirstLienLTVLPSIndividualLoanamountdividedbyappraisedvalueofhomeHypotheticalFRMInterestRateLPSIndividualAverageinterestrateon30-yrFRMwithinmonth,state,conforming,LTV,andFICObucketsRenanceLPSIndividualReornotCondoLPSIndividualCondopropertyornotInvestmentPropertyLPSIndividual2ndhomeorinvestmentSubprimeLPSIndividualSubprimeindicatorastheservicerbelieves;doesnotincludeAlt-APrepaymentPenaltyLPSIndividualFlagforprepaymentpenaltyalongPrepaymentPenaltyTermLPSIndividualLengthinmonthsofprepaymentpenaltyPercentageaboveConformingLPSIndividualFlagforconformingloan.
ShareGovernmentSecuritizedLPSIndividualSecuritizationagafter1yrofloanlifeSharePrivateSecuritizedLPSIndividualSecuritizationagafter1yrofloanlifeHousePriceChangePrior3YearsFHFACBSA-QtrHousepricechangeinthepast3yearsDecreaseinHousePricesPrior10YearsFHFACBSA-QtrIndicatorvariableforwhethertherewere4quartersofhousepricedepreciationinthepast10yearsShareCollegeorMoreCensusZip(static)Proportionof2000populationwithcollegeeducationorbetterNon-RecourseGhentandStateStateswhererecourseinresidentialmortgagesislimitedbyKudlyak(2010)thevalueofthecollateralsecuringtheloan.
CurrentLTVLPSandFHFAIndividualThemortgageloanamountattheendofthepriorperioddividedbythecurrenthomevalue.
ThecurrenthomevalueisestimatedbyadjustingthehomevalueatoriginationbythehousepriceappreciationattheMSAlevelsincetheorigination.
UnemploymentLevelBLSCBSA-QtrUnemploymentrateIncomeGrowthfromOriginationBEACBSA-QtrGrowthrateofpercapitapersonalincome410.
000.
100.
200.
300.
400.
500.
600.
700.
800.
901.
0019951997199920012003200520072009CumulativeProportionFRMARMCMFigure1:CompositionofMortgageProducts.
TheguredepictsthecompositionbetweenFixedRateMortgages(FRM),AdjustableRateMortgages(ARM),andComplexMortgages(CM)overtheperiodbetween1995and2009.
42-0.
06-0.
04-0.
0200.
020.
040.
06199820002002200420062008QuarterlyHousePriceAppreciationQ5Q1Q3Figure2:QuarterlyHousePriceChangesbyComplexityQuintileThisguredepictsthequarterlyhousepricechangesofMSAsquintilessortedaccordingtotheproportionofcomplexmortgagesin2004.
Q1,Q3,andQ5correspondtothemeanappreciationlevelsofMSAintherst,third,andfthquintileaccordingtothecomplexshare.
43PanelA:Value-to-IncomeRatio00.
10.
20.
30.
40.
50.
60.
70.
80.
91012345678ValuetoIncomeRatioCumulativeDistributionFRMCMARMPanelB:FICOCreditScore00.
10.
20.
30.
40.
50.
60.
70.
80.
91500550600650700750800FICOScoreCumulativeDistributionFRMCMARMFigure3:CumulativeDistributionFunctionsbyMortgageTypeTheseguresdepictthecumulativedistributionfunctionsofthevalue-to-incomeratio(VTI)andFICOcreditscoresforFixedRateMortgages(FRM),AdjustableRateMortgages(ARM),andComplexMortgages(CM)overtheperiodbetween1995and2009.
PanelA:ComplexMortgagesin2002PanelB:ComplexMortgagesin2005PanelC:ComplexMortgagesin2008Figure4:GeographicDistributionofComplexMortgagesTheseguresdepictthegeographicdistributionofcomplexmortgagesin2002,2005,and2008.
PanelA:MortgagePaymentAfterOneYearRelativetoFRM00.
0050.
010.
0150.
020.
0250.
030.
0350.
040.
0450.
0500.
20.
40.
60.
811.
21.
41.
61.
82ActualMortgagePaymentafter1YearRelativetoFRMDistributionARMCMPanelB:MortgagePaymentAfterThreeYearsRelativetoFRM00.
010.
020.
030.
040.
050.
0600.
20.
40.
60.
811.
21.
41.
61.
82ActualMortgagePaymentafter3YearsRelativetoFRMDistributionARMCMPanelC:MortgagePaymentAfterFiveYearsRelativetoFRM00.
010.
020.
030.
040.
050.
060.
070.
0800.
20.
40.
60.
811.
21.
41.
61.
82ActualMortgagePaymentafter5YearsRelativetoFRMDistributionARMCMFigure5:MortgagePaymentRelativetoFRMTheseguresdepicttheactualmortgagepaymentsforAdjustableRateMortgages(ARM)andforComplexMortgages(CM)one,three,andveyearsafteroriginationrelativetothemortgagepaymentsofaFixedRateMortgages(FRM)withsimilarborrowercharacteristics.
PanelA:RemainingBalanceAfterOneYear00.
10.
20.
30.
40.
50.
60.
70.
80.
80.
850.
90.
9511.
051.
1RemainingMortgageBalanceAfterOneYearRelativetoOriginalBalanceCumulativeDistributionFRMCMARMPanelB:RemainingBalanceAfterThreeYears00.
050.
10.
150.
20.
250.
30.
350.
40.
80.
850.
90.
9511.
051.
1RemainingMortgageBalanceAfterThreeYearsRelativetoOriginalBalanceCumulativeDistributionFRMCMARMPanelC:RemainingBalanceAfterFiveYears00.
050.
10.
150.
20.
250.
80.
850.
90.
9511.
051.
1RemainingMortgageBalanceAfterFiveYearsRelativetoOriginalBalanceCumulativeDistributionFRMCMARMFigure6:RemainingMortgageBalancesTheseguresdepicttheremainingmortgagebalancesafterone,three,andveyearsrelativetotheinitialbalancesforFixedRateMortgages(FRM),AdjustableRateMortgages(ARM),andComplexMortgages(CM).
00.
0020.
0040.
0060.
0080.
010.
0120.
0140102030405060MonthsAfterOriginationHazardRateFRMCMARMFigure7:ProportionofMortgageDelinquenciesbyMonthAfterOriginationTheguredepictstheproportionofsurvivingloansthataredelinquentbymonthafterorig-nationforFixedRateMortgages(FRM),AdjustableRateMortgages(ARM),andComplexMortgages(CM)overtheperiodbetween2003and2009.
1WorkingPaperSeriesAseriesofresearchstudiesonregionaleconomicissuesrelatingtotheSeventhFederalReserveDistrict,andonfinancialandeconomictopics.
RiskTakingandtheQualityofInformalInsurance:GamblingandRemittancesinThailandWP-07-01DouglasL.
MillerandAnnaL.
PaulsonFastMicroandSlowMacro:CanAggregationExplainthePersistenceofInflationWP-07-02FilippoAltissimo,BenotMojon,andPaoloZaffaroniAssessingaDecadeofInterstateBankBranchingWP-07-03ChristianJohnsonandTaraRiceDebitCardandCashUsage:ACross-CountryAnalysisWP-07-04GeneAmrominandSujitChakravortiTheAgeofReason:FinancialDecisionsOvertheLifecycleWP-07-05SumitAgarwal,JohnC.
Driscoll,XavierGabaix,andDavidLaibsonInformationAcquisitioninFinancialMarkets:aCorrectionWP-07-06GadiBarlevyandPietroVeronesiMonetaryPolicy,OutputCompositionandtheGreatModerationWP-07-07BenotMojonEstateTaxation,Entrepreneurship,andWealthWP-07-08MarcoCagettiandMariacristinaDeNardiConflictofInterestandCertificationintheU.
S.
IPOMarketWP-07-09LucaBenzoniandCarolaSchenoneTheReactionofConsumerSpendingandDebttoTaxRebates–EvidencefromConsumerCreditDataWP-07-10SumitAgarwal,ChunlinLiu,andNicholasS.
SoulelesPortfolioChoiceovertheLife-CyclewhentheStockandLaborMarketsareCointegratedWP-07-11LucaBenzoni,PierreCollin-Dufresne,andRobertS.
GoldsteinNonparametricAnalysisofIntergenerationalIncomeMobilityWP-07-12withApplicationtotheUnitedStatesDebopamBhattacharyaandBhashkarMazumderHowtheCreditChannelWorks:DifferentiatingtheBankLendingChannelWP-07-13andtheBalanceSheetChannelLamontK.
BlackandRichardJ.
RosenLaborMarketTransitionsandSelf-EmploymentWP-07-14EllenR.
RissmanFirst-TimeHomeBuyersandResidentialInvestmentVolatilityWP-07-15JonasD.
M.
FisherandMartinGervais2WorkingPaperSeries(continued)EstablishmentsDynamicsandMatchingFrictionsinClassicalCompetitiveEquilibriumWP-07-16MarceloVeraciertoTechnology'sEdge:TheEducationalBenefitsofComputer-AidedInstructionWP-07-17LisaBarrow,LisaMarkman,andCeciliaElenaRouseTheWidow'sOffering:Inheritance,FamilyStructure,andtheCharitableGiftsofWomenWP-07-18LeslieMcGranahanIncompleteInformationandtheTimingtoAdjustLabor:EvidencefromtheLead-LagRelationshipbetweenTemporaryHelpEmploymentandPermanentEmploymentWP-07-19SainanJin,YukakoOno,andQinghuaZhangAConversationwith590NascentEntrepreneursWP-07-20JeffreyR.
CampbellandMariacristinaDeNardiCyclicalDumpingandUSAntidumpingProtection:1980-2001WP-07-21MeredithA.
CrowleyHealthCapitalandthePrenatalEnvironment:TheEffectofMaternalFastingDuringPregnancyWP-07-22DouglasAlmondandBhashkarMazumderTheSpendingandDebtResponsetoMinimumWageHikesWP-07-23DanielAaronson,SumitAgarwal,andEricFrenchTheImpactofMexicanImmigrantsonU.
S.
WageStructureWP-07-24MaudeToussaint-ComeauALeverage-basedModelofSpeculativeBubblesWP-08-01GadiBarlevyDisplacement,AsymmetricInformationandHeterogeneousHumanCapitalWP-08-02LuojiaHuandChristopherTaberBankCaR(BankCapital-at-Risk):AcreditriskmodelforUScommercialbankcharge-offsWP-08-03JonFryeandEduardPelzBankLending,FinancingConstraintsandSMEInvestmentWP-08-04SantiagoCarbó-Valverde,FranciscoRodríguez-Fernández,andGregoryF.
UdellGlobalInflationWP-08-05MatteoCiccarelliandBenotMojonScaleandtheOriginsofStructuralChangeWP-08-06FranciscoJ.
BueraandJosephP.
KaboskiInventories,LumpyTrade,andLargeDevaluationsWP-08-07GeorgeAlessandria,JosephP.
Kaboski,andVirgiliuMidrigan3WorkingPaperSeries(continued)SchoolVouchersandStudentAchievement:RecentEvidence,RemainingQuestionsWP-08-08CeciliaElenaRouseandLisaBarrowDoesItPaytoReadYourJunkMailEvidenceoftheEffectofAdvertisingonHomeEquityCreditChoicesWP-08-09SumitAgarwalandBrentW.
AmbroseTheChoicebetweenArm's-LengthandRelationshipDebt:EvidencefromeLoansWP-08-10SumitAgarwalandRobertHauswaldConsumerChoiceandMerchantAcceptanceofPaymentMediaWP-08-11WilkoBoltandSujitChakravortiInvestmentShocksandBusinessCyclesWP-08-12AlejandroJustiniano,GiorgioE.
Primiceri,andAndreaTambalottiNewVehicleCharacteristicsandtheCostoftheCorporateAverageFuelEconomyStandardWP-08-13ThomasKlierandJoshuaLinnRealizedVolatilityWP-08-14TorbenG.
AndersenandLucaBenzoniRevenueBubblesandStructuralDeficits:What'sastatetodoWP-08-15RichardMattoonandLeslieMcGranahanTheroleoflendersinthehomepriceboomWP-08-16RichardJ.
RosenBankCrisesandInvestorConfidenceWP-08-17UnaOkonkwoOsiliandAnnaPaulsonLifeExpectancyandOldAgeSavingsWP-08-18MariacristinaDeNardi,EricFrench,andJohnBaileyJonesRemittanceBehavioramongNewU.
S.
ImmigrantsWP-08-19KatherineMeckelBirthCohortandtheBlack-WhiteAchievementGap:TheRolesofAccessandHealthSoonAfterBirthWP-08-20KennethY.
Chay,JonathanGuryan,andBhashkarMazumderPublicInvestmentandBudgetRulesforStatevs.
LocalGovernmentsWP-08-21MarcoBassettoWhyHasHomeOwnershipFallenAmongtheYoungWP-09-01JonasD.
M.
FisherandMartinGervaisWhydotheElderlySaveTheRoleofMedicalExpensesWP-09-02MariacristinaDeNardi,EricFrench,andJohnBaileyJones4WorkingPaperSeries(continued)UsingStockReturnstoIdentifyGovernmentSpendingShocksWP-09-03JonasD.
M.
FisherandRyanPetersStochasticVolatilityWP-09-04TorbenG.
AndersenandLucaBenzoniTheEffectofDisabilityInsuranceReceiptonLaborSupplyWP-09-05EricFrenchandJaeSongCEOOverconfidenceandDividendPolicyWP-09-06SanjayDeshmukh,AnandM.
Goel,andKeithM.
HoweDoFinancialCounselingMandatesImproveMortgageChoiceandPerformanceWP-09-07EvidencefromaLegislativeExperimentSumitAgarwal,GeneAmromin,ItzhakBen-David,SouphalaChomsisengphet,andDouglasD.
EvanoffPerverseIncentivesattheBanksEvidencefromaNaturalExperimentWP-09-08SumitAgarwalandFayeH.
WangPayforPercentileWP-09-09GadiBarlevyandDerekNealTheLifeandTimesofNicolasDutotWP-09-10FranoisR.
VeldeRegulatingTwo-SidedMarkets:AnEmpiricalInvestigationWP-09-11SantiagoCarbóValverde,SujitChakravorti,andFranciscoRodriguezFernandezTheCaseoftheUndyingDebtWP-09-12FranoisR.
VeldePayingforPerformance:TheEducationImpactsofaCommunityCollegeScholarshipProgramforLow-incomeAdultsWP-09-13LisaBarrow,LashawnRichburg-Hayes,CeciliaElenaRouse,andThomasBrockEstablishmentsDynamics,VacanciesandUnemployment:ANeoclassicalSynthesisWP-09-14MarceloVeraciertoThePriceofGasolineandtheDemandforFuelEconomy:EvidencefromMonthlyNewVehiclesSalesDataWP-09-15ThomasKlierandJoshuaLinnEstimationofaTransformationModelwithTruncation,IntervalObservationandTime-VaryingCovariatesWP-09-16BoE.
HonoréandLuojiaHuSelf-EnforcingTradeAgreements:EvidencefromAntidumpingPolicyWP-09-17ChadP.
BownandMeredithA.
CrowleyToomuchrightcanmakeawrong:SettingthestageforthefinancialcrisisWP-09-18RichardJ.
Rosen5WorkingPaperSeries(continued)CanStructuralSmallOpenEconomyModelsAccountfortheInfluenceofForeignDisturbancesWP-09-19AlejandroJustinianoandBrucePrestonLiquidityConstraintsoftheMiddleClassWP-09-20JeffreyR.
CampbellandZviHercowitzMonetaryPolicyandUncertaintyinanEmpiricalSmallOpenEconomyModelWP-09-21AlejandroJustinianoandBrucePrestonFirmboundariesandbuyer-suppliermatchinmarkettransaction:ITsystemprocurementofU.
S.
creditunionsWP-09-22YukakoOnoandJunichiSuzukiHealthandtheSavingsofInsuredVersusUninsured,Working-AgeHouseholdsintheU.
S.
WP-09-23MaudeToussaint-ComeauandJonathanHartleyTheEconomicsof"RadiatorSprings:"IndustryDynamics,SunkCosts,andSpatialDemandShiftsWP-09-24JeffreyR.
CampbellandThomasN.
HubbardOntheRelationshipbetweenMobility,PopulationGrowth,andCapitalSpendingintheUnitedStatesWP-09-25MarcoBassettoandLeslieMcGranahanTheImpactofRosenwaldSchoolsonBlackAchievementWP-09-26DanielAaronsonandBhashkarMazumderCommenton"LettingDifferentViewsaboutBusinessCyclesCompete"WP-10-01JonasD.
M.
FisherMacroeconomicImplicationsofAgglomerationWP-10-02MorrisA.
Davis,JonasD.
M.
FisherandToniM.
WhitedAccountingfornon-annuitizationWP-10-03SvetlanaPashchenkoRobustnessandMacroeconomicPolicyWP-10-04GadiBarlevyBenefitsofRelationshipBanking:EvidencefromConsumerCreditMarketsWP-10-05SumitAgarwal,SouphalaChomsisengphet,ChunlinLiu,andNicholasS.
SoulelesTheEffectofSalesTaxHolidaysonHouseholdConsumptionPatternsWP-10-06NathanMarwellandLeslieMcGranahanGatheringInsightsontheForestfromtheTrees:ANewMetricforFinancialConditionsWP-10-07ScottBraveandR.
AndrewButtersIdentificationofModelsoftheLaborMarketWP-10-08EricFrenchandChristopherTaber6WorkingPaperSeries(continued)PublicPensionsandLaborSupplyOvertheLifeCycleWP-10-09EricFrenchandJohnJonesExplainingAssetPricingPuzzlesAssociatedwiththe1987MarketCrashWP-10-10LucaBenzoni,PierreCollin-Dufresne,andRobertS.
GoldsteinDoesPrenatalSexSelectionImproveGirls'Well‐BeingEvidencefromIndiaWP-10-11LuojiaHuandAnalíaSchlosserMortgageChoicesandHousingSpeculationWP-10-12GadiBarlevyandJonasD.
M.
FisherDidAdheringtotheGoldStandardReducetheCostofCapitalWP-10-13RonAlquistandBenjaminChabotIntroductiontotheMacroeconomicDynamics:Specialissuesonmoney,credit,andliquidityWP-10-14EdNosal,ChristopherWaller,andRandallWrightSummerWorkshoponMoney,Banking,PaymentsandFinance:AnOverviewWP-10-15EdNosalandRandallWrightCognitiveAbilitiesandHouseholdFinancialDecisionMakingWP-10-16SumitAgarwalandBhashkarMazumderComplexMortgagesWP-10-17GeneAmromin,JenniferHuang,ClemensSialm,andEdwardZhong