Aggressive Lending and Real Estate Markets

Transcription

1 Aggressive Lending nd Rel Ese Mrkes December 20, 2006 Andrey Pvlov The Whron School, Universiy of Pennsylvni nd Simon Frser Universiy E-mil: Susn Wcher The Whron School Universiy of Pennsylvni E-mil:

2 Aggressive Lending nd Rel Ese Mrkes This pper esblishes heoreicl nd empiricl link beween he use of ggressive morgge lending insrumens nd he underlying house price voliliy. Wihin he conex of generl equilibrium model wih borrowing consrins, we demonsre h he supply of ggressive lending insrumens, such s non-morizing low-equiy morgges, increses he sse prices in he underlying mrke becuse borrowers use hese insrumens o furher leverge heir curren income. Thus, he ggressive lending insrumens effecively relx he borrowing consrin fced by prospecive homeowners. Furhermore, in our model lenders rionlly re-price ll morgge insrumens following negive demnd shock. We show h he relive use of ggressive lending insrumens declines following negive demnd shock wheher his re-pricing is niciped or no. These wo resuls provide for he imporn policy implicion h he vilbiliy of ggressive morgge lending insrumens mgnifies he rel ese cycle nd he effecs of lrge negive demnd shocks. Using boh locl nd nionl price index d we empiriclly confirm he predicions of he model. In priculr, we find h neighborhoods nd ciies h experienced high concenrion of ggressive lending insrumens heir respecive rel ese mrke peks suffered more severe price declines nd lower supply of ggressive insrumens following negive demnd shock. Overll, we find h he flucuion of supply of ggressive lending insrumens increses he voliliy of he underlying sse prices over he course of he mrke cycle. 2

3 Inroducion This pper esblishes link beween he vilbiliy of ggressive morgge lending insrumens nd underlying sse mrke prices. Indusry sources sugges h ggressive lending insrumens, such s ineres only lons, negive morizion lons, low or zero equiy lons, nd eser-re ARMs, ccouned for nerly wo-hirds of ll U.S. lon originions since The wide-spred populriy of hese insrumens hs no escped he enion of he policy mkers in hese counries. As erly s 2004, he U.S. Federl Reserve Chirmn he ime, Aln Greenspn, expressed concern h mny homeowners ook ou ineresonly morgges or opion-djusble re morgges o buy propery hey oherwise could no fford. "In he even of widespred cooling in house prices, hese borrowers, nd he insiuions h service hem, could be exposed o significn losses," Greenspn sid. More recenly, Lingling Wei wroe in he Wll Sree Journl (December 5, 2006) h lenders hve hrd ime selling heir sub-prime morgge lending unis becuse of concerns over he subprime sellers' cosly obligion of hving o buy bck he lons lredy sold in he secondry mrke due o borrowers' defuls. In he sme issue, Ruh Simon nd Jmes Hgery repor h delinquency res on subprime morgges origined in he ps yer hve sored o he highes levels in decde. While hese delinquencies re sill oo low levels o ffec he generl economy, hey hve 1 FDIC Oulook: Breking New Ground in U.S. Morgge Lending. December 18 h, 2006 <hp:// 2 Nonprime morgge originions rose n even pce from 2001 hrough 2003 o rech beween $25 billion nd $30 billion in Jnury Originions ccelered in 2004 before peking in Mrch 2005 in rnge beween $60 billion o $70 billion nd declined since hen o rech pproximely $50 billion in Augus Recen d on subsequen monhs re no fully vilble nd re subjec o revision. < hp:// Refer lso o ppendix bles A1 hrough A4. 3

4 promped credi-ring gencies o insruc issuers of morgge-bcked bonds o se side ddiionl funding o cover losses. Even if delinquencies remin modes, his incresed cpil requiremen will ulimely force lenders o re-price heir lons. In his pper we provide boh heoreicl nd empiricl suppor for hese policy concerns nd indusry recions. Wihin he conex of generl equilibrium model we demonsre h he exisence of ggressive lending insrumens, such s ineres-only morgges, increses sse prices in he underlying mrke becuse borrowers re ble o furher leverge heir curren income or welh. The inbiliy o borrow gins humn cpil imposes consrin on mny young US households, especilly if boh he personl nd he economy-wide produciviy re expeced o grow overime. The biliy o increse he leverge of curren welh nd income relxes his consrin, which, in urn, increses he housing expendiures of hese households. This ddiionl source of demnd is hen rnsled ino higher mrke prices, especilly in mrkes of fixed or inelsic supply. We obin his resul wih ll lending insrumens, ggressive or conservive, being firly priced. This effec cn in heory cuse one-ime price increse which reflecs he relxed borrowing consrin for mny households wih no furher consequences. However, more generl model h llows mrke pricipns o revise he probbiliy of negive demnd shock from heir hisoricl experience suggess h morgge insrumens ge repriced following such shocks. In his cse, he price premium of ggressive insrumens is correled wih ps relizions. This correlion increses he overll voliliy of rel ese mrkes nd mgnifies negive demnd shocks. We obin his resul even if ll lending insrumens re firly priced, nd he re-pricing of ggressive insrumens following negive demnd shock is fully niciped. We use locl-level rnscion-bsed price indices nd concenrion of ggressive insrumens in Los Angeles Couny o empiriclly invesige he hypohesized link. We lso use he OFHEO meropolin res indices nd concenrion of ggressive insrumens cross ciies o invesige his link on nionl level. The empiricl resuls from hese wo d ses provide re consisen wih our heoreicl conclusions described bove. Firs, we find h neighborhoods or ciies h benefied from high concenrion of ffordble ggressive insrumens he op of heir mrke cycle experienced deeper price declines following heir respecive negive demnd shock. Furhermore, he neighborhoods or he ciies h experienced he lrges price declines hd he smlles concenrion of ggressive insrumens he boom of he mrke. These wo findings re consisen wih he prevlence of ggressive insrumens h enbles recen relizions of he mrke nd mgnifies he effecs of negive demnd shocks. This mgnifying effec on he downside is presen even in he bsence of sizeble deful res. In oher words, i is he flucuion of he use of ggressive insrumens h excerbes mrke downurns, no he fc h such insrumens genere relively higher deful res. 4

5 The mrke shre of ggressive insrumens in our model is endogenously deermined hrough he inercion of borrower segmenion nd he re-pricing of ll morgges. Therefore, he flucuion in he use of ggressive insrumens hroughou he mrke cycle is no enirely conrolled by lenders. However, if lender recion dds o he pricing flucuion of ggressive insrumens he resul my be dded price voliliy nd ccompnying desbilizion of mrkes. While he empiricl evidence is no ye vilble o confirm his poenil impc, he concepul frmework mkes eviden secondry policy implicion. In priculr, he lending indusry regulors should be creful no o ke seps h ler mrkes fer he fc, rher he gol should be o minin vilbiliy of morgge lending insrumens in boh rising nd flling mrkes. Ours is no he firs sudy o invesige he link beween lending nd sse mrkes. Allen nd Gle (1998 nd 1999), Herring nd Wcher (1999), nd Pvlov nd Wcher (2002, 2005) show h underpricing of he deful risk in bnk lending leds o infled sse prices in mrkes of fixed supply. Furhermore, Pvlov nd Wcher (2002, 2005, 2006) show h underpricing of he deful risk excerbes sse mrke crshes. One unifying feure of his prior lierure on he link beween lending nd sse mrkes is h he sse-bcked lons re mispriced, eiher rionlly or no. Our poin of deprure in his pper is h ll lons re ssumed firly priced. Lenders rec o curren informion on risks which my chnge over he cycle. Perceived risks chnge s mrke condiions chnge, hence, he flucuions in he risk pricing of ggressive insrumens, if ny, re correcly niciped. In he firs insnce, wh drives mrke price inflion bove is fundmenl vlue in his model is he evolving consrin fced by borrowers. I is he ime vriion of his consrin, ogeher wih he dynmic re-pricing of he lending insrumens, h generes our finding h ggressive lending mgnifies he effec of negive demnd shocks. A hndful of empiricl invesigions direcly sudy he impc of ggressive lending on rel ese, wheher hese insrumens re priced correcly or no. Hung nd Tu (2006) find h he increse of he use of djusble re morgges in Cliforni is ssocied wih n increse in medin home prices. They mke no commen on wheher his increse is emporry nd will reverse wih he business cycle or wheher i is one-ime permnen posiive shock. Similrly, he Sepember 2004 IMF repor on he World Economic Issues suggess h counries wih higher use of djusble-re morgges hve more volile housing mrkes (Chper II, pge 81). The mechnism hey conjecure o explin his finding is h higher use of ARM-like insrumens mkes rel ese mrkes more sensiive o ineres re chnges. This repor does no consider he flucuion in vilbiliy of ARMs nd oher ggressive insrumens hroughou he rel ese mrke cycle. Even hough he empiricl findings of boh sudies do no provide direc es of our model, hey re indeed consisen wih is implicions. We proceed s follows. Secion 2 develops he link beween lending nd sse mrkes in heoreicl model. Secion 3 presens he d nd resuls using Los Angeles nd nionl d. Secion 4 concludes wih brief summry nd suggesions for fuure reserch. 5

6 2 Model This secion presens model of borrower demnd nd lending behvior in he presence of boh rdiionl morgges nd ggressive lending insrumens in he conex of compeiive rel ese mrke wih fixed supply. 3 A ech poin in ime, borrowers decide on heir housing expendiure, lenders offer firly priced ggressive nd conservive morgge lons, nd he price for rel ese is se o cler he mrke. 2.1 Borrower demnd We presume here re wo disinc ypes of borrowers on he mrke conservive nd ggressive. We cll he firs ype conservive, s hey would choose he rdiionl morgges ll he ime s long s boh insrumens re priced firly. The budge consrin for rel ese of he conservive nd ggressive borrowers is given respecively by: ( r + γ ) P Q = H c c c rpq = H (1) where γ represens he required morizion pymen on he conservive morgge (no, required for he ggressive morgge, r c denoes he ineres pymen on ech, morgge, presumed o be pid he beginning of ech period, Q c denoes he quniy c of rel ese purchsed by ech borrower group, nd H nd H denoe he budge lloced o rel ese by ech group. 4 Define H = H + H, α = c H H (2) We inroduce unceriny over ime in he budge lloced o rel ese by boh groups: dh = µ d gdq (3) 3 Numericl soluions of model wih supply response suggess h our resuls hold s long s he elsiciy of supply is no infinie. 4 These budge consrins re he resul of he lenders resricion h pymen o income rio cnno exceed cerin pre-deermined level. Thus, borrowers re off of heir demnd curve nd re consrin by his requiremen. However, hese llocions o rel ese cn in heory lso be obined by opimizing seprble uiliy funcion of consumpion nd housing, wih insnneous budge consrin of he following form: C+ rpq= I, where C denoes consumpion of oher goods, nd rp denoes he morgge pymen. Wih logrihmic uiliy funcion, he llocions in Equion (1) re exc. Wih isoelsic uiliy, he funcionl form of he llocions is more compliced, bu heir behvior wih respec o ineres res, prices, nd quniies is he sme. In his cse, he price of rel ese cnno be solved for explicily, bu our comprive sics resuls hold. 6

7 where µ d is expeced growh in budge over d, gdq is compound Poisson process, wih, 0 wih probbiliy (1 δ ) d dq = (4) 1 wih probbiliy δ d g is he rndom moun wih men λ 0 nd vrince β by which demnd flls in he cse of negive demnd shock. Summing he demnd of he ggressive nd conservive borrowers nd equing i o he ol fixed supply of rel ese, Q, we obin: A C (1 α) H αh Q= Q + Q = + c ( r + γ ) P r P (5) Solve for P : P H 1 α α = + Q r + γ r c (6) Assuming he prices of lons is unchnged hrough ime (n ssumpion we relx below), he expeced chnge in price in cse of negive jump is: H 1 α α (1 g ) H 1 α α E + E c + c = ( gp ) = λp Q r + γ r Q r + γ r (7) This resul suggess h losses re proporionl o he sse price s long he supply of ggressive insrumens does no chnge following negive demnd shock. 2.2 Lender Behvior We ssume compeiive risk-neurl lender who offers boh he conservive nd he ggressive morgges nd ses he res so h he expeced reurn for ech insrumen is he risk-free re, ssumed zero. The lender is ssumed o experience losses only if negive jump in demnd occurs. The expeced reurn for he ggressive nd he conservive insrumens is: c c ER ( ) = r δ ( λ γ) = r + δγ ER ( ) = r δλ (8) 7

8 Seing hese expeced reurns o zero, we find he ineres res on he wo insrumens: r c r = δ ( λ γ) = δλ (9) Subsiue (9) ino (6) o obin he price in period : P H 1 α α H δλ+ (1 δ) γα = + = Q r + γ r Q δλ( δλ + (1 δ ) γ ) c (10) P Clerly, 0, which demonsres he price inflion effec of ggressive lons. α 2.3 Dynmic shre of ggressive lending nd myopic lenders Wih unceriny gens re no likely o know he probbiliy of negive demnd shock. Thus in his secion, we ssume h gens esime he probbiliy of negive demnd shock from exising hisoricl d nd upde hese probbiliies wih experience. Any resonble economeric mehod would produce higher esime for he probbiliy of negive demnd shock, ˆ δ, immediely following n observed shock. For insnce, he simples esime of he probbiliy of shock is he number of ps shocks divided by he ime lengh of he d smple, ˆ δ = # shocks / T. Clerly his esime will increse immediely following negive demnd shock, since he numeror will increse by 1, while he denominor will no chnge. Since he lender is ssumed risk-neurl, we cn replce he probbiliy of negive demnd shock, δ, wih is esime nd he resuling prmeer unceriny will no ler he ineres res nd sse price derived in Equions (9) nd (10). For now we furher ssume h lenders re myopic in he sense h hey do no nicipe he upwrd revision of he shock probbiliy following negive demnd shock. In oher words, lenders hink h he probbiliy of negive demnd shock will no be revised even hough in he fermh of n cul even i will indeed be revised upwrd. We relx his ssumpion in he nex secion. An upwrd revision of he shock probbiliy, δ, following negive demnd shock hs wo consequences h excerbe he shock. Firs, he sse price flls more hen proporionlly o he decline in demnd. Equion (9) shows h boh ineres res increse wih n increse in he shock probbiliy. Equion (10) hen shows h he P price is declining funcion of he shock probbiliy, < 0. Thus, following negive δ demnd shock, he sse price needs o djus no only o ccoun for he new lower ol demnd, H, bu lso o incorpore he higher probbiliy of fuure negive shocks. 8

9 The second consequence of n upwrd revision of he shock probbiliy, δ, is h he composiion of he ggressive nd conservive lons chnges. Modify Equion (10) o show h: rp H (1 ) α r = + α c Q r + γ (11) Differenie wih respec o δ: rp H (1 α) λδ ( ( λ γ) + γ) δλλ ( γ) H (1 α) λγ = = > δ Q + γ + γ c c ( r ) Q ( r ) (12) Q Then, using Equion (1), i is immedie h < 0. Similrly, i cn be verified h δ c Q > 0. In oher words, he shre of ggressive morgges declines following δ negive demnd shock. P Furhermore, he price effec, < 0 is mgnified for mrkes wih higher presence of δ 2 P ggressive lending. The cross-derivive is: α δ 2 P H 1 1 = + = αδ δ Q δλ δγ+ γ δλ H λ γ λ = Q ( δλ δγ + γ ) ( δλ ) (13) which is negive since he posiive erm hs smller numeror nd lrger denominor. Therefore, for mrkes wih higher presence of ggressive lending, mesured by higher α, he sse price is more sensiive o chnges in he probbiliy of negive demnd shock, δ. In oher words, mrkes wih high concenrions of ggressive lending insrumens experience lrger hn proporionl (o he shock) price declines following negive demnd shock. 2.4 Dynmic shre of ggressive lending nd sregic lenders In his secion we ssume lenders correcly nicipe n increse in he jump probbiliy esime following negive demnd shock. We find h he sse price sill declines 9

10 more hn proporionlly o he decline in demnd nd his resul remins sronger for lrger concenrion of ggressive insrumens. Le subscrips B nd A denoe he ime immediely before nd fer he crsh, respecively. The ineres re on he ggressive insrumen before he crsh is given by: or, or, Pr = δ ( P P) (14) B B B B A ( δ r ) P = δ P (15) B B B B A PA ( δb rb) PB = ( δb δbλ) (1 λ ) (16) The ddiionl risk from he re-pricing of ll insrumens following negive demnd shock rises he ineres re before he shock, r > δ λ. Therefore, B B (1 λ) P B P A (17) In oher words, he decline in price is more hen proporionl o he decline in demnd following negive demnd shock. Thus, even if lenders correcly nicipe he repricing of ll insrumens following negive demnd shock, ggressive insrumens mgnify he sse price impc of negive demnd shocks. Furhermore, his effec is sill lrger for mrkes wih high concenrion of ggressive insrumens, lhough pr of he difference is reduced by he sregic behvior of he c lenders. I is esily verified h r = r δγ. Subsiue his ino Equion (10) nd suppress he ime subscrips, P H 1 α α = + Q r + (1 δγ ) r (18) Noe h r is no longer given by Equion (9), bu is sill n incresing funcion of he shock probbiliy, δ. Differenie (18) wih respec o α nd δ: 10

11 2 P H 1 1 = + = αδ δ Q r + (1 δγ ) r r r γ H = δ δ Q ( r + (1 δγ ) ) ( r ) (19) The cross-derivive in Equion (19) is negive becuse he posiive erm hs smller numeror nd lrger denominor. Therefore, he sensiiviy of he sse price o chnges in he shock probbiliy is lrger for mrkes wih higher demnd for he ggressive insrumen. In oher words, mrkes wih high concenrion of ggressive lending insrumens experience lrger price declines following negive demnd shocks of consn mgniude. Noe h Expression (19) reduces o Equion (13) if r = δλ. However, s discussed bove, r is higher due o he lender s nicipion of re-pricing fer negive shock. The second consequence of n upwrd revision of he shock probbiliy, δ, is h he composiion of he ggressive nd conservive lons chnges. Modify Equion (10) o show h: rp H (1 ) α r = + α c Q r + γ (20) Differenie wih respec o δ: r r r ( r + (1 δγ ) ) r ( γ) (1 δγ ) + r γ rp H (1 α) δ δ H (1 α) = = δ > 0 (21) 2 2 δ Q + δγ + δγ ( r (1 ) ) Q ( r (1 ) ) Q Then, using Equion (1), i is eviden h < 0. Similrly, i cn be verified h δ c Q > 0. In oher words, he shre of he ggressive morgges declines following δ negive demnd shock. In summry, he bove model suggess he following empiricl implicions: 1. The sse price is n incresing funcion of he demnd for ggressive lending insrumens. 2. If lenders do no observe he rue probbiliy of negive demnd shock bu esime i from hisoricl d, following negive demnd shock. he esimed probbiliy of negive demnd shock increses 11

12 b. he sse price declines more hen proporionlly o he decline in demnd c. he decline in sse price is lrger for mrkes wih high concenrion of ggressive lending insrumens d. he use of ggressive insrumens declines 3. These implicions hold even if lenders correcly nicipe he chnge in shock probbiliy, or is esime, following negive demnd shock, lhough heir mgniude my be prilly reduced by he sregic behvior of lenders. 3.0 Empiricl evidence In his secion we es he empiricl implicions of our concepul frmework using wo disinc d ses, described below. In priculr we show h sse prices decline more in mrkes wih high concenrion of ggressive insrumens nd h he use of ggressive insrumens declines he mos for mrkes h experience he lrges price declines. 3.1 Los Angeles Empiricl Evidence Using d from he rel ese mrke downurn in Souhern Cliforni we consider wheher he hisoricl evidence is consisen wih our hypohesis. We uilize rnscion d from DQuick, compny specilizing in collecing rel ese rnscion d. The underlying d comes from he Couny Recorder. Following Pvlov (2001) nd Deng, Pvlov, nd Yng (2005) we divide Los Angeles Couny ino 22 res h cpure o gre exend he heerogeneiy of he Los Angeles rel ese mrke. Then, we compue he ol percen decline for ech of he regions beween My, 1990 nd Ocober, 1995, which represen he op nd he boom of he Los Angeles rel ese mrke cycle, respecively. We use ll rnscions which occurred wihin 3 monhs of he op nd he boom of he mrke o esime hedonic regression of he following form: 22 22, (22) ln( V ) = α ζ + β ζ + X γ + ε i z zi z zi i i i z= 1 z= 1 where V i denoes he vlue of rnscion i, ζ zi is n indicor vrible which kes he vlue of 1 if he propery is loced in zone z nd zero oherwise, i is n indicor vrible which kes he vlue of 1 if he rnscion occurred beween Augus, 1995 nd Jnury, 1996, nd zero oherwise, X i denoes horizonl vecor of physicl chrcerisics of he propery, α, β, nd γ re prmeers of he model nd ε i is he esimion error. The physicl chrcerisics we include re number of bedrooms nd bhrooms, size of he lo nd of he building, yer buil, nd wheher he propery hs pool or no. Given he bove equion, he esimed prmeers β z re esimes of he percen decline in propery vlues from 1990 o 1995 for ech of he 22 neighborhoods. Tble 1 provides summry sisics for he Los Angeles price d. The medin percen decline during h period ws jus over 21% for he enire meropolin re, rnging from 7% o 35% for ech of he 22 neighborhoods. 12

13 We furher use lon originion d from Wells Frgo Morgge h conins privelbel securiized morgge lon originions, spnning period from 1988 o The d is of high quliy nd is consisen hrough ime. This d ccouns for over 20% of ll lon originions in Los Angeles Couny nd conins he posl zip code of he underlying propery. This llows us o spilly ssign he originions o ech of he 22 neighborhoods. While no ineres only or exended morizion morgges were vilble he ime, he le 1980 s ws he period when djusble-re morgges becme populr in he U.S. While ARMs re no priculrly ggressive insrumens, hey do hve ll he chrcerisics of hese insrumens when compred o he rdiionl fully morizing fixed-re morgges. For exmple, ARMs llow borrowers o spend more on rel ese purchse, holding heir housing expendiure consn. This benefi cme he cos of incresed risk for some borrowers. ARMs were new he ime, so heir pricing nd fuure vilbiliy ws uncler. Finlly, ARMs becme incresingly populr during he le 1980s run-up. Summry sisics of he lon d is lso repored in Tble 1. Tble 2 repors single vrible regression of he op o boom price decline, mesured in bsolue erms, in neighborhood s funcion of he ARM shre in h neighborhood he op of he mrke. We firs repor he resuls for lons used for purchsing homes. The proporion of ARMs he op of he mrke is ssocied wih lrge nd significn impc on he subsequen price decline. For ech one percen higher shre of ARMs in 1990, he price decline increses by 1.3 percen for h neighborhood. This finding is consisen wih our heoreicl implicion h he presence of ARMs he op of he mrke mgnifies he subsequen negive demnd shock. Furhermore, s expeced, our resuls re wek nd insignificn for lons used for refinncing. Our model suggess h ggressive lending insrumens llow borrowers o purchse homes hey oherwise cnno fford. Clerly, his effec is no operionl for refinncing lons, due o he fc h he borrower is lredy n owner. The insignificn bu in he righ sign coefficiens we find my be due o he refinncing civiy shdowing he originion lons. In oher words, borrowers who used n ARM for purchse my be more likely o use n ARM for refinncing or equiy ou purposes becuse hey hve higher comfor level wih hese insrumens. Tble 3 repors he resuls of single vrible regression of he chnge in proporion of ARM originions during he 1990 o 1995 period on he percen decline in ech neighborhood. The model implies h res h suffer he lrges price declines during crsh re hose in which ARM originions decline he mos. The resuls repored in Tble 3 re consisen wih his, s he decline in ARM originions is ssocied wih he percen decline in prices from op o boom. This finding is significn for ll lons nd lons used for purchse, lhough s expeced no significn for lons used for refinncing. Aggressive insrumens pper o be ho money. Their prevlence pus he mrke greer risk s heir originions end o decline on relive bsis fser hn he rdiionl more conservive insrumens in he fce of negive demnd shock in he underlying mrke. 13

14 3.2 Nionl-level Empiricl Evidence We furher es our heoreicl implicions using nionl dse of house price chnges nd he prevlence of ggressive insrumens. We obin meropolin re price indices from OFHEO. We selec ll meropolin res in he US which hve experienced ol coninuous nominl price decline of les 5% ny ime in he ps. This includes he following en ciies: Boson, Dlls, Denver, Honolulu, Los Angeles, New York, Phoenix, Sl Lke Ciy, Sn Diego, nd Sn Frncisco. Tble 4 provides summry sisics for his d. We obin lon originion d from he Federl Housing Finnce Bord. 5 This d is from FHFB s Monhly Survey of Res nd Terms on Convenionl Single-Fmily Nonfrm Morgge Lons. The repored informion is bsed on fully morized morgge lons used o purchse single-fmily non-frm homes nd excludes non-morized lons, blloon lons, nd lons used o refinnce houses. The survey repors only convenionl morgges, nd hus excludes morgge lons insured by he Federl Housing Adminisrion (FHA) or gurneed by he Veerns Adminisrion (VA). The FHFB d se conins ll originions s well s he proporion of ARM originions hrough ime. Since ech mrke reched is op nd boom differen imes, we use he difference beween he proporion of ARM originions in ech ciy nd he proporion of ARM originions cross he nion. This is he excess ARM originions bove he nionl verge for ech ciy which djuss he d for he seculr rend of incresed use of ARMs. We lso repor he chnge in he excess ARM originions in Tble 4. Even jus he descripive sisics repored in Tble 4, provide suppor for our heoreicl predicions. The proporion of ARM originions in mos mrkes h experienced lrge negive demnd shock ws bove he nionl verge he respecive peks of hese mrkes. Furhermore, he proporion of ARM originions fell below he nionl verge following he negive demnd shock in ech ciy. Tble 5 repors single vrible regression of he op o boom decline, mesured in bsolue erms, in meropolin re funcion of he ARM originions shre in h re in excess of he nionl verge originions shre he op of he mrke. The proporion of ARMs on he op of he mrke hs lrge nd significn impc on he subsequen price decline. This finding is consisen wih our heoreicl implicion h he presence of ARMS he op of he mrke mgnifies he subsequen negive demnd shock. The effec using nionl d repored in Tble 5, is smller in mgniude hn he effec for he Los Angeles neighborhoods repored in Tble 2. This is no surprising, s meropolin res experience reducion in he vriion of price declines s resul of lrge ggregion. 5 Federl Housing Finnce Bord December 18 h, 2006 <hp:// (Tble 12) 14

15 Tble 6 repors he regression resuls of he op o boom chnge in he proporion of ARM originions in excess of he nionl verge chnge in proporion s funcion of he percen decline in ech meropolin re h experienced decline. While his resul is mrginlly significn ( he 10% level), i is of subsnil mgniude nd in he expeced sign. This evidence is consisen wih our heoreicl implicion h he shre of ARM originions experience lrger declines in mrkes h experience lrger price declines. We furher invesige he impc of ARMs cross he nion boh in rising (RISING) nd flling mrkes. To his end, we firs regress he OFHEO price chnge for ech yer beween 1986 nd 2002 on he proporion of ARM originions he previous yer for ll meropolin res in he OFHEO dse. We hen regress he slope esimes from ech of he 22 regressions on he verge nionl ppreciion re for h yer. The posiive nd significn coefficien repored in Tble 7 indices h high shre of ARM orgnizions hve posiive impc on subsequen price chnges during up mrkes nd negive impc during down mrkes. In oher words, mrkes wih relively high concenrion of ggressive insrumens experience lrger price flucuions over he mrke cycle, which is consisen wih he heoreicl findings of our model. 3.3 Alernive Explnions Boh he nionl nd he Los Angeles empiricl evidence is srongly consisen wih our heoreicl predicions. However, wo lernive mechnisms cn poenilly genere similr empiricl findings. Firs, ffordbiliy consrined mrkes re supply inelsic, nd, herefore experience lrger price increses nd declines wih he mrke cycle (see Linnemn nd Wcher (1989)). In his cse, high concenrion of ARMs on he mrke op is response o he ffordbiliy consrin, nd he decline in ARM shre is demnd led response o he reduced ffordbiliy consrin he mrke boom. While perfecly plusible, his explnion does no in ny wy conrdic our hypohesis h ggressive lending insrumens mgnify he mrke cycles. As long s ARMs help llevie he ffordbiliy consrin in he inelsic mrkes he op, heir presence pushes rnscion prices even higher on he op of he mrke. Absen ggressive insrumens, he highly binding ffordbiliy consrin he op of he mrke would hve curiled he price run-ups nd, herefore, reduced he mgniude of he mrke cycle. This is excly s prediced by he heoreicl model. Noneheless, he mgniude of he impc of ggressive lending insrumens we esime from he d would be lessened wih his lernive mechnism, if i is work. Second, i is possible h exubern borrowers he op borrow using ggressive morgge insrumens nd lso bid up prices. In his cse, even if ggressive lending insrumens did no exis, hose sme overly-opimisic borrowers would hve bid up prices o he sme exen. While his explnion my poenilly genere he firs effec we find, i.e., h prices fll more in ARM-rich neighborhoods or ciies, i is unlikely o genere he second, h ARM shre declines more for hrder hi neighborhoods. We find his resul for lons used for purchse only, s well s in he enire dse. I is hrd 15

16 o imgine h somebody purchsing home is no opimisic bou he fuure of he mrke or heir personl fuure even following negive demnd shock. Th is, hey re voiding ARMs he boom of he mrke while noneheless purchsing homes becuse hey re ferful of he fuure nd hey re more ferful in res where prices declined more. In ny cse if his were he ol explnion we would no find nonsignificn response for refi borrowers. Buyers re usully opimisic, negive demnd shock is jus ssocied wih fewer number of opimiss in he mrke. Noneheless, we pln o direcly es hese wo lernive implicions in fuure work. In priculr, we pln o include mesure of mrke elsiciy s n explnory vrible for he price decline in ddiion o he shre of ARMs. Furhermore, we pln o invesige rejecion res cross mrkes nd hrough he mrke cycle o es if home buyer s senimen owrds ggressive lending chnges. 4.0 Conclusion In his pper we show, boh heoreiclly nd empiriclly, h he presence of ggressive lending insrumens mgnifies rel ese mrke cycles. Mrkes wih high concenrion of ggressive lending insrumens re risk of relively lrger price declines following negive demnd shock. A he sme ime, mrkes h decline he mos following negive demnd shock, end o suffer greer wihdrwl of ggressive lending. The five mrkes h currenly hve highes concenrion of ggressive lending insrumens re Florid, Arizon, Disric of Columbi, Nevd, nd Cliforni for prime lons, nd Illinois, Uh, Cliforni, Arizon, nd Nevd for sub-prime (Appendix ble 1). Our findings predic h hese mrkes re likely o experience relively he lrges mrke declines should negive demnd shock occur. 16

17 References Allen, F Presidenil Address: Do Finncil Insiuions Mer? The Journl of Finnce. 56: Allen, F. nd D. Gle Innovions in Finncil Services, Relionships, nd Risk Shring. Mngemen Science. 45: Allen, F. nd D. Gle Opiml Finncil Crises. Journl of Finnce. 53: Brh, J. R., e l Governmens vs. Mrkes. Jobs nd Cpil, VII (3/4), Cse, K, nd R. Shiller Is There Bubble in he Housing Mrke? Brookings Ppers on Economic Aciviy (Brookings Insiuion), 2003:2, Edelsein, R Explining he Boom Cycle, Speculion or Fundmenls? The Role of Rel Ese in he Asin Crisis. M.E. Shrpe, Inc. Publisher Edelsein, R., Y. Dokko, A. Lcyo, nd D. Lee Rel Ese Vlue Cycles: A Theory of Mrke Dynmics. Journl of Rel Ese Reserch. 18(1): Eichholz, P., N. degrf, W. Ksrop, nd H. Veld Inroducing he GRP 250 propery shre index. Rel Ese Finnce. 15(1): Herring, R. nd S. Wcher Rel Ese Booms nd Bnking Buss-An Inernionl Perspecive. Group of Thiry, Wsh. D.C. Hung, S. nd C. Tu An exminion of house price ppreciion in Cliforni nd he impc of ggressive morgge producs. Working pper Inernionl Monery Fund. Sepember World Economic Oulook: The Globl Demogrphic Trnsiion. 2: <hp:// Kriner, J. nd C. Wei House Prices nd Fundmenl Vlue. FRBSF Economic Leer Krugmn, P Th Hissing Sound. The New York Times: Augus 8. Lemer, E Bubble Trouble? Your Home Hs P/E Rio Too. UCLA Anderson Forecs. McCrhy, J. nd R. Pech Are Home Prices he Nex Bubble? FRBNY Economic Policy Review. 10 (3):

18 Mer, K. nd B. Renud Asi s Finncil Crisis nd he Role of Rel Ese. M.E. Shrpe Publishers. Pvlov, A. nd S. Wcher Robbing he Bnk: Shor-erm Plyers nd Asse Prices. Journl of Rel Ese Finnce nd Economics. 28:2/3, Pvlov, A. nd S. Wcher The Anomy of Non-recourse Lending. Working Pper. Sio, H The US rel ese bubble? A comprison o Jpn. Jpn nd he World Economy, 15, Shiller, R The Bubble s New Home. Brron s: June 20. Simon, R. nd J. Hgery More Borrowers wih Risky Lons re Flling Behind. Wll Sree Journl: December 5. Smih, M, G. Smih, nd C. Thompson When is Housing Bubble no Housing Bubble? Working Pper. Shun, C An Empiricl Invesigion of he role of Legl Origin on he performnce of Propery Socks. Europen Docorl Associion for Mngemen nd Business Adminisrion Journl, 3, Wei, L Subprime Lenders re Hrd o Sell. Wll Sree Journl: December 5. 18

19 Tble 1: Descripive Sisic of he Los Angeles Price nd Lon D Vrible Men Medin Minimum Mximum S. Dev Price decline 21.4% 21.1% 6.8% 34% 7.7% by region 1990 proporion of ARMs 7% 6.5% 1.2% 12.4% 2.9% All lons 1995 proporion of ARMs 10% 9% 2.1% 20% 3.8% All lons Chnge in proporion of 3% 3% -2.1% 14% 3.7% ARMs, Tble 1 provides summry sisics for he Los Angeles price nd lon d. The price decline is compued using he following equion: 22 22, (23) ln( V ) = α ζ + β ζ + X γ + ε i z zi z zi i i i z= 1 z= 1 where V i denoes he vlue of rnscion i, ζ zi is n indicor vrible which kes he vlue of 1 if he propery is loced in zone z, i is n indicor vrible which kes he vlue of 1 if he rnscion occurred beween Augus, 1995 nd Jnury, 1996, X i denoes horizonl vecor of physicl chrcerisics of he propery, α, β, nd γ re prmeers of he model nd ε i is he esimion error. The physicl chrcerisics we include re number of bedrooms nd bhrooms, size of he lo nd of he building, yer buil, nd wheher he propery hs pool or no. Given he bove equion, he esimed prmeers β z re esimes of he percen decline in propery vlues from 1990 o 1995 for ech of he 22 neighborhoods. The medin percen decline during h period ws jus over 21% for he enire meropolin re, rnging from 7% o 35% for ech of he 22 neighborhoods. Lon originion d from Wells Frgo Morgge conins prive-lbel securiized morgge lon originions, spnning period from 1988 o The summry sisics provided in Tble 1 re for ll originions in 1990 nd 1995, nd he chnge beween 1990 nd While he ARM shre incresed on verge cross he nion during he period, here ws gre dispersion in he growh res in differen LA neighborhoods. In fc, bou qurer of our neighborhoods sw no increse or decrese in he shre of ARMs during he period despie he posiive seculr rend. 19

20 Tble 2: Los Angeles Price Decline nd Aggressive Lending in 1990 Dependen vrible in bsolue vlue Originion Lons Type of lons Consn % ARM, 1990 R 2 Percen price decline All lon originions My 1990 Oc 1995 (3.19) (2.56) Percen price decline Lons for purchse only My 1990 Oc 1995 (4.9) (2.34) Percen price decline Lons for refinncing nd My 1990 Oc 1995 Equiy only (3.67) (1.8) Tble 2 repors single vrible regression of he op o boom decline in neighborhood s funcion of he ARM shre in h neighborhood he op of he mrke. The ol decline is mesured in bsolue erms. We repor he resuls for ll lons s well s lons used for purchse nd refinncing seprely. The proporion of rms on he op of he mrke hs lrge nd significn impc on he subsequen price decline. For ech one percen increse in shre of ARMs in 1990, he price decline incresed by 1.3 percen for h neighborhood. This finding is consisen wih our heoreicl implicion h he presence of ARMs he op of he mrke mgnifies he effec of he subsequen negive demnd shock. Furhermore, s expeced, our resuls re wekes for lons used for refinncing. Our model suggess h ggressive lending insrumens llow borrowers o purchse homes hey oherwise cnno fford. Clerly, his effec is wekened for refinncing lons, since he borrower is lredy n owner. Noneheless, ggressive lons hve posiive impc on refinncing becuse he borrower my be more willing o pospone sle of heir home if hey cn wihdrw lrge porion of he equiy hey hve. 20

21 Tble 3: Chnge in Aggressive Lending nd Los Angeles Price Declines Dependen Vrible Consn Percen price decline, My 1990 Oc 1995 (bsolue vlue) R 2 Chnge in proporion of ARM originions , All lons (4.93) (-3.48) Chnge in proporion of ARM originions , Purchse only (3.4) (-3.43) Chnge in proporion of ARM originions , Refinnce or equiy ou only (3.86) (-1.43) Tble 3 repors he resuls of single vrible regression of he chnge in proporion of ARM originions during he 1990 o 1995 period on he percen price decline in ech neighborhood. Our model predics h ARM originions decline he mos in res h suffered he lrges price declines during crsh. The resuls repored in Tble 3 re consisen wih his implicion, s he percen decline from op o boom hd negive impc on he chnge in ARM originions. This finding hs lrge implicions nd is significn for ll lons s well s lons used for purchse. As expeced, he effec is weker bu sill presen for refinncing nd equiy ou lons. Aggressive insrumens pper o be ho money. Their prevlence pus he mrke risk s heir originions end o decline on relive bsis fser hn he rdiionl more conservive insrumens in he fce of negive demnd shock in he underlying mrke. 21

22 Tble 4: Descripive Sisic of he Nionl Price nd Lon D Vrible Men Medin Minimum Mximum S. Dev. Price decline, op o boom 11.8% 11.37% 6.69% by ciy Proporion of ARMs Nionl verge proporion Top of he mrke Proporion of ARMs Nionl verge proporion Chnge from op o boom of he mrke 13.8% 13.5% -13.2% -16.5% -10% 35% % 11% 4.13 Tble 4 provides summry sisics for he nionl price nd lon d. We use he OFHEO price index o mesure price declines. We selec ll meropolin res in he US which hve experienced ol coninuous nominl price decline of les 5% ny ime in he ps. This includes he following en ciies: Boson, Dlls, Denver, Honolulu, Los Angeles, New York, Phoenix, Sl Lke Ciy, Sn Diego, nd Sn Frncisco. We obin lon originion d from he Bnker s Associion. Their d se conins ll originions nd he proporion of ARM originions hrough ime. Since ech mrke reched is op nd boom differen imes, we use he difference beween he proporion of ARM originions in ech ciy nd he proporion of ARM originions cross he nion. This is he excess ARM originions bove he nionl verge for ech ciy. This djuss he d for he seculr rend of incresed use of ARMs. We lso repor he chnge in he excess ARM originions in Tble 4. Even jus he descripive sisics repored in Tble 4 provide some suppor for our heoreicl predicions. The proporion of ARM originions in mos mrkes h experienced lrge negive demnd shock ws bove he nionl verge he respecive peks of hese mrkes. Furhermore, he proporion of ARM originions fell below he nionl verge following he negive demnd shock in ech ciy. 22

23 Tble 5: Nionl Price Decline nd Aggressive Lending Dependen vrible in bsolue vlue Consn % ARM op minus % ARM nionl verge R 2 Percen price decline Top o boom (4.8) (2.07) Tble 5 repors single vrible regression of he op o boom decline in meropolin re S funcion of he ARM originions shre in h re in excess of he nionl verge originions shre he op of he mrke. The ol decline is mesured in bsolue erms. The proporion of ARMs on he op of he mrke hs lrge nd significn impc on he subsequen price decline. This finding is consisen wih our heoreicl implicion h he presence of ARMs he op of he mrke mgnifies he subsequen negive demnd shock. The effec using nionl d is smller in mgniude hn he effec for he Los Angeles neighborhoods repored in Tble 2. This is no surprising, s meropolin res experience reducion in he vriion of price declines s resul of lrge ggregion. 23

24 Tble 6: Chnge in Aggressive Lending nd Nionl Price Declines Dependen Vrible Consn Percen price decline, Top o boom (bsolue vlue) R 2 Chnge in proporion of ARMs minus Nionl proporion of ARMs (.3) (-1.65) Tble 6 repors he regression resuls of he op o boom chnge in he proporion of ARM originions over he nionl verge chnge in proporion s funcion of he percen decline in ech meropolin re h experienced decline. While his resul is mrginlly significn ( he 10% level), i s of subsnil mgniude nd in he righ sign. This evidence is consisen wih our heoreicl implicion h he shre of ARM originions flls more in mrkes h experience lrger price declines. 24

25 Tble 7: The effec of ggressive lending during up nd down mrkes Dependen Vrible Consn Averge Appreciion Re R 2 Slope of price chnge on ARM originions shre (0) (5.76) For ech yer beween 1986 nd 2002 we firs regress he OFHEO price chnge during h yer on he proporion of ARM originions he previous yer for ll meropolin res in he OFHEO dse. We hen regress he slope esimes from ech of he 22 regressions on he verge nionl ppreciion re for h yer. The posiive nd significn coefficien repored in Tble 7 indices h high shre of ARM originions hve posiive impc on subsequen price chnges during up mrkes nd negive impc during down mrkes. In oher words, mrkes wih relively high concenrions of ggressive insrumens experience lrger price flucuions over he mrke cycle, which is consisen wih he heoreicl findings of our model. 25

26 Appendix Tbles Tble A1: Firs Liens, coun-weighed, 2006 originions only source: clculions of Federl Reserve economiss, Andres Lehner e l. Prime Lons Boom 5 Top 5 Se ARM % Se ARM% Oklhom Florid Arknss Arizon Louisin Disric of Columbi Norh Dko Nevd Mississippi Cliforni Subprime Lons Boom 5 Top 5 Se ARM % Se ARM% Oklhom Illinois Wes Virgini Uh Tennessee Cliforni Mississippi Arizon Ohio Nevd Percenge of Adjusble Re Lons h re Subprime Lons 60 Subprime Lons (%) Yer 26

27 Tble A2 Recen Collerl Trends in Lending for Ineres-Only nd Py-Opion Adjusble Re Morgges: Combining Higher-Risk Lon Feures Resuls in Risk Lyering nd Heighens he Overll Level of Credi Risk Yer Low or No Documenion Lon o Vlue b Credi Score b Invesor Shre c Prepymen Penly % % 50.50% % % 51.90% % % 59.20% Clculed s percenge of ol ineres-only or py-opion djusble-re morgge originions. b Originl combined lons o vlue nd credi scores re weighed verges. c Clculed s nonowner nd second home originions. Source: LonPerformnce Corporion (Al-A nd B&C morgge securiies dbse). hp:// 27

28 Appendix Tble A3, underlying d follow Nonprime morgge originions d re securiized originions of Al-A nd subprime produc. D on nonprime morgge originions re no fully vilble fer Augus 2005 nd re no displyed. Source: LonPerformnce Corporion (Al-A nd B&C morgge securiies dbse). hp:// 28

29 Appendix TbleA3 (underlying d) Monh nd Yer Nonrdiionl Producs Help Homebuyers Bridge he Affordbiliy Gp Ineres-Only Shre of Nonprime Originions (percen) Py Opion: Negive Amorizion Shre of Nonprime Originions (percen) 1-Dec 1.06% 1.11% 2-Jn 0.75% 0.79% 2-Feb 1.36% 1.39% 2-Mr 2.21% 2.29% 2-Apr 1.71% 1.82% 2-My 1.72% 1.80% 2-Jun 2.86% 3.02% 2-Jul 3.33% 3.48% 2-Aug 3.65% 3.77% 2-Sep 4.15% 4.25% 2-Oc 5.01% 5.18% 2-Nov 5.84% 6.00% 2-Dec 6.31% 6.53% 3-Jn 6.16% 6.44% 3-Feb 6.60% 6.79% 3-Mr 7.07% 7.26% 3-Apr 6.25% 6.46% 3-My 8.28% 8.59% 3-Jun 8.65% 9.03% 3-Jul 8.74% 9.06% 3-Aug 9.23% 9.60% 3-Sep 9.42% 10.19% 3-Oc 10.92% 11.95% 3-Nov 12.66% 14.06% 3-Dec 15.01% 16.76% 4-Jn 15.66% 17.66% 4-Feb 18.80% 20.17% 4-Mr 22.30% 23.91% 4-Apr 24.45% 25.95% 4-My 26.58% 28.51% 4-Jun 29.54% 32.87% 4-Jul 29.96% 34.61% 4-Aug 28.62% 34.68% 4-Sep 29.99% 35.78% 4-Oc 28.83% 35.14% 4-Nov 29.33% 35.23% 4-Dec 29.44% 36.84% 5-Jn 27.83% 36.69% 29

30 5-Feb 28.34% 40.08% 5-Mr 29.88% 45.36% 5-Apr 29.80% 47.68% 5-My 33.42% 51.62% 5-Jun 33.77% 53.60% 5-Jul 36.33% 53.34% 5-Aug 35.97% 53.97% 5-Sep 32.26% 55.07% 5-Oc 30.09% 48.82% 5-Nov 29.04% 52.35% 30