Are morgage lenders guily of he housing bubble? A UK perspecive Aricle Acceped Version Xiao, Q. and Devaney, S. (26) Are morgage lenders guily of he housing bubble? A UK perspecive. Applied Economics, 48 (45). pp. 427 429. ISSN 466 4283 doi: hps://doi.org/.8/36846.26.5623 Available a hp://cenaur.reading.ac.uk/6552/ I is advisable o refer o he publisher s version if you inend o cie from he work. Published version a: hp://www.andfonline.com/doi/full/.8/36846.26.5623 To link o his aricle DOI: hp://dx.doi.org/.8/36846.26.5623 Publisher: Taylor & Francis All oupus in CenAUR are proeced by Inellecual Propery Righs law, including copyrigh law. Copyrigh and IPR is reained by he creaors or oher copyrigh holders. Terms and condiions for use of his maerial are defined in he End User Agreemen. www.reading.ac.uk/cenaur CenAUR Cenral Archive a he Universiy of Reading
Reading s research oupus online
Are Morgage Lenders Guily of he Housing Bubble? A UK perspecive Qin Xiao and Seven Devaney 2 This is an Acceped Manuscrip of an aricle published by Taylor & Francis in Applied Economics on 3/3/26, available online: hp://www.andfonline.com/doi/full/.8/36846.26.5623 Absrac: Exising heoreical models of house prices and credi rely on coninuous raionaliy of consumers, an assumpion ha has been frequenly quesioned in recen years. Meanwhile, empirical invesigaions of he relaionship beween prices and credi are ofen based on naional-level daa, which is hen esed for srucural breaks and asymmeric responses, usually wih subsamples. Meen (999) argues ha local markes are srucurally differen from one anoher and so he coefficiens of any esimaed housing marke model should vary from region o region. We invesigae differences in he price-credi relaionship for 2 regions of he UK. Markov-swiching is inroduced o capure asymmeric marke behaviours and urning poins. Resuls show ha credi abundance had a large impac on house prices in Greaer London and nearby regions alongside a srong posiive feedback effec from pas house price movemens. This impac is even larger in Greaer London and he Souh Eas of England when house prices are falling, which are he only insances where he credi effec is more prominen han he posiive feedback effec. A srong posiive feedback effec from pas lending aciviy is also presen in he loan dynamics. Furhermore, bubble probabiliies exraced using a discree Kalman filer nealy capure marke urning poins. Key words: regional house prices, housing credi, Markov-swiching, asymmeric responses, urning poins JEL code: G2, G2, R5 Corresponding auhor: Hull Universiy Business School, Universiy of Hull, Coingham Road, Hull, HU6 7RX, phone: +44 () 482 46372, email: q.xiao@hull.ac.uk 2 Henley Business School, Universiy of Reading, Reading, RG6 6AH, phone: +44 () 8 378 6657, email: s.devaney@reading.ac.uk
2. Inroducion A generally acceped view is ha irresponsible lending was he prime culpri behind he seep ascen of house prices and he ensuing 27-8 financial crisis ha affeced he US and many oher developed economies. Excessive and poorly allocaed credi also holds a prominen place in explanaions of numerous oher financial and economic crises ha he world has suffered over he previous cenury (BIS, 2, Wong, 2, Gerlach and Peng, 24, Borio and Lowe, 22, IMF, 2, Herring and Wacher, 22, Herring and Wacher, 999, Drees and Pazarbaşioğlu, 998). Alernaive perspecives on explanaory facors are far less prominen, excep in rare cases (see (Reinhar and Rogoff, 28, Kahn, 29)). Therefore, sudying he impac of lending on asse prices and house prices, in paricular, is an imporan exercise, bu he exploraion of oher plausible explanaions also maers for enriching our undersanding of recurren financial crises. This is an area in which he curren sudy hopes o conribue. The Unied Kingdom (UK) has experienced phenomenal growh in housing wealh in recen decades, from 364.7 billion a he beginning of 983 o over 4 rillion a he end of 27 (Oxford Economics (OE) daa). This is a enfold increase over a period when reail prices rose by only 53 per cen i and he physical dwelling sock expanded by less han 25 per cen. ii These numbers imply ha increases in house prices are he primary conribuor o he rise in his counry s housing wealh, a view suppored by published house price indices such as he Halifax Price Index, which rose by 558 per cen over he same period, equaing o 7.2% per annum. This makes he UK an ineresing case sudy for esing he role of credi in driving price increases and he possible presence of housing bubbles. Empirical sudy of he UK is faciliaed by he accessibiliy of good qualiy house price and morgage lending daa. House prices in he UK have received a grea deal of aenion in recen decades from policymakers and economic commenaors. A major reason for his aenion derives from he belief ha rising house prices drive consumpion upwards. iii I is now recognised ha increasing house prices have a significan wealh effec on consumpion and he major models of he UK economy now incorporae housing wealh alongside financial wealh in heir consumpion funcions. The wealh effec of housing refers o he fac ha households, who aemp o smooh consumpion over he life cycle, will spend and borrow more when he value of heir housing asse increases. I is ofen claimed ha house price appreciaion can redisribue wealh bu no increase i in aggregae, and ha he wealh effec on consumpion and invesmen is ambiguous a priori. A permanen increase in house prices will have a posiive wealh effec on landlords and home-owners; i will neverheless have a negaive income effec on enans and prospecive firs-ime buyers (Aoki e al., 24, Goodhar and Hofmann, 28). This view is valid when he majoriy of ransacions ake place among he regular residens of an economy. However, in an open economy such as he UK, where foreign buyers consiue a major driver of cerain housing markes (e.g. Cenral London) in recen years, redisribuion can occur in a rising marke from foreigners o he regular residens. Besides a wealh effec, scholars also recognize he imporance of a collaeral effec. Higher house prices increase housing demand from borrowing-consrained households, leading o credi expansion by financial insiuions. A rise in credi supply may, in urn, have repercussions on house prices because i lowers he lending rae and simulaes economic aciviy. As a resul, prices may rise in he shor-run wih speculaion encouraged by higher
3 expeced capial gains. iv This makes he housing marke prone o subsanial price swings. The cyclicaliy of house prices can hus lead o considerable variaions in households collaeral posiion over he marke cycle. I has been observed ha he amoun of secured borrowing is closely relaed o his collaeral posiion and ha he spread of morgage raes over he risk-free rae varies wih he collaeral posiion of each household (Aoki e al., 24). Financial sysems in indusrialized counries have undergone a process of liberalizaion and deregulaion since he 97s. Goodhar e al. (24) argue ha hese insiuional and regulaory changes are likely o have increased he pro-cyclicaliy of financial sysems by nururing pro-cyclical lending pracices of banks, hus srenghening he links beween he financial secor and propery prices. Goodhar and Hofmann (28) observe ha many indusrialized counries have experienced exraordinarily srong growh in money and credi accompanied by srong growh in house prices in recen decades. The Unied Kingdom wen hrough significan liberalisaion of he morgage marke in he 98s. Beween 98 and 986, a series of regulaory measures were removed in he UK o improve compeiion in he morgage marke. Building socieies were allowed o access wholesale funding markes; banks were allowed o compee direcly wih building socieies in he morgage marke; oher non-banks such as deparmen sores and insurance companies were increasingly able o offer morgage producs. These changes have resuled in increased compeiion and financial innovaion ha made wihdrawing housing equiy easier when house prices rise. Aoki e al. (24) show ha, prior o he mid-98s, here was lile relaionship beween housing equiy and morgage equiy wihdrawal. They became more closely linked from he lae 98s as new morgage producs made refinancing easier and cheaper. Anoher imporan developmen in he morgage marke is he securiisaion of consumer credi. Lowe e al. (22) argue ha securiisaion of morgages creaed a new circui of global capial, while naional morgage markes became he condui hrough which home owners were conneced o his wave of globally sourced capial. In he UK, equiy sored in owner-occupied propery became much more fungible because of he very liberal morgage marke. Doms and Krainer (27) show ha innovaions in he morgage marke afer 2 allowed morgage lenders o lower down-paymen requiremens; and ha coinciding wih his and oher developmens in he morgage marke has been a marked increase in homeownership in he US. To dae, mos sudies on he price-credi nexus focus on he naional housing marke. Sudies of UK regional housing markes have largely cenred on he exisence or oherwise of a ripple effec. The ripple effec may be described by four disincive feaures: i) regional differenials in house price growh are much greaer han can be explained by incomes; ii) he regional differenial in price growh is highly cyclical alongside he business cycle; iii) he relaive house prices of differen regions converges o a se of consans in he long-run; iv) here is an epicenre ha always leads he ups and downs of regional house price gaps, and which is always more volaile han he res (Holmans, 99, Meen, 999). In his ques for he ripple effec, Meen (999) noiced ha here are significan regional differences in he way ha house price growh reacs o economic condiions. He poins ou ha he housing marke of a naion may be bes characerised as a series of inerlinked local markes raher han a single naional marke. He refers his feaure as spaial dependence. These local markes are srucurally differen owing o differences in economic condiions
4 and household composiion. Hence, he coefficiens of any esimaed housing marke model mus vary from region o region his is referred o as coefficien heerogeneiy. Our sudy makes he following conribuions o he lieraure. As far as we are aware, his is he firs scholarly aricle invesigaing he relaionship beween house prices and housing credi based on regional raher han naional house prices. We also make some heoreical conribuions in explaining he conundrum surrounding he role of credi in he deerminaion of house prices by consrucing a model based on marke opporuniies. Like in Aoki e al. (24) and in Oralo-Magne and Rady (26), he crucial ransmission mechanism in our model is credi consrain. Unlike hese models, ours does no require households o be coninuously raional. The research communiy has become increasingly aware ha he assumpion of raionaliy, a sandard feaure of mainsream economic models, may no hold for all households in all imes (see (Xiao, 2, TheGuardian, 25) for a lis of references). For insance, paricipans in housing markes are ofen subjec o marke psychology raher han making raional deliberaions beween non-housing consumpion and housing invesmen. Empirically, he inclusion of a Markov-swiching sae variable enables us o endogenize srucural changes indicaed by earlier lieraure (Goodhar and Hofmann, 28), as well as asymmerical adjusmen in differen phases of marke cycles (Cook, 23). 2. Review of he lieraure Goodhar and Hofmann (28) poin ou ha opimal porfolio adjusmen is he ransmission mechanism beween money and house prices. This mechanism also suggess a wo-way ransmission. An expansion of money reduces he marginal uiliy of liquid asses relaive o he marginal uiliy of oher asses. Agens rebalance asse porfolios in an aemp o resore equilibrium, resuling in increases in a broad range of asse prices (Melzer, 995; Nelson, 23). By he same oken, an increase in house prices alers he relaive marginal uiliy of housing asses, riggering a porfolio rebalancing which involves a higher demand for moneary asses (Greiber and Sezer, 27). Goodhar and Hofmann (28) furher argue ha here are poenially mulidirecional links beween money, credi, house prices and he wider economy, and ha no definie conclusions can be drawn in he absence of a fully-fledged general-equilibrium heoreical model. There are some general equilibrium models in he heoreical lieraure which consider a subse of he variables menioned in he preceding passage. Based on he financial acceleraor model of Bernanke e al. (999), Aoki e al. (24) consruc a general equilibrium model of house prices, housing invesmen and consumpion. The model, based on he uiliy maximizaion principle, describes how credi channels amplify and propagae shocks when households choose beween non-housing consumpion and housing invesmen. In a similar vein, Oralo-Magne and Rady (26) propose a model of he housing marke wih a credi consrain and a propery ladder leading from sarer homes o rade-up homes. In heir model, he abiliy of young households o afford he down paymen on a sarer-home is a powerful driver of he housing marke. They highligh a channel whereby changes in income may yield overreacion, especially in he price of rade-up homes. The empirical lieraure has been fascinaed by he ofen observed coincidence beween house price growh and credi expansion. Does his coincidence reflec merely he effecs of some common driving force, such as economic policy or he business cycle, or does i reflec a direc link beween he wo variables? If here is a direc link, does i run from house prices o housing credi or from housing credi o house prices, or in boh direcions?
5 Hoffman (2), using a vecor error correcion model (VECM), found ha he long-run developmen of privae secor credi canno be explained in he majoriy of he 6 counries invesigaed wihou including an index of average residenial and commercial propery prices. He furher finds ha, in mos counries, he ineracion beween credi and propery prices is bi-direcional. Gerlach and Peng (25) also adop a VECM framework, in heir sudy of Hong Kong. They found ha bank lending appears o adjus o propery prices in he long-run, no he reverse. In he shor-run, propery prices also drive bank lending, bu lending does no appear o influence he shor-run dynamics of prices. In he conex of he Irish economy, Fizparick and McQuinn (27), again using VECM, found a muually reinforcing long-run relaionship beween house prices and housing credi. In he shor-run, hey found ha he conemporaneous value of credi growh has a posiive and significan effec on house price growh; bu he reverse is no rue. The lagged level of credi (loan-oincome raio) also conribued o he rae of house price growh. In an aemp o assess he lead-lag relaionships beween money, credi, house prices and he macro-economy, Goodhar and Hofmann (28) performed an analysis for a panel of 7 indusrialized counries using fixed-effecs panel vecor auoregression (panel VAR). Their sample spanned from 973 o 26, bu hey also re-esimaed heir model over a shorer period from 985 o 26 because of suspeced srucural changes. Their empirical analysis shows ha money growh has a significan effec on house prices and credi, credi influences money and house prices, and house prices influence boh credi and money. This mulidirecional link is found o be sronger over 985 o 26 han over he full sample period. They aribue his o financial liberalizaion in indusrialized counries, which was discussed earlier. They furher found ha shocks o house prices, credi and money all have significan repercussions on GDP, CPI and ineres raes. Shocks o he laer group of macroeconomic variables in urn have significan effecs on house prices, money and credi. Furhermore, he impac of shocks o money and credi on house prices are sronger when house prices are rising han oherwise. Davis and Zhu (2) consruc a simple model of propery demand, sock adjusmen and new consrucion. Credi consrain is incorporaed in he demand funcion. Their model suggess ha lending is closely relaed o propery prices and propery markes can develop cycles given plausible assumpions. Their empirical work, using VECM framework, covers 7 indusrialized counries spanning he period 97 23. Their resuls show ha rising commercial propery prices cause credi expansion in some counries and ha increasing lending booss commercial propery prices in ohers. They furher found ha propery prices show paricularly srong links o credi in counries where banking crises were associaed wih propery losses during 985 995. Boh prices and credi are srongly influenced by GDP growh. Finally, using daa from OECD counries from 92 o 2, Bordo and Landon-Lane (23) esimae a panel VAR in order o idenify shocks ha can be inerpreed as loose moneary policy shocks, low inflaion shocks, bank credi shocks and house price shocks. They show ha loose moneary policy played an imporan role in housing booms along wih he oher shocks. They also show ha, during boom periods, here is a heighened impac of all hree policy shocks wih he bank credi shock playing an imporan role. However, when hey look a individual house price boom episodes, he cause of he price boom is no so clear. while on average i plays an imporan role, he bank credi shock is no imporan for he US house price boom of he 99s and 2s. (p.37) Loose moneary policy and low inflaion played some role, bu he unexplained residual is he mos imporan shock
6 explaining he US house price boom. The auhors argue ha he residual could be picking up oher unspecified facors such as financial innovaion and he shadow banking sysem. 3. The Model 3.. Capial marke arbirage condiion as a graviy pull A housing marke, like any oher marke, is subjec o arbirage. From an invesor s poin of view, a housing asse should generae he same rae of reurn as any oher asse obainable in he wider capial marke, excep for a difference in risk premium. This risk premium could reflec a liquidiy risk, an operaional risk or a risk of capial loss. Allowing for he fac ha housing has a consumpion value, individuals may be willing o pay a consumpion premium which is over and above he price ha can be jusified purely from an invesmen poin of view. The above argumens imply a varian of he usual asse marke arbirage condiion ha should hold in he long-run. Hence P E[ P R C ] I () where P = he price of one uni of housing asse a he beginning of period ; R = he impued ren incurred a he end of period ; I = he ime varying marke required rae of reurn; and C = a composie variable capuring any oher explanaory variables, such as credi consrain, risk premium and consumpion premium. The coefficien shows how C is relaed o P. To economize on noaions, we will se = in he heoreical derivaion. Define hence i I log ; (2) i E P R C lnp log. (3) i If he ransversaliy condiion, lim E p i i, is saisfied, and if he impued ren grows a a consan rae, i can be shown (see (Xiao and Huang, 2)) ha he fundamenal house price saisfies: p f ( ) j j E r j c j. < < (4) where x=ln(x), and,, and are funcions of he underlying srucural parameers and are imeinvarian under he consan growh assumpion. As c includes he influence of credi consrains, relaxaion of a binding consrain would have a posiive impac on house prices. If expecaions are adapive/pah-dependen (Nerlove, 958, Feige, 967) v, his fundamenal price will be a funcion of he hisory of he variables involved in equaion 4 hese migh include he lagged values of impued ren and morgage advances.
7 3.2. The forces of speculaion under uncerainy The pah of house prices implied by he arbirage condiion would be observed a all imes if housing markes were efficien in exploiing all available informaion. Empirical evidence shows ha sock and bond markes are semi-srong efficien (Brennan and Schwarz, 982, Pesando, 978, Hochkiss and Ronen, 22, Fama, 97, Fama e al., 969, Fama and French, 988, Jaffe, 974, Hall, 2, Shiller, 98), bu real esae markes are differen. Real esae asses are lumpy and are infrequenly raded in privae markes wih high ransacion coss. They are also locaion specific, requiring subsanial local knowledge. This suggess ha informaion inefficiency is an inheren feaure of real esae markes. Furhermore, hey can be subjec o exensive inervenion from local and naional governmen, which creaes an addiional source of uncerainy in he form of legislaive risk. As a resul, real esae asse prices can deviae sysemaically from fundamenals for a prolonged period. This deviaion can generae profi opporuniies, hence inviing speculaion (Case and Shiller, 989, Gunermann and Norrbin, 99, Maier and Herah, 29, Beracha and Skiba, 2, Xiao, 2, Xiao and Huang, 2). Consider a posiive general produciviy shock which leads o a boom in he real secor of he economy. Demand rises, supply adjuss o capure he profi opporuniy creaed by he excess demand and firms hire more workers and seek more producion space. Wage raes rise in he labour marke where firms compee o arac workers wih he righ amoun of skills. The reurns and values of commercial real esae asses increase wih he rising demand for space. Wih inelasic supply vi, house prices also climb up as more households can afford o own housing asses wih income generaed from heir human and capial asses. Produciviy growh is hus ranslaed ino a housing marke boom. In his changed and changing environmen, i is hard for buyers and sellers of housing asses o assess heir inrinsic value. The impued ren migh have changed and he housing asse premium migh have shifed. Such informaion is no easily accessible by an average marke paricipan. House prices a any given ime are hus se by he flow demand and he flow supply, which only parly bear he signaure of a raional calculaion based on equaion (4). Under he uncerainy, individuals may resor o momenum rading, creaing a posiive feedback effec. Furhermore, researchers indicae ha emoion influences our decisions in many imporan maers (Ashkanasy and Humphrey, 2, Ashkanasy, 23, Fisher, 28, Huy, 22). Therefore, a any ime, house prices may conain a bubble elemen b. p f p b (5) 3.3. Credi consrain as a ransmission mechanism Suppose ha borrower ypes are uniformly disribued beween and. Type- borrowers always repay he loan, while ype-s never do. A lender knows he disribuion of borrower ypes, bu canno easily disinguish one from anoher. In order o minimize he risk of defaul, hey se one of wo consrains on every borrower: M PH, (6) Or
8 M Y (7) whichever is lower. The Greek leer denoes he loan-o-value raio, wih (, ), and he Greek leer he income muliple, wih >. Boh and are se by he lender. The Roman leer M denoes he size of he morgage a household can ake ou, Y is income, and H he household s desired unis of housing services, which is normalized o uniy from now on o avoid cluered noaion. If neiher of he consrains is binding, a household can borrow as much as i wishes a he going ineres rae and he loan marke would clear a he required rae of reurn. When one of he wo consrains is binding, he amoun of loan available is deermined by eiher of he above wo equaions. Mahemaically, M S f S I M D f ' M S M wih unbinding D Oherwise consra in Wih I D M f D f ' D (8) where f = funcion, S M = morgage supply, and D M = morgage demand. Morgage supply may also be influenced by lender confidence on op of he mechanical rules se ou above. We will gauge lender confidence wih reference o morgages in arrears in he empirical secion. Equaion 8 implies ha if he consrain is no binding, dm M S S d di I d df di I M S (9) df I Noice ha S is he elasiciy of morgage supply wih respec o he ineres rae. In di M his case, house prices and income will have no bearing on morgage supply. If he income muliple consrain is binding, hen dm M S S d db d B () dy d Y In his case a boom in he labor marke will be ransmied ino a boom in he housing marke hrough he relaxaion of he credi consrain in equaion 4. If he collaeral consrain is binding, hen
9 dm M S S d da d A () dp d P In his case, a shock o he housing marke will be ransmied o he credi marke. By he same oken, a bubble in he housing marke will also spill over ino he credi marke. The heoreical derivaions imply an empirical sysem of he following form: Where y X E E (2) ' V P y, M P X x M x, The row vecor x P consiss of explanaory variables for house prices, and he row vecor x M consiss of explanaory variables for housing credi. I is possible ha agens in he housing and/or morgage marke behave asymmerically in differen phases of a marke cycle. The phase of a cycle can be capured by a laen sae variable and he parameers of he above model are sae-dependen. The exac sae ha he marke is in is no observable, bu he probabiliy of he sae can be esimaed. In his case, he above sysem can be rewrien as s y X E E (3) s s s ' V s s where s is a sae variable following a firs-order wo-sae Markov chain, wih ransiion probabiliies Pr s Pr s j s i, s j s i i,..., (4) q ij i, j,2 where is he informaion se available a ime and qij is he probabiliy ha sae i will be followed by sae j given s i, s i,..., and. Equaion 4 saes ha he probabiliy disribuion of s depends on pas evens alone hrough he value of s vii.
4. Empirical Analyses 4.. Daa and mehodology Our full sample consiss of quarerly daa for UK regional housing markes over he period 983Q - 22Q. We begin our analysis in he mid-98s owing o daa availabiliy, bu his also coincides wih significan liberalizaion of he UK residenial morgage marke. From 983Q o 27Q4, house prices increased by 558. per cen in he UK as a whole and by 683.5 per cen in Greaer London (Halifax, all buyers, seasonally adjused price). The marke umbled aferwards. By 22Q, house prices had dropped by 8. per cen in he UK and 2.4 per cen wihin Greaer London (see Figure ). In order o assess is reliabiliy given he price movemens described above, he sysem se ou in equaions 3 and 4 is esimaed using he subsample of daa spanning 983Q 27Q4. The parameers obained are used hen o assess he ou-of-sample performance of he model during 28Q 22Q. Twelve geographical regions are covered in his sudy: Norh (N), Yorkshire & Humber (Y&H), Norh Wes (NW), Eas Midlands (EM), Wes Midlands (WM), Eas Anglia (EA), Souh Wes (SW), Souh Eas (SE), Greaer London (GL), Wales (W), Scoland (S) and Norhern Ireland (NI). The locaions of hese regions wihin he UK are shown in Figure 2. For comparison, a separae model is also esimaed for UK average house prices. As he sudy period covers wo known housing marke cycles, we have a sample feauring muliple peaks and roughs wih which o assess he accuracy of he esimaed sae ransiion probabiliy as a predicor of acual urning poins. INSERT FIGURES &2 ABOUT HERE INSERT TABLE ABOUT HERE Table displays variables included in he regression equaions a he model selecion sage. Halifax sandardized regional house prices are used for P viii. Ideally, regional level morgage advances should be used in he empirical model, bu he auhors have no access o such daa. Insead, naional ne morgage advances are used as a proxy for regional M. Regional renal prices are also approximaed by UK impued renals of owner-occupiers for he same reason. In he price equaion, we include in he composie variable, C, he affordabiliy of morgages (morgage raes), he influence of ownership (owner-occupied housing sock per person), and he impac of unemploymen (regional claiman coun rae). In he morgage equaion, we include morgages in arrears as an addiional explanaory variable. This variable may negaively impac lender confidence regarding he fuure sae of affairs. I may also limi he capaciy of lenders o issue new loans. In he regression, house prices were divided by gross disposable income per head. The resul is a house price series ha is independen of any income effec. This is henceforh referred o as he price-income muliple. Meanwhile, ne morgage advances and impued renals were divided by gross disposable income o creae a morgage-income raio and a renal-income raio, respecively. Finally, owner-occupied housing sock was divided by populaion o give sock-per-person. This process, on op of incorporaing he influences of income and populaion, has he added advanage ha differencing he ime series, which is usually required o aain saionariy, became unnecessary. Imporan long-run informaion is hus reained and he exraced sae probabiliies are more likely o reflec he cyclical movemens of he differen regional markes. All he variables presened in Table have passed Augmened
Dickey-Fuller (ADF) and Phillips and Perron (PP) uni roo ess a he convenional significance levels. ix To reduce compuaion coss, he sysem equaion models are pre-seleced using no-swiching hree-sage-leas-squares (3SLS). Variables are dropped one a a ime when insignifican a he 95 per cen level a his iniial sage. Seleced models are hen re-esimaed using Markovswiching FGLS SUR. The FGLS esimaor is appropriae when he disurbance erm is heeroscedasic. The erm SUR refers o he esimaion mehod whereby he cross-equaion covariance is explicily esimaed and incorporaed. The smoohed probabiliy of he unobservable sae variable s for a given sample of size T, Pr s i, is inferred using discree Kalman filer (Hamilon, 994). In esimaing Pr s i T T, we assume ha he DGP s parameers, are known, when in ruh hey need o be esimaed. This can be achieved hrough maximizing he log likelihood funcion of he observed daa using EM algorihm, as EM algorihm is efficien, simple and sable (Dellaer, 22, Xiao and Tan, 26). Simulaions are used o esablish he 99, 95 and 9 per cen confidence inervals for he parameer esimaes, as he criical values for sandard -ess do no apply in his case (Hamilon, 994, Horowiz, 2, Soffer and Wall, 99). The join significance of he sysem of equaions is esed using he Wald saisic. As par of he diagnosic checks, he esimaed parameers are applied o observaions beween 28 Q and 22 Q. The model performed well in boh in-sample fi and ou-of-sample forecas. The excellen fi has been achieved in many cases as a resul of including lagged dependen variables. There is, herefore, a srong posiive feedback effec in boh house prices and housing credi (see Table 3 and Figure 3 below). 4.2. Parameer esimaes, model evaluaion, and discussion 4.2.. The morgage marke As explained earlier, UK ne morgage advances are used as a proxy for regional morgage advances. As a resul, he morgage equaion has he same explanaory variables wih similar parameer esimaes across differen regions (Table 2). However, he esimaes are no idenical as he covariance srucures beween house prices and morgage advances are heerogeneous across differen regions. On he whole, he morgage-income raio is negaively and significanly affeced by conemporaneous year-on-year changes in he morgage rae. x I is also significanly and negaively affeced by he conemporaneous value of he percenage of morgages in arrears for 2 monhs or longer. Thus, afer conrolling for income, rising morgage coss and increasing morgage arrears deer lending aciviy hrough heir effecs on poenial buyers and lenders. Ineresingly, he impac of he second facor (arrears) is much larger han he impac of he firs (morgage raes). For insance, for he UK as a whole, a one-uni increase in he year-on-year change of morgage rae reduces he raio by.67 in sae one (he sae associaed wih he expansionary phase of he marke) or by.352 in sae wo (associaed wih a conracing marke). In conras, his raio would go down by.224 (in sae one) or.766 (in sae wo) if he percenage of morgages in arrears were o increase by one uni. The impac of average house prices on he morgage-income raio is significan and posiive, as expeced, bu i is quie small in magniude. For example, if he price-income muliple increases by one uni, he morgage-income raio would increase by merely.7 (in sae one) or.79 (in sae wo). Finally, he lagged value of he morgage-income raio is an imporan facor; if he value of his raio increased by one
2 uni in he previous quarer, is curren value increases by.698 (in sae one) or.73 (in sae wo). INSERT TABLE 2 ABOUT HERE I is obvious from Table 2 ha he magniude of impac of each explanaory variable varies no only across regions bu also across differen phases of a cycle. In a conracing marke, as opposed o an expanding one, he supply of morgages is more responsive o ineres rae changes in all regional markes excep for EM, WM, NW, SE and GL, and he dampening effec of morgages in arrears on morgage lending is far larger in all regions excep for EM, WM, NW and GL. Thus, wih he excepion of he regional markes menioned, our resuls sugges ha lending aciviy is more sensiive o loan pricing and poenial losses when he underlying collaeral is deerioraing in value. Meanwhile, excep for NW, GL and SE, he posiive-feedback-effec represened by he lagged morgage-income raio is smaller in a conracing marke, and, apar from NW and GL, he posiive impac of price on morgage advances is slighly larger. Thus lenders may become more cauious in following he lending momenum and slighly more aware of price movemens when he housing marke is collapsing. The above observaions indicae ha NW and GL have behaved consisenly in he way morgage supply responding o all marke influences; EM and WM are consisen wih each oher in heir morgage supply response o ineres raes and morgages-in-arrears; SE, on he oher hand, is consisen wih NW and GL in he response o ineres raes and price-income changes. EM and WM are righ nex o each oher geographically, bu NW is separaed from GL and SE by EM and WM. Thus, geography has a role o play, bu here is no simple geographical paern wih which o convey he enire sory. 4.2.2. The housing marke The wo mos imporan effecs on house prices are credi and posiive-feedback effecs. The resuls in Table 2 show ha he impac of ne morgage advances on house prices (he credi effec) is ypically posiive, bu varies significanly across differen regions. The impac is very large (.4.956) in he souh (GL, SW, SE and EA), modes (.29.368) in he middle par of he counry (W, WM and EM), relaively small (.99.256) in he norh (Y&H, N, NW and S), and insignifican in Norhern Ireland. The credi effec also varies across he wo saes. In all cases excep GL and he SE, morgage advances have a larger effec on house prices in an expansionary housing marke. This finding concurs wih Goodhar and Hofmann (28), showing ha he souhern regions are likely o collapse quicker han he res of he counry in a ighened credi environmen. For insance, he parameer esimaes of morgageincome raio (M()) for GL and EM are respecively.465 and.368 in sae one, and.956 and.329 in sae wo. Across boh saes, he posiive feedback effec dominaes he credi effec in all regional markes, wih he excepion of a conracing GL. The parameer esimaes of he lagged priceincome muliple (which represens a posiive-feedback effec) range from.84 o.86 in sae one, and from.728 o.25 in sae wo. Wih he excepion of GL, SE, and SW, he feedback effec is slighly smaller in an expanding marke. As he posiive feedback effec reflecs price bubbles, his observaion indicaes ha he souhern regions are more prone o price bubbles han he res of he counry when he marke is booming.
3 Three oher facors appear o have some impac on he house price-income muliple in a leas one of he welve regions. Firs, he sock of owner-occupied housing per person has a large and posiive impac across boh saes for GL and WM, and in one of he wo saes for SW. This suggess ha housing ownership has a visible impac only in a very limied number of regional markes. Second, he claiman coun has a marginally significan negaive effec a he 99 per cen level on he house price-income muliple in one of he wo saes in NI, bu no in oher regions. Third, he only region where house prices appear o have been direcly negaively affeced by ineres raes is EA. This impac only appears o be imporan in expanding markes and is nearly negligible in a conracing marke. The esimaion resuls indicae ha he UK housing marke may be roughly divided ino hree major submarkes: one formed by SW, SE and GL (he souhern submarke); one by W, WM, EM and EA (he middle submarke); and he final one by NW, Y&H, N and S (he norhern submarke). This is consisen wih he conclusions of previous researchers who find ha he UK housing marke is segmened along roughly he same lines (MacDonald and Taylor, 993, Meen, 999). Regions wihin he same submarkes exhibi similar price responses o he availabiliy of naional morgage credi, as well as o heir own pas sae. However, NI is a separae eniy from he res. For he price equaion, i has a differen se of explanaory variables and hese variables are significan a he convenional levels only in one of he wo saes. The model is joinly significan for all regions as indicaed by he Wald saisics (Table 3). INSERT TABLE 3 ABOUT HERE 4.2.3. The sae of he UK regional housing markes The erm sae in he curren conex refers o he expansion or he conracion phase of a housing marke cycle. Table 2 indicaes ha house price growh has a sligh posiive impac on credi expansion, ha credi expansion produces furher credi expansion, which has a large posiive impac on house price growh, and ha house price growh breeds furher price growh. These rounds of price growh, excep perhaps for he iniial shock, can occur wihou any changes o fundamenals. Hence, a price bubble is more likely o form in a rising marke and so he erm sae may refer o he bubble sae of he marke. The smoohed probabiliies of he sae are exraced using he discree Kalman filer. If he esimaed probabiliy ha he price Pr s s i q, i =, 2, is greaer han.5, we conclude ha he housing is in sae one, i marke is more likely o be in a bubble sae han no, and vice versa. Thus, hey are inerpreed as he esimaed probabiliies ha he house price conains a bubble elemen (EPB). Bear in mind ha q i does no need o rise when prices rises, as he laer can do so for fundamenal reasons as well. However, because of a srong feedback effec, a price bubble would grow wih a rising price. Hence, i q should peak when he corresponding house price peaks and rough when he price roughs. When an esimaed urning poin misses he acual one by one, wo, hree, four or more quarers, i is marked wih one, wo, hree or four aserisks correspondingly. Among he 2 regions, he esimaes correcly prediced he urning poins 36.7% of he ime, missed he urning poins by one quarer 6.3% of he ime, by wo quarers.2% of he ime, by hree quarers 8.2% of he ime. Tha is o say ha he esimaes capure he urning poins wihin +/- hree quarers 7.4% of he ime. The esimaes indicae ha mos regions experienced wo bubble periods beween 983Q2 and 27Q4: one in he lae 98s and one in he years before he end of 27 (Table 4 and Figure 4).
4 INSERT FIGURE 4 ABOUT HERE INSERT TABLE 4 ABOUT HERE The firs expansion of he housing marke wihin he sample period began in he souhern submarke righ a he sar of he period (983Q2). This was followed wo years laer by EA, and hree years laer by he res of he middle submarke and by he NW region. Expansion in Y&H and N did no begin unil 987Q and 988Q, respecively. The expansion ended firs in he souhern submarke and EA in 988Q4. WM followed sui in 989Q, he EM and W regions in 989Q2, and N, NW and Y&H in 989Q4. Neiher S nor NI was caugh in his cycle according o our resuls. The second expansion sared in GL in 995Q4. This was followed by SE in 996Q2, EA in 998Q2, and SW in 999Q2. S joined his expansion in 2Q and N in 2Q4. EM, WM, NW and Y&H were relaive laecomers, beginning expansion in 2Q. W hen joined in 2Q4 and NI evenually jumped on o he bandwagon in 23Q. The sign of a downurn appeared as early as 27Q in NW, SW and W, and in N and NI one quarer laer. By 27Q3, mos of he remaining regional markes also sared urning down. Only wo markes, S and WM, seemed unyielding o he downward pressure by he end of he esimaion sample (27Q4). If we are o compare regional prices wih he UK naional average, he expansion / conracion paern refleced in he laer would fail o pick up he rich opography of regional house price movemens ha our analysis has indicaed. This shows he danger of relying on a single naional indicaor for gauging he sae of he housing marke in he Unied Kingdom and, more broadly, he need for housing marke analyses o be aware of poenial regional differences. 5. Conclusion The curren sudy aemps o esablish imporan conribuing facors o he rapid ascen of UK housing wealh prior o 27 and he ensuing economic hardship. A priori, a growh in naional disposable income (he income effec), a shifing of wealh from oher asse markes ino housing (he porfolio effec), an influx of foreign money (he open-marke effec) and an expansion of domesic housing credi (he credi effec) could have all played some par in ha drama. To examine all of hese effecs, one needs an all-inclusive general equilibrium framework, as argued by Goodhar and Hofmann (28). In he curren sudy, we consruc a parial equilibrium model focusing on he las facor he expansion of domesic credi. I is generally believed ha a malfuncioning credi marke was he culpri for he housing bubble prior o he 27-8 financial crisis and is primarily responsible for he crisis iself. This sudy aemped o idenify he exen o which he housing loan marke could be held responsible. We chose o examine his quesion in he UK raher han in he USA or oher counries because of he pronounced price movemens in he UK over his period and he availabiliy of high qualiy daa. To dae, mos sudies on he price-credi nexus focus on he naional housing marke. Meen (999), in his ques o examine a ripple effec in prices, noiced ha here are significan regional differences in he way house price growh reaced o economic condiions. Thus, we sudy his issue by looking a regional housing markes insead. An analyical framework incorporaing a bubble ransmission mechanism beween housing and credi markes was firs se up. The model was hen esimaed using hisorical daa for he
5 period 983Q o 27Q4. Geographical differences were examined using regional house prices and srucural shifs were allowed for by applying a Markov-swiching echnique o differeniae beween expanding and conracing markes. Boh in-sample and ou-of-sample daa (28Q 22Q) fi very well wihin he model. The heoreical model suggess ha credi consrains ac as a propagaion mechanism, ransmiing shocks occurring in he labour and/or housing marke o he morgage marke. Ye our resuls for UK regional housing markes sugges ha he loan-o-value raio does no appear o be an imporan explanaory variable. The empirical oucomes show ha he price-income muliple is significan, bu has a raher small explanaory power for ne morgage advances, while he variable represening unemploymen is no usually significan a all. Thus, credi consrains appear o have limied power in ransmiing shocks in he labour and housing markes o he morgage marke. The variable which has he larges impac on lending aciviy is morgages-in-arrears. The large coefficien on his variable may primarily reflec lender confidence raher han lending capaciy consrain, as oal morgages in arrears reached only 3.5 per cen a is previous peak in 992 and.4 per cen a is recen peak in 29 xi. We furher found ha, in a falling (as opposed o a rising) marke and in mos regions, boh ineres-raes and morgages-in-arrears have a larger negaive effec and he price-income muliple has a slighly larger posiive impac on morgage lending, bu he momenum effec of lending drops slighly. The larger dampening effec of morgage in arrears is paricularly desabilising in a declining marke given is size. This asymmeric response may reflec procyclical lending pracices of banks, as prediced by Goodhar e al. (24), and insiuional consrains faced by lenders. The larger negaive impac of ineres raes in a down marke suggess ha moneary policy can be more poen in such a marke, which is of ineres for policymaking. However, he supply of morgages in some regions (e.g. EM, WM, NW, GL and SE) does no always respond o economic and marke influences in he same way as in oher regions. Thus, a policy or regulaory measure designed o help he marke may have differenial effecs across differen regions and he aggregae oucome may no be as anicipaed. The heoreical model suggess ha ren is one of he mos imporan drivers of house prices. Empirically, impued ren failed o show any significan impac on he price. Ineres raes reflecing he discoun facor also failed o have any direc impac on prices in he majoriy of regional markes. Neverheless, i has an indirec impac via he credi effec. Insead, house prices are shown o be primarily driven by a posiive-feedback effec and a credi effec. Furhermore, he feedback effec is far more imporan ha he credi effec. Thus, alone, irresponsible lending in he housing credi marke could no have creaed one of he bigges housing bubbles in he known economic hisory of he Unied Kingdom. We conribue o he lieraure heoreically by explicily modeling credi consrain as a bubble ransmission mechanism ha is based on marke opporuniies raher han uiliy opimizaion. We also conribue o he lieraure empirically by examining regional differences across he Unied Kingdom as well as variaions in differen phases of marke cycles in hese regions. We discovered ha he esimaes of he sae probabiliies have he poenial o serve as a marker for daing housing marke cycles, a baromeer for measuring he evoluion of housing bubbles, and a predicor of marke urning poins. Fuure work may seek o esablish a general equilibrium model capuring all four effecs discussed earlier in his secion, and improve upon our resuls by employing regional explanaory variables.
6 References AOKI, K., PROUDMAN, J. & VLIEGHE, G. 24. House prices, consumpion, and moneary policy: a financial acceleraor approach. Journal of Financial Inermediaion, 3, 44-435. ASHKANASY, N. M. 23. Emoions in organizaions: A mulilevel perspecive. In: DANSEREAU, F. & YAMMARINO, F. J. (eds.) Research in muli-level issues. Oxford, UK: Elsevier/JAI Press. ASHKANASY, N. M. & HUMPHREY, R. H. 2. A muli-level view of leadership and emoions: Leading wih emoional labor. In: BRYMAN, A., COLLINSON, D., GRINT, K., JACKSON, B. & UHO-BIEN, M. (eds.) Sage Handbook of Leadership. UK: Sage. BARKER, K. 24. Review of Housing Supply. Available: hp://image.guardian.co.uk/sysfiles/guardian/documens/24/3/7/barker.pdf BERACHA, E. & SKIBA, H. 2. Momenum in residenial real esae. Journal of Real Esae Finance and Economics, 43, 299-32. BERNANKE, B., GERTLER, M. & GILCHRIST, S. (eds.) 999. The financial acceleraor in a quaniaive business cycle framework.: Norh-Holland. BIS 2. 7s Annual Repor. Unpublished manuscrip. BORDO, M. D. & LANDON-LANE, J. 23. Wha explains house price booms?: hisory and empirical evidence. NBER working paper 9584. Naional Bureau of Economic Research. BORIO, C. & LOWE, P. 22. Asse prices, financial and moneary sabiliy: Exploring he nexus. BIS working paper no. 4. BRENNAN, M. J. & SCHWARTZ, E. S. 982. An equilibrium model of bond pricing and a es of marke efficiency. Journal of Financial and Quaniaive Analysis, XVII, 3-329. CASE, K. E. & SHILLER, R. J. 989. The Efficiency of he Marke for Single-Family Homes. The American Economic Review, 79, 25-37. COOK, S. 23. The convergence of regional house prices in he UK. Urban Sudies, 4, 2285-2294. DAVIS, E. P. & ZHU, H. B. 2. Bank lending and commercial propery cycles: Some crosscounry evidence. Journal of Inernaional Money and Finance, 3, -2. DELLAERT, F. 22. The Expecaion Maximizaion Algorihm. GVU Cener; College of Compuing; Georgia Tech, GIT-GVU-2-2,. DOMS, M. & KRAINER, J. 27. Innovaions in morgage markes and increased spending on housing
7 Federal Reserve Bank of San Francisco Working Paper No. 5. DREES, B. & PAZARBAŞIOĞLU, C. 998. The Nordic banking crisis: Pifalls in financial liberalizaion? IMF occasional paper no. 6. FAMA, E. F. 97. Efficien capial marke: a review of heory and empirical work. Journal of Finance, 25, 383-47. FAMA, E. F., FISHER, L., JENSEN, M. C. & ROLL, R. 969. The adjusmen of sock prices o new informaion. Inernaional Economic Review,, -2. FAMA, E. F. & FRENCH, K. R. 988. Dividend yields and expeced reurns on sock reurns. Journal of Poliical Economy, 22, 3-25. FEIGE, E. L. 967. Expecaions and Adjusmens in he Moneary Secor. The American Economic Review 57, 462-473. FISHER, C. D. 28. Wha if we ook wihin-person performance variabiliy seriously?. Indusrial and Organizaional Psychology: Perspecives on Science and Pracice, 2, 85-89. FITZPATRICK, T. & MCQUINN, K. 27. House prices and morgage credi: Empirical evidence for Ireland. Mancheser School, 75, 82-3. GERLACH, S. & PEN, W. S. 25. Bank lending and propery prices in Hong Kong. Journal of Banking & Finance, 29, 46-48. GERLACH, S. & PENG, W. 24. Bank lending and propery prices in Hong Kong. Hong Kong Moneary Auhoriy working paper. GOODHART, C. & HOFMANN, B. 28. House prices, money, credi, and he macroeconomy. Oxford Review of Economic Policy, 24, 8-25. GOODHART, C., HOFMANN, B. & SEGOVIANO, M. 24. Bank regulaion and macroeconomic flucuaions. Oxford Review of Economic Policy, 2, 59-65. GUNTERMANN, K. & NORRBIN, S. 99. Empirical ess of real esae marke efficiency. The Journal of Real Esae Finance and Economics, 4, 297-33. HALL, R. 2. Sruggling o undersand he sock marke---he 2 ely lecure. American Economic Review 9, -. HAMILTON, J. D. 994. Time series analysis, Princeon Universiy Press. HERRING, R. J. & WACHTER, S. 999. Real esae booms and banking buss: An inernaional perspecive The Group of 3, occasional papers. Washingon, D.C. HERRING, R. J. & WACHTER, S. (eds.) 22. Asse price bubbles: Implicaions for moneary, regulaory and inernaional policies: MIT Press.
8 HOFFMAN, B. 2. The deerminans of privae secor credi in indusrialised counries: Do propery prices maer?. Working Paper 8. Bank of Inernaional Selemens. HOLMANS, A. 99. House prices: changes hrough ime a naional and sub-naional level. Governmen Economic Service Working Paper No.. HOROWITZ, J. L. 2. The boosrap. In: HECKMAN, J. J. & LEAMER, E. (eds.) Handbook of Economeric 2. Elsevier Science. HOTCHKISS, E. S. & RONEN, T. 22. The Informaional Efficiency of he Corporae Bond Marke: An Inraday Analysis. Review of Financial Sudies, 5, 325-354. HUY, Q. 22. Emoions and sraegic organizaion: Opporuniies for impacful research. Sraegic Organizaion,, 24-247. IMF 2. World Economic Oulook. JAFFE, J. F. 974. Special informaion and insider rading. Journal of Business, 47, 4-428. KAHN, J. A. 29. Produciviy swings and housing prices. Curren Issues in Economics and Finance, 5. LOWE, S. G., SEARLE, B. A. & SMITH, S. J. 22. From housing wealh o morgage deb: he emergence of Briain's asse-shaped welfare sae. Social policy and sociey,, 5-6. MACDONALD, R. & TAYLOR, M. P. 993. Regional house prices in Briain : Long-run relaionships and shor-run dynamics. Scoish Journal of Poliical Economy, 4, 43-55. MAIER, G. & HERATH, S. 29. Real Esae Marke Efficiency: A Survey of Lieraure. SRE- Discussion 29/7. Available: hp://epub.wu.ac.a/42//documen.pdf. MARCH, J. G. 988. Decisions and organizaions, New York, NY, Blackwell. MEEN, G. 999. Regional house prices and he ripple effec: A new inerpreaion. Housing Sudies, 4, 733-753. NERLOVE, M. 958. Adapive Expecaions and Cobweb Phenomena. The Quarerly Journal of Economics, 72, 227-24. ORTALO-MAGNE, F. & RADY, S. 26. Housing marke dynamics: On he conribuion of income shocks and credi consrains. Review of Economic Sudies, 73, 459-485. PESANDO, J. E. 978. On he Efficiency of he Bond Marke: Some Canadian Evidence. Journal of Poliical Economy, 86, 57-76. REINHART, C. M. & ROGOFF, K. S. 28. Is he 27 U.S. sub-prime financial crisis so differen? An inernaional hisorical comparison. NBER Working Paper 376.
9 SHILLER, R. J. 98. Do sock prices move oo much o be jusified by subsequen changes in dividends? The American Economic Review 7, 42-436. STOFFER, D. S. & WALL, K. D. 99. Boosrapping sae-space models: Gaussian maximum likelihood esimaion and he Kalman filer. Journal of he American Saisical Associaion, 86, 24-33. THEGUARDIAN. 25. Emoions, psychology or personaliy wha's your spending rigger? [Online]. The Guardian. [Accessed 6 Jan 26 26]. WONG, K.-Y. 2. Housing marke bubbles and he currency crisis: The case of Thailand. The Japanese Economic Review, 52, 382-44. XIAO, Q. 2. Crashes in real esae prices: Causes and predicabiliy. Urban Sudies, 47, 725-744. XIAO, Q. & HUANG, W. 2. Risk and predicabiliy of Singapore s privae residenial marke. Quaniaive Finance,, 529-543. XIAO, Q. & TAN, G. 26. Markov-swiching uni roo ess: a sudy of propery price bubbles in Hong Kong and Seoul. The 35h Ausralian Conference of Economiss.
2 Table - Variables Examined During Regression Analysis All daa are sourced from DaaSream, excep for he Halifax sandardised regional house prices, hese being obained from he websie for Lloyds Banking Group. All variables in he able were included in he regression equaions a he model selecion sage. Variables were dropped from he analysis if hey were saisically insignifican a he 95 per cen level. () indicaes a conemporary value and (-) a lagged value. Explained variables Halifax sandardised regional house price /UK gross disposable income per head UK morgage ne advances/uk gross disposable income Explanaory Variables UK morgage ne advances/uk gross disposable income (M()) YOY change of UK morgage rae (%) (I()) UK Impued renals of owner-occupiers/uk gross disposable income (R()) UK morgages 2 monhs or more in arrears (%) (A()) YOY change of UK morgage rae (%) (I()) UK median morgage percen advance (LTV()) UK owner occupied housing sock (Vol)/Populaion (O()) UK house price/uk gross disposable income per head (UK_P()) YOY change in regional claiman coun rae (%) (C()) Lagged UK morgage ne advances/uk gross disposable income (M(-)) Lagged regional house price/uk gross disposable income per head (P(-))
2 Table 2: Parameer Esimaes (Sample 983Q 27Q4) Parameers, along wih he covariance marices, are esimaed using ieraive FGLS, wih confidence inervals (C.I.) esablished using Boosrap simulaion wih replicaions. If an esimae falls ouside he 9, 95 or 99 per cen C.I., i is marked wih a * or ** or *** respecively. For insance, he coefficien of he lagged price/income raio of Eas Anglia falls ouside he 99 percen C.I., which implies ha he probabiliy ha he rue parameer akes he value of.843 is less han per cen. Eas Anglia (EA) Price Morgage Sae one M() P(-) I() I() A() UK_P() M(-).644**.843*** -.225 -.343 -.722.79.7 Sae wo M() P(-) I() I() A() UK_P() M(-).4.943 -.6 -.55-3.699..6 Eas Midlands (EM) Price Morgage Sae one M() O() P(-) I() A() UK_P(T M(-).368 6.259*.84 -.347 -.698.76.725 ) Sae wo M() O() P(-) I() A() UK_P(T M(-) ).329-3.238*.976** -.77 -.33.78.669 Greaer London (GL) Price Morgage Sae one M() O() P(-) I() A() UK_P() M(-).465 5.637.94 -.356 -.75.77.79 Sae wo M() O() P(-) I() A() UK_P() M(-).956 7.53.736 -.9* -.964.58.745 Norh (N) Price Morgage Sae one M() P(-) I() A() UK_P() M(-).284*.935 -.33 -.2.63.746 Sae wo M() P(-) I() A() UK_P() M(-).9.965 -.34 -.737.78.78 Norhern Ireland (NI) Price Morgage Sae one C P(-) I() A() UK_P() M(-) -.72***.86*** -.24 -.26.5.822 Sae wo C P(-) I() A() UK_P() M(-) -.4**.998** -.348 -.725.76.722 Norh Wes (NW) Price Morgage Sae one M() P(-) I() A() UK_P() M(-).235*.942 -.332 -.69.76.722
22 Sae wo M() P(-) I() A() UK_P() M(-).7***.967*** -.55 -.48.67.76 Scoland (S) Price Morgage Sae one M() P(-) I() A() UK_P() M(-).257***.949** -.3 -.538.63.782 Sae wo M() P(-) I() A() UK_P() M(-).99.972 -.344 -.726.77.77 Souh Eas (SE) Price Morgage Sae one M() O() P(-) I() A() UK_P() M(-).63-6.232**.2 -.356 -.73.75.726 Sae wo M() O() P(-) I() A() UK_P() M(-).839 8.952*.728 -.8-2.94.2.627 Souh Wes (SW) Price Morgage Sae one M() O() P(-) I() A() UK_P() M(- ).749 -.72***.93* -.65 -.923.64.687 Sae wo M() O() P(-) I() A() UK_P() M(- ).664.472.777 -.359 -.769.78.75 Wales (W) Price Morgage Sae one M() P(-) I() A() UK_P() M(- ).33.933 -.29 -.29.56.797 Sae wo M() P(-) I() A() UK_P() M(- ).65.95 -.354 -.748.78.77 Wes Midlands (WM) Price Morgage Sae one M() O() P(-) I() A() UK_P() M(- ).37 5.66.872 -.346 -.678.74.73 Sae wo M() O() P(-) I() A() UK_P() M(- ).29-6.54*.25** -.98 -.48.89.68 Yorkshire & Humber (Y&H) Price Morgage Sae one M () P (-) I() A() UK_P() M(- )
23.22.955 -.79 -.239.57.785 Sae wo M () P (-) I() A() UK_P() M(- ).7.962 -.35 -.742.77.79 All UK Price Morgage Sae one M() O() UK_P(-) I() A() UK_P() M(- ).256 -.948.965* -.67 -.224.7.698 Sae wo M() O() UK_P(-) I() A() UK_P() M(- ).362 6.89.865 -.352 -.766.79.73
24 Table 3: Diagnosics for Regression Models In-sample (983Q 27Q4) and ou-of-sample (28Q 22Q). Region In-sample Fi Ou-of-sample Fi R 2 Adj R 2 Wald saisic Sae R 2 Sae Adj R 2 Sae 2 R 2 Sae 2 Adj R 2 EA.997.996 95.79.959.949.984.979 EM.996.996 26.28.977.97.983.978 GL.999.998 25.34.992.99.96.949 N.996.996 29.7.98.976.979.974 NI.994.993 24.38.945.933.97.965 NW.997.997 227.85.982.978.986.983 S.997.997 224.68.986.982.983.979 SE.998.998 22..982.978.966.956 SW.997.997 22.26.96.949.966.957 W.996.996 223.29.984.98.98.977 WM.996.996 28.39.986.982.985.982 Y&H.997.996 25.55.986.983.98.977 UK.998.998 23.25.99.988.985.98
25 Table 4 Cycles and Bubbles in UK Regional Housing Markes (c.f. Figure 6) If he esimaed probabiliies ha he explained variable is in sae one, Prs s i qi, i =, 2, is greaer han.5, we conclude ha he housing marke is more likely o be in a bubble sae han no, and vice versa. Noe ha q i need no o rise when he price rises as i may do so for fundamenal reasons. However, if accurae, qi should peak when he corresponding house price peaks and rough when he price roughs. When an esimae misses he corresponding observed urning poin by one, wo, hree, four or more quarers, i is marked wih one, wo, hree, or four aserisks correspondingly. Refer o figure 4 for crossverificaion. Region Bubble period Observed and esimaed phases of cycles (Prob(s =) >.5) Observed Esimaed EA 983Q2-984Q4 Conrac 983Q2 985Q2 983Q2 988Q4**** 989Q2-992Q2 Expand 985Q2 988Q4 988Q4 989Q2** 22Q - 27Q4 Conrac 988Q4 998Q2 989Q2 998Q* Expand 998Q2 27Q3 998Q 27Q4* EM 983Q2-987Q2 Conrac 983Q2 986Q2 983Q2 988Q4**** 989Q3-999Q3 Expand 986Q2 989Q2 988Q4 99Q*** 2Q4-27Q4 Conrac 989Q2 2Q 99Q 2Q Expand 2Q 27Q3 2Q 27Q4* GL 983Q2-989Q Expand 983Q2 988Q4 983Q2 988Q4 996Q3-27Q4 Conrac 988Q4 995Q4 988Q4 992Q2**** Expand 995Q4 27Q3 992Q2 27Q3 Conrac 27Q3 27Q4 27Q3 27Q4 N 988Q4-989Q4 Conrac 983Q2 988Q 983Q2 988Q 22Q3-25Q3 Expand 988Q 989Q4 988Q 989Q2** Conrac 989Q4 2Q4 989Q2 2Q* Expand 2Q4 27Q2 2Q 24Q3**** Conrac 27Q2 27Q4 24Q3 27Q4 NI 24Q3-27Q2 Conrac 983Q2 23Q 983Q2 23Q
26 Region Bubble period Observed and esimaed phases of cycles (Prob(s =) >.5) Observed Esimaed Expand 23Q 27Q2 23Q 27Q2 Conrac 27Q2 27Q4 27Q2 27Q4 NW 983Q2-984Q Conrac 983Q2-986Q2 983Q2 986Q3* 989Q3-27Q4 Expand 986Q2-989Q4 986Q3 989Q2** Conrac 989Q4-2Q 989Q2 996Q3**** Expand 2Q - 27Q 996Q3 27Q4*** Conrac 27Q 27Q4 **** S 24Q2 27Q Conrac 983Q2 2Q 983Q2 24Q**** Expand 2Q 27Q4 24Q 27Q*** Conrac 27Q 27Q4**** SE 983Q2-989Q Expand 983Q2 988Q4 983Q2 989Q* 24Q4-27Q4 Conrac 988Q4 996Q2 989Q 2Q4**** Expand 996Q2 27Q3 2Q4 27Q4* Conrac 27Q3 27Q4 **** SW 984Q4-99Q Expand 983Q2 988Q4 983Q2 989Q* 23Q4-27Q Conrac 988Q4 999Q2 989Q 2Q2**** Expand 999Q2 27Q 2Q2 27Q Conrac 27Q -27Q4 27Q 27Q4 W 988Q2-989Q2 Conrac 983Q2 986Q2 983Q2 986Q2 22Q4-24Q3 Expand 986Q2 989Q2 986Q2 988Q4** 26Q4-27Q Conrac 989Q2 2Q4 988Q4 2Q4 Expand 2Q4 27Q 2Q4 27Q Conrac 27Q -27Q4 27Q 27Q4
27 Region Bubble period Observed and esimaed phases of cycles (Prob(s =) >.5) Observed Esimaed WM 989Q2-986Q2 Conrac 983Q2 986Q2 983Q2 988Q3**** 989Q2-27Q4 Expand 986Q2 989Q 988Q3 99Q**** Conrac 989Q 2Q 99Q4 2Q Expand 2Q 27Q4 2Q 27Q4 Y&H 988Q3-989Q3 Conrac 983Q2 987Q 983Q2 987Q 22Q - 25Q4 Expand 987Q 989Q4 987Q 989Q2** Conrac 989Q4 2Q 989Q2 2Q2*** Expand 2Q 27Q3 2Q2 24Q4**** Conrac 27Q3 27Q4 24Q4 27Q4 UK 986Q2-989Q2 Expand 983Q2 989Q 983Q2 989Q 26Q4-27Q Conrac 989Q 2Q 989Q 992Q**** Expand 2Q 27Q3 992Q 27Q** Conrac 27Q3 27Q4 27Q 27Q4
Q 983 Q 985 Q 987 Q 989 Q 99 Q 993 Q 995 Q 997 Q 999 Q 2 Q 23 Q 25 Q 27 Q 29 Q 2 Ln ofindex (Q 983 = ) Q 983 Q 985 Q 987 Q 989 Q 99 Q 993 Q 995 Q 997 Q 999 Q 2 Q 23 Q 25 Q 27 Q 29 Q 2 Nominal Price (s) 28 Figure : Trends in Key Economic Variables (a) House prices and he house price-income muliple: UK sandardised prices Sandardised Price Price-Income Raio 2 7 8 6 4 2 8 6 4 2 6 5 4 3 2 Price-Income Raio Source: Halifax House Price Index, Lloyds Banking Group (b) Trends in lending, inflaion and economic growh MFIS ne lending Loans on dwellings GNP House prices Inflaion (RPI) 7.5 7. 6.5 6. 5.5 5. 4.5 4. Source: Daasream
29 Figure 2: UK Saisical Regions as a 3s December 2 Map reproduced under he OS Open Daa agreemen: Conains Ordnance Survey daa Crown copyrigh and daabase righ 23. The shaded areas denoe regions of England while Scoland, Wales and Norhern Ireland are separae counries wihin he Unied Kingdom. Boundary changes mean ha here are some differences beween Norh and Norh Wes used in he paper and Norh Eas and Norh Wes as shown on he map. Similarly, here are some differences beween Eas Anglia and Souh Eas used in he paper and Eas of England and Souh Eas as shown on he map.