HOUSE PRICE SHOCKS, NEGATIVE EQUITY AND HOUSEHOLD CONSUMPTION IN THE UK IN THE 1990s

Transcription

1 February 2002 HOUSE PRICE SHOCKS, NEGATIVE EQUITY AND HOUSEHOLD CONSUMPTION IN THE UK IN THE 1990s by Rchard Dsney*, Andrew Henley and Davd Jevons Abstract We examne the mpact of housng captal gans on savngs behavour durng the 1990s Brtsh housng market cycle usng mcrodata from the Brtsh Household Panel Survey and county-level house prce data. We condton the models on household real fnancal captal gans usng Famly Resources Survey data. We fnd a margnal propensty to consume out of housng wealth of between 0.01 and 0.03, dependng on specfcaton. Among our novel fndngs are asymmetrc behavour between perods of house prce rses and falls, wth stronger consumpton response durng perods of house prce ncreases, and a dsproportonate mpact on savng f the household has negatve housng equty. Keywords: Savng, Housng wealth, House prces, Negatve equty JEL Classfcaton: D91, E21, R31 * School of Economcs, Unversty of Nottngham, and Insttute for Fscal Studes School of Management and Busness, Unversty of Wales Aberystwyth Oxford Economc Research Assocates (OXERA) Correspondence: Professor Andrew Henley, School of Management and Busness, Unversty of Wales Aberystwyth, SY23 3DD, Wales, UK. Te l: +44 (0) , emal: andrew.henley@aber.ac.uk. The research was supported by Economcs and Socal Research Councl fundng, grant R Thanks to partcpants n a semnar at the Bank of Sweden for comments. The data used n ths paper were made avalable through the UK Data Archve. The Brtsh Household Panel Survey data were orgnally collected by the Insttute for Socal and Economc Research at the Unversty of Essex, funded by the ESRC. The usual dsclamers apply.

2 1 HOUSE PRICE SHOCKS, NEGATIVE EQUITY AND HOUSEHOLD CONSUMPTION IN THE UK IN THE 1990s 1. Introducton The paper examnes the mpact of shocks to the value of housng wealth on household consumpton n the UK. Unlke prevous studes for the UK, we use long dfferences n household panel data to examne the mpact of house prce shocks on consumpton, followng the basc method utlsed on Unted States data by Sknner (1989, 1996) and Engelhardt (1996). We have several fndngs. Our basc fndng suggests an average margnal propensty to consume out of housng wealth of between 0.01 and 0.03, dependng on the specfcaton comparable to the modal estmates on US mcrodata. However, we also have novel fndngs relatve to the US lterature. Frst we examne the argument that consumpton responses to house prce shocks are asymmetrc, as also suggested by both Engelhardt and Sknner. We fnd the reverse result from the US evdence, namely that consumpton mpacts of house prce changes are stronger when house prces are rsng. We also have a new result concernng negatve equty that the elastcty of consumpton wth respect to house prce shocks s greatest when house prces are rsng for households that had zero or negatve equty values n ther housng stock. The straghtforward nterpretaton of ths last result s that negatve equty nduces precautonary savng so that house prce nflaton that lfts households out of negatve equty nduces a dsproportonately large consumpton response (here, an average margnal propensty to consume out of housng wealth for such households of 0.04 to 0.06). In followng the US lterature n usng mcrodata to obtan these results, our results contrast wth exstng UK studes of the housng wealth-consumpton relatonshp that have typcally used tme seres analyss whether at the natonal or regonal level (Muellbauer and Murphy, 1990, 1995; Carruth and Henley, 1990a, 1990b; Mles, 1992, 1993a; Davey, 2001). However, our fndngs and methods also depart from the US studes n several respects. Frst, we use local house prce varaton over tme from county-level surveys of house prces rather than self-reported values of housng wealth as our proxy for changes n household housng wealth, for two reasons. Such data can be constructed on a consstent bass n the UK (unlke the US, where regonal house prce data are typcally too aggregated to capture local market varatons). In a ddton, there s

3 2 a danger that self-reported values correlate wth savng behavour. Moreover n the estmates we allow for the accumulaton (decumulaton) of fnancal wealth through captal gans (losses) rather than relyng smply on a measure of actve savng. 1 The plan of the remander of the paper s as follows. The next secton brefly dscusses the ssues and the exstng lterature. Secton 3 descrbes the data sets, and how we utlse them. Secton 4 descrbes the econometrc procedures and the man results. Secton 5 concludes. 2. Issues and prevous lterature In the basc lfe cycle model of savng, ncreases n the value of housng wealth should ncrease consumpton, even when there are bequest motves and borrowng constrants (Sknner, 1994). A smlar story arses n the permanent ncome hypothess, f housng wealth s treated as annutsable wealth, and also n a smple story where the household attempts to mantan a constant ncome-wealth rato. There has been dscusson n the Unted Kngdom lterature as to whether the house prce boom of the md-1980s fuelled the consumpton boom n the economy at that tme (Muellbauer and Murphy, 1990; Attanaso and Weber, 1994) and, more recently, as to the contrbuton of realsed captal gans both on fnancal wealth and housng wealth to the declne n the UK household savng rato snce the md-1990s from around 10% to less than 4% of post-tax ncome by md-2000 (Davey, 2001). Smlar arguments have been dscussed n the Unted States see, for example, Bosworth, Burtless and Sabelhaus (1991). Housng wealth s ntrnscally less lqud than fnancal wealth. Thus whle actve savng n fnancal assets seems to be hghly responsve to real captal gans and losses on fnancal assets, 2 actve savng may be less responsve to changes n housng wealth. Housng bequest motves may be strong, and households may be unwllng to extend ther debt, or to move, n order to release housng wealth. 3 Moreover, t has been argued that ndvduals partton ther wealth nto fungble and non-fungble accounts as a form of self-control mechansm (Thaler, 1990). Thus the magntude of the propensty to consume out of housng wealth, and the crcumstances and types of household where 1 It s nconsstent to model the mpact of captal gans (losses) n housng wealth whlst gnorng captal gans (losses) n fnancal wealth n the calculaton, partcularly as there s much evdence from the US that the declne n the measured actve savng rate has been drven by rsng fnancal wealth, especally n the equty market see below. 2 For US evdence: see, nter ala Gale and Sabelhaus (1999), Poterba (2000) and Dynan and Mak (2001) for recent surveys. For the UK, see Davey (2001).

4 3 the greatest response s observed, are ultmately emprcal ssues. Nevertheless, gven that home ownershp s a key form of household wealth-holdng n the Unted Kngdom, even a small response may generate large aggregate mpacts on actve fnancal savng. Tme seres evdence for the Unted States based on aggregate consumpton and aggregate housng wealth, whether estmated as a specfc consumpton functon or through an Euler equaton, generally gves a margnal propensty to consume out of housng wealth of 0.04 to 0.05 (Peek, 1983; Bhata, 1987; Sknner, 1994, 1996). However, US evdence based on household data sets gves very mxed results. Typcally these studes use a self-reported measure of actve savng and a measure of self-reported housng wealth (or equty, net of mortgage) less any reported value of home mprovement work, snce there may be a negatve correlaton between actve savng and the latter. For example, frst dfferences of the Panel Survey of Income Dynamcs from n Sknner (1989) gve a postve effect of housng wealth on consumpton across households, but controllng for selecton of movers and allowng for taste dfferences seems to elmnate these effects. There are few Unted Kngdom studes that provde a coeffcent that can be nterpreted as a margnal propensty to consumer from housng wealth. Nevertheless, Muellbauer and Murphy (1995) obtan an average elastcty of Carruth and Henley (1990b) obtan estmates that mply an elastcty of Mles (1993a) obtans a smulated elastcty of 0.02, but Mles (1997), usng cross secton data, gets hgher responses but the coeffcents are very unstable. A further emprcal fndng from the US lterature s that there seems to be an asymmetry of response between gans and falls n housng wealth. One explanaton for possble asymmetrc behavour n response to housng captal gans s that housng gans may be antcpated, but losses not so (Sknner, 1993). Sknner (1996) fnds greater senstvty of consumpton to falls n housng wealth ndeed there s no sgnfcant effect of rsng house values on consumpton n hs work, whereas for falls the margnal propensty to consume s 0.1. However Engelhardt (1996) gets sgnfcant responses to shocks n both drectons (0.004 to 0.08 to rses, 0.04 to 0.13 for falls) albet agan wth greater responsveness to house prce falls. Agan, however, results are hghly senstve to estmaton method, such as the choce of treatment of outlers. To our knowledge, there are no studes of house prce-consumpton response asymmetres for the UK. There s 3 Especally elderly households: see, for example, Vent and Wse (1990). But see, for the UK, Dsney, Henley and Stears (2002).

5 4 also evdence that households may respond dfferently to housng shocks whch mply nomnal wealth losses, as opposed to changes n real prces, although such studes focus prmarly on the mpact of nomnal losses on housng moblty (Engelhardt, 2001; Genosove and Mayer, 2002). The amount of equty contaned n the housng also has an effect on moblty (Genosove and Mayer, 1997). Snce moblty s a major source of equty release, ths s an mportant ndrect determnant of the mpact of housng wealth on consumpton, wth Dsney, Henley and Stears (2002) fndng clear dfferences n fnancal savng responses to house prces shocks between movers and non-movers among elderly households n the UK. Thus t s mportant to check estmated results for condtonng on movers versus non-movers, and to nvestgate selecton effects n movng behavour. 3. Data We utlse household level longtudnal data on savngs and holdngs of fnancal wealth n order to dentfy the cross-sectonal varaton n house prce movements on household savng behavour over tme. The Brtsh Household Panel Survey (BHPS) provdes longtudnal data on actve savng; that s non-negatve transfers from current ncome nto fnancal wealth. The BHPS s an annually repeated panel survey of around 5000 households resdent n Great Brtan to the south of the Caledonan Canal, commencng n The precse form of the queston on savng asked of each ndvdual n the household s do you save any amount of your ncome for example by puttng somethng away now and then n a bank, buldng socety, or Post Offce account other than to meet regular blls? Note that ths form of the queston purges holdngs of fnancal assets of those that arse smply because of the dscreteness of the payment perod. The average amount saved n ths way each month over the last year s recorded. From Wave 2 onwards a smlar supplementary queston s asked about monthly contrbutons to prvate penson schemes. We defne actve savng as the sum of these two aggregated across all adult household members. Summary nformaton on ths savng measure for each wave from 1992 to 1999 s reported n Table 1. Just under a half of all households report some actve savng. The average for those who save s 180 per month, and for all households s 83 per month. Turnng to real housng gans, we have n prncple two alternatve methods for calculatng the real apprecaton n the value of the household s home over the perod of

6 5 analyss. The frst s to make use of each respondent s annual estmate of the current market value of hs or her home. However ths measure of housng captal gan s contamnated by expendture on mprovements and addtons and by movng behavour, both of whch are lkely to correlate wth measures of fnancal savng. 4 The BHPS provdes only ncomplete nformaton on home mprovement actvty. 5 An alternatve approach, adopted here, s to use changes n average market value n the localty n whch the household was resdent at the start of the perod. More precsely we use the change n the average market prce of a sem-detached resdence n the 1993 county of resdence (out of 65 countres deflated by the UK retal prce, excludng mortgage repayments, ndex, as reported on a quarterly bass by the Halfax Bank. 6 ). Table 2 reports the dstrbuton of real house prce shocks measured n ths way across the BHPS sample of owner-occupers over the perod 1993 to The average real housng gan over the full perod s 21,207, although as the lower medan value ndcates there s some skewness n the dstrbuton of gans. There s wdespread varaton n the scale of the house prce shock across households. In fact 15.6% of owner-occupers experenced a real fall n house prces over ths perod. The perod 1993 to 1995 was a perod of slowng fallng house prces after the bust at the begnnng of the 1990s. Real losses were wdespread wth 87.7% of the sample experencng a real fall n house prces. The average negatve shock up to 1995 was over The perod from 1995 to 1999 was one of recovery, culmnatng n qute sharp rses n % of the sample experenced real housng apprecaton, and the average gan was over 23,000. As dscussed n Secton 2, n order to estmate the true relatonshp between a real house prce shock and actve savng, account must be made of any correlaton between the housng shock and other fnancal components of the household balance sheet, by controllng for the effect of passve savng (Engelhardt 1996) whch occurs through the real apprecaton or deprecaton of fnancal wealth. Ths has not typcally been done 4 Home mprovements (rasng housng equty) may be negatvely correlated wth fnancal savng measures, f partally or wholly funded out of fnancal wealth. Movng wll nvolve transactons costs, whch may agan be borne out of fnancal wealth, but may also be utlsed to release housng equty, especally among older households. 5 Informaton s avalable where mprovements and addtons were fnanced through an addtonal mortgage or loan advance. There s some evdence of overestmaton of house values partcularly durng the early 1990s when the housng market was n recesson, and ths further complcates the use of self-reported gans. Nevertheless, as Engelhardt (1996) notes, self-reported or perceved gans may be a more nformatve drver of savngs decsons than actual gans. 6 The Halfax Bank (formerly Halfax Buldng Socety) s the largest lender for home purchase n the UK.

7 6 n studes of ths knd. Unfortunately, the BHPS does not queston respondents on an annual bass about levels of fnancal wealth. 7 Consequently for nformaton on fnancal wealth we use the annually repeated cross-sectonal Famly Resources Survey (FRS). 8 The model estmated on FRS s a reduced form to explan fnancal asset holdngs n terms of the demographc characterstcs and economc actvty status of the head of household and spouse (f present), household composton and housng tenure status. Ths model s used to mpute fnancal asset holdngs for each ndvdual n owneroccuper households n the BHPS from 1993, and household fnancal wealth s then obtaned by aggregatng wthn households. 9 In effect therefore we construct an nstrumental varable for (unobserved) fnancal wealth for BHPS households usng the rght-hand sde varables as nstruments. We use data from seven avalable surveys up to the fscal year 1999/2000. The Famly Resources Survey s collected to nform government polcy on the socal securty system, and the questonnare s desgned to llct detaled nformaton on ndvdual level wealth wthn a crtcal wealth band of 1,500 to 20,000, wth censorng outsde ths band. Table 3 therefore reports a tobt estmaton usng data pooled from all seven avalable cross-sectons, wth upper and lower censorng ponts mposed to concde wth the questonnare desgn. The results show that nearly all the demographc varables ncluded have very well determned coeffcents, wth sensble sgns and values. Lqud wealth ncreases wth age, but at a declnng rate. Wealth peaks at 58.7 years of age. Chldren dran fnancal resources, and at a hgher rate as they are older. Dvorce, wdowhood and ethnc status each have a substantal depressant effect. The economcally nactve have hgher fnancal wealth. The achevement of a unversty degree s assocated wth substantally hgher wealth. Relatve to the base of outrght ownershp, other forms of housng tenure are assocated wth lower wealth, especally rental status and n partcular publc sector rental. 7 Informaton on fnancal wealth holdng was gathered n wave 5, but ths exercse has only been repeated n wave 10 (at the tme of wrtng unavalable). 8 The FRS covers the same geographcal area as the BHPS, and has acheved a sample for each fscal year (Aprl to March) of 23,000-25,000 households, snce 1993/1994. The questonnare ncludes detaled nformaton on fnancal asset balances. Fnancal wealth s defned as ncludng money left n any current account at the end of the month, the balance of any form of nterest-bearng depost or savngs account and the value of stocks, shares, natonal savngs and premum bond ssues. Assets held by chldren n the household are also ncluded. However t should be noted that ths measure gnores the labltes sde of the household balance sheet as not nformaton on (non-mortgage) debt s solcted. 9 The FRS survey s conducted throughout the year on a fscal year (Aprl to March) cycle. The BHPS s conducted n the fnal quarter of each year, broadly n the mddle of each FRS survey perod, and so FRS 1993/1994 s matched to BHPS 1994 etc.

8 7 Fnally the data reveal, ceters parbus, sgnfcant regonal varatons n wealth, wth households n Wales farng partcularly badly and households n Scotland and Northern England also showng lower wealth (relatve to the London base). The rankng of the coeffcents on these regonal dummy varables correlates closely and postvely wth average regonal house prces, pontng to the domnance of a scale rather than a portfolo effect. 4. Estmatng the Impact of Housng Gans on Savng Defnng cumulatve household actve savng from perod 0 to the current perod as s a, and household fnancal assets as A, then the dentty relatonshp between the change n fnancal assets and actve and passve savng over the same perod (captal gans on holdng fnancal wealth), s p, for household s as follows: (1) A A s + s A 0 a p All varables are expressed n real terms. In order to nvestgate the effect of housng gans on actve savng we specfy an estmatng equaton wth cumulatve actve savng as the dependent varable and rght-hand sde varables n long dfferences to control for both the stock adjustment mpact of the ntal level of fnancal assets and the scale of passve savng, and to model the mpact of real housng shocks: (2) s a = α + α Aˆ + α s + α h + α h. D + Z' γ + ε p 3 4 ne 0 where h s the real house prce shock, ne D 0 s a dummy varable to nvestgate the dfferental mpact of a shock on a household ntally n negatve equty, control varables, αj and γ are coeffcents and Z' are other ε s an error term. ^ denotes that the asset varable s mputed. Equaton (2) as t stands cannot be estmated because we cannot drectly observe or mpute change n fnancal assets, Â, to gve: s p. We therefore replace s p wth our mputed (3) s a = β ˆ ˆ + ne 0 + β1 A0 + β2 A + β3 h + β4 h. D0 + Z' γ ε'

9 8 Usng the dentty n (1) the underlyng margnal propensty to consume from housng wealth (gven by α3 n equaton 2) s now β3 /( 1 β2 ), and that for a household n negatve equty (gven by α + ) n equaton 2) s now β + β )/(1 ). ( 3 4 β2 ( 3 α4 Table 4 reports estmates of equaton (3). All estmates are obtaned usng a robust medan regresson estmator, n order to deal wth problem of the censorng of actve savng at zero, and to mtgate the mpact of ts dstrbutonal skewness. All monetary varables are converted to 1995 prces usng the UK retal prce (excludng mortgage repayments) ndex. Estmaton ncludes control varables as follows: real household ncome n each year covered by the perod n queston, a bnary varable for ntal outrght home ownershp (.e. no mortgage), a quadratc n the ntal age of the household head, the ethnc mnorty status, gender and ntal martal status (marred/cohabtng, separated/dvorced or wdowed) of the household head, bnary varables for a change n martal status, bnary varables for the ntal economc status of the household head and spouse (employed, self-employed, retred), bnary varables for the retrement of the head or spouse durng the perod, bnary varables for educaton to degree level of the head and the spouse, ntal number of adults n the household and numbers of chldren n dfferent age bands (0-2, 3-4, 5-11, 12-15, 16-18), bnary varables for a head and a spouse smoker, and a bnary varable for membershp of an occupatonal (company) penson scheme. 10 Column (1) reports a base specfcaton that ncludes the ntal level of real fnancal assets and the real housng gan on the rght hand sde. The real house prce shock has a statstcally sgnfcant negatve effect on the level of real household savng between 1993 and 1999, and ndcates a margnal propensty to consume from housng wealth of However, as column (2) reveals, ths estmate s based downwards by the omsson of the effect of passve savng (the change n fnancal assets) on actve savng. Once the change n real fnancal assets s ncluded the coeffcent on the real housng gan rses n absolute sze by half and the mpled margnal propensty to consume s now Ths varable attracts a consstent, negatve coeffcent n the actve savng regressons mplyng that occupatonal penson scheme membershp reduces total savng between 1993 and 1999 by 540 at the mean. Ths suggests, unsurprsngly, that occupatonal penson scheme membershp s at least a partal substtute for personal penson savng and perhaps actve savng generally.

10 9 Households n negatve equty Households may make a greater adjustment to ther savng behavour n response to housng captal gans f those gans correspond to a reducton n household balance sheet dstress. Durng the early part of the 1990s nomnal house prce falls were common and a sgnfcant mnorty of households experenced negatve housng equty (.e. ther loan to house value rato exceeded unty). The proporton of owner-occuper households n negatve equty n the estmaton sample n 1993 s 7%. Column (3) ncludes the nteracton of the real housng captal gan wth the ntal negatve equty dummy varable. 11 The nteracton effect attracts a negatve coeffcent that s statstcally sgnfcant at 9 per cent. The results suggest that the margnal propensty to consume from real housng gans s almost three tmes hgher for households n negatve equty. It should also be noted that the nteracton coeffcent s more precsely determned for the perod durng whch house prces were rsng (statstcally sgnfcant at 0.3%). Ths s dscussed further below. Selecton bas from the ncluson of mover households The sample used so far ncludes both households who stay at the same address between 1993 and 1999 and those who move (at least once) durng ths perod. House movng by owner-occupers may be assocated wth sgnfcant portfolo adjustments between housng and fnancal wealth. Dsney, Henley and Stears (2002) fnd sgnfcant dfferences n passve savng behavour between movers and non-movers n a sample of older (over 50 years of age) households. Movng may provde a ready vehcle through whch to release housng captal gans. Gven ths t s mportant to consder the queston of the mpact of mover-stayer selecton bas on our estmates. Column (4) re-estmates the specfcaton n column (3) for non-movers on ther own allowng for a selectvty correcton. The movng decson s dentfed usng addtonal nformaton about household composton, resdence type and the sze of the household relatve to that of the resdence n 1993, along wth a dummy varable capturng a stated preference n 1993 for movng house. As the sample sze nformaton shows 71% of the 1993 sample of owner-occupers remaned at the same address through to The coeffcent on the selectvty correcton term s statstcally sgnfcant at 6 per cent, and there s some reducton n the sze and sgnfcance of real housng gan coeffcents, partcularly for the negatve equty effect. Nevertheless for 11 Henley (1998) descrbes the constructon of ths varable for the BHPS data.

11 10 owner-occupers wth postve equty we stll fnd evdence for a housng gan offset on savngs, even f those households do not actually move to realse that gan. Asymmetrc behavour n response to movements n housng wealth As dscussed n secton 2, US research has suggested that owner-occuper households behave asymmetrcally to nomnal housng gans and losses. The results presented n columns (5) and (6) of Table 4 nvestgate ths ssue. Rather than attemptng to dfferentate the effects of upward and downward house prce movements n crosssecton, we estmate separate regressons for actve savng for 1993 to 1995, when as Table 2 showed most households experenced real housng losses and 1995 to 1999 when real housng losses were mnmal. The results confrm that households behave very dfferently n response to real housng gans compared to losses. However the drecton of the effects are the reverse of those found by Engelhardt (1996) for the Unted States. The effect of the real house prce shock s statstcally sgnfcant for the perod of housng gans but not for losses. Furthermore the negatve equty nteracton effect s now strongly sgnfcant durng the perod of gans. For owner-occupers wth postve equty the mpled mpc from a postve housng wealth gan s 0.009, but for those n negatve equty the mpc s almost fve tmes larger at and more robust than n column (3). Thus the response for households wth ntal negatve equty s also asymmetrc. 5. Concluson Macroeconomc research has suggested that personal sector spendng and savng behavour n the UK may have become closely related to movements n house prces snce fnancal deregulaton of the housng fnance ndustry durng the 1980s. Lfe-cycle theory would suggest that unantcpated housng gans ought to result n offsettng effects on savng, although n practce these may be small and attenuated by captal market rgdtes, bequest motves and the effects of precautonary savng. Ths paper has presented the frst attempt at a mcro-econometrc nvestgaton of ths ssue usng Brtsh data on owner-occuper households. Our results suggest that the margnal propensty to consume from real housng gans durng the 1990s was between 0.01 and 0.03 for the medan household, that s a 1000 real housng captal gan may have resulted at the medan n an 10 to 30 reducton n actve savng over the

12 11 perod 1993 to Ths result s comparable to Engelhardt s (1996) estmate of an mpc of 0.03 for Unted States owner-occupers over the perod 1984 to Furthermore we fnd that consderably stronger effects of real housng shocks on savng are to be found for households ntally n negatve equty, especally when house prces are rsng. At the medan our mpc estmates are between 0.04 and Fnally our results suggest that households behave dfferently n response to postve and negatve real housng wealth shocks, and pont to a mportant dfference between Brtsh and North Amercan households. Brtsh households reduce savng as real housng wealth ncreases but do not appear to save sgnfcantly more n response to a fall n house values. Agan ths s partcularly so for households wth ntal negatve equty. These results may partally explan the observed macroeconomc correlaton between house prces and consumpton. In Great Brtan n mllon owneroccupers 12 had housng wth an average value of 79,000. The average real gan between 1995 and 1999 was 11%, translatng to an ncrease n real housng wealth of 139 bllon. Wth an estmated mpc of 0.01 ths translates n to an ncrease n consumpton of 1.4 bllon, or a 0.3% growth n the 1995 total. 13 Whle ths result goes some way to explan the observed macroeconomc assocaton between real house prces and personal sector spendng, t s rather smaller than earler macroeconomc estmates for the UK. The reconclaton of the results remans an mportant avenue for further research. 12 Source: Offce for Natonal Statstcs, Housng Statstcs n Great Brtan, September households fnal consumpton expendture was 455 bllon, source Offce for Natonal Statstcs.

13 12 Table 1: Actve Savng by Households % of households wth actve savng or contrbuton towards a Personal Penson Mean non-zero actve savng (1995 s per month) Mean actve savng of all households (1995 s per month) Year All Years Source: computed from BHPS Waves 2 to 9, usng sample household weghts Table 2: Dstrbuton of Real House Prce Shocks across Owner-Occuper Households 1995 prces <0 15.6% 87.7% 2.4% 0 to % 12.3% 34.7% to % % to % % > % % Average change 21,207-2,381 23,179 Medan change 10,700-2,982 14,235 Source: computed from BHPS Waves 3, 5 and 9 and county house prce data from Halfax plc.

14 13 Table 3: Reduced Form Model of Household Fnancal Assets Dependent varable: ndvdual fnancal Coeffcent Standard Error assets Age Age squared Female No. of adults n household No. of chldren 0-4 No. of chldren 5-10 No. of chldren No. of chldren Marred/Cohabtng couple Separated/dvorced Wdowed Ethnc mnorty Self-employed Unemployed FT educaton/nactve Long term sck Retred Degree Spouse wth degree Home mortgaged Home rented (prvate sector) Home rented (publc sector) Regon: North Yorkshre & Humber North West East Mdlands West Mdlands East Angla South East South West Wales Scotland Intercept *** 0.07*** 35.41*** 26.77*** *** 45.94*** 75.96*** 60.00*** 86.88*** 79.64*** *** 70.19* *** *** 75.71** 56.43*** 73.21*** 49.22*** 74.67*** 59.20*** 95.56*** 82.38*** 77.53*** 80.87*** 78.12*** 95.03** 66.42* 77.67** *** 83.23*** *** Year dummy 1994/ / / / / / *** 62.64* 62.77*** 63.04*** 62.88*** 61.56*** Sgma *** N 54,721 Log lkelhood Source: computed from pooled annual Famly Resource Surveys, 1993/1994 to 1999/2000 Estmaton: Tobt wth censorng at 1,500 and 20,000 Note: * denotes sgnfcance at 10%, ** at 5%, *** at 1%

15 14 Table 4: Medan Regresson Savng Estmates (1) (2) (3) (4) (5) (6) Non-movers A (0.126)*** (0.099)*** (0.149)*** (0.178)*** (0.048)*** (0.089)*** ÄA (0.085)*** (0.128)*** (0.166) (0.055)** (0.086)*** ÄH (0.005)** (0.003)*** (0.005)** (0.006)** (0.015) (0.003)** ÄH.D ne (0.015)* (0.025) (0.044) (0.009)*** Selectvty (1231.5)* N Pseudo R Mpc mpc D ne = Source: computed from BHPS Waves 3, 5 and 9 Notes: Standard errors n brackets; * denotes sgnfcance at 10%, ** at 5%, *** at 1%; regressons nclude addtonal demographc and other controls as descrbed n the text. Full results avalable on request. Selectvty correcton n column (4) uses year of start of current resdence spell n 1993, dummy for stated preference for movng n 1993, household sze n 1993, excess rooms (number of man rooms/household sze) n 1993 and dummes for detached, sem-detached and apartment propertes as nstruments. Reported results are from two-step estmaton and reported selectvty coeffcent s for the nverse Mlls rato.

16 15 Fgure 1: Annual Change n Average UK House Prces per cent Average % change n real house prce Average % change n nomnal house prce

17 16 Bblography O. Attanaso and G. Weber (1994), The UK consumpton boom of the late 1980s: aggregate mplcatons of mcroeconomc evdence, Economc Journal, 104: K.B. Bhata (1987), Real estate assets and consumer behavour, Quarterly Journal of Economcs, 102: B. Bosworth, G. Burtless and J. Sabelhaus (1991), The declne n savng: some mcroeconomc evdence, Brookngs Paper on Economc Actvty, 26: A. Carruth and A. Henley (1990a), Can exstng consumpton functons forecast consumer spendng n the late 1980s?, Oxford Bulletn of Economcs and Statstcs, 52: A. Carruth and A. Henley (1990b), The housng market and consumer spendng, Fscal Studes, 11(3): M. Davey (2001), Savng, wealth and consumpton and Mortgage equty wthdrawal and consumpton, Bank of England Quarterly Bulletn, Sprng, R. Dsney, A. Henley and G. Stears (2002), Housng costs, house prce shocks and savngs among older households n Brtan, Regonal Scence and Urban Economcs, 32 (forthcomng). K. Dynan and D. Mak (2001), Does stock market wealth matter for consumpton? mmeo, Washngton: Board of Governors of the Federal Reserve System. G. V. Engelhardt (1996), House prces and home owner savng behavour, Regonal Scence and Urban Economcs, 26: G.V. Engelhardt (2001), Nomnal loss averson, housng equty constrants and household moblty: evdence from the Unted States, mmeo, Centre for Polcy Research, Syracuse Unversty, NY. W. Gale and J. Sabelhaus (1999), Perspectves on the household savng rate, Brookngs Papers on Economc Actvty, 1, D. Genosove, D. and C. Mayer, C. (1997) Equty and tme to sale n the real estate market Amercan Economc Revew 87. June, D. Genesove and C. Mayer (2002), Loss averson and seller behavour: evdence from the housng market, Quarterly Journal of Economcs (forthcomng). A. Henley (1998), Changes n the dstrbuton of housng wealth n Great Brtan, , Economca, 65: D. Mles (1992), Housng markets, consumpton and fnancal lberalsaton n the major economes, European Economc Revew, 36,

18 17 D. Mles (1993a) House prces, personal sector wealth and consumpton: some conceptual and emprcal ssues, Manchester School, 61, June, D. Mles (1997) A household level study of the determnants of ncomes and consumpton, Economc Journal, 107, January, J. Muellbauer and A. Murphy (1990), Is the UK balance of payments sustanable?, Economc Polcy, J. Muellbauer and A. Murphy (1995) Explanng regonal consumpton n the UK mmeo, Nuffeld College, Oxford. J. Peek (1983), Captal gans and personal savng behavor, Journal of Money, Credt and Bankng, 15: J. Poterba (2000), Stock market wealth and consumpton, Journal of Economc Perspectves, 14, Sprng, J. Sknner (1989) Housng wealth and aggregate savng Regonal Scence and Urban Economcs, 19, J. Sknner (1994) Housng and savng n the Unted States n: Y. Noguch and J.M. Poterba (eds) Housng Markets n the Unted States and Japan, Chcago: Chcago Unversty Press, for Natonal Bureau of Economc Research. J. Sknner (1996) Is housng wealth a sdeshow? n D. Wse (ed) Advances n the Economcs of Agng, Chcago: Chcago Unversty Press, for Natonal Bureau of Economc Research. R. Thaler (1990) Anomales: Savng, fungblty and mental accounts, Journal of Economc Perspectves, 4, S. Vent and D. Wse (1990) But they don t want to reduce housng equty, n D. Wse (ed) Issues n the Economcs of Agng, Chcago: Chcago Unversty Press for Natonal Bureau of Economc Research.