74110__791548072

download 74110__791548072

of 35

Transcript of 74110__791548072

  • 8/13/2019 74110__791548072

    1/35

    PLEASE SCROLL DOWN FOR ARTICLE

    This article was downloaded by:

    On: 17 August 2010

    Access details: Access Details: Free Access

    Publisher Routledge

    Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

    Housing StudiesPublication details, including instructions for authors and subscription information:http://www.informaworld.com/smpp/title~content=t713424129

    Understanding Neighbourhood Housing Markets: Regional ContextDisequilibrium Sub-markets and SupplyGLEN BRAMLEYa; CHRIS LEISHMANb; DAVID WATKINSaaSchool of the Built Environment, Heriot-Watt University, Edinburgh, UK bDepartment of UrbanStudies, University of Glasgow, UK

    To cite this ArticleBRAMLEY, GLEN , LEISHMAN, CHRIS and WATKINS, DAVID(2008) 'Understanding NeighbourhoodHousing Markets: Regional Context, Disequilibrium, Sub-markets and Supply', Housing Studies, 23: 2, 179 212

    To link to this Article: DOI: 10.1080/02673030701875113URL: http://dx.doi.org/10.1080/02673030701875113

    Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

    This article may be used for research, teaching and private study purposes. Any substantial or

    systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply ordistribution in any form to anyone is expressly forbidden.

    The publisher does not give any warranty express or implied or make any representation that the contentswill be complete or accurate or up to date. The accuracy of any instructions, formulae and drug dosesshould be independently verified with primary sources. The publisher shall not be liable for any loss,actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directlyor indirectly in connection with or arising out of the use of this material.

    http://www.informaworld.com/smpp/title~content=t713424129http://dx.doi.org/10.1080/02673030701875113http://www.informaworld.com/terms-and-conditions-of-access.pdfhttp://www.informaworld.com/terms-and-conditions-of-access.pdfhttp://dx.doi.org/10.1080/02673030701875113http://www.informaworld.com/smpp/title~content=t713424129
  • 8/13/2019 74110__791548072

    2/35

    Understanding Neighbourhood HousingMarkets: Regional Context, Disequilibrium,Sub-markets and Supply

    GLEN BRAMLEY*, CHRIS LEISHMAN** & DAVID WATKINS**School of the Built Environment, Heriot-Watt University, Edinburgh, UK, **Department of Urban Studies,

    University of Glasgow, UK

    (Received November 2006; revised December 2007)

    ABSTRACT Neighbourhood housing markets play a pivotal role in the evolution of thedemographic, social and economic functioning of neighbourhoods, and hence are necessarily a keyconcern for national and local policy makers. This paper examines propositions about thedeterminants of neighbourhood housing market outcomes in England, particularly price changesover the last 10 20 years. These concern the influence of (sub-)regional economic and demographic

    forces, the nature and persistence of disequilibrium, the existence of sub-markets and the influence of

    supply. Data from a wide range of sources at ward and local levels underpins models of price leveland change set within a multi-level structure. Results are discussed in relation to the role ofneighbourhood in an understanding of the micro-structures of housing markets, as well as in relationto contemporary policies.

    KEY WORDS: Housing market, changing demand, neighbourhoods, sub-markets

    Introduction

    This paper draws on evidence and analysis from a recent project to examine propositionsabout the drivers of neighbourhood housing market outcomes in England, particularly price

    changes over the last 1020 years. It uses a set of linked models at neighbourhood and sub-

    regional levels to examine issues about how housing markets operate at neighbourhood level,

    in particular the importance of (sub-)regional economic and demographic factors, the nature

    and persistence of disequilibrium, the existence of sub-markets and the influence of supply.

    This may be seen as providing a bridge between macro-economic analyses of housing

    Housing Studies,

    Vol. 23, No. 2, 179212, March 2008

    Downl

    oade

    d

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    3/35

    concern for national and local policy makers. The English Housing Market Renewal

    Programme (HMRP) is trying to achieve a sustained improvement in the conditions and

    market performance of neighbourhoods which have been afflicted by low demand and

    failing markets. A viable housing market is one crucial and necessary condition for

    transforming the prospects of disadvantaged areas (ODPM, 2003). The recent increased

    emphasis on owner occupation in British housing policy (ODPM, 2005), including in

    regeneration and relatively low income contexts, further underlines the importance ofthese issues.

    While there is an increasing volume of data about the geographically micro level of

    the housing market, there is as yet a lack of strongly-founded understanding of the

    determinants of market behaviour and outcomes at this level which take adequate account

    of the full range of relevant factors. Certain traditional analytical furrows have been

    well-ploughed, for example, hedonic house price models and the role of sub-markets,

    reviewed briefly in the second section, but these studies have often lacked the full breadth

    of social and environmental impacts and have rarely addressed the issue of change. This

    leads into the identification of the key hypotheses examined in this paper.The general modelling framework is put forward in the third section. Use is made of

    panel model results for housing market areas, but the main focus is on cross-sectional

    models of price level and change at neighbourhood (ward) level set within a two-level

    structure. This enables market area processes to be distinguished from neighbourhood

    processes, while recognising the links between them. Data from a wide range of sources at

    ward and local level, described in the fourth section, underpins these empirical models

    of price level and change. The following section presents results of the analysis and

    modelling, while the final section draws together conclusions.

    Existing Research

    A starting point for understanding housing market variations within urban regions must be

    the classical theory of urban land rent, which goes back to Ricardo but is chiefly associated

    with the work of Alonso (see also Evans, 1973; Muth, 1969; Richardson, 1978). However,

    this theorys assumptions about travel costs and the concentration of economic activity in

    city centres are increasingly called into question by modern trends, particularly road-based

    travel and freight logistics. Nevertheless, urban market models should take account of the

    influence of location and access, including the possibility of change.Subsequent work within this tradition has recognised that real cities have more

    complexity about their structures. In particular, real physical/geographical features may

    induce different social and physical development patterns in particular zones in one

    historical era, which then (through a process of path-dependence) are reflected in the

    subsequent character of these areas. The role of past public housing interventions are

    particularly significant in this respect in the UK These features may be strongly reinforced

    180 G. Bramleyet al.

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    4/35

    (Hirsch, 1977). It is difficult to fully untangle causality here, as for example in the debate

    about the role of good schools in pushing up house prices, when it is the middle class

    background of the pupils which makes these schools good (Cheshire & Sheppard, 2004).

    The effect of differing socio-economic composition is the focus of Meen et al.(2005a)

    study of economic segregation, and currently there is considerable policy interest in the

    issue of whether poverty is becoming more or less concentrated (i.e. segregated) (Berube,

    2005; Green, 1994; Greenet al., 2005; Lupton, 2005). There is strong interest in policies topromote socio-economic mixing, and related concern about the effects of the

    residualisation of social rented housing (Forrest & Murie, 1983; Malpass, 2005). Less

    attention has been paid to segregation in terms of household demography, although

    Bramley & Morgan (2003) suggest that this is also significant.

    Hedonic models is the term most commonly used for statistical modelling of

    variations in house prices or rents between individual properties or neighbourhoods within

    cities. Private housing is a highly heterogeneous or composite good and its demand is seen

    as derived from the demand for its component attributes. Hedonic models, initially

    developed by Rosen (1974) following Lancaster (1966), may be used for the purpose ofestimating these implicit or hedonic prices of individual attributes comprising a composite

    good. There are numerous published hedonic models focused on the identification and

    estimation of positive and negative externality effects such as greenspace or traffic/aircraft

    noise (e.g. review by Nelson, 2004). Such studies have also assessed the effects of

    transport accessibility (e.g. Gibbons & Machin, 2004).

    Hedonic studies are most often undertaken for single cities, and it may be argued that

    their results are non-generalisable, because price premia/discounts vary markedly between

    studies. We certainly expect the prices of housing and neighourhood attributes to vary

    in absolute value depending upon the local/city-region context, but it may still bereasonable to posit a general model which accounts for the variation in city/regional

    context. However, collinearity between amenity, locational and transportation

    accessibility measures in these studies can result in either unstable parameter estimates

    or, when the test conditions are more strictly controlled, non-generalisable results. It is

    also the case that hedonic models are essentially cross-sectional, and thus do not generally

    focus on change over time.

    This leads into the topic of sub-markets. Some argue that, even within a single city-

    region, different parts of the market operate semi-independently, because the buyers in the

    different sectors are different people and they are not willing to switch to other types ofhousing if they cannot get their preferred choice; thus the balance between supply and

    demand may differ persistently between sub-markets. Sub-markets may be defined with

    reference to either or both of contiguous geographical sub-areas or types of dwelling. Tests

    for their existence focus on whether the regression models fitted separately are statistically

    significantly different from a model fitted to all the data together (Jones et al., 2004; Pryce

    & Gibb 2006) Overall there is no consensus that sub-markets really do operate

    Understanding Neighbourhood Housing Markets 181

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    5/35

    aggregates over time. In other words, they are concerned with housings role in economic

    cycles and fluctuations. Although this subject has received serious attention from leading

    economists, until recently there has been little connection between this work and the

    urban/hedonic modelling described above.

    House prices are seen as driven mainly by real incomes, in terms of long-run

    fundamentals, with a cyclical tendency driven by the bubble-building momentum effect

    ultimately curbed by the bubble-bursting correction effect (Muellbauer, 2004).Recognition of the potential of booms (and busts) to overshoot assumes that expectations

    are at least partly adaptive and not wholly rational, consistent with other evidence that

    housing markets are not perfectly efficient (e.g., Meen, 1999). Meen (1998) added to

    the longer-run fundamentals the following terms (in addition to income): ratio of

    dwelling stock to number of households; a measure of personal wealth; the user cost

    of capital (a function of interest and tax rates as well as house price growth); and

    repossessions. Variations on this model appear in Meen & Andrew (2003), Barker

    (2004), and Meenet al.(2005b). General findings of high responsiveness to incomes and

    low responsiveness of supply help to explain the strongly pro-cyclical behaviour ofBritish house prices, and the high level of prices in economically buoyant regions. The

    Barker (2004) review of housing supply has been primarily concerned with this

    phenomenon

    Quite a number of the macro-studies of British housing have applied a similar model

    structure to regional time series (Ashworth & Parker, 1997; Giussani & Hadjimatheou,

    1991; Muellbauer & Murphy, 1997; Munro & Tu, 1996). Much attention has been given to

    the so-called ripple effect, whereby certain regions appear to lead the market while

    others follow with varying lags. This literature clearly shows that factors such as income,

    previous price levels and growth rates, interest and tax rates, the balance of demand andsupply, unemployment and other labour market variables will all play a part in market

    outcomes. Some of these will have a different pattern over time in different regions, while

    some will only be meaningful at national level (e.g. interest and tax rates). The models

    described below incorporate these factors at a sub-regional scale.

    Modelling Neighbourhood Market Change

    Hypotheses

    The central aim of this analysis is to identify and quantify the effects of a range of factors

    on housing market outcomes at neighbourhood level, with a particular emphasis on house

    price changes. This paper investigates in particular the following hypotheses:

    (a) Neighbourhood housing markets are systematically influenced by national and

    regional economic factors, as well as by more specific local conditions;

    (b) Disequilibrium in local and neighbourhood markets may be identified from

    182 G. Bramleyet al.

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    6/35

    (f) Increasing the supply of private housing will lower prices at market area level,

    but may have ambiguous effects at neighbourhood level;

    (g) Increasing the supply of social housing is likely to increase the concentration of

    poverty and reduce prices at neighbourhood level.

    This paper relates to the broader theme of this special issuemicrostructures of housing

    marketsby providing a bridge between individual housing choices, the options and

    constraints facing different consumers, and the broader local and regional market context

    (the macro-structures). Hypothesis (a) is important for establishing the link between the

    macro and micro. One motive for looking in new ways at market structures may be a

    concern that markets may fail, in some sense, as an orderly and predictable mechanism.

    Hypotheses (b) and (c) look at disequilibrium and its persistence as one important way in

    which markets might fail. The notion of sub-markets, in its strong form, is a challenge to

    the notion of a unified market, and this is the focus of hypotheses (d) and (e). Public policy

    impinges on neighbourhoods most obviously through allowing or promoting the building

    of new housing, and this will have effects on the market through supply and demand, but

    also through changing the neighbourhood environment and its social structure. Therefore,

    these effects are the focus of hypotheses (f) and (g). More generally, the choice of housing,

    in terms of type, size, tenure etc, is always packaged jointly with the neighbourhood, and

    the way neighbourhood attributes are valued is crucial (location, location, location, as

    the estate agents would say). How these values may be changing is of particular interest,

    both as a monitor of urban performance but also, given housings investment/wealth

    aspect, a crucial determinant of opportunities for asset accumulation.

    Formal Statement of Price Model

    At the level of the local housing market area (HMA), house price is determined by an

    inverted demand function of the general form:

    Pk PlR; Nk; Vk; Qk; Hk; Ek; Yk; Zk; Wk; dk 1

    where time subscripts are omitted, k subscripts refer to HMA areas, R refers to national

    macro-economic or monetary factors (e.g. interest rate) in the relevant time period, N

    refers to demographic population or household numbers, V is vacancies, Q is (total or new)

    supply, H measures the type and quality of housing units, E is employment level andchange, Y is household income, Z are other socio-economic characteristics of the

    population (including poverty/deprivation, class/occupation/education) and W are general

    environmental attributes (e.g. climate, scenery). Bramley & Leishman (2005) illustrate

    such a model, showing how this equation would be part of a local supply-and-demand

    system alongside other equations for several endogenous elements on the right hand side,

    particularly N (through migration) and Q (new supply) and possibly V (which may be

    Understanding Neighbourhood Housing Markets 183

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    7/35

    and therelative positionof that sub-area in the pecking order of neighbourhoods within

    the local market. An appropriate general function for neighbourhood price would then be:

    Pjk Pk:Pn{Hjk2Hk; Zjk2 Zk; Ujk2 Uk; Gjk2Gk; Wjk2 Wk} 2

    where j subscripts refer to neighbourhoods, G are geographical locational/access factors,

    and U are urban form attributes of neighourhoods (e.g. density and building form).

    Equation (2) may be seen as a particular formulation of the familiar hedonic price modelapplied to neighbourhood-grouped data. Thus, neighbourhood price is determined by

    market area (predicted) price and the differences between neighbourhood values of the

    other variables, particularly dwelling composition factors, and the corresponding market

    area mean value. It should also be noted that certain variables are assumed to operate only

    at market area (e.g. labour market factors, E, Y; quantity supply/demand variables Q),

    while other factors only operate at neighbourhood level (e.g. U and G). For variables

    that are included at both levels, the coefficients in these two models might well be

    different at the two levels. This is the essence of the multi-level perspective we bring to

    this study.When changes over time are considered, the HMA price change function becomes:

    DPk PlDR; DNk; DVk; DQk; DHk; DEk; DYk; DZk; dk 3

    It is assumed that the environmental attributes are unchanging over time; for shorter time

    periods this assumption might also apply to the social attributes (Z) and the housing

    type/quality mix (H). Equation (3) also assumes that the marginal impacts of changes in

    variables such as N and E are the same in a context of change as they are in a context of

    explaining differences between market areas. The corresponding change function atneighbourhood scale is given by:

    DPjk DPk:Pn{DHjk2Hk; DZjk2 Zk; DUjk2 Uk} 4

    It is assumed here that the geographical/access and environmental factors do not change

    over the relevant timescale, so they drop out. Therefore, relative price performance at

    neighbourhood scale is driven by the balance between changes in relative poverty/social

    conditions, affecting the demand side, and changes in the relative supply of housing of

    different types and qualities (including differences in urban form such as density) on the

    supply side.

    Errors and Disequilibrium Indicators

    Observed price data will deviate somewhat from the levels predicted by such models as (1)

    and (2). Such errors may be viewed as arising from several distinct sources: (a) random

    variation in the mix of dwellings traded in terms of unmeasured aspects of quality relative

    184 G. Bramleyet al.

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    8/35

    data. Category (d) deviations we have allowed for in our model through the term d; it is

    assumed that these arise in particular time periods but tend to dissipate over time. Errors of

    type (e) generally reflect the limitations of assuming constant response functions over time

    and space; they may arise particularly at the neighbourhood level, for example, if it turns

    out to be the case that response functions are non-linear or vary depending on conditions at

    the higher market area level. Varying response functions can be explored, particularly in a

    multi-level modelling framework or simply by partitioning the data, although this can addquickly to the complexity of the model and this may have a cost in terms of transparency.

    Formal panel modelling can provide a framework for untangling some of these effects,

    particularly area fixed effects and disequilibrium error correction. However, the data

    available to this study does not fully meet the requirements of formal panel modelling, in

    terms of having a fully consistent set of variables observed for all areas over a consecutive

    series of annual time periods. However, it is possible to go part way towards identifying or

    allowing for these different elements of error. For example, the average value of the

    residual price error for area k over a series of estimates for different time periods may be

    taken as an estimate of the area fixed effect. In addition, as explained in the next section,the study has access to a panel model of sub-regional housing markets in England over the

    period 19832004 (Bramley & Leishman, 2005; Bramley et al., 2007). Relationships

    calibrated within this model at the higher HMA level may be incorporated into the models

    of market outcomes at local and neighbourhood level for particular points in time or for

    changes over defined time periods.

    Therefore, the modelling procedure adopted is something of a hybrid approach, drawing

    on elements of both a panel approach and a multi-level approach. The paper starts by

    estimating log price level equations corresponding to the combination of Equations (1) and

    (2) for five different time periods between 1988 and 2005. The residuals from theseequations are then partitioned into several distinct components. First, the average residual

    across all time periods at HMA level is treated as a market area level fixed effect. Second,

    the difference between the average HMA residual in each time period and this fixed effect

    is treated as a measure of market area level disequilibrium at that time, dkt. Under this

    interpretation, the actual observed price deviates from that which would be predicted from

    the values of the determinant variables assuming the average response to these variables.

    This could arise because of a recent change in determinant conditions (a shock), which the

    market has not fully responded to, or because actors have set prices on the basis of

    expectations which are out of line with actual current conditions. Disequilibrium impliessome discrepancy in the quantities demanded and supplied, at the current price, but it is

    more difficult to observe these precisely. In practice, local and neighbourhood markets

    adapt to such discrepancies through a variety of mechanisms, including delays in sale and

    purchase or moves to other areas. While this study interprets these price deviations as

    indicative of disequilibrium, it is also possible that they partly reflect measurement error

    (e g sample selection effects) They may also be interpreted as indications of the presence

    Understanding Neighbourhood Housing Markets 185

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    9/35

    is derived from the difference between the higher and lower status average residuals. This

    is interpreted as a measure, for each time period, of whether the local market is

    characterised by having more excess demand for higher quality housing (a positive slope),

    or vice versa. It provides a summary measure of neighbourhood level market imbalance

    for each area, assuming that this can be characterised as having one main dimension of

    variation in quality.

    The second stage of the modelling is to estimate equations for price change, expressedas log differences, between time periods. These equations correspond to a combination of

    Equations (3) and (4) above. Broadly speaking, the explanatory variables in these

    equations are in change form. Where this is of particular interest, as in the case of new

    supply variables, these are split into HMA-level average and neighbourhood differential

    components. As noted earlier, some of the explanatory factors are assumed not to change

    over this timescale and drop out of the model. However, some geographical features are

    retained in the model to pick up systematic trends associated with these (e.g. the possibility

    that rural areas might experience a general relative price trend).

    The disequilibrium terms estimated as above for the base period are then incorporatedinto the model. If, as is expected in hypothesis (c), local housing markets adjust

    progressively to achieve a balance, then it would be expected that the dktterm would have

    a negative coefficient, and the closer this is to unity then the more complete the

    adjustment. Thus, over longer time periods this coefficient would expect to be closer

    to 21.0, with smaller values over shorter periods. The ward slope indicator skt is

    multiplied by the difference between the predicted ward price level and the HMA mean

    predicted price (P^jt

    P^kt

    ) to provide a ward level imbalance indicator in log price

    difference units. Again, a negative coefficient is expected on this variable if

    neighbourhood housing markets have a tendency to adjust to imbalances.

    Other Modelling Issues

    A further complication that may arise relates to the possibility that certain variables that

    we wish to utilise in the models are potentially endogenous. For example, there may be an

    interest in the role of Z-type variables, such as the percentage of poor households, in

    influencing house price levels at neighbourhood level. But that share may itself be a joint

    outcome of the same market process generating the price level. Similarly, if there were a

    wish to include the variable V (vacancies) in the neighbourhood level model, as a driver ofhouse prices, its potential endogeneity would have to be recognised. Therefore, the paper

    follows the standard instrumental variables approach and estimates predictive equations

    for changes in these endogenous variables, substituting the predicted values in the price

    change equations. This procedure has the advantage of making it possible to trace indirect

    effects of key policy-influenced variables relating to new housing supply via poverty

    shares or vacancy rates

    186 G. Bramleyet al.

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    10/35

    factors involved and the great variation in these factors between markets and

    neighbourhoods. However, within such a framework it is possible to test for evidence

    of any tendency for the market to actually operate in a more balkanised fashion. Given

    some hypothesised basis for the structuring of sub-markets, it is possible to divide the data

    up in this way and compare (a) the simple descriptive price trends for the different

    subgroups; (b) the relative importance of different drivers for different subgroups; and

    (c) the overall explanation provided by the disaggregated versus the aggregated model.Evidence on such tests is offered by looking at a disaggregation in terms of region, general

    demand pressure (tight versus slack markets), poverty versus affluence, and

    density/housing type.

    Geographical Framework and Data Sources

    Small Area and Higher Level Geographies

    Neighbourhood is defined for this study primarily on the basis of ward (2004 StastisticalWard,n 8000 in England). The main reason for this was to have sufficient size of unit

    to contain viable numbers of market transactions to analyse. A secondary reason was that

    some data sources were more readily available at this level. Some would argue that a

    smaller unit, such as Census Super Output Areas (SOAs) might be preferable, in

    representing the concept of a neighbourhood, with its implications of relative familiarity

    and homogeneity. However, in this instance the pragmatic reasons have been the dominant

    factors in the choice.

    The higher level geography entails invoking the concept of the housing market area

    (HMA). This is the area at which it is assumed the forces of supply and demand workthemselves out and the level at which the key economic and demographic drivers operate.

    Again for reasons of practicality, this paper relies on a compromise set of areas to

    represent the HMA concept: these are former health authority (HA) areas (n 90), which

    comprise single large authorities (metropolitan districts) or groupings of smaller

    authorities (contiguous London boroughs by sector, districts within former counties).

    These larger units were the basis for the development of a migration model known as

    MIGMOD for government (ODPM, 2002), and provide the basis for the annual panel

    dataset constructed by the researchers, which now covers a period of 20 years (1984

    2004) (Bramleyet al., 2007). While as spatial units the HAs are imperfect (some countiesare too large, while some met districts might perhaps be better grouped with neighbours),

    it is thought this is offset by the value of the panel model as framework for handling HMA

    level economic, environmental and demographic influences. Local authorities (LAs) are

    most convenient as a basis for data compilation, with a wide range of measures available,

    and LAs nest within the larger HA units.

    Understanding Neighbourhood Housing Markets 187

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    11/35

    Particularly important are the data on housing market outcomes, specifically market

    pricesand transactions. This paper draws on two key sources here:

    (1) The Nationwide Building Society (NBS), an important national mortgage

    lender, which has maintained detailed transaction records (including many

    property attributes, and location) since 1980 and which regularly publishes

    regional indices.

    (2) The HM Land Registry (HMLR), which has compiled data on all housing

    transactions since 1995, with very limited attribute data attached (four house

    types, new/second-hand, and location).

    The Nationwide (NBS) data has the advantage of covering a much longer period of time

    and providing a richer set of attributes, but the disadvantage of representing a variable

    minority segment of the market (currently rather over 10 per cent of mortgaged

    transactions). The NBS data are primarily used to measure house prices and new building

    in the earlier part of the study period (before 1995). The first time period used, 198889,

    was chosen to represent the peak of the previous housing market cycle in England; thus,changes between 1988 and 2004 or 2005 are of particular interest in representing change

    over a whole market cycle (more or less). The primary focus of analysis is on price levels

    and changes for a typical/average house type, represented by a second-hand semi-detached

    house (HMLR) or a composite type 3 house1 (NBS).

    Another key challenge is to incorporate data on the quantity, location and characteristics

    ofnew housing development. Four main data sources are relevant here.

    (1) The HMLR data provide measures of the number of new homes sold.

    (2) The Continuous Recording System (CORE) maintains a record of new first time

    lettings of social housing units, postcoded since 1998 in England (this source also

    provides additional demand and profile indicators for the social rented sector).

    (3) Comparisons of 1991 and 2001 Censuses on a common geography can identify

    measures of net change in dwellings by tenure.

    (4) Net changes in stock can be measured at ward level up to 2004 or 2005 using new

    LA returns to the Neighbourhood Statistics data system.

    This paper reports models using a combination of these sources. It can be seen that the data

    available provide a fuller picture for the period since the late 1990s than for the earlierperiod.

    The most important source for measures of demographic and socio-economic structureof the population and housing profiles in terms of type and tenure are the Censuses of 1991

    and 2001. Considerable use is made of measures of change between these two dates. Other

    data sources used to provide indicators at small area level include the following:

    . Data from an ODPM survey of town centres, giving distance to nearest centre

    above successive size thresholds.

    188 G. Bramleyet al.

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    12/35

    . Housing vacancy and second home rates by ward (2004, 2005), from a special

    return generated for Neighbourhood Statistics (NeSS), which may be compared

    with earlier Census values.

    . Measures of geographical access to jobs based on 1991 and 2001 Census

    workplace and travel to work data, developed by Coombes & Raybould (2005).

    Additional data available only at LA level included: public sector housing relets,

    vacancies, waiting list and homelessness numbers; job growth and unemployment rates;

    proxy-based income estimate; household growth and migration. Where used these are

    aggregated up to HMA level.

    Composite Measures Derived from Panel Model

    The panel model reported here is essentially that reported in Bramleyet al.(2007, Ch. 3), a

    development of work by Bramley & Leishman (2005) and Bramley (2002), which itself

    built on a substantial database originally created in an ODPM (2002) project to build amigration model for England. This has been enhanced here in a number of respects,

    particularly by including more factors on the land supply side of the model. The time

    duration of the panel has also been extended from what was previously only about 10 years

    to more than 20 years, from 1983 to 2004, thereby covering more than one whole cycle and

    bringing the story up to date.

    The independent explanatory variables used in the model comprise a small number of

    factors which are essentially national time series (interest rates, GDP growth), and a larger

    number of variables which vary over both time and space (income, unemployment, job

    growth, job workplace: resident workforce ratios, total population, social rented housingstock and new supply, social sector relets, new planning permissions, stock of outstanding

    planning permissions). In addition, another set of variables comprise measures of local

    characteristics which do not change rapidly and which are measured at one point in time

    (generally 1991), i.e. cross-sectional variables. These include measures of air pollution,

    climate, commuting, crime, poor housing quality, ethnicity, urbanisation, occupational

    mix, household types, vacant and derelict land.

    In order to reflect the HMA-level market influences within the ward level models

    discussed below, some composite indicators are constructed to capture the effects of groups

    of exogenous variables. These are based on a slightly simplified reduced form version of thehouse price equation. The groups of factors used are as follows: environmental; economic

    (including labour market); social housing supply/turnover; new supply; and a residual factor

    capturing regional market dynamics (e.g. the ripple effect) (see Appendix A for definitions

    and weightings, and Figure 1 for a schematic representation). Different versions of these are

    constructed to reflect average relative conditions for the whole period and changes over sub-

    periods of five years (so that changes for the relevant periods can be included in ward price

    Understanding Neighbourhood Housing Markets 189

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    13/35

    time-varying or not down the page. Figure 1 also indicates how the composites indicators

    just described fit into the overall scheme. Non-time varying factors play an important role

    in the models for price level. However, in the models of price change, most of the variables

    used are time-varying and expressed in change form.

    Measures used in Disaggregation

    For the purposes of dividing the data according to various hypothetical sub-markets, andalso for testing various possible endogenous or interactive effects, various composite

    indictors were used. The first of these (NewLDInd) refers to the general demand status of

    the area, and in particular the presence of symptoms of low demand, and was developed

    in related research (Bramleyet al., 2007, Ch. 2). This indicator reflects earlier research on

    low demand housing (Bramleyet al., 2000) and utilises a factor analysis approach to boil

    down 39 individual demand measures (at ward or LA level) to a set of seven dimensions

    Figure 1. Schematic map of variables included in models at different levels by type of factor and

    whether time-varying

    190 G. Bramleyet al.

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    14/35

    To measure poverty in a consistent way in both 1991 and 2001, a proxy formula using

    nine 2001 Census indicators was established by regressing these on the official Low

    Income score index from the Governments 2004 Indices of Multiple Deprivation (IMD),

    which actually refers to 2002 data. Equivalent variables from the 1991 Census were then

    used to compute an equivalent score for that year, with poverty change being the

    percentage point difference between these two. The proxy indicators are: no car; not owner

    occupier; limiting long-term illness; unemployment; children; non-employment; one-person households; social renting; overcrowding.

    To measure urban form and housing type a simple breakdown in terms of density and house

    type was used. A more suburban type of ward was identified on the basis of lower density

    (under 25 dwellings per hectare gross, average log of plotsize in m2 . 6.0), more detached

    homes (.30 per cent), less terraces (,30 per cent) and flats (,10 per cent). This was

    contrasted with wards with more flats (.25 per cent), and the remainder group in the middle.

    For each of these groupings, there is a roughly 25:50:25 distribution of

    wards/populations.

    The regional breakdown used distinguishes four groupings: London, the rest of theSouth, the Midlands and the three Northern regions

    Descriptive and Model Findings

    Descriptive Pattern of Price Changes

    First, there is a description of house price changes, broken down by regions and sub-

    categories of neighbourhood for different time periods. Apart from establishing the point

    that there are differences in price trends between these areas, this evidence is chieflyrelevant to hypothesis (d) concerning sub-markets. Table 1 presents price changes in log

    differences for two different but overlapping time periods, broken down by four regions

    and a three-way breakdown by three different neighbourhood categorisations. There is

    some interesting detail in this Table, for example, that house price growth over the longer

    period was higher in poorer and more urban areas, but also in higher demand areas;

    whereas in the more recent (second half) of that period the increases were slightly greater

    in more affluent and suburban areas (there are further differences in the most recent period,

    not shown here). Differences between regions (within, say, middle categories) are of a

    similar order of magnitude.Table 2 summarises the average extent of the differences in these descriptive measures

    of price growth, in percentage point terms. Over the longer period 1, the differences in

    trend were most marked between urban (flats) and suburban type areas, followed by

    high/low demand areas, but in each case the differences exceed 10 per cent points.

    In period 2 the figures were rather smaller, as expected, but again the suburban/urban

    physical category difference was greatest This pattern repeats again in the most recent

    Understanding Neighbourhood Housing Markets 191

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    15/35

    Table 1. Mean price change by period, neighbourhood category and broad region (log differences)

    Reg Social Period 1 Period 2 Phys Period 1 Period 2 Market Period 1 Period 2Categ 88 04 96 04 Categ 88 04 96 04 Categ 88 04 96 04

    Nth Affluent 1.002 0.909 Suburb 0.920 0.901 Hi Dem 1.274 0.850Middle 1.055 0.804 Middle 1.086 0.729 Middle 1.096 0.787Poor 1.133 0.681 Urban 1.339 0.882 LowDem 1.045 0.711Total 1.087 0.757 Total 1.087 0.757 Total 1.087 0.757

    Mid Affluent 0.948 0.952 Suburb 0.894 0.979 Hi Dem 1.193 0.921Middle 1.003 0.897 Middle 1.031 0.846 Middle 1.011 0.883Poor 1.069 0.807 Urban 1.262 0.927 LowDem 0.957 0.850Total 1.016 0.877 Total 1.016 0.877 Total 1.016 0.877

    Sth Affluent 0.911 0.967 Suburb 0.864 1.020 Hi Dem 1.102 0.987Middle 0.954 0.989 Middle 0.943 0.955 Middle 0.933 0.976Poor 1.026 0.962 Urban 1.131 1.006 LowDem 0.887 0.977Total 0.955 0.978 Total 0.955 0.978 Total 0.955 0.978

    Lond Affluent 1.028 0.939 Suburb 0.649 0.969 Hi Dem 1.647 1.000Middle 1.354 0.986 Middle 1.042 0.975 Middle 1.139 0.986Poor 1.497 1.004 Urban 1.531 0.998 LowDem 1.224 0.951Total 1.406 0.992 Total 1.406 0.992 Total 1.406 0.992

    Eng Affluent 0.944 0.951 Suburb 0.881 0.987 Hi Dem 1.348 0.966Middle 1.040 0.928 Middle 1.022 0.847 Middle 1.017 0.911Poor 1.180 0.825 Urban 1.380 0.986 LowDem 0.983 0.821Total 1.072 0.896 Total 1.072 0.896 Total 1.072 0.896

    192

    G.Bramleyetal.

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    16/35

    importance in the market may vary between localities and over time, although it might

    also reflect changing images and preferences for different urban locations, and the effects

    of regeneration investment. Differences between high and low demand areas are not

    unexpected, given that lower demand areas show a range of evidence of slack demand,

    unpopularity or excess supply towards the end of the periods 1 and 2. Differences on the

    poverty-affluence dimension would be consistent with accounts which emphasis socio-

    economic segmentation, including the segregation perspective mentioned earlier.

    Price Level Models

    Price level models are not reported in detail as the main interest is in price changes;

    however, price level models play a role in identifying base period residuals from which

    various measures of fixed area and disequilibrium effects are calculated. Cross-sectional

    regression models are fitted to data on ward price levels for the whole of England for five

    time periods: 198889; 199597; 200002; 200304; 200506. The dependent variable

    is in log form so the coefficients measure proportional impacts on prices. In the first period

    the price data are from NBS and given the smaller sample numbers only rather more than

    half the wards are included in the model ( n 4955). Less explanatory variables are

    available for this period and the fit is less good with only 73 per cent of the variance

    explained. For the remaining periods, HMLR data are used and most wards are included,

    with higher proportions of variance explained (between 85 per cent and 93 per cent).

    Explanatory variables used relate to the nearest available time period; for example, for

    198889 quite a lot of 1991 Census variables are used, for the last three periods 2001

    Census variables are used, while for the 199597 period a mixture of values from the two

    Censuses are employed. The model for 200002, for example, contains 46 variables of

    which 10 are derived from the HMA level panel, together with 4 regional dummies, 6

    urban form indicators, 3 housing type mix variables, 4 market variables, 5 social

    indicators, 7 location/access measures and 6 development variables.

    Table 2. Average difference in price trend between categories within regions by period andcategory type (percentage point changes)

    Period 1 Period 2 Period 3Category 88 04 96 04 01 05

    Affluent-poor 11.0 6.3 3.8Type/density 27.4 8.9 4.0

    High/low demand 17.2 3.5 3.3

    Understanding Neighbourhood Housing Markets 193

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    17/35

    The HMA level fixed effects are by definition reflective of unmeasured characteristics

    of areas that appear to have some ongoing effect on their housing market price levels.

    Looking at the identity of authorities with particularly high or low fixed effects may help

    provide some clues as to the types of factors involved. High values are found with central

    London boroughs, some non-central city metropolitan districts, and some counties (e.g.

    Somerset, Wiltshire). Low values are associated with some northern metropolitan central

    cities (Sheffield, Manchester, Newcastle), former mining areas (Doncaster, Nottingham-shire, Derbyshire), parts of London (Brent/Harrow, Lambeth/Southwark/Lewisham), and,

    interestingly, some high demand southern counties (Cambridgeshire, Surrey).

    The HMA-level disequilibrium terms (dkt) derived from this analysis do change from

    year to year. Areas which appear over-priced in 200506 include several central/inner

    London boroughs (Southwark, Camden, Wandsworth, Kensington), the city of Liverpool

    (European City of Culture hype, perhaps), some other metropolitan districts (Gateshead,

    Bury, Tameside, Kirklees), and some previously low priced counties (Derbyshire,

    Staffordshire). The areas which appear under-priced include some other London

    boroughs to the east of the city centre, some metropolitan cities (Newcastle, Sunderland,Leeds, Birmingham, Coventry), non-core city LAs (Dudley), and some low demand

    declining industrial areas (Cleveland). If one were looking for an immediate practical

    application of this model, then these would appear to be areas to invest in!

    The third measure derived from analysis of price level residuals is what is termed here

    as the slope indicator (skt), which expresses the extent to which prices vary more or less

    than predicted across the quality spectrum within each authority. Again, this can vary

    between time periods. The areas with a high positive slope (prices varying more than

    expected between wards) may be variously characterised as (a) experiencing highly

    uneven demand internally, and/or (b) lacking sufficient choice/diversity on the supply sideto match the range of demand, and/or (c) experiencing an incomplete process of

    adjustment favouring the better quality wards. Areas in this category include some London

    boroughs (Haringey/Enfield, Hammersmith/Ealing/Hounslow, Barking/Havering), some

    provincial cities (Sheffield, Leeds, Bradford, Coventry), some other metropolitan districts

    (Sefton, Calderdale, Dudley, Stockport), and Nottinghamshire This indicator is generally

    positively associated with higher average poverty levels, a higher range of poverty

    variation, and more composite low demand.

    Authorities with low negative slopes (prices varying less than expected) may be

    variously characterised as (a) experiencing a generous supply of up-market neighbour-hoods, relative to high income demand, and/or (b) experiencing a shortage of supply of

    cheaper neighbourhoods, alias an affordability problem, and/or (c) experiencing an

    incomplete process of adjustment favouring the lower quality wards. It is particularly

    interesting that some rural/coastal locations recently highlighted for having exceptional

    affordability problems for locals, partly due to external demand, feature on this list

    (Cornwall Dorset) Several London boroughs appear on this list together with a number

    194 G. Bramleyet al.

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    18/35

    higher and lower status wards within HMAs (skt) accounts for a similar share of the

    residual variation, without the same tendency to fall over these time periods.

    This part of the analysis confirms hypothesis (b), that it is possible to identify elements

    of apparent disequilibrium from price level models, both for HMAs as a whole and in

    terms of the general price structure within them. These account for a significant share ofthe overall residual variation.

    Price Change Models

    There is now a turn to the central focus of this paper, models which attempt to explain

    variations in house price change over time at local and neighbourhood level. These

    constitute the second stage of the modelling as explained in the third section. Before

    reviewing findings for the main focus of interest, price change, it is worth noting that

    drivers of change have also been modelled in two key endogenous variables previouslyidentified, neighbourhood poverty and housing vacancy levels. These may act as key

    transmission mechanisms (or intervening variables) linking exogenous economic or

    demographic drivers, or policy-influenced new development patterns, to market outcomes.

    For example, changes in poverty at neighbourhood level are significantly associated with

    changes in social versus private housing provision, as well as employment conditions,

    migration/ethnic mix, urban/rural location, region and other factors. The same is true to a

    degree for changes in vacancies, although these are more difficult to model. There has

    been much recent concern about speculative investment in private renting in certain areas,

    particularly city centres but also some low demand areas (House of Commons, 2006). Thisis alleged to have led to more stock being held vacant, including some newly built stock;

    some of the findings are consistent with this story.

    The price change models fitted correspond to a combination of equations (3) and (4)

    above. Table 4 compares models for two time periods in terms of overall level of

    explanation, while Tables 56 report results in terms of individual variable effects in those

    two periods The first of these periods covers the 15-year period from 1988/9 to 2003/4

    Table 3. Components of residual variation in price levels at different dates (standard deviations ofresidual components in log of house price units)

    HMA HMA diseq Ward slope Ward OverallYear fixed delta sigma other Residual

    1988 89 0.045 0.054 0.032 0.171 0.2101995 97 0.045 0.040 0.038 0.120 0.170

    2000 02 0.045 0.020 0.027 0.126 0.1612003 04 0.045 0.024 0.027 0.120 0.1562005 06 0.045 0.026 0.030 0.125 0.163

    Understanding Neighbourhood Housing Markets 195

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    19/35

    time period or where those tested were insignificant. The dependent variables are log

    differences, and therefore coefficients may be compared across models.

    The overall fit of these models is as expected less than that achieved with models forprice level, (between 43 per cent and 52 per cent of the variance can be explained,

    compared with 73 per cent93 per cent). Some of the explanatory power associated with

    cross-sectional variables is lost, while the level of data noise is more substantial relative to

    the magnitude of changes.

    Table 4 shows a comparison, in the first block, between a model using only HMA-level

    variables, a model using only ward-level variables, and a model using both. This shows

    that, in the first (longer) period the ward level variables explain rather more than the HMA

    level factors, but this is partly because there are more ward variables and these necessarily

    capture between-HMA effects as well as within-HMA effects. In the second period, thetwo groups of variables provide a similar level of explanation. However, in both periods,

    combining both sets of variables leads to a significant enhancement in the explanatory

    power of the model. This provides strong evidence in support of hypothesis (a), that

    neighbourhood housing markets are systematically influenced by national and regional

    economic factors, as well as by more local conditions. However, the corollary, which is

    also supported is that neighbourhood market changes are not simply a scale model of

    Table 4. Summary of levels of explanation of house price change provided by level of variables andtype of disaggregation (r-squared statistics, log difference in house price, wards in England)

    Model & area grouping Period 1 Period 2all areas 1988 2004 1996 2004

    HMA level vars only 0.292 0.313Ward level vars only 0.348 0.330

    Full model both levels 0.438 0.425Demand levelHigh demand 0.579 0.232Middle demand 0.369 0.428Low demand 0.346 0.559Combined disagg 0.515 0.454Physical formLow density/detached 0.126 0.247Middle form/type 0.207 0.532More flats 0.496 0.173Combined disagg 0.484 0.458

    Socio-economic levelAffluent 0.158 0.209Middle income 0.404 0.417Poor 0.486 0.482Combined disagg 0.462 0.470

    196 G. Bramleyet al.

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    20/35

    Table 5. Summary comparison of price change model for all areas versus disaggregation by income level, 1988 2004 (coefficients and significance levels,ward level England, log difference in house price)

    All Wards Affluent Mid-Income Poor

    Var Name Variable Description Coeff Sig Coeff Sig Coeff Sig Coeff Sig

    (Constant) 0.764 0.929 0.731 0.485delta88 HMA disequilm d1988 21.071 **** 20.907 **** 20.936 **** 21.332 ****reldem88 Internal imbalance s 22.162 **** 22.490 *** 22.587 **** 21.782 ****lhpsupch HMA D supply (-ve) 0.115 20.221 0.246 0.169lhpecoch HMA D income & emp 0.681 **** 0.369 *** 0.708 **** 0.741 ****lhpsohch HMA D social hsg (-ve) 0.318 **** 0.037 0.452 **** 0.286 ****lhpresch HMA D residual price 0.363 **** 0.311 **** 0.399 **** 0.271 ***pdemol14 Demolitions % 91 04 0.001 * 0.003 ** 0.001 20.001pchdet Ddetached dwg % pt 0.002 ** 20.002 0.002 * 0.004 *pchter Dterraced dwg % pt 0.004 *** 20.003 0.004 ** 0.008 **

    pchflat Dflat dwg % pt 0.007 **** 0.002 0.011 **** 0.007 **pchsec14 D2nd home 9104% pt 20.006 0.024 20.011 20.018prpchvac Pred D Vacancies 9104 20.003 20.002 0.000 20.003lispov91 Income Poor hhd % 1991 20.001 20.002 0.003 0.024prpchpov2 Pred D Poverty 91 01 20.005 20.052 0.007 0.067pchsocr Dsocial rent % pt 20.004 0.010 20.008 * 20.013pchemp Demployment rate % pt 20.002 20.009 0.000 0.011pchprfmg Dprof & mgt occs % pt 20.001 0.002 0.001 20.015 ****pchnwh D non-white ethnic % pt 20.005 ** 0.003 20.007 ** 20.011 **ldistnrc Log dist nearest cent 20.033 **** 20.031 *** 20.024 *** 20.046 ***ldjobacl Dlog accessible jobs 0.402 *** 0.007 0.523 *** 0.841 **

    geogbar IMD access (rural proxy) 20.064 **** 20.027 20.070 **** 20.046 *London London regional dummy 0.105 **** 0.062 0.093 ** 0.081north Northern regional dummy 0.081 **** 0.101 *** 0.081 *** 0.079 **mids Midland regional dummy 0.050 *** 0.079 ** 0.043 0.042swee Sth West & East dummy 0.011 0.033 20.004 20.005pnewd9304nl New priv build HMA % hhd 20.439 **** 20.516 **** 20.389 **** 20.443 ****pnewd9304nd New priv build ward diff 20.009 20.050 20.010 0.099 **pscmp24l New social HMA % hhd 0.087 0.278 20.037 0.347 **pscmp24d New social ward diff 20.012 0.065 20.019 20.076 *

    UnderstandingNeighbourhoodH

    ousingMarkets

    197

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    21/35

    Table 5. Continued

    All Wards Affluent Mid-Income Poor

    Var Name Variable Description Coeff Sig Coeff Sig Coeff Sig Coeff Sig

    Adjusted r-squared 0.438 0.158 0.404 0.486Std Error estimate 0.252 0.220 0.247 0.272F ratio 123.7 7.2 53.7 44.8Number of cases 4561 961 2255 1343

    198

    G.Bramleyetal.

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    22/35

    Table 6. Summary comparison of price change model for all areas versus disaggregation by income level, 1996 2004 (coefficients and significance levels,ward level England, log difference in house price)

    All Wards Affluent Mid-Income Poor

    Var Name Variable Description Coeff Sig Coeff Sig Coeff Sig Coeff Sig

    (Constant) 0.881 0.903 0.751 1.000delta96 HMA disequilm d1996 20.739 **** 20.536 **** 20.670 **** 20.767 ****reldem96 Internal imbalance s 21.384 **** 20.069 21.564 **** 20.947 ***lhpsup24 HMA D supply (-ve) 20.202 *** 0.158 20.066 20.408 **lhpeco24 HMA D income & emp 0.194 **** 0.205 **** 0.270 **** 0.094lhpsoh24 HMA D social hsg (-ve) 20.028 0.129 * 0.149 *** 20.156 **lhpres24 HMA D residual price 0.071 **** 0.217 **** 0.202 **** 20.058pchter Dterraced dwg % pt 20.001 ** 20.003 **** 0.000 20.003 **pchflat Dflat dwg % pt 0.000 0.000 0.001 * 20.002pchsec14 D2nd home 91 04% pt 0.009 *** 0.010 ** 0.009 ** 0.010

    prpchvac Pred D Vacancies 9104 20.007 **** 20.003 20.006 **** 20.002lispov91 Income Poor hhd % 1991 20.021 **** 20.026 **** 20.014 ** 20.026 **prpchpov2 Pred D Poverty 9101 20.055 **** 20.075 **** 20.042 ** 20.070 **pchsocr Dsocial rent % pt 0.009 **** 0.014 **** 0.007 *** 0.012 **pchemp Demployment rate % pt 20.008 **** 20.010 **** 20.006 *** 20.010 *pchprfmg Dprof & mgt occs % pt 20.002 *** 0.001 20.002 ** 20.005 *pchnwh Dnon-white ethnic % pt 0.006 **** 0.002 0.005 *** 0.008 ***ldistnrc Log dist nearest cent 0.010 **** 20.001 0.012 **** 0.008c1503 km Near major city centre 0.030 **** 0.021 0.032 **** 0.030 *ldjobacl D log accessible jobs 20.140 * 20.457 **** 20.093 20.007geogbar IMD access (rural proxy) 0.021 **** 0.034 **** 0.018 *** 0.020

    london London regional dummy 0.025 0.066 *** 20.023 0.039north Northern regional dummy 20.103 **** 20.020 20.090 **** 20.183 ****mids Midland regional dummy 0.017 * 0.046 *** 0.012 20.025swee Sth West & East dummy 0.057 **** 0.061 **** 0.049 **** 0.047 **pdemol14 Demolitions % 91 04 0.000 0.000 0.000 0.000pnewd9304nl New priv build HMA % hhd 20.100 **** 0.016 20.036 20.178 ****pnewd9304nd New priv build ward diff 20.037 **** 20.056 **** 20.031 ** 20.021pscmp24 l New social HMA % hhd 0.090 ** 20.217 *** 20.191 *** 0.332 ***pscmp24d New social ward diff 0.057 **** 0.032 * 0.045 *** 0.052 **

    Under

    standingNeighbourhoodH

    ousingMarkets

    199

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    23/35

    Table 6. Continued

    All Wards Affluent Mid-Income Poor

    Var Name Variable Description Coeff Sig Coeff Sig Coeff Sig Coeff Sig

    Adjusted r-squared 0.425 0.209 0.417 0.482Std Error estimate 0.136 0.090 0.117 0.187F ratio 200.4 19.8 97.3 60.3Number of cases 7824 2071 3901 1850

    200

    G.Bramleyetal.

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    24/35

    The resulting models achieve r-squared values of 0.376 and 0.357 in the two periods,

    markedly below those for the preferred models reported here. While some of the

    coefficients are similar, these simpler models provide less insight into the impact of new

    supply, poverty/social change and market imbalance.

    Evidence on sub-markets

    The rest of Table 4, as well as Tables 56, are designed in part to test hypotheses (d) and

    (e) concerning sub-markets. A single model applied to all wards is compared with models

    where the responsiveness of included variables (i.e. their coefficients) are allowed to vary

    between different categories of ward. As in the earlier descriptive analysis, three distinct

    ways of breaking down wards are compared: by level of demand, by physical form/house

    type, and by relative affluence/poverty. There is no further breakdown by region, although

    it should be pointed out that the models contain regional dummies. Table 4 shows the level

    of explanation achieved for each subgroup, and the overall explanation from combining

    the predictions of the disaggregated models.The results in Table 4 provide some measure of further support for the sub-markets

    thesis. For each disaggregation, at least one of the subgroups achieves a higher level of fit

    (r-squared) than the overall model. More important, in every case the recombined

    predictions of the disaggregated models somewhat outperform the single overall model.

    For example, in period 1, whereas the overall model explains 44 per cent of the variance,

    disaggregated models can achieve levels of 46 per cent (affluence/poverty), 48 per cent

    (physical form) or 52 per cent (demand level). In period 2 the improvement is rather more

    modest, the best being 47 per cent for affluence/poverty-based disaggregation versus 43

    per cent for the single overall model.It is suggested that although this evidence provides support for the sub-markets view, it

    is not overwhelming evidence for that, partly because the increased model fit performance

    is modest, partly because only some of the subgroups perform markedly better, and partly

    because it is not clear which disaggregation is preferred. Whereas on the descriptive

    evidence in Tables 12 above one might have plumped for the physical form/type-based

    disaggregation, on the basis of Table 4 one might prefer the demand level-based

    disaggregation, if looking at period 1, or the affluence/poverty one if looking at period 2.

    Looking at explanation levels within the subgroups, also, there is no consistency between

    the time periods. For example, the explanation is better for high demand areas in period 1but low demand areas in period 2; better for flatted areas in period 1 but middling areas in

    period 2. At least under the affluence/poverty disaggregation there is consistency between

    the two periods, with poorer areas being best explained and affluent areas least well

    explained. The relative lack of stability in terms of the performance of the subgroups

    suggest that sub-markets, insofar as they exist, are not permanent phenomena but rather

    more transient or contingent This brings the concept closer to the notion of disequilibrium

    Understanding Neighbourhood Housing Markets 201

    Downl

    oa

    ded

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    25/35

    more similar than different. Relatively few variables (which are statistically significant)

    change sign or even change dramatically in size of effect.

    In period 1, the following differences are worthy of comment:

    . Housing type mix changes affect prices in poorer but not more affluent areas;

    . Changes in social housing have significant if offsetting effects, mainly in poorer

    areas at HMA level;

    . The negative association with non-white ethnic population is mainly seen inpoorer areas;

    . The ward-based job accessibility change measure mainly affects poorer areas;

    . New private building is positive at ward level in poorer areas.

    Broadly speaking, these differences are explicable and in accordance with expectations.

    The differences observable in period 2 are not all the same ones.

    . HMA level factors often seem to work differently on poorer areas from their more

    general effect, for example in the case of supply and income/employment;

    . The internal imbalance correction effect does not seem to work in affluent wards;

    . Building new private housing has a negative HMA-level effect on poorer areas

    (they are perhaps more vulnerable to oversupply), whereas the neighbourhood

    effect is distinctly negative in affluent areas (as feared by NIMBYs!);

    . New social housing provision seems to have a positive HMA-level impact on

    poorer areas but a negative effect elsewhere.

    At least some of the differences observable can be rationalised, and hence could in

    principle be included within a more sophisticated model, for example, by including

    interaction terms or thresholded variables. This implies that, rather than thinking in terms

    of sub-markets, there should be a look at making the models less crude, with more

    allowance for the effects of context. Overall, the conclusions from this section are that the

    evidence provides only moderate and qualified support for hypothesis (d), on the existence

    of sub-markets. At the same time, the findings provide considerable support for hypothesis

    (e), that most price variation can be accounted for through a common model framework,

    although this finding also has to be qualified by the observation that the overall explanation

    of price changes is less than that achievable with price levels.

    It was suggested earlier that the concepts of disequilibrium and sub-markets might

    overlap. Therefore, it is of interest to comment on the performance of the disequilibrium

    indicators in these change models. The clear finding is that these indicators, based on the

    levels model residuals, work well and as expected (negative coefficients) in all time

    periods and for nearly all subgroups. Local markets appear to be capable of adjusting

    over time. The coefficients around 21.0 for delta88 in Table 5 imply that HMA-level

    disequilibrium in 1988/9 was fully corrected by 2003/04; c.70 per cent of the

    disequilibrium apparent in c 1996 was eliminated by 2003/04 (Table 6); 65 per cent of

    202 G. Bramleyet al.

    Downl

    oa

    ded

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    26/35

    This finding concurs with the previous finding that such sub-market structures that may

    exist do not appear to be that stable.

    Specific Influences on Price Change

    It has already been shown that adding HMA level economic and other variables to themodel greatly enhances its explanatory performance. It is also clear from Tables 5 and 6

    that economic and labour market conditions at the wider level provide a strong explanation

    for some of the price changes, particularly using the relevant composite change indicator

    from the panel model (lhpecoch/lhpeco24), also reinforced in the longer period by the job

    access change indicator (ldjobacl). This provides further support for hypothesis (a).

    Looking at house types, we can see that increased presence of flats was a positive

    factor, over the longer periods, but declining in period 2 and apparently negative in

    the most recent period.4 Drawing on established urban economic theory there is an

    expectation of both a greater prevalence of flats and a higher land rent/price level in higherdemand markets and more central/accessible places, and the positive relationship probably

    reflects this. If this variable is split into an HMA part and a ward part, the ward effect is

    generally weaker or negative; this may be picking up the more negative neighbourhood

    environmental effect of concentrations of flats (town cramming), while the recent

    negative turn in this coefficient may also reflect oversupply of this type of housing in

    recent years, partly as a result of government planning policies.

    Both of the variables treated as endogenous, changes in poverty and vacancies, work

    generally in the expected negative direction, although these effects are clearer and more

    significant in period 2 (Table 6) than in period 1. Prices rose less where poverty wasinitially higher, and less where poverty increased rather than reduced. There was a general

    fall in poverty, according to the measure here, over the period 19912001, and this was

    particularly great in many previously poorer areas. Therefore, this factor accounts

    for some of the recent relative and absolute rise in prices in poorer neighbourhoods.

    A significant negative association with vacancy change is indicated (as expected) for

    period 2 in Table 6, overall and in middle-income areas. Detailed results for the most

    recent period (200105) are not reported, but it is worth mentioning that in this period

    some of these relationships with vacancies and poverty appear to change direction.5

    Although non-white populations were negative in the levels models, and in some period1 change models, they appear to work positively on period 2 price change (Table 6).

    The effects are as noted more pronounced in poorer areas, which tend to have more ethnic

    concentrations. The recent positive shift may reflect the general market buying power of

    these populations, and also the pressure of international in-migration which has increased

    in recent years. Although not included in the models reported, LA-level international

    migration indicators do have an impact on recent period price change models When

    Understanding Neighbourhood Housing Markets 203

    Downl

    oa

    ded

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    27/35

    to speculative oversupply or over-investment). Rural wards saw lower price growth over

    the longer period but positive growth in period 2, lessening recently.

    The Impact of New Supply

    Before reviewing the empirical findings it is worth clarifying what the impact of new

    housing investment would be expected to be on local housing markets, on the basis of

    economic theory and the literature reviewed in the second section. There are several

    qualitatively different impacts, and these may operate differently at different geographical

    levels, over different time periods, and in differing general market conditions.

    . Supply and demand effects. If there is more supply in a market area in a given time

    period then, other things being equal, this may be expected to have a negative

    impact on prices. This supply and demand effect is more likely to operate at HMA

    level and less likely to operate at a small neighbourhood scale; and more likely to

    apply with private sector supply than with social renting, which has non-market

    pricing and rationing.. Environmental quality effects. New building should upgrade the average quality

    of an area, especially where the existing housing is of relatively poor quality; this

    effect is less likely in an area of existing high quality, such as many mature,

    affluent suburbs. The environmental effects may be expected to impact chiefly at

    neighbourhood level, and over the longer term, with the possibility of disruptive

    effects in the short term.

    . Social effects. New development may affect the social profile of an area, and

    indirectly influence factors such as crime or school quality, particularly at

    neighbourhood level. While new social housing may represent physicalupgrading, it may still be perceived negatively in terms of social mix. New

    owner-occupier housing is more likely to be perceived positively and this may be

    particularly important in areas previously dominated by rented housing.

    . Confidence effects. New building may be taken as a signal that an area has a

    positive future, because people are demonstrably investing in it and moving into

    it, and that signal may encourage others, especially in areas which have

    previously been experiencing decline.

    When data from actual markets is observed, what is seen is the combined effect of all ofthese processes and perceptions. The question is how they balance out. In some cases

    positive and negative effects may cancel out; in other cases, the positives or the negatives

    may reinforce each other and act cumulatively.

    The new supply variables may be considered alongside the tenure change variables and

    relevant panel model composites. New private building has a substantial negative impact

    at HMA level (Tables 56); this is the expected supply-demand effect There is generally

    204 G. Bramleyet al.

    Downl

    oa

    ded

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    28/35

    be noted that the size of this effect is a lot smaller than the effect of increasing poverty

    (which would be the likely corollary of increased social renting). In other words, the

    indirect effects of social renting via poverty may be as important as the direct effects.

    Finally, the panel composite variable lhpsohch/lhpsoh24 indicates that the size and

    turnover of the social rented sector also has a mainly consistent impact (the positive signs

    here effectively indicate a negative relationship; see note 6). The only exception is poorer

    wards in period 2.

    The effects of new supply variables in price level models are also worth considering at

    this point; these are summarised in Table 7 for four time periods. The story is fairly

    consistent. The significant positive signs on the relsup variable indicate that HMAs

    with more private new supply have lower prices.6 Similarly, the negative signs on the

    pnewd9304nl variables indicate a similar effect at HMA level which is significant in the

    more recent periods (to which the data relate better). At neighbourhood level, new private

    build has mixed effects, positive in the first period and small negative effects in the later

    periods. This underlines the mixed effects mentioned above tending to partially cancel out.

    However, there is also a generally negative association with the total amount of new

    development as measured by the net change in dwellings 1991 2004 (pchdwg14), so the

    predominant effect is negative. Positive signs on relsoh indicate that areas with a large

    stock and turnover of social housing have lower prices (see note 6 again). However, actual

    recent new social building has mixed effects, positive at LA level in two periods, while at

    neighbourhood levels the effects are positive in some periods, negative in others but

    insignificant in two cases.

    The finding that, on the whole, additional private housing output has the effect of

    reducing house prices, particularly at HMA level, supports hypothesis (f) and is consistent

    with earlier applications of the panel market model by Bramley & Leishman (2005) and

    Bramley (2002), as well as with the findings of Meen et al.(2005b). This in turn provides

    some basis for the policy strategy represented by the Barker report (2004), which seeks to

    increase overall housing output substantially to ease affordability problems.

    Table 7. Effects of supply variables in price level models (effect on log price of semi-detachedhouse of selected variables by period, wards in England)

    Supply

    1995/7 2000/02 2003/4 2005/6

    Variable Coeff t stat Coeff t stat Coeff t stat Coeff t stat

    relsup 2.101 14.5 1.540 9.3 0.571 4.4 0.630 4.4lhpsup23/34 0.088 0.8 20.296 22.6 0.503 3.3 0.638 3.9Relsoh 0.423 9.5 1.056 17.1 0.953 19.7 0.890 17.5lhpsoh23/34 0.237 3.6 20.373 25.9 20.094 21.6 20.237 23.7pchdwg14 20.0001 21.5 20.0005 25.7 20.0003 24.0 20.0003 24.2

    Understanding Neighbourhood Housing Markets 205

    Downl

    oa

    ded

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    29/35

    At the same time, new building may have more positive effects at neighbourhood level;

    even where the coefficient is negative, it is usually much less so than the wider market area

    effect. These effects, which are of particular concern in the context of regeneration areas,

    are complicated by indirect effects via changes in social composition, particularly the

    incidence of poverty. Allowing for this, social renting investment is more likely to have a

    net negative effect, and private housing a positive effect, at neighbourhood level, as

    suggested in hypothesis (g).

    Concluding Discussion

    This paper has developed an approach to modelling local housing market outcomes in

    England, with a particular emphasis on house price changes over the medium term at

    neighbourhood level. While the underlying motivation for the research is to inform active

    policies of market renewal in some areas and the promotion of supply and affordability

    in others, the particular focus of this paper is on certain aspects of our understanding of

    how local housing markets work. These relate to the relationship of neighbourhoods totheir wider sub-regional market context, the nature and persistence of disequilibrium,

    the existence and significance of sub-markets, and the role and impact of new housing

    supply. The work provides a bridge between macro-economic analyses of housing markets

    and micro-level perspectives on individual choices over mobility, tenure and location.

    The work draws on traditional urban economic models of urban structure, hedonic house

    price models and macro/regional housing market models, while recognising that the

    connections between these approaches have not been fully developed. The particular

    contribution of this study lies, in part, in its assembly and use of a unique database, and in

    part on certain analytical innovations. These include a focus on modelling price change overthe medium term, a structured approach to the issues of disequilibrium/imbalance in local

    markets, a two-level structure of neighbourhoods nested in market areas, the importation of

    composite indicators from a panel model at the higher market area level, and an examination

    of indirect as well as direct effects of new investment and supply change. The work clearly

    has some implications for our understanding of the nature and existence of sub-markets in

    housing, as is drawn out from both descriptive and modelling results.

    The particular challenge of modelling/predicting price change is tackled in the second

    stage of the modelling. Models of change tend to be less well defined, afflicted by a

    relatively greater degree of data noise and less extensive sets of explanatory variablesavailable in change form. Despite these problems, it is thought that the change models here

    are relatively successful in providing plausible explanations accounting for approaching

    half of the variance in rates of change. One valuable contributor to this success is the

    ability to import composite measures of higher market area level change derived from a

    panel model. Another key factor is the use of the disequilibrium and internal imbalance

    206 G. Bramleyet al.

    Downl

    oa

    ded

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    30/35

    achieved by models containing variables from one level or the other, or both, as

    well as from the strong effect of HMA economic variables specifically.

    (b) Disequilibrium in local and neighbourhood markets may be identified from

    price models. It is believed that the partitioning of residuals from price level

    models here provides an approximate basis for identifying disequilibrium for

    HMA units as a whole, and across the quality spectrum within HMAs, at

    different points in time. It is shown that the scale of these measures is non-trivial

    as a share of the overall residual variation in prices.

    (c) The price change models demonstrate that such local disequilibria do correct

    progressively over time.

    (d) That sub-markets may operate to some extent within local housing markets,

    with differing pricing structures and trends for areas which differ in physical

    form or social profile, is a hypothesis received from some of the literature. This

    proposition receives some support, particularly from descriptive price change

    data, and in a more qualified way from the modelling of price change. Sub-

    markets are tested in relation to different demand levels, physical form/house

    type characteristics and relative levels of poverty versus. affluence, while also

    allowing for regional differences. Although disaggregated models perform

    somewhat better overall, there is no clear evidence favouring particular

    disaggregations. The detailed modelling results show more similarities than

    differences, although there are different response functions for some variables.

    (e) Nevertheless, most differences in price level and trends can be accounted for

    through a common model framework. Although there are some differences in

    response functions, these may themselves be time period dependent, and may

    be captured via more generalised models which take account of context through

    non-linear, interactive or threshold terms.

    (f) Increasing the supply of private housing will lower prices at market area level,

    but may have ambiguous effects at neighbourhood level. This expectation is

    broadly borne out by both price level and price change models. Supply-demand

    effects predominate at market area level, while at neighbourhood level

    environmental, confidence and social composition effects are at work

    (potentially in different directions, depending on the type of area). There is

    some support for the expectation that these neighbourhood effects are more

    likely to be positive in poorer areas.

    (g) Increasing the supply of social housing is likely to increase the concentration of

    poverty and reduce prices at neighbourhood level. This proposition is broadly

    supported, although the effect works mainly indirectly through the poverty rate.

    Some of the direct effects of social housing investment (e.g. environment,

    confidence) appear to be positive.

    Understanding Neighbourhood Housing Markets 207

    Downl

    oad

    ed

    At:18

    :4917

    August2010

  • 8/13/2019 74110__791548072

    31/35

    complacency. With regard to the theme of microstructures, it might be argued that what

    the models fail to explain may be what other research approaches can begin to illuminate.

    However, the study has investigated a couple of micro-structural issues relating to sub-

    markets and disequilibrium, as well as providing quite a rich account of market

    determinants and outcomes at the micro level of neighbourhoods. The findings tend to

    cast doubt on stronger versions of the sub-markets thesisthat localised markets are

    separate, operate in different ways, and do not converge or reach an economic equilibrium

    even in the medium to longer term. Nevertheless, some qualified support is found for

    notions of a degree of segmentation and disequilibrium, certainly in particular time

    periods. This is saying that context matters, but that it may be possible to model this.

    Acknowledgements

    The authors acknowledge the financial and practical support provided for this research by the Joseph Rowntree

    Foundation and the former Office of the Deputy Prime Minister. Responsibility for the analysis and interpretation

    of the data and for any views expressed rests with the authors.

    Notes

    113 bedrooms, not detached, with central heating.

    2 The mobility rate is first adjusted for the expected effects of tenure, household demography etc on

    mobility rates, using a simple regression model, so as to focus on mobility which is higher than

    expected or vice versa.3 It is true that the way we have defined HMA disequilibrium, as distinct from HMA fixed effect, means

    that correction of the longer period is almost inevitable; however, Table 3 shows that the relative

    magnitudes of these components makes this a non-trivial finding.4

    The change in flats share indicator refers to 19912001.5 This is believed to relate to speculative investment in private renting, particularly city centre apartments

    and cheaper property in former low demand areas, associated with the Buy to Let phenomenon.6

    With the panel composite indicators relsup and relsoh, higher supply has a negative weight reflecting its

    impact on price in the panel model; hence a positive sign in the ward price change model is consistent

    with a negative effect at market area level. Similar effects apply to change versions of these indicators,

    such as lhpsohch/lhpsoh24.

    References

    Ashworth, J. & Parker, S. C. (1997) Modelling regional house prices in the UK, Scottish Journal of Political

    Economy, 44, pp. 225246.Barker, K. (2004) Review of Housing Supply: Delivering Stability: Securing Our Future Housing Needs . Final

    Report, Recommendations HM Treasury.

    Berube, A. (2005) Narrowing the Gap? The Trajectory of Englands Poor Neighbourhoods, 19912001. CASE

    Brookings Census Briefs No. 4 (London: London School of Economics, Centre for the Analysis of Social

    Exclusion).

    Bramley, G. (2002) Planning reg