Property valuation and the structure of urban housing markets
| Pages | 157-175 |
| Published date | 01 May 1999 |
| DOI | https://doi.org/10.1108/14635789910258543 |
| Date | 01 May 1999 |
| Author | Craig Watkins |
| Subject Matter | Property management & built environment |
Academic papers:
Property
valuation
157
Journal of Property
Investment & Finance,
Vol. 17 No. 2, 1999, pp. 157-175.
#MCB University Press, 1463-578X
Received 21 July 1998
Revised 21 December
1998
ACADEMIC PAPERS
Property valuation and the
structure of urban housing
markets
Craig Watkins
Centre for Property Research, Department of Land Economy,
University of Aberdeen, Old Aberdeen, UK
Keywords Housing market, Model, Regression analysis, Valuation
Abstract Since the 1980s UK academics have promoted the use of multiple regression analysis
in property valuation. Recently, however, there has been growing recognition that regression
models will be subject to aggregation bias if they fail to accommodate the existence of housing
market segmentation (submarkets). In this study, we compare the empirical performance of a
standard hedonic house price regression model for the city of Glasgow with a segmented model
which recognises the importance of understanding the underlying market structure and, in
particular, the existence of submarkets for different dwelling types. The results show that the
(weighted) standard error of the segmented model is significantly lower than that of the market
wide model. Consequently, we propose a two-stage approach to the application of MRA techniques
to residential valuation. First, following traditional institutional analysis of housing markets, the
market should be subdivided into distinct structurally differentiated market segments. These
segments can usefully be identified by principal components factor analysis which allows the
identification of the most important common components in the housing bundle. Second, separate
house price equations should be estimated for each market segment. Although the best-fit
equation may vary from sector to sector this is likely to reflect the behavioural realities of the
property market, and will provide the basis for more accurate valuations.
Introduction
UK practitioners have lagged behind valuers in the US in applying multiple
regression analysis to property valuation problems. In the US the application of
regression analysis to the estimation of property values can be traced to the
1950s when the increased commercial availability of computers enabled
property professionals to employ regression techniques (see Pendleton, 1965).
By regressing the physical and locational characteristics of properties on the
selling price, the technique allows valuers to estimate the value of individual
property attributes and, from these, to construct estimates of the value of
properties whose characteristics are known. This approach is analogous to the
comparable method of valuation and has been shown to be both cost effective,in
valuing large numbersof properties, and accurate (Dodgson and Topham, 1990).
Despite this, it is only since the 1980s that UK academics have begun
tentatively to promote the adoption of statistical methods of analysis in
The author would like to acknowledge the support provided by a University of Paisley
Research Grant. Thanks are also due to Colin Jones who provided helpful comments and
guidance throughout this research project. The usual disclaimer applies.
The research register for this journal is available at
http://www2.mcb.co.uk/mcbrr/jpif.asp
The current issue and full text archive of this journal is available at
http://www.emerald-library.com
JPIF
17,2
158
valuation more widely (see Adair and McGreal, 1988; Greaves, 1984). More
recently, however, with the further development of user friendly statistical
packages and the increasing availability of detailed data on open market
transactions, we have seen renewed enthusiasm for this approach (see, for
example, Dunse and Jones, 1998; Antwi, 1995; Adair et al., 1996; Gallimore and
Ward, 1992; Adair, 1991).
Even with this support for MRA in property valuation, some authors have
noted potential problems. Most commonly these have focused on the statistical
properties of regression equations (Mark and Goldberg, 1998). On occasion,
however, they have been based on what it is known about the characteristics of
property markets. Lusht (1976) and Newsome and Zeitz (1992), for example,
argue that markets are segmented. Thus in order to ensure better model
performance, it is important to segment the market according to size bands and
property type prior to calibrating regression equations. Although, this is
undoubtedly an important point, the sub-divisions employed in these papers
are essentially arbitrary. Adair et al. (1996) produce a more rigorous
explanation of the importance of market segmentation. With recourse to the
economic theory of urban housing markets, they explain that by estimating
regression equations across segments (or housing submarkets), valuers will
produce price estimates which will be subject to aggregation bias. This is
supported by evidence from an empirical study of the Belfast housing market.
In this paper we extend Adair et al.'s (1996) analysis. Specifically, we set out
to compare the empirical performance of a multiple regression equation
compared with models based on a classification of different dwelling
structures. The paper is developed in the following stages. In the next section,
we explore the theoretical basis for applying multiple regression analysis to
property valuation. We show how regression analysis will benefit from both
insights from neo-classical economists' hedonic house price theory and
institutionalists' concept of the ``housing submarket''. In doing so, we explore
how submarket existence can give rise to aggregation bias in hedonic
regression equations unless these market cleavages have been taken into
account. The third section introduces the data and research methods employed
in this study. This precedes our introduction, in section four, of the use of
principal components factor analysis as a means of classifying properties into
groups with similar structural characteristics. In section five, we summarise
our empirical results. Finally, in the last section, we conclude that, on the basis
of these results and in order to improve valuation accuracy, MRA must be
preceded by the identification of submarkets for distinct property structures.
Hedonic house price models and market segmentation
The development of local housing market analysis owes much to the analytical
framework developed by a group of economists at the Columbia University.
Given the special characteristics of housing as an economic commodity
(heterogeneity, spatial immobility and durability), this group postulated that
housing markets were best conceptualised as a set of quasi-independent
To continue reading
Request your trial