Enhancing information use to improve predictive performance in property markets

Date01 December 2001
DOIhttps://doi.org/10.1108/14635780110406851
Published date01 December 2001
Pages472-497
AuthorPatrick J. Wilson,John Okunev
Subject MatterProperty management & built environment
JPIF
19,6
472
Journal of Property Investment &
Finance, Vol. 19 No. 6, 2001,
pp. 472-497. #MCB University
Press, 1463-578X
ACADEMIC PAPERS
Enhancing information use to
improve predictive
performance in property
markets
Patrick J. Wilson
School of Finance and Economics, University of Technology, Sydney and
Visiting Scholar,University of Wollongong, Gwynneville, Australia, and
John Okunev
School of Banking and Finance, University of New South Wales,
Sydney, Australia
Keywords Forecasting, Property markets
Abstract Over the last decade or so there has been an increased interest in combining the
forecasts from diff erent models. Pool ing the forecast outc omes from different m odels has been
shown to improve out -of-sample foreca st test statistics be yond any of the individ ual component
techniques. The discussion and practice of forecast combination has revolved around the
pooling of results f rom individual forec asting methodolog ies. A different approa ch to forecast
combination is foll owed in this paper. A method is used in which negativ ely correlated forecasts
are combined to see if t his offers improved o ut-of-sample fore casting performanc e in property
markets. This is co mpared with the outcome from both the o riginal model and with benchmark
naõÈve forecast s over three 12-month o ut-of-sample per iods. The study will look at securitised
property in three international property markets ± the USA, the UK and Australia.
Introduction
Food for thought: economists have forecast nine out of the last five recessions.
The search for accurate forecasting techniques in property and other markets is
perennial. While causal basedtechniques are likely to represent the most popular
forecasting methodology, various univariate techniques such as ARIMA models
and spectral forecasting models, or combined univariate and multivariate
techniques such as MARIMA (transfer function) models, Kalman filters and the
like have shown that they are all capable of producing consistently acceptable
forecastingresults in certain situations.But which technique should the property
professionaluse in which situation?
The research register for this journal is available at
http://www.mcbup.com/research_registers
The current issue and full text archive of this journal is available at
http://www.emerald-library.com/ft
This paper was presented at the European Real Estates Society (ERES) meeting, Bordeaux, June
2000. The authors wish to thank the referees for several comments that improved the
information content of this paper. The research has been supported by various UTS Research
Grants. The research was undertaken while the first named author was Visiting Scholar,
Department of Economics, University of Wollongong
Academic papers:
Enhancing
information use
473
It is unfortunately the case that different forecasting techniques may produce
different outcomes when appliedto the same data sets. The basic explanation for
this is that differenttechniques extract different information aboutthe behaviour
of the series under consideration. A dilemma facing the property market and
other professionals, therefore, revolves around selection of the most appropriate
forecasting method. As a means of dealing withthis problem, especially over the
last decade or two, there has been an increased interest in combining or pooling
the forecasts from different models. Clemen (1989) has undertaken a very useful
survey of the literature on combining forecasts and the broad conclusion is that
combining does improve forecast accuracy. However an underlying assumption
here is that, while the models being combined may well be extracting different
information, their predictive abilities are broadly similar. That is, one model is
not consistentlymore accurate than another[1].
Winkler (1989) has likened the concept of combining forecasts to the concept
of diversification to reduce risk ``...a combined forecast can be thought of as
having a smallerrisk of an extremely large errorthan an individual forecast''.We
might also think of the pooling of forecasts as reducing the risk of over-reliance
on the outcomes ofa single model[2]. Faced with a choice of models the property
analyst may not be in a strong position to judge which one will perform best ex
ante. Pooling provides a degree of comfort for risk averse investors in the sense
that they are less likely to feel exposed if the consensus is wrong than if they
have backed a single model which produces a similarly inaccurate but counter-
consensualresult. Such ``hedging of bets'' across more thantwo models, however,
may not be overly beneficial since research by Makridakis and Winkler (1983)
and Bopp (1985) has shown that if there are more than two sets of forecasts
included inthe combination, there willbe diminishing returns.
The research of Clemen (1989), Diebold (1989), Makridakis (1989) and others
has revolved around the combination of forecasts from different forecasting
models. In this paper we are interested in whether an approach pursued by
Ridley (1997, 1999), in which negatively correlatedforecasts are combined, offers
improved out-of-sample forecasting performance in property markets. In a
broader sense,since it is not unusual to have as much as 10 percent or more of an
investment portfolio given over to tactical asset allocation, an improvement in
short-run forecasting outcomes can have a beneficial effect on general portfolio
performance. In this paper we find that there is some evidence to support the
notion that combining negatively correlated forecasts in property markets can
reduce forecast error. The remainder of the paper is divided as follows: the
following section provides an overview of the literature on the combining of
forecasts; the next section discusses the concept of combining original and
antithetic forecasts; the subsequent section discusses data sources and presents
results of the analyses while the final section draws some conclusions. In
addition, the Appendix presents a brief overview of spectral regression as used
in the present analysis.

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT