Real estate portfolio construction and estimation risk
| DOI | https://doi.org/10.1108/14635780510599458 |
| Published date | 01 June 2005 |
| Pages | 234-253 |
| Date | 01 June 2005 |
| Author | Stephen Lee,Simon Stevenson |
| Subject Matter | Property management & built environment |
Real estate portfolio construction
and estimation risk
Stephen Lee
School of Business, Centre for Real Estate Research (CRER),
The University of Reading, Reading, UK, and
Simon Stevenson
Department of Banking and Finance, Graduate School of Business,
University College Dublin, Blackrock, Ireland
Abstract
Purpose – The use of modern portfolio theory (MPT) in the construction real estate portfolios has
two serious limitations when used in an ex ante framework: the intertemporal instability of the
portfolio weights; and the sharp deterioration in performance of the optimal portfolios outside the
sample period used to estimate asset mean returns. Both problems can be traced to wide fluctuations in
sample means. Aims to prove that the use of a procedure that ignores the estimation risk due to the
uncertain in mean returns is likely to produce sub-optimal results in subsequent periods.
Design/methodol ogy/approach – This study extends previous ex ante-bas ed studies by
evaluating ex post optimal portfolio allocations in subsequent test periods by using methods that
have been proposed to reduce the effect of measurement error on optimal portfolio allocations.
Findings – While techniques designed to handle estimation risk in capital market studies have
yielded promising results, they are not completely successful in improving out-of-sample performance
in this case. It is hypothesised that such results are due to the cyclical nature of property and that the
contrarian and mean-reversion effects picked up in studies of stocks and bonds are not captured when
an asset such as direct property is examined. This conclusion is also supported by the strong
performance of the tangency portfolios, and in particular the classical unadjusted Sharpe portfolio,
over the shorter horizons, which would be consistent with a cyclical momentum effect.
Originality/value – The results suggest that the consideration of the issue of estimation risk is
crucial in the use of MPT in developing a successful real estate portfolio strategy.
Keywords Optimizationtechniques, Portfolio investment,Risk management
Paper type Research paper
1. Introduction
Investors in real estate have typically attempted to diversify portfolios through a
process of naive diversification. Recently modern portfolio theory (MPT) has been
advocated as a more rational approach to the construction of real estate portfolios to
identify the “best” combination of assets to hold (Lee, 1992). In this approach, the
importance of each asset is evaluated in terms of its individual relative risk and return
characteristics, as measured by its mean and standard deviation, and its portfolio risk
as characterised by its correlation with other assets. Given these parameters MPT will
find that combination of assets that, for each level of risk, will offer the highest level of
return. Such work typically uses historic ex post data to test the effectiveness of such
portfolio strategies. However, historic data by its nature is certain; consequently the
portfolio holdings are the “best” that could have been achieved in the past. This is
equivalent to playing the portfolio investment game with a marked deck, Madura and
The Emerald Research Register for this journal is available at The current issue and full text archive of this journal is available at
www.emeraldinsight.com/researchregister www.emeraldinsight.com/1463-578X.htm
JPIF
23,3
234
Received February 2004
Accepted October 2004
Journal of Property Investment &
Finance
Vol. 23 No. 3, 2005
pp. 234-253
qEmerald Group Publishing Limited
1463-578X
DOI 10.1108/14635780510599458
Abernathy (1985). Fund managers, however, are hired to construct portfolios, which
yield high ex ante rather than ex post risk-adjusted returns, and are therefore forced to
play with an unmarked deck. However, when the application of MPT has been tested in
ex ante framework the ex post results tend to perform poorly.
The classical approach to portfolio construction using MPT has two serious defects:
(1) the intertemporal instability of the portfolio weights (Lee, 1998); and
(2) the sharp deterioration in performance of the optimal portfolios outside the
sample period used to estimate asset mean returns (Jorion, 1985).
In effect, due to the fact that the inputted parameters are unstable, the estimated
optimal allocations can differ markedly between periods. This is made even more acute
as optimiser typically produce portfolios with extreme holdings in a limited number of
assets with some assets taking zero weights while others have very large allocations.
Black and Litterman (1992) refer to these as corner solutions. Although the resul ting
portfolios are optimal in the statistical sense, the results would be unacceptable to any
prudent portfolio manager (Jorion, 1985). Such corner solutions portfolios quickly
become sub-optimal with changes in the means over time, leading to a further
reduction in ex ante performance. In addition such extreme portfolio allocations assets
seem to be against the spirit of diversification, Michaud (1989). One way to control for
such extreme holdings is to place constraints (upper and lower bounds) on the amount
any one asset, or group of assets, can have in the optimum portfolio (Byrne and Lee,
1995; Stevenson, 2000a). Indeed papers such as Frost and Savarino (1988) and Chopra
(1993) suggest portfolio optimisations, which are subject to such constraints, have
better ex ante performance compared with unconstrained solutions. However, any
constraints are likely to be arbitrary, leading to the results being hard to generalise. For
example, one possible constrained portfolio is the equally weighted naı
¨ve portfolio.
Morrell (1993), however, argues that that it is generally not possible for property funds
achieve equal-weighting in a portfolio and at the same time be represented in key
market segments. In addition fund managers typically desire to maintain weights
similar to a benchmark portfolio. Also at a practical level due to the indivisibility of
property and the marked differences in lot size between say the office and industrial
sectors an equal-weighted portfolio strategy would be impossible to implement. Thus
an equal-weighted portfolio is therefore probably not a realistic, or even a desirable
goal of fund managers. In addition such an approach fails to tackle the fundamental
reason for the major shifts in portfolio allocations over time, the instability in the
sample means. In contrast the estimation error in variances and covariances is no t as
much of concern since these parameters are relatively stable over time and therefore
are more precisely estimated. Studies such as Kalberg and Ziemba (1984), Chopra and
Ziemba (1993) and Stevenson (2001a) have found similar results. Thus the use of a
portfolio selection procedure based on historical parameters that ignores the estimation
risk due to the uncertain in mean returns is likely to produce sub-optimal results in
subsequent periods. Indeed previous work on the application of MPT to the real estate
portfolio shows this to be the case (Myer and Webb, 1991; Mueller and Laposa, 1995;
Pagliari et al., 1995).
The above discussion suggests that the consideration of the issue of estimation risk
is crucial in the use of MPT in developing a successful real estate portfolio strategy.
Therefore, following Eun and Resnick (1988) and Kwok (1990), this study extends
Real estate
portfolio
235
To continue reading
Request your trial