Está en la página 1de 27

J Real Estate Finan Econ (2010) 41:53–79

DOI 10.1007/s11146-009-9230-y

Alpha and Persistence in Real Estate Fund Performance

Shaun A. Bond & Paul Mitchell

Published online: 5 January 2010


# Springer Science+Business Media, LLC 2009

Abstract This paper investigates whether fund managers investing in the direct real
estate market can systematically and persistently deliver superior risk-adjusted
returns. The research that has been published has typically focused on the
performance of managers trading public real estate securities. Our study draws on
a unique data set of commercial real estate funds collated by the Investment Property
Databank (IPD) in the United Kingdom, covering up to 280 funds over the period
1981 to 2006. The widespread finding is that very few managers appear to be able to
generate excess risk-adjusted returns. Furthermore, there is little evidence of
performance persistence in either fund returns or risk-adjusted fund returns.

Keywords Commercial real estate investment . Fund manager performance

JEL Classification R33 . G11 . G23

Introduction

Alpha is a powerful concept in investment and fund management. It relates to the


delivery of superior risk-adjusted returns, either from an active fund manager or
from an asset class. Investors choose fund managers on the basis of their potential to
deliver alpha, and fund managers are often rewarded on this basis. However, there
has also been a long-running debate across asset classes as to whether or not active
fund management can systematically add value.

S. A. Bond (*)
Department of Finance and Real Estate, College of Business, University of Cincinnati, Cincinnati,
OH 45221-0195, USA
e-mail: shaun.bond@uc.edu

P. Mitchell
Paul Mitchell Real Estate Consultancy Ltd, London E11 2SA, UK
54 S.A. Bond, P. Mitchell

There has been comparatively little research on the extent to which real estate
fund managers can systematically and persistently deliver superior risk-adjusted
returns, and on the magnitude of any such performance. While there has been
research on real estate mutual fund managers trading public real estate securities
(Chiang et al. 2008), and the performance of managers of public real estate
companies (Brounen et al. 2007), few studies have investigated the performance of
large institutional real estate fund managers investing directly in commercial
property assets. In many countries, such as the United Kingdom, this segment of
the investment market dominates the value of publicly traded real estate funds or
mutual fund-type products aimed at retail investors (see discussions in Chun et al.
2004; Bond et al. 2007).
Why study real estate fund managers? Aside from the interest in this topic by the
professional investment community, there are other reasons to consider that this is a
suitable topic for academic research. Commercial real estate markets seem ideal to
consider the implications of the Grossman and Stiglitz (1980) hypothesis given the
widely held perception of the value of private information in these markets.1
Furthermore, there is considerable evidence to suggest that the returns on
commercial real estate assets exhibit persistence (Lee and Ward 2000; Devaney et
al. 2007). This is potentially important in two respects. Firstly, it is possible that the
artificial smoothing of property valuations may induce persistence in fund returns.2
This could lead to managers being rewarded for high levels of persistent
performance when the apparent persistence is really an artefact of the data
aggregation process. Secondly, as suggested by Key and Marcato (2005), knowledge
of the underlying persistence may point to profitable momentum trading strategies
that could be employed to achieve superior risk-adjusted performance. However, is
not clear that this persistence could be profitably exploited given the high transaction
costs and illiquidity that underlies trading in commercial property assets. Also, such
a study may shed light on the performance of managers of other non-publicly traded
assets, for example private equity fund managers and infrastructure fund managers.
An extensive proprietary database of real estate fund manager performance has
been used in this study. This data on annual fund performance from 1981 to 2006,
which is longer and more extensive than many other studies on real estate fund
performance, allows us to investigate the extent to which real estate fund managers
exhibit a high level of (risk-adjusted) performance and also whether they are able to
maintain good performance, if it is found. By necessity, a secondary objective of this
paper is to investigate the most appropriate form of risk-adjustment in real estate
markets. It is recognised, however, that the form of risk-adjustment adopted in this
paper may not be universally accepted and for this reason performance relative to a
benchmark is also examined in this study.

1
Grossman and Stiglitz (1980) argue that if information is costly, the usual assumption of the efficient
markets hypothesis, that prices reflect all available information, may not hold. It is only worthwhile for
investors to acquire information if they obtain a return from doing so.
2
The appraisal-smoothing problem in real estate has been extensively discussed in the literature, see Bond
and Hwang (2007) for a summary of the literature. However, given the long-term nature of real estate
investors and the sample horizons used in this study, such short term persistence effects may not be
relevant to the analysis.
Alpha and Persistence in Real Estate Fund Performance 55

The next section of this paper presents a literature review on the key academic
studies on alpha and persistence in commercial real estate markets and other asset
classes. “Methodology” details the methodologies adopted in this study for assessing
alpha and persistence in UK property and for the risk-adjustment model. “The Data”
describes the IPD database and the specific samples of data used and types of
analysis undertaken in the study. “Performance, Alpha and Persistence in UK
Property Fund Management” presents the detailed quantitative analysis of alpha and
persistence and the factors behind it. The conclusions are presented in “Conclusion”.

Literature Review

One of the earliest attempts to evaluate the performance of mutual fund managers
was by Jensen (1968). Using the Capital Asset Pricing Model (CAPM) as a basis,
Jensen showed how a manager’s superior performance could be captured by the
intercept term (alpha) in the model. In a sample of mutual funds Jensen found that
the average value of alpha was negative, implying that managers underperformed
relative to a risk-adjusted benchmark. A large number of studies followed the work
of Jensen. In a highly cited article, Grinblatt and Titman (1989) identified the
presence of a skilled set of fund managers. They noted that this superior performance
was only found in gross returns. After allowing for the effects of fees the superior
performance diminished. The dampening effect of fees on performance is a recurring
theme in the literature.
The question of whether the performance of a manager continues over different
time periods was considered by Hendricks et al. (1993). They found that it was
possible to generate economically significant returns by developing a trading
strategy based on the past performance of fund managers. Showing strongly in the
research was that the performance of managers with “icy hands” also persists. In
fact, Hendricks et al. point out that the persistence in returns for the poorest
performers was much more prevalent that the persistence in performance for the top
managers. Persistence in performance was also found in studies by Goetzmann and
Ibbotson (1994) and Brown and Goetzmann (1995).
Carhart (1997) provides a critical appraisal of the Hendricks et al. study. In
particular he adjusts for momentum effects in the underlying equity market data and
potential survivorship bias in the data (something which the subsequent article by
Carhart et al. 2002 and a number of articles on hedge funds, notably by Malkeil and
Saha 2005, explore further). While noting some evidence of superior performance,
Carhart’s overwhelming conclusion is that “...most funds under-perform by about the
magnitude of their investment expenses”.
More recent evidence using bootstrap techniques has provided support for the
notion that there is a small group of superior fund managers and the superior
performance of these managers persists over time. Kosowski et al. (2006) used
bootstrap techniques to allow for the non-normal distribution of fund alphas and
found evidence that the large positive alphas of the top 10% of fund were unlikely to
arise by chance. Furthermore, the authors found that this performance persisted over
time. In addition Kosowski et al. (2007) found similar evidence of a small group of
elite hedge fund managers that showed persistence in their level of performance.
56 S.A. Bond, P. Mitchell

Busse and Irvine (2006) also found support for performance persistence using
Bayesian statistical methods. An interesting additional finding was that Bayesian
alphas were useful for predicting future outperformance by fund managers.

Performance and Persistence in Real Estate Funds

One of the few studies that evaluated the topic of performance persistence for
property fund managers was provided by Hahn et al. (2005). Using data on real
estate opportunity funds and a methodology similar to that employed in this study,
the researchers found some evidence of persistence among real estate opportunity
fund managers. In Hahn et al’s study, performance persistence was measured across
different funds raised by the manager. The research indicates that as much as 20–
24% of subsequent fund performance may be related to past fund performance.
However, due to the nature of fund raising that takes place, it may be difficult for
investors to profitably exploit this finding because it can be several years before the
performance of a fund is clearly determined.
While not a study on fund performance, Ling (2005) directly addressed the
question of whether participants in the direct real estate market could capitalise on
the perceived market inefficiencies of commercial real estate to produce accurate
market forecasts. He found no evidence that a published consensus forecast,
compiled from the forecasts of institutional real estate owners and managers, was
helpful to investors. While there are many caveats in such as study,3 it does not
appear to support the idea that participants in the commercial real estate market can
exploit informational inefficiencies.
A more extensive collection of research is available on the performance of
managers of funds focused on publicly traded real estate securities. Kallberg et al.
(2000) examined the performance of mutual funds that invest only in the REIT
sector. They found that managers do add value through active portfolio management
and estimate that an extra two percentage points annually is added to performance
relative to passive funds. Surprisingly, they found little evidence of performance
persistence. Published at a similar time, Gallo et al. (2000) also found positive
abnormal performance for real estate mutual funds. However, their sample was
smaller than Kallberg et al. and only covered a six-year period. It also provided little
evidence from the multi-factor performance benchmarking models that are now
commonly used in the literature.
Interestingly, other studies, using different time periods and alternative bench-
marking models, have failed to find evidence of positive abnormal performance
(O’Neal and Page 2000; Lin and Yung 2004). While not finding significant positive
performance, Lin and Yung did note evidence of short-term performance persistence
among both high and low-performance funds.
Chiang et al. (2008) and Hartzell et al. (2009) have attempted to reconcile the
apparent differences in these earlier studies. Using different approaches, but focusing
closely on the specification of the benchmarking models employed, both sets of
authors conclude that there is little evidence of outperformance by real estate mutual

3
That is, the best forecasters may keep their forecasts private, and owners and managers may have other
objectives in submitting their forecasts, for instance, to make their region look more attractive to investors.
Alpha and Persistence in Real Estate Fund Performance 57

fund managers. Hartzell et al. in particular provides an extensive discussion of


benchmarking and highlights the importance of introducing factors closely
associated with the real estate market.
Eichholtz et al. (2009) provide a comprehensive study on the performance of
global (listed) real estate mutual funds. Using a similar methodology to that used in
this paper, they found that global and European fund managers were able to add
value, whereas Asian, Australian and North American managers did not. This raises
interesting questions about the way markets operate in these areas. When
performance benchmarks were corrected for investment style, it was found that in
addition to outperformance by global and European managers, the Asian managers
also demonstrated positive performance in excess of their benchmarks.
The literature on the performance of REIT managers has been well summarised in
Brounen et al. (2007). Their study focused on the impact of trading intensity and
acquisitions in understanding the investment performance of publicly-traded real
estate firms in the US, UK and Australia. The findings of the study suggest that it is
difficult for managers in these companies to generate outperformance based on an
active trading strategy. However, their conclusions are sensitive to the way in which
trading activity was measured.
While a number of studies have focused on real estate mutual funds, there are
three key points of difference between that set of literature and the current study:
& the nature of the funds examined—institutional investment in direct real estate
assets as compared to products which invest in publicly-traded real estate
securities aimed predominately at retail investors;
& the size and depth of the sample studied—this study contains a much larger
sample of funds; and,
& the frequency of the data—this study uses annual performance data as opposed to
monthly data in the mutual fund studies.4

Methodology

This section outlines the main methodological approaches adopted for analyzing the
performance of fund manager returns in this study. An interesting dilemma
encountered with this research is that the multi-factor approach to measuring risk-
adjusted performance is not well developed for commercial real estate funds
investing directly in the real estate market (as opposed to real estate mutual funds
investing in REITs). In keeping with other performance studies where the approach
to benchmarking is not well established (such as hedge funds and private equity
funds), we evaluate funds using both raw returns (no risk adjustment applied), and at
least one form of risk-adjusted returns series. The following discussion of
methodology is based on Brown and Goetzmann (1995) and Carhart (1997).

4
It would have been preferable to use higher-frequency data. However, in the UK such data is much less
extensive than annual data, relating to a small number of funds on a monthly basis (around 70 compared to
over 200 annually) and to a short period (2001 onwards) on a quarterly basis. There are also indications
that higher-frequency data would not have enhanced the analysis.
58 S.A. Bond, P. Mitchell

Performance and its Persistence

The first stage of the analysis focuses on sorting the funds into quartile groupings
based on performance in a ranking period. Such “ranking” periods are based on
three, five or ten year time periods. The performance of the funds is then observed in
the subsequent “evaluation” period (the immediately following three, five or ten year
period). It can then be noted how many funds that rank in the top quartile during the
ranking period remain in the top quartile during the evaluation period or, if it occurs,
how many migrated to lower performing quartiles. From this information transition
matrices between quartiles can be calculated.
In each case when persistence is assessed, two main statistical tests are employed.
The first is a chi-square test to examine whether the transition probabilities are equal
across cells. A test statistic that is not statistically different from zero implies each fund
manager has an equal chance of falling into any quartile during the evaluation period.
This is consistent with the notion that fund managers do not possess any specific ability.
The second test is the same as that put forward by Brown and Goetzmann (1995) and
used by Hahn et al. (2005). This test amalgamates the first and second quartiles to
represent “winner” performance (an above median result), and groups the lower two
quartiles to represent “loser” (below median) performance. A fund that is in the
“winner” group in the ranking period and retains that position in the subsequent
evaluation period is designated a winner-winner (WW). A fund delivering a below
median performance in both periods is designated a loser-loser (LL), and similarly a
fund which either moves from a below median result to an above median result or an
above median performance to a below median performance is indicated by the label
loser-winner (LW) or winner-loser (WL), respectively. The test statistic is calculated
as the natural log of the cross-product ratio (CPR), i.e.:

lnðCPRÞ ¼ lnððWW  LLÞ=ðWL  LW ÞÞ;

which is asymptotically normal with a standard deviation of:


rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 1 1 1
þ þ þ :
WW LL WL LW
In each case this statistical test is applied to both raw returns and risk-adjusted
performance measures.
In addition to these tests, we also follow the evaluation methods displayed in
Carhart (1997) and show the subsequent return performance of each portfolio of funds
grouped by their initial ranking. That is, we calculate the performance in the evaluation
period of all funds listed in the first quartile during the ranking period. This allows us
to observe, if as a group, the performance of each quartile reverts to a common mean or
whether a performance differential remains during subsequent periods.

Risk-Adjusting Property Fund Manager Performance

The use of factor models in assessing the performance of fund managers is discussed
in this section. In order to derive a measure of risk-adjusted performance (in this case
Jensen’s alpha), it is necessary to know what factors drive property returns.
Alpha and Persistence in Real Estate Fund Performance 59

However, the selection of models for consideration is limited by the low frequency
of the data set (in this case annual returns).
Two methods dominate the literature on benchmarking, characteristics-matched
benchmarks and regression-based benchmarks.5 However, the application of either
approach to assessing risk-adjusted performance for real estate fund managers has
been limited. The one exception to this has been the extensive research on risk
models for real estate mutual funds discussed in the literature review above.
A characteristics-matched approach, while theoretically interesting, is not possible
with the proprietary data set available for this study. Also it is not clear what
characteristics should be used in the sorting procedure. Hence in this study we focus
on regression-based models to assess risk-adjusted performance.
For the period 1981–2006, three models of the return in excess of the risk-free
interest rate6 were investigated using performance quartiles comprising each year’s
corresponding funds. Each of these models is described below:
1. A single index model (using the IPD Universe excess total returns index as a
benchmark). The beta in this model simply measures the sensitivity of the fund’s
return (in excess of the risk-free rate) to that of the IPD Universe.7
2. A version of Sharpe’s index model (Sharpe 1992) using excess returns in the
IPD office, retail, industrial and the “other” property sectors. In this case, as
fund managers cannot go short in the property market and the sectors they can
invest in are reasonably defined, the coefficients of the Sharpe index model are
constrained to be positive and also to sum to one, thereby ensuring that the fund
is fully invested.
Sharpe’s model can be written as:
X
K
rpt ¼ g p0 þ g pj Ijt þ upt ð1Þ
j¼1

Where
g pj  0; j ¼ 1; ::; K
PK
j¼1 g pj ¼ 1
 
and υpt is serially uncorrelated and distributed as N 0; s 2u .
Annual excess returns on fund p in year t are represented by the variable rpt.
Annual excess returns on the four main real estate sectors (office, retail, industrial,
and other) in year t are shown by the variable Ijt (where j indexes sector type, K is
the total number of sectors). The four estimated coefficients for fund p (g pj, j =1 to 4)
in this model represent average portfolio exposures for the fund manager over the
estimation period to the respective sectors.

5
See Chan et al. (2006) for a detailed survey of risk-adjusted benchmarking of fund manager
performance.
6
The risk-free rate for each year is determined by the quarterly average three-month Treasury bill rate.
7
As explained in “The Data”, the IPD indices are derived from details voluntarily submitted to the IPD by
funds for benchmarking and performance measurement and attribution.
60 S.A. Bond, P. Mitchell

3. A model using returns on publicly traded securities (the European Public Real
Estate Association [EPRA] index and a corporate bond index), as factors in a
multifactor model. In this case the coefficients of the factors were not
constrained in the way they were for the Sharpe index model.

In all three models, alpha represents the (expected) excess return, that is, the
overall return (over the risk-free rate) less that due to risk. However, because the risk
factors are different, its value is likely to differ across the three models.
In applying the models it is found that the Sharpe index model (using four
property sectors) provides the highest level of explanatory power. This model is
adopted for the analysis of individual fund returns over ten-year horizons. However,
for five-year horizons, there are insufficient degrees of freedom to apply the Sharpe
Index model, so we use the single-index model, which was narrowly behind the
four-factor model in terms of goodness of fit (R2 typically greater than 0.75) in the
initial investigations.

Attributes of Performance and Persistence

As a final evaluation measure we employ the persistence test of Hendricks et al.


(1993) to determine whether risk-adjusted performance for each fund in the ranking
period is a predictor of risk-adjusted performance in the evaluation period. While
this test has been criticized by Carhart (1997) because of the potential econometric
problems of using a previously estimated variable in a secondary regression, it has
some intuitive appeal and is included along with a range of other fund
characteristics. This regression is intended to investigate whether any known fund
characteristics can be used to predict subsequent fund performance. If such a
regression had strong explanatory power, it may be possible to develop a
profitable trading strategy for picking top performing funds. More discussion of
this approach is included in “Performance, Alpha and Persistence in UK Property
Fund Management”.

The Data

This section introduces the source of the data used in “Performance, Alpha and
Persistence in UK Property Fund Management” to analyse the persistence of property
performance and the characteristics of the sample of data used. It also considers the
extent of any bias associated with the data.

The IPD Fund Database

In common with most studies of this type, our analysis focuses on the performance
of specific funds rather than fund management houses or individual fund managers.
In doing this, it draws on the records collated by the Investment Property Databank
(IPD) since 1981. Funds voluntarily submit their details to the IPD for independent
performance measurement and benchmarking. The IPD (2007) estimates that its
records cover 55% of professionally managed investment property in the UK. This
Alpha and Persistence in Real Estate Fund Performance 61

definition excludes, amongst others, small private landlords, owner-occupied


properties, and funds that own the operating business as well as the property (e.g.
hotels, hospitals).
Potential biases in such “manager universe” indices are widely noted in the
practitioner and academic literature. In the IPD’s database, such biases may be both
positive and negative—for example, resulting from some collective skill that
professionally managed funds add or, alternatively, from those who eschew
benchmarking (such as opportunistic or absolute return funds). In this later respect,
Hahn et al’s (2005) conclusion that some US real estate opportunity fund managers
persistently out-perform is notable.
The histories of funds newly entering the database may be retrospectively added.
Similarly, the historic records of funds expiring are also retained up to their last full
calendar year. Funds “expire” not only when they wind up but also when they merge
or split, the portfolio management company changes, or when there is a substantial
change to the name of the fund. Unfortunately the database does not identify the
reason why funds wind up.
The IPD performance data relate to the underlying property assets and in this
respect are ungeared (unlevered). However, funds’ exposures to indirect holdings,
which may be geared (levered), such as property unit trusts, REITs etc, are also
included. Property unit trusts, REITs and other listed property companies are
included in the database but, in the same way as other portfolios, the data relate to
the unlevered performances of the underlying property assets.
Returns are measured net of property management costs (letting, rent review
and general property management costs). However, they do not take account of
fund management costs, for which no estimates are available. For a general,
balanced fund, interviews with investors and investment consultants suggest that
fund management costs are typically in the range 25–40 basis points, with
performance-related fees increasing and (less commonly) reducing these very
marginally. Other than this marginal effect from performance-related fees, inter-
views with investors suggest that the best fund managers charge only marginally
higher fees (5–10 bps).
The growing number of specialist vehicles and value-added funds (albeit, in
accounting for about 14% of funds, still in the minority) tend to charge slightly
higher base fees than the general funds; performance-related fees are also understood
to be higher. Surveys by the European Association for Investors in Non-listed Real
Estate Vehicles (INREV), suggest that fund management fees for valued-added
funds are marginally higher at around 60 bps.
Given this small variation in fees amongst the type of funds in the IPD universe,
we do not believe the inability to net fund management fees out of our estimates of
relative performance and alpha will significantly alter our conclusions.

Quantifying Fund and Benchmark Performance

In contrast to IPD practice, in this study funds are combined on an unweighted basis
to form sample averages. This is consistent with the approaches adopted in studies of
other asset classes and treats each fund equally, avoiding biases brought about by
differences in size.
62 S.A. Bond, P. Mitchell

All funds in the IPD database are included in the analysis until they expire, but on
the condition that they are in existence for the duration of a qualifying period (i.e.
the ranking period defined below).
Relative performances of funds against the benchmark are calculated in line with
the IPD’s methodology, i.e. as the ratio of the fund return to the benchmark return.

Ranking and Evaluation Periods

This analysis is based on fund performance over three, five and ten-year periods.
Performance over one particular period is then compared with the following period.
The initial period is termed the ranking period, the following one the evaluation
period. In total, 12 sets of data are examined:
& one set of ten-year data (1987–1996 ranking period vs 1997–2006 evaluation
period);
& four sets of five-year data (1982–1986 vs 1987–1991, 1987–1991 vs 1992–1996,
1992–1996 vs 1997–2001, 1997–2001 vs 2002–2006); and,
& seven sets of three-year data (1983–1985 vs 1986–1988, 1986–1988 vs 1989–
1991, 1989–1991 vs 1992–1994, 1992–1994 vs 1995–1997, 1995–1997 vs
1998–2000, 1998–2000 vs 2001–2003, and 2001–2003 vs 2004–2006).
Funds are ranked into quantiles (specifically quartiles and deciles) according to
their performance over each ranking period and each evaluation period. Performance
of the set of funds in each quantile is calculated as the unweighted fund average.
A benchmark return for all the three, five, and ten-year periods is calculated as the
annual average of the unweighted average of all funds in the sample under
investigation. Such benchmarks will differ to the IPD Universe, primarily because
this study does not include all funds in each period and because fund returns are
unweighted.

Fund Numbers, Attrition, Creation and Bias

Previous studies of persistence in other asset classes (see, for example Carhart et al.
2002) have identified the survivorship bias generated by excluding funds which
subsequently “die”. This study limits survivorship bias by including some funds
which subsequently expire. However, some bias still may exist on account of the
following:

1. Funds are excluded at the outset if they do not survive throughout the whole of
the ranking period. As Table 1 shows, the number of funds affected is relatively
small up until the mid-1990s but this increases and is more substantial
(representing about 35% of the number of funds) from the late 1990s. In
general, the worst performing funds have a greater probability of expiring, so the
benchmark return used will be marginally inflated and the ranking of the funds
in this study’s sample could be overstated.
2. Funds not in existence at the beginning of each period’s analysis are excluded
from the analysis. Table 1 shows that the number of funds excluded in this way
is large, comparable to the total number which qualify for inclusion in the
Alpha and Persistence in Real Estate Fund Performance 63

Table 1 Number of funds in the sample compared to number in the IPD Universe

10-year horizon 5-year horizons

1987–2006 1982– 1987– 1992– 1997–


1991 1996 2001 2006

Funds existing at start of analysis 200 141 200 242 277


Funds without full ranking period history 30 0 5 29 88
Funds in starting sample 170 141 195 213 189
Funds ceasing to exist during evaluation period 85 4 25 77 49
Surviving funds at end of analysis 85 137 170 136 140
Funds starting during the entire analysis period 195 105 107 134 140
(net of deaths)
Funds in existence at end of analysis 280 242 277 270 280

The first row of the table shows, for each window of analysis, the number of possible funds that exist in
the IPD fund database. Row two of the table shows how many funds are excluded because they do not
cover the entire history of the ranking period. Hence, when the data in row two is subtracted from row
one, the number of funds included in the sample for the ranking period is shown in row three. The fourth
row displays the number of funds that cease to exist during the evaluation and row fifth shows how many
funds are still in existence at the end of the period of analysis. Row six shows the number of funds started
over the period of analysis (this included both the ranking period and the evaluation period) and this row is
added to row five to show how many funds exist in the IPD fund database at the end of the period of
analysis (shown at the top of each column). Note that while funds are not included if they start up during
one part of the analysis, the fund may be included in a subsequent horizon window

sample. Bias could arise because newly formed funds appear initially to out-
perform mature funds8; hence if the newly formed funds are excluded from the
analysis, it will tend to flatter the ranking and the relative performance of mature
funds in this study’s sample. Of those included in the analysis, the proportion of
funds in the upper quartiles is likely to be exaggerated, and the proportion in the
lower quartiles understated; the estimates of persistence will be correspondingly
affected.

A detailed discussion on the impact of these omissions on the data used in the
study is contained in the appendix.

Characteristics of Funds

The IPD Universe encompasses a range of fund types. Table 2 presents details of
these for a number of the samples used in the analysis. Segregated pensions are the
largest fund type throughout, and have broadly maintained their weighting.
However, the weight of the second and third largest fund types in the 1980s—the
life funds and the unit-linked life and pension funds—has diminished over the last
decade or so. The pooled pension funds have also declined in significance.

8
This is inferred by comparing the performances of the evaluation and ranking periods of the same period
(e.g. the set of funds in the evaluation period 1992–1997 with those of the ranking period 1992–1997)—
the latter includes the funds created during the previous ranking period.
64 S.A. Bond, P. Mitchell

Table 2 Distribution of funds by type in the sample

Funds continuously in existence during:

10 years 1987–96 5 years 1997–2001 3 years 2001–2003

Life and general insurance funds 18% 15% 14%


Unit-linked life & pension funds 26% 20% 18%
Segregated pension funds 37% 37% 34%
Pooled pension funds 8% 5% 3%
Unregulated PUTs & other unitised funds 4% 13% 18%
Other (property companies, charities etc.) 7% 10% 13%
TOTAL 100% 100% 100%

To demonstrate the types of funds in the sample, this table shows the composition of the sample by fund
type and for the different sample periods shown in the table. Unlike studies, such as Brounen et al. (2007),
property companies form only a small part of the sample. The largest category of funds is segregated
pension funds. These are separate accounts of pension funds, either managed in-house or by a third party;
pooled pension funds are co-mingled schemes run by fund managers. Unit-linked life and pension funds
are schemes associated with life insurance and pension policies run by the life insurance companies and
whose performance is linked directly to that of the property fund; these contrast with the life and general
funds of the insurance companies where property forms part of a multi-asset exposure and where policy
holders’ returns are calculated differently. Unregulated property unit trusts (PUTs) etc. are collective
investment schemes marketed at institutional and retail investors

By contrast, the unregulated property unit trusts and other unitized funds and, to a
lesser extent, the “other” category (which covers property companies, charities, and
traditional institutions such as the Crown) has grown in importance.

Performance, Alpha and Persistence in UK Property Fund Management

This section assesses the extent of performance persistence in UK real estate funds
and quantifies the magnitude of any out-performance (alpha). It also examines the
characteristics of funds according to performance and the factors behind perfor-
mance. A brief overview of the ranges of annual performance across funds is
presented, followed by an examination of the subsequent performances of cohorts of
funds ranked according to their initial performance.
The bulk of the analysis examines performance over horizons longer than 1 year,
i.e. three, five, and ten-year periods. The first part considers the persistence of
relative performance and the magnitude of such performance and follows the
methodology described in “Methodology”.
The subsequent part considers the persistence of risk-adjusted performance,
having first quantified the extent of Jensen’s alpha across property funds. This is
done because some out-performance may be compensation for risk rather than
reflecting genuine fund manager skill. This analysis draws on the risk models
introduced in “Methodology”. Finally, the characteristics of and factors behind
performance and alpha are explored.
Alpha and Persistence in Real Estate Fund Performance 65

Relative Performance and its Persistence

Figure 1, Panels A–D illustrate the performance of funds grouped according to their
initial ranking. For example, all the funds in the IPD Universe in 1987 are ranked
according their performance in 1987 and allocated to a quartile; the (unweighted)
performance for each quartile in 1987 and for the same set of funds in subsequent
years is then calculated. Note that there will be some attrition in the number of funds.
Similar calculations are undertaken using 1992, 1997 and 2002 as starting points.
The results are presented relative to the unweighted average performance of all
funds.
By definition, the relative performance in the starting year of the top quartile is
relatively strong, with the converse applying to the bottom quartile. For the 1987 and
1992 cohorts, Panels A and B reveal that this initial relative performance very
quickly dissipates, to the extent that each cohort’s performance fluctuates closely
around the average after one or two years. The same is true for the 1982 cohort,
which is not illustrated. In these cases, one year’s strong (or weak) performance

A : Subsequent relative performance of quartile funds B : Subsequent relative performance of quartile funds
in 1987 in 1992
10 Top quartile
8
8 2nd quartile 6
Relative performance (%)

Relative performance (%)

6 3rd quartile 4
4 Bottom quartile 2
2
0
0
-2
-2
-4 Top quartile
-4 2nd quartile
-6 3rd quartile
-6
-8 Bottom quartile
-8
-10 -10
1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 1992 1994 1996 1998 2000 2002 2004 2006

C : Subsequent relative performance of quartile funds D : Subsequent relative performance of quartile funds
in 1997 in 2002
6.0 8.0
Top quartile Top quartile
2nd quartile 6.0 2nd quartile
4.0
Relative performance (%)

Relative performance (%)

3rd quartile
3rd quartile
Bottom quartile 4.0
2.0 Bottom quartile
2.0
0.0
0.0
-2.0
-2.0

-4.0 -4.0

-6.0 -6.0
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2002 2003 2004 2005 2006

Fig. 1 Performance and Persistence in Real Estate Fund Returns: The graphs above show the
performance of fund grouped by quartiles across the sample period. Panel A displays the average
(unweighted) relative-performance of funds grouped into quartiles based on the performance in 1987. In
subsequent years the average performance of the funds based on the 1987 sorting is calculated and the
different colour lines trace the path of the performance of the quartiles. Similarly Panel B displays the
performance of quartiles based on a sorting of funds existing in 1992. Panel C displays the initial
subsequent performance of funds existing in 1997 and sorted into quartiles based on performance in that
year. Interestingly there is evidence that the performance ranking is maintained for the first 4 years of the
sample and this ranking is also found to hold in the final 2 years of the sample. Persistence in initial return
rankings is also found in Panel D based on funds existing in 2002. However, by 2005 the initial fund
performance has converged and the return for funds across quartiles is virtually indistinguishable
66 S.A. Bond, P. Mitchell

tends not to persist. However, this is less the case for the 1997 and 2002 cohorts
where the performance of the top and bottom quartiles seems to persist for longer (2
to 3 years), after which it fluctuates around the average.
There are similar findings at the more extreme ends of the return distribution—top
and bottom decile funds eventually fluctuate around the average fund performance,
reaching such a state quickly in the 1980s and first half of the 1990s but taking
longer from the second half of the 1990s.
This graphical analysis presents visual clues on the extent of persistence in UK
real estate fund performance. The following sections explore this more robustly.

The Persistence of Medium Term Relative Performance

The analysis examines the persistence of relative performance over three, five, and
ten-year horizons. The three and five-year horizons correspond to the periods which
investors typically judge property performance, while the ten-year horizon is a more
demanding test encompassing a number of property market cycles.
The study does not examine in detail persistence over a one-year horizon as it can
be both impractical (given property’s illiquidity) and inefficient (on account of
transaction costs) to rebalance portfolios over such a short horizon. It might also be
expected that appraisal smoothing of asset values may bias one-year persistence
tests. Consistent with this rationale, and the findings of Lee (2003), Figure 1
suggests (particularly for the 1997 and 2002 cohorts) relatively high levels of
persistence over one-year horizons for both the best and the worst performers.
The findings on medium-term performance persistence are presented in varying
degrees of detail. Tables 3 and 4 below present the averages of each set of horizons
(i.e. of all the sets of three, five, and ten-year periods which are examined). Full

Table 3 Proportions remaining in top quantile rankings—relative performance

Top decile in Top quartile in Top half in


both periods both periods both periods

Including expired funds


10 year horizon 12% 16% 26%
All 5 year horizons 16% 30% 44%
All 3 year horizons 15% 32% 49%
Excluding expired funds
10 year horizon 29% 35% 48%
All 5 year horizons 19% 36% 53%
All 3 year horizons 17% 34% 54%

This table compiles information from the individual transition matrices for each ranking and evaluation period
across the sample period (covering all ten-year, five-year and three-year horizons). The first column of data
shows the proportion of funds that remain in the top decile for both the ranking and evaluation periods.
Similarly the second column of data shows the percent of funds that remain in the top quartile in the ranking
and evaluation periods, and the final column shows the proportion of funds with above median performance in
each of the ranking and evaluation periods. The first three rows of data present information from the transition
matrices that explicitly accounts for the funds that terminate during the evaluation period. The final three rows
show information only from those funds that remained in operation during the evaluation period
Alpha and Persistence in Real Estate Fund Performance 67

Table 4 Proportions remaining in bottom quantile rankings—relative performance

Bottom decile in Bottom quartile in Bottom half in


both periods both periods both periods

Including expired funds


10 year horizon 0% 5% 21%
All 5 year horizons 8% 20% 41%
All 3 year horizons 13% 22% 45%
Excluding expired funds
10 year horizon 0% 13% 46%
All 5 year horizons 13% 27% 53%
All 3 year horizons 17% 27% 53%

This table compiles information from the individual transition matrices for each ranking and evaluation
periods across the sample period (covering all ten-year, five-year and three-year horizons). The first
column of data shows the proportion of funds that remain in the bottom decile for both the ranking and
evaluation periods. Similarly the second column of data shows the percent of funds that remain in the
bottom quartile in the ranking and evaluations periods, and the final column shows the proportion of funds
with below median performance in each of the ranking and evaluation periods. The first three rows of data
present information from the transition matrices that explicitly accounts for the funds that terminate during
the evaluation period. The final three rows show information only from those funds that remained in
operation during the evaluation period

details for each of the specific three, five, and ten-year periods are available in
supplementary tables from the contact author’s website.
Table 3 shows the proportion of funds in the top 10%, 25% and 50% in the first
three, five, or ten-year period retaining such rankings in the following three, five, or
ten-year period. Table 4 shows the corresponding proportions for the poorest
performing funds.
Tables 3 and 4 show a mixed picture of persistence. The statistical tests emphasise
this. None of the ten-year tests are significant. The overall distributions are
significant in only one (out of four) of the five-year horizons, and in two (out of
seven) of the three-year horizons; at the aggregated level, both the three and the
five-year distributions are significant at the 5% level. Excluding “expired” funds
from the transition probabilities results in none of the five-year horizons being
significant and only one of the three-year horizons being significant; the aggregated
three and five-year horizons remain significant. Only two (out of seven) of the
three-year cross-product ratios are significant at the 5% level, although the aggregated
ratio for the three-year horizons is significant. The indications are strongest for the best
performers over three and five-year horizons where the transition probabilities for
those consistently in the top quantile are relatively high.
Furthermore, Table 3 indicates that those experiencing top-decile performance
have a particularly high probability of repeating such good performance. Persistence
also exists amongst those in the top quartile although this is less striking than
amongst the top decile performers (subsequent checks show that the second decile
performers have a lower probability of staying in the top two deciles than the top
decile performers). There is no evidence of persistence within the top 50% of funds.
68 S.A. Bond, P. Mitchell

Amongst poor performers, Table 4 indicates that evidence of persistence is less


compelling. The proportions continuing to perform poorly are in most cases around
or less than the expected level (for the quantile groups the expected level is 25%
when expired funds are excluded from the analysis); they are also lower than the
proportions (shown in Table 3) persisting with good performance. Over the ten-year
horizon, the suggestion is that the worst categories of “losers” subsequently reverse
their initial poor performance. The impression, therefore, is that persistence in
property fund relative performance exists only amongst the very top performers.
It is difficult to detect any cyclical pattern to persistence, although the better
performing funds in the second half of the 1990s did subsequently display relatively
high levels of persistence. However, this did not continue in the most recent (2001–
2003) three-year cohort, so any evidence of a strengthening in persistence in recent
years is mixed.

The Magnitude of Persistent Relative Out-Performance

Table 5 presents the average relative performance of funds in each performance


quantile, over horizons of ten, five and three years. The first column of each horizon
shows the performance of groups of funds ranked by quantile; the second column
shows the relative performance of the same set of funds in the following (ten, five
and three-year) period. The figures shown are the averages across horizons (i.e. one
set of ten-year horizons, the average of four sets of five-year horizons and the
average of seven sets of three-years horizons).
The most striking observation from the above table is how the subsequent
performances of the top and bottom performing sets of funds converge with the
benchmark. While the funds which previously performed well on average
maintained an advantage in the following period and the poorest performing funds
typically continued to suffer, the magnitudes are modest. The exception is that over

Table 5 Relative annual average performance according to ranking period quantile

Ranking period quantile Horizon averages

10 years 5 years 3 years

Ranking Evaluation Ranking Evaluation Ranking Evaluation

Top decile 3.9 0.0 5.1 0.5 6.5 0.5


Top quartile 2.6 −0.1 3.3 0.4 4.2 0.4
2nd quartile 0.4 −0.1 0.5 0.2 0.7 0.0
3rd quartile −0.6 −0.2 −0.9 −0.3 −1.0 −0.1
Bottom quartile −2.4 0.5 −3.0 −0.2 −3.9 −0.3
Bottom decile −3.3 1.2 −4.1 −0.4 −5.6 −0.4

In addition to showing the top and bottom deciles, this table shows the (unweighted) average performance
of funds by quartile (including top and bottom decile) in both the ranking and evaluation periods for each
performance horizon. The important issue to note is the general convergence of funds toward a similar
value regardless of their performance in the ranking period
Alpha and Persistence in Real Estate Fund Performance 69

the ten-year horizon the lowest performing funds outperformed during the evaluation
period.
An interesting interpretation of the performances in the columns entitled
“evaluation” is that they represent the relative return of investing, costlessly, in the
previous period’s hierarchy of funds. For example, a strategy of investing in the top
quartile group of funds from the previous five-year period would have yielded only
0.4% per annum in excess of the benchmark. It is clear that, after accounting for
transaction costs (which in the UK total 7.5% of capital value for a round-trip), a
strategy of switching investment into a fund which performed well in the previous
period on average will not have paid off.

Risk-Adjusted Performance and its Persistence

The above results provide an analysis of the persistence in property returns not
adjusted for risk. This evidence is important in understanding the dynamics of total
returns generated by fund managers. However, many investors are interested not
only in total returns but in total returns adjusted for risk. “Methodology” considered
possible models that could be used in adjusting for the risk exposures of fund
managers. There is evidence that even simple models can account for a large amount
of the variation in individual fund manager performance. Accordingly, the factor
models identified in “Methodology” are employed in the following analysis.
The estimates of Jensen’s alpha from the ten and five-year models described in
“Methodology” are summarised in the following section. Then, as in the section above,
this information is used to examine the transition between performance quantiles.
The first stage in the analysis applied the four property sector Sharpe model to
two ten-year periods (1987–96 and 1997–2006) to evaluate the performance of UK
real estate fund managers. The four IPD sectors (retail, offices, industrials, and other
property) represent the benchmarks for the model. Each equation is estimated for a
ten-year period. Hence, there is a general caveat that the model has limited degrees
of freedom. Table 6 shows the resulting estimates of alpha over the ranking period
(1987–1996) and the evaluation period (1997–2006). Interestingly, only the top

Table 6 Estimates of 10-year alpha for UK property fund managers

Quartile ranking 1987 to 1996 1997 to 2006

Quartile 1 0.92 1.86


Quartile 2 −0.63 0.51
Quartile 3 −1.49 −0.29
Quartile 4 −3.27 −1.27
Top decile 1.93 2.63
Bottom decile −4.39 −1.74

Using the style model of Sharpe (1992) estimates of fund outperformance (alpha) are shown over two ten-year
horizons. All funds are allocated to quartiles and the (unweighted) average outperformance measure is
computed for each quartile along with the top and bottom decile. In the first period, only the funds in the top
quartile showed evidence of outperformance. In the second ten-year period both the first and second quartiles
recorded positive alpha values
70 S.A. Bond, P. Mitchell

quartile (and decile) funds have an average alpha value that is positive in the first
period. In the second period both the first and second quartiles have an average value
of alpha that is greater than zero.
Table 6 shows the average alpha values for each quartile, which are determined
based on performance levels in the ranking and evaluation period. That is, in each
period the alpha measure is calculated and each fund ranked accordingly. While
some funds will appear in the same quartile for the evaluation period as they did in
the ranking period, many will not.
Figure 2 below tracks the performance of managers in each quartile from the
ranking period to the evaluation period. A strong tendency towards convergence is
noted in the chart. The general ordering of managers by quartile was maintained
(deciles are shown by the darker lines), with the top quartile managers in the ranking
period, as a group, still outperforming the other managers. However the dispersion
of outcomes between groups was markedly reduced.
One problem with using the Sharpe Index model is the requirement to use ten-year
data periods to estimate the model. This period length is chosen to balance the
demands to provide sufficient degrees of freedom with ensuring at least two periods to
analyze a manager’s performance. To allow for greater analysis of manager
performance by period, a set of models using a single index model (the IPD Universe
index) is estimated to obtain a five-year alpha measure. By estimating the models in
five-years periods, a greater number of intervals can be studied with a larger number of
managers included in the sample. The same caveat applies to this analysis as applied
above; estimating an econometric model with only a small number of observations

3.0%

2.0%

1.0%

0.0%
Alpha

-1.0%
Top decile
-2.0% Top quartile
2nd quartile
-3.0%
3rd quartile
-4.0% Bottom quartile
Bottom decile
-5.0%
Ranking period Evaluation period

Fig. 2 Average fund alpha, according to initial (“ranking period”) performance quantile, 10 year horizons.
This chart shows the average fund alpha by quartile in the ranking and evaluation periods. That is, the
(unweighted) average of fund alpha based on the quartile assigned in the ranking period is compared to the
(unweighed) average of the funds in those same quartiles in the second ten-year evaluation period. The alpha
measures are generated from a Sharpe (1992) style model and capture fund performance against four
benchmarks of property-type sector performance. The chart shows considerable convergence of fund
performance between the ranking and evaluation periods. However, there is some evidence that the
general ordering of the initial quartiles is maintained
Alpha and Persistence in Real Estate Fund Performance 71

will give rise to large confidence intervals around each parameter estimate. The
magnitude of alpha is not estimated with a great deal of precision.
Table 7 shows the estimated magnitudes of alpha by quartiles for UK property
fund managers, based on the single index model and using five-year time windows.
The results presented show the average value of alpha ranked for each quartile when
managers are sorted by performance. The orderings established in the ranking
periods are maintained when calculating average alpha values for the corresponding
evaluation periods.
There is some evidence to suggest that managers’ performance, as measured by
alpha, is counter-cyclical. Alpha is higher during the boom periods of 1987–91 and
2002–2006 and lowest for the period covering the early 1990s (i.e. 1992–96). Apart
from the 1992–96 period, the top quartile managers, as a group, went on to record
the highest average level of performance in the evaluation period (albeit at a lower
level than before). This did not occur with the top decile managers, who had average
performance measures below in the bottom decile group for subsequent performance
in the recent 1992–96 and 1997–2001 evaluation periods. This result must be
carefully interpreted as the number of surviving funds in each decile is small, and the
results may be driven by just one or two adverse cases.
Figure 2 and Table 7 therefore portray a similar pattern of convergence in alpha as
in the earlier analysis of raw, relative performance. The convergence in Jensen’s
alpha between five-year horizons is greater than that for relative performance but
such convergence is less pronounced for Jensen’s alpha than relative performance for
the ten-year horizons.

The Persistence of Risk-Adjusted Performance

To highlight trends in the persistence of risk-adjusted performance, we calculate the


same types of transition matrices used earlier to examine raw relative performance.
Table 8 provides some indication of persistence in alpha. Compared to the transition

Table 7 Estimates of alpha for UK property fund managers based on a single index model

Ranking 1982–1986 sample 1987–1991 sample 1992–1996 sample 1997–2001 sample


period
quantile Ranking Evaluation Ranking Evaluation Ranking Evaluation Ranking Evaluation

Top decile 6.86 0.73 8.76 1.18 4.12 −0.22 6.38 −1.75
Top quartile 4.08 1.00 5.94 0.65 3.05 −0.11 4.55 2.64
2nd quartile 0.46 0.31 1.69 0.20 1.05 0.79 1.79 0.80
3rd quartile −1.63 0.57 −0.24 0.23 −0.38 −0.30 −0.25 0.11
4th quartile −4.55 0.07 −3.03 −0.59 −2.96 1.28 −3.88 0.34
Bottom decile −5.93 −0.10 −4.46 0.26 −4.63 1.58 −6.20 0.53

This table shows the (unweighted) average fund alpha over a five year horizon calculated from a single
factor model. Each fund’s alpha is ranked by quartile, along with the top and bottom decile, in the ranking
period. Each funds’ subsequent performance is traced through to the evaluation period. For example, the
first two columns show the average fund alpha by quartile for the 1982–1986 period. The evaluation
period reports the average alpha of the funds (calculated from the quartile developed during the ranking
period), during the period 1987–1991
72 S.A. Bond, P. Mitchell

Table 8 Transition probabilities for fund manager risk-adjusted performance: 10-year samples, 1987–
1996 to 1997–2006

Evaluation period quartile

Top 2nd 3rd Bottom Expired Total

Ranking period quartile Top 21% 23% 7% 9% 40% 100%


2nd 17% 12% 24% 7% 40% 100%
3rd 7% 14% 9% 12% 58% 100%
Bottom 7% 0% 10% 21% 62% 100%
All 13% 12% 12% 12% 50% 100%

Table 8 shows the transition probabilities for fund manager performance based on 10-year horizons and
alpha derived from the Sharpe index model discussed in “Methodology”. For each fund in the sample,
alpha is estimated and the funds sorted into quartiles, based on the period 1987–1996. The alpha of each
fund is then calculated for the evaluation period (1997–2006), and the funds are sorted into quartiles based
on performance in the evaluation period. From the frequency counts of funds being sorted into new
quartiles given their performance in the ranking period, a transition probability can be calculated. For
example, in this table the first cell shows that a top quartile fund in the ranking period had a 21%
probability of remaining a top quartile fund in the evaluation period. The table also includes a column,
titled expired, to account for funds that operated in the ranking period but did not operate (or ceased to
operating during) the evaluation period

matrix for raw relative performance, the transition probabilities for above- and
below-average funds are higher on a risk-adjusted basis.
For example, the probability that a top quartile manager will remain an above
median manager (first or second quartile performance) is 44%. Compare this to the
probability that a top quartile manager would turn in a below median performance, at
15% and an element of persistency is observed.9 Similarly, the probability of a bottom
quartile manager remaining a bottom quartile manager in the evaluation period is
21%. The probability that a bottom quartile manager will return an above median
performance in the evaluation period is only 7% (44% for a top quartile manager).
Two tests of persistence are reported for the ten-year performance analysis.
Firstly, the cross-product ratio test shows that the performance persistence apparent
on visual inspection of Table 8 is confirmed. The cross-product ratio is 2.84, which
is statistically significant at the 5% level of significance. Secondly, a chi-square test,
is highly significant (chi-square = 27.9, df = 12).
A similar analysis is undertaken for the five-year results based on alpha derived
from a single index model (results not shown). The tests of persistence are conducted
using both the individual five-year periods and also an aggregated set of results
covering all five-year periods. The cross-product ratio test on the aggregated results is
not statistically significant (1.33, p-value=0.08), implying no persistence in manager
performance. However, the chi-square test is significant (chi-square 22.9, df = 12,
p-value=0.02), although it should be noted that when the expired funds are excluded
from the test the results change and show no indication of persistence in quartile
rankings (chi-square 11.1, df=9, p-value=0.27). It is also notable that persistence in

9
If there was no persistence, the probabilities of remaining above or below median is 25%. It does not
equal 50% in this calculation as expired funds are included in the analysis.
Alpha and Persistence in Real Estate Fund Performance 73

alpha amongst the very best (top decile) performers over five years is less than in the
analysis of raw, relative performance.
When the statistical tests are conducted on each five-year sub-sample, the results
show some sensitivity to the time frame chosen. There is strong evidence of
persistence in manager performance for funds in the ranking period 1997–2001,
when evaluated over the period 2002–2006. In the earlier periods no test of
persistence is statistically significant at the 5% significance level.
Over five-year horizons, evidence of persistence is therefore weaker for risk-
adjusted returns than for relative performance.

The Attributes and Predictability of Performance and Alpha

The results from the previous section suggest that, at least for the ten-year
performance interval, there is evidence of persistence in alpha but not in relative
performance. For the five-year results, the evidence is even more mixed, with the
conclusion of persistence dependent on the time period chosen. If this information is
to be useful to investors, it would be helpful to know whether there is any way top
performing managers can be identified. In this section we investigate a model of the
predictability of fund alpha and relative performance. To avoid bias from
contemporaneous relationships between the variables in the model we only use
lagged values of the explanatory variables. The model estimated is:

aij ¼ g 0 þ g 1 LCVj1
i
þ g 2 DLFji þ g 3 DSPFji þ g 4 DPPFji þ g 5 DULij þ g 6 DUNji þ g 7 STRUCTji
þ g 8 PROPSCORij1 þ g 9 HIj1
i
þ g 10 EYj1
i
þ g 11 DEVRATj1
i
þ g 12 NETINVj1
i
þ eij

Where:
j a j subscript indicates the time period of a variable. Period j-1 refers
to the 1987 to 1996 interval and j refers to the 1997 to 2006 interval
i is a superscript indicating each property fund selected for analysis
LCV is the natural log of a fund’s capital value (direct & indirects)
averaged over the relevant time period (eg. 1987–1996)
DLF is a dummy variable to indicate that the fund is a Life & General
Insurance Funds (IPD type codes 2 & 3)
DSPF is a dummy variable to indicate that the fund is a Segregated Pension
Funds (IPD type code 4)
DPPF is a dummy variable to indicate that the fund is a Pooled Pension
Funds (IPD type code 5)
DUL is a dummy variable to indicate that the fund is a Unit Linked Life or
Pension Funds (IPD type code 7)
DUN is a dummy variable to indicate that the fund is an Unregulated PUTs
and other unauthorised funds (IPD type codes 6 & 9)
STRUCT is a variable created by IPD representing the return attributed to
allocation to 11 segments (shopping centres, City of London offices,
South East industrial etc) for each fund averaged over the each period
PROPSCOR is a variable created by IPD representing the return attributed to stock
selection for each fund averaged over the period
74 S.A. Bond, P. Mitchell

HI is a Herfindahl index to measure the “investment focus” (or


specialisation) of each fund. The index for each fund is created using
the following formula:
X
N
HIi ¼ Sl2
l¼1

where Sl is the fund’s annual percentage capital value weighting to


sector l averaged over the period
EY is a variable comprising the average all-property equivalent yield for
each fund (averaged over the period). The equivalent yield is the
discount rate, which equates the future income flows to the gross
capital value
DEVRAT is a variable created as the ratio of total development expenditure
divided by the total capital value of each fund. Then take the average
of this ratio over the period examined, e.g. 1987–1996
NETINV is a variable defined as the ratio of net investment divided by the total
capital value of each fund. Then take the average of this ratio over
each of the ten-year periods    
e is a random error term that is N 0; s 2e ; and Cov eij ; ekj ¼ 08i 6¼ k
The variables included in the model are limited to those available in the IPD
database. Management house, individual fund manager, investment process etc and
changes in these might all be factors associated with and predictive of performance, but
it is not possible to incorporate this information due to limitations in the data available.
The model is estimated for the 1997 to 2006 period, but the value of the
explanatory variables all relate to the 1987 to 1996 period. This regression is helpful
if an analyst wishes to determine, on an ex-ante basis, which managers may
outperform in the subsequent ten-year period. Surprisingly the model shows a
reasonable level of explanatory power. Approximately 35% of the variation in fund
alpha (37% for raw performance) is directly attributable to the fund characteristics
established in the earlier period. The model can also be interpreted as a version of
the Hendricks et al. (1993) paper, in which the test of significance of the lagged
alpha term can be viewed as a test of persistence.
Two interesting conclusions are apparent from Table 9. First, the equivalent yield
of the fund in the earlier period is a significant predictor both of fund alpha and of
fund performance in the following period. As mentioned earlier this may indicate an
uncontrolled for risk factor. Second, the IPD property score measure is significant
and negative for alpha (albeit not significant for performance). This may indicate
mean-reverting behaviour in the performance of specific assets.
The regression analysis in this section is repeated using the single-index model
over five-yearly periods (results not shown but available from the contact author).
Generally the five-year regression results do not support the hypothesis that alpha or
performance persists in fund manager returns.10 The equivalent yield of a fund was

10
The lagged value of alpha is only significant (at the 10% level) in one of the four periods considered for
analysis; lagged performance was significant at the 10% level in two out of the four 5-year periods
considered.
Alpha and Persistence in Real Estate Fund Performance 75

Table 9 Determinants of fund alpha/performance, as a function of lagged alpha/performance and lagged


fund characteristics: 1997 to 2006 sample period

Alpha Performance

Coefficient t-ratio Coefficient t-ratio

Intercept 0.47 0.12 6.28 1.20


alpha(−1)/performance(−1) 0.12 1.31 −0.10 0.85
Size (capital value) −0.03 0.20 0.17 0.77
*
Fund type :
Life & general insurance −3.50 4.70 −4.31 4.39
Segregated pension −3.15 4.67 −3.53 3.96
Pooled pension −3.61 4.54 −4.37 4.16
Unit-linked life & pension −3.49 4.79 −4.13 4.29
Unregulated PUTs & other unauthorised −3.71 3.94 −4.38 3.52
IPD attribution scores:
Structure 0.11 0.56 −0.03 0.13
Property −0.20 2.15 −0.16 1.34
Investment focus −0.60 0.23 1.50 0.44
Equivalent yield 0.41 2.48 0.84 3.83
Development exposure 0.86 0.09 −4.45 0.35
Net investment −0.62 0.45 −1.37 0.76
R2 0.35 0.37

Coefficients significant at 5% level shown in bold. The table shows estimation results for a cross-sectional
equation explaining fund performance. Two sets of results are provided, the first uses fund alpha as the
dependent variable (shown in first two columns), the second uses total fund performance as the dependent
variable. In each case the estimated coefficients of the variables are shown along with the t-statistics. All
explanatory variables are measured with respect to the previous ten-year period (1987 to 1996), compared
to the value of the dependent variable (either fund alpha or performance), which is measured for the 1997
to 2006 period. The dummy variable for fund type omits the fund type property companies, traditional
institutions and charities

found to be a consistent predictor both of fund alpha and of fund performance in


subsequent five-year periods. The variable net investment was also found to be a
significant predictor in three of the four five-year periods examined for alpha (less so
for performance). However, the sign of its coefficient is not consistent, with both
negative and positive values being observed.

Conclusion

This paper finds that evidence of systematic medium or long-term out-performance


(alpha) amongst real estate fund managers is not compelling. It is periodic and at
best focused on a small number of funds. Evidence of persistent poor performance is
also not compelling. There is no clear evidence that these tendencies have changed
over the past few years although any developments associated with the recent
expansion in real estate fund management may take time to be realized.
76 S.A. Bond, P. Mitchell

In line with these conclusions, the subsequent performances of top and second-
quartile funds are, on average, less exceptional. For the three and five-year horizons,
there is still out-performance (and alpha) but it is relatively modest. Over 10-year
horizons, the earlier relative performance advantage is totally eroded, whereas for
risk-adjusted alpha it is still there but marginal. This conclusion is tentative, as the
most appropriate way to benchmark fund performance remains an area for further
research.
In investigating whether fund characteristics may explain either risk-adjusted or
relative performance, it was found that the portfolio yield was an important indicator
of future fund performance. However, it is recognised that this variable may proxy
for an unidentified source of risk and this reinforces the need for further research on
the most appropriate form of risk adjustment. A high development exposure for a
fund is not found to impact fund performance in a statistically significant way.
Finally, there are no strong conclusions to emerge about fund manager type. The
preponderance of segregated pension funds (and traditional institutions) with
persistent top-decile performance is interesting and may be signalling something
different about the style of some of this group.

Acknowledgements The research was funded and commissioned under the auspices of the Investment
Property Forum’s Research Programme (2008–2009). The authors acknowledge the cooperation of the
Investment Property Databank, in particular Malcolm Frodsham and Roberto Diaz. The authors are also
thankful for helpful comments received from Katrina Bond, Kevin Chiang, Piet Eichholtz, John Glascock,
Charles Ward, and an anonymous referee, as well as participants at the European Real Estate Society
Annual Conference 2008, the Maastricht-MIT Real Estate Symposium 2008, and the American Real
Estate and Urban Economics Association Annual Meeting 2009.

Appendix Potential Survivorship Bias in the Dataset

This appendix provides additional information about the impact of the exclusion
of funds without a full history in the ranking period. Table 1 provides a break-
down of the data set to show the number of funds omitted from the ranking period
for each of the five and ten-year intervals used in this study because they do not
survive for the complete time of the ranking period (e.g. in the 1987–1991 ranking
period five funds were omitted for this reason). In addition, new funds entering the
sample are not included in the analysis if they are not in existence at the start of the
ranking period; however, if the funds survive for the full length of the evaluation
period they will be included in the subsequent window of analysis.
To explore the impact such exclusions may have on this study we first examine
the average returns of the funds excluded to see if they differ from the funds
remaining in the sample. For each set of ranking and evaluation periods Table 10
shows the average of the funds excluded to assess if there is a systematic bias
associated with the omissions. The first three columns focus on the five-year ranking
periods (1982–1986, 1987–1991, 1992–1996, 1997–2001), with column one
displaying the (unweighted) annual averages for the funds included in the analysis.
The second column displays the annual averages for those funds existing in the first
year of the ranking period but subsequently excluded because they did not survive
for the complete five years of the ranking period (the number of funds involved is
Alpha and Persistence in Real Estate Fund Performance 77

Table 10 Assessing survivorship influences in the sample

Funds in Analysis Ave of funds Annual average Funds in Analysis Ave of funds starting Annual average
(ranking period not surviving of funds (evaluation period after first year of of funds
average) ranking including non- average) ranking with full including new
period surviving funds evaluation history funds omitted

1982 8.5 All survived 8.5 Not applicable Not applicable Not applicable
1983 8.2 8.2
1984 9.8 9.8
1985 9.4 9.4
1986 10.3 10.3
1987 22.6 16.1 22.4 22.0 23.7 22.5
1988 30.6 25.9 30.6 29.7 32.7 30.6
1989 17.1 2.5 17.0 16.5 18.1 17.0
1990 −7.6 −7.6 −7.8 −7.1 −7.6
1991 0.1 0.1 −1.0 2.7 0.1
1992 −0.7 −3.0 −0.9 −0.9 −0.7 −0.9
1993 19.4 23.0 19.9 20.0 19.4 19.9
1994 12.7 6.5 12.2 12.3 12.7 12.4
1995 3.6 −0.9 3.4 3.0 4.8 3.4
1996 9.4 9.4 9.4 9.3 9.4
1997 16.3 14.4 15.7 15.0 17.1 15.5
1998 12.1 10.1 11.6 11.1 13.3 11.6
1999 14.6 13.4 14.4 14.1 14.6 14.3
2000 11.0 11.0 11.0 10.9 11.1 10.9
2001 7.3 7.3 7.2 7.7 7.3
2002 Not applicable Not applicable Not applicable 9.8 Not applicable 9.8
2003 11.0 11.0
2004 18.6 18.6
2005 20.1 20.1
2006 19.2 19.2

The first three columns of this table examine whether the funds not included in the ranking period (due to
expiring before the end of the ranking period), have a systematic impact on the overall average returns of
the funds in the study. Column one shows the (unweighted) average annual return for funds currently
included in the study. Column two displays the (unweighted) average annual return of those funds existing
in the first year of the ranking period but ceasing to exist before the end of the five-year ranking period.
Column three shows the overall (unweighted) average annual return in each year including the funds that
were omitted in column two. On average the change is small but clearly the funds omitted from the study
have lower average return than the funds remaining in the sample. Columns four to six examine the
possible impact from omitting funds from the evaluation period, that were not in existence at the start of
the ranking period. Column four shows the (unweighted) average annual fund return in the evaluation
period for the funds included in the analysis in “Performance, Alpha and Persistence in UK Property Fund
Management”. The (unweighted) average annual return in the evaluation period for new funds that did not
exist at the start of the ranking period (but had a complete history in the evaluation period) is shown in
column five. On average the newer funds have a higher average annual return than the existing funds in
the data sample. The final column shows the average annual fund return including both the existing funds
in the study and the funds omitted because of their start date. Table 1 provides additional information
about the number of funds entering and exiting the data set
78 S.A. Bond, P. Mitchell

shown in Table 1). Inspection of the columns shows that the excluded funds do have
a lower average return than the funds remaining in the analysis. Column three shows
the corrected annual average return for all funds including the previously omitted
funds. Generally the impact of excluding funds without a full history in the ranking
period is minor, as the number of non-surviving funds was small relative to the
overall number of funds during the first part of the sample. The greatest impact is
seen to occur in the late 1990s period (the 1997–2001 horizon). Note that funds that
have a full history for the ranking period but subsequently expire during the
evaluation period have been included in all analyses reported in this paper.
The final three columns of Table 10 assess the impact of the decision to exclude
funds that begin after the first year of the ranking period and have a full history for
the subsequent evaluation period (for example, a fund that commences in 1999 and
remains in operation until 2006 will be excluded from the 1997–2001 ranking
period). Inspection of columns four and five shows that excluding the newer funds
from the analysis may bias downward the average performance in the evaluation
period (as the newer funds have a higher average performance). Column six shows
the corrected fund averages during the ranking period.
The slight difference in fund average performance between columns three and six
reflects the fact that a small number of funds may have existed for a short period of
time outside of the key starting dates for the five-year windows (e.g. a fund that
starts in 1998 and lasts only two years).
Therefore, even though the economic impact of the omissions appears to be small,
there is clearly the potential for bias in the tests of persistence. In particular, the
omission of funds that performed below average in the ranking period, and those that
performance above average in the evaluation period, may bias tests of persistence
downward. To evaluate this potential bias, we focus on the five-year period that is
likely to be most affected by the bias (ranking period 1992–1996 and evaluation
period 1997–2001). In calculating the chi-square test and cross-product ratio test, the
chi-square test showed most impact (rising from 7.19 to 15). However, neither the
revised chi-square test nor the cross-product ratio test (which showed little change),
altered the conclusions drawn from the main part of the study.

References

Bond, S. A., Hwang, S., Mitchell, P., & Satchell, S. E. (2007). Will Private Equity and Hedge Funds
Replace Real Estate in Mixed-Asset Portfolios?. Journal of Portfolio Management, Special Real
Estate Issue, 74–84.
Bond, S. A., & Hwang, S. (2007). Smoothing, nonsynchronous appraisal and cross-sectional aggregation
in real estate price indices. Real Estate Economics, 35, 349–382.
Brounen, D., Eichholtz, P., & Ling, D. C. (2007). Trading intensity and real estate performance. Journal of
Real Estate Finance and Economics, 35, 449–474.
Brown, S., & Goetzmann, W. N. (1995). Performance persistence. Journal of Finance, 50, 679–698.
Busse, J. A., & Irvine, P. J. (2006). Bayesian alphas and mutual fund persistence. Journal of finance, 61,
2251–2288.
Carhart, M. M. (1997). On persistence in mutual fund performance. Journal of Finance, 52, 57–82.
Carhart, M. M., Carpenter, J. N., Lynch, A. W., & Musto, D. K. (2002). Mutual fund survivorship. Journal
of Finance, 15, 1439–1463.
Chan, K. C., Dimmock, S. G., & Lakonishok, J. (2006). Benchmarking Money Manager Performance:
Issues and Evidence, NBER Working Paper 12461.
Alpha and Persistence in Real Estate Fund Performance 79

Chiang, K. C. H., Kozhevnikov, K., Lee, M.-L., & Wisen, C. H. (2008). Further evidence on the performance
of real estate funds: the case of real estate mutual funds. Real Estate Economics, 36, 47–62.
Chun, G. H., Sa-Aadu, J., & Shilling, J. D. (2004). The role of real estate in an institutional investor’s
portfolio revisited. Journal of Real Estate Finance and Economics, 29, 295–320.
Devaney, S. P., Lee, S. L., & Young, M. S. (2007). Serial persistence in individual real estate returns in
the UK. Working Paper, University of Reading.
Eichholtz, P., Kok, N., & Margaritova, A. (2009). Investment Style and Performance in The Global Real
Estate Mutual Fund Market, paper presented at the RERI annual meeting, April 2009.
Gallo, J. G., Lockwood, L. J., & Rutherford, R. C. (2000). Asset allocation and the performance of real
estate mutual funds. Real Estate Economics, 28, 165–184.
Goetzmann, W. N., & Ibbotson, R. G. (1994). Do winners repeat? Patterns in mutual fund return behavior.
Journal of Portfolio Management, 20, 9–18.
Grinblatt, M., & Titman, S. (1989). Mutual fund performance: an analysis of quarterly portfolio holdings.
Journal of Business, 62, 393–416.
Grossman, S. J., & Stiglitz, J. E. (1980). On the impossibility of informationally efficient markets. The
American Economic Review, 70, 393–408.
Hahn, T. C., Geltner, D., & Gerardo-Lietz, N. (2005). Real estate opportunity funds. Journal of Portfolio
Management, Special Real Estate Issue, 143–153.
Hartzell, J. C., Muhlhofer, T., & Titman, S. (2009). Alternative benchmarks for evaluating REIT mutual
fund performance. Real Estate Economics.
Hendricks, D., Patel, J., & Zeckhauser, R. (1993). Hot hands in mutual funds: short-run persistence of
relative performance 1974–1988. Journal of Finance, 48, 93–130.
Investment Property Databank. (2007). The IPD Index Guide, Edition 2.
Jensen, M. C. (1968). The performance of mutual funds in the period 1945–64. Journal of Finance, 23,
389–416.
Kallberg, J., Liu, C. L., & Trzcinka, C. (2000). The value added from investment managers: an
examination of REIT funds. Journal of Financial and Quantitative Analysis, 35, 387–408.
Key, T., & Marcato, G. (2005). Direct investment in real estate: momentum profits and their robustness to
trading costs. Journal of Portfolio Management. Special Real Estate Issue, 55–69.
Kosowski, R., Naik, N. Y., & Teo, M. (2007). Do hedge funds deliver alpha? A Bayesian and bootstrap
analysis. Journal of Financial Economics, 84, 229–264.
Kosowski, R., Timmermann, A., Wermers, R., & White, H. (2006). Can mutual fund “Stars” really pick
stocks? New evidence from a bootstrap analysis. Journal of Finance, 61, 2551–2595.
Lee, S. L., & Ward, C. W. R. (2000). Persistence of UK real estate returns: A Markov Chain analysis,
Working Paper, Department of Land Management, University of Reading.
Lee, S. (2003). The Persistence of Real Estate Fund Performance paper presented at the 19th Annual
American Real Estate Society Meeting, April 2003.
Lin, C. Y., & Yung, K. (2004). Real estate mutual funds: performance and persistence. Journal of Real
Estate Research, 26, 69–93.
Ling, D. C. (2005). A random walk down main street: can experts predict returns on commercial real
estate. Journal of Real Estate Research, 27, 137–154.
Malkeil, B. G., & Saha, A. (2005). Hedge funds: risk and return. Financial Analysts Journal, 61, 80–88.
O’Neal, E., & Page, D. (2000). Real estate mutual funds: abnormal performance and fund characteristics.
Journal of Real Estate Portfolio Management, 6, 239–47.
Sharpe, W. F. (1992). Asset allocation, management style and performance measurement. Journal of
Portfolio Management, 7–19.