market timing and investment selection: evidence...
TRANSCRIPT
Market Timing and Investment Selection:
Evidence from Real Estate Investors∗
Yael V. HochbergKellogg School of Management
Northwestern University & NBER
Tobias MuhlhoferKelley School of Business
Indiana University
September 13, 2012
Abstract
The ability to select outperforming geographic and property-type submarkets is often regardedby practitioners as a primary form of portfolio manager value-added in the commercial real-estate market. In this paper, we examine fund managers’ abilities to generate abnormal profitsthrough selection of property sub-market segments and contrast this to their ability to generateabnormal profits through market timing. Using data on publicly traded REIT portfolios as wellas portfolios of private entities, we find that the vast majority of both public and private portfoliomanagers exhibit little market timing ability. In contrast, portfolio managers exhibit substantialvariation in their ability to successfully select property sub-markets, with nearly half exhibitingpositive selection ability. Both market timing and sub-market selection ability exhibit significantpersistence in the center of the performance distributions, but not in the top deciles.
∗We thank Jim Clayton, Jeff Fisher, Shaun Bond, Jay Hartzell, Sheridan Titman, Russ Wermers, Barney Hartman-Glaser and seminar and conference participants at the Western Finance Association Annual Meetings, the Universityof Texas, the University of Indiana and the RERI Annual Conference for helpful discussions and suggestions. Wethank NCREIF for provision of data on private property holdings. Both authors gratefully acknowledge fundingfrom the Real Estate Research Institute. Hochberg additionally acknowledges funding from the Zell Center for RiskResearch at the Kellogg School of Management. Address correspondence to [email protected] (Hochberg),[email protected] (Muhlhofer). Portions of this research were conducted while Muhlhofer was a visiting facultymember at the University of Texas at Austin.
1 Introduction
Active money management has long had a significant role in the financial services industry. A large
literature, beginning with Jensen (1968), has been dedicated to determining whether the substantial
fees charged by professional portfolio management companies–mutual funds in particular–are offset
by value-added provided by these active management entities. More recently, researchers have begun
to examine similar questions regarding the ability of managers to generate abnormal returns in
alternative asset classes, such as hedge funds, private equity, and venture capital. In this paper, we
examine the ability of portfolio managers to generate abnormal profits in a large alternative asset
market: Commercial Real Estate.
The commercial real estate market represents a significant portion of the investment universe,
rivaling the size of the publicly traded equities market, with the total value of the asset class estimated
at $11 trillion as of the end of 2009 (Florance, Miller, Peng and Spivey (2010)).1 In contrast to the
mutual fund setting, where markets are thought to be relatively efficient and evidence on managerial
value-added is mixed at best, the commercial real-estate market represents a relatively inefficient
market with greater scope for informational advantages. Like many alternative asset classes, the
commercial real estate market involves transactions that primarily occur in relatively illiquid and
opaque private asset markets, and thus may provide greater opportunities for managerial skill, value-
added and informational advantages to lead to abnormal profits.
Here, we examine commercial real estate portfolio managers’ abilities to generate abnormal profits
through two specific forms of value-added ability: the ability to generate abnormal profits through
selection of property sub-market segments, and the ability to generate abnormal profits through
market timing. In the commercial real estate market, the primary investment decisions made by a
portfolio manager are typically regarded as the choice of geographic or property-type submarkets in
which they invest. Thus, portfolio managers’ ability to generate abnormal profits is largely viewed
by practitioners as their ability to select a city (CBSA) and property type in which properties
1By comparison, Wilshire estimates the total market capitalization of US publicly traded equities at $12 trillion atthe same point in time.
1
will outperform the broader commercial real estate market.2 Evidence in the academic literature,
however, suggests market timing may play also a role in portfolio manager performance. Existing
studies of the real estate market suggest that property markets display evidence of predictability
(see e.g. Plazzi, Torous and Valkanov (2010), Liu and Mei (1992, 1994), Barkham and Geltner
(1995), Case and Shiller (1990), Case and Quigley (1991)). Furthermore, studies employing simulated
technical trading strategies suggest that market timing profits can be made in the real property
market (e.g. Geltner and Mei (1995) and Muhlhofer (2012)).
To assess portfolio managers’ abilities to generate abnormal profits, we classify their holdings
and trades by geographic and property-type segment. We then compute weighted differences in
returns resulting from exposure to property sub-markets either across time (to assess market timing
ability) or relative to broader market benchmarks (to assess sub-market selection ability).3 Our
timing measure captures a portfolio manager’s tendency to tilt the composition of their broader
portfolio towards certain sub-markets when these outperform and to tilt the portfolio away from
these submarkets when they underperform. Our selection measure captures a portfolio manager’s
ability to consistently select submarket categories (or ‘styles’) which outperform broader benchmarks.
Our analysis is thus similar in spirit to the style-based analysis of Fung and Hsieh (2002, 2004), who
perform such an analysis for hedge funds.4
We utilize a complete dataset of property trades and holdings by institutional-grade REITs,
augmented by a dataset of property trades made by portfolio managers of private entities, such as
commingled real-estate funds, who have legally committed to disclose complete trading information
2Evidence for this statement can be found, for example, in the annual reports and 10-K filings of REITs. Asan example, consider Simon Properties (currently the largest REIT) and its 10-K for the year 2010. Simon, in itsportfolio description (p.13), characterizes its investment choices primarily by subtype and location. In its propertytable (pp.14–32), once again the primary attributes for the firm’s investment properties are size and location (CBSA).The discussion of the company’s development pipeline (starting p. 81) also characterizes investment choices exclusivelyby city. Similarly, Camden, a large apartment REIT, lists on page 6 of its 2010 annual report, some highlights of itsportfolio. In this listing, investment choices are only defined by city and state (i.e. CBSA). Property type is superfluous,as Camden only invests in apartment complexes. Starting, page 10 of its 10-K filing for the same year, in the “PropertyTable”, all investment choices are characterized primarily by size and location.
3Our measures are similar in nature to those of Daniel, Grinblatt, Titman and Wermers (1997), utilizing sub-marketreturns where DGTW employ individual investment returns and using a broader style benchmark return where DGTWemploy a characteristic benchmark.
4Fung and Hsieh (2002, 2004) argue that investment classification along style-based lines (here, sub-markets) isboth better-suited and economically warranted for alternative asset classes relative to analysis using the common assetpricing factors, as is frequently done for mutual funds.
2
to a private data collector under a strict non-disclosure agreement. We are thus able to identify
and analyze individual real estate property holdings for a large set of public and private portfolio
managers.5 We supplement the holdings data with commonly-used benchmark return series for
commercial real-estate geographic and property-type segments at varying levels of aggregation –
CBSA-level, state-level, divisional-level, regional-level and U.S. national-level, each interacted with
property type classes. Our data represent approximately half of the equity invested into the asset
class as a whole, with the remaning half encompassing, to a large extent, sub-institutional grade
properties and owner-occupiers who do not engage in delegated portfolio management.
Directly evaluating the portfolio of property holdings of REITs and private investors allows a
more micro-level of analysis than the oft used analysis of mutual funds of REITs enables (see e.g.
Kallberg, Liu and Trzcinka (2000), Hartzell, Muhlhofer and Titman (2010)), in that it allows us to
paint an exhaustive picture of the actual investment decisions in the privately traded commercial
real estate market. As we are able to observe the individual property characteristics, we can readily
calculate the portfolio weights of each geographic and property-type submarket in the portfolio,
and select benchmarks that are appropriate for computing timing and selection ability measures.
Knowing the timing of individual property transactions also allows us to more accurately compute
portfolio weights across time.
Consistent with the practitioner view, we find that the vast majority of both public and private
portfolio managers exhibit little market timing ability, little or even negative ability to successfully
time their investments vis a vis the market, regardless of the aggregation level of benchmark specified.
A small number of top quartile managers, however, do appear to possess statistically significant ability
to time the market at all levels of benchmark specialization. Both private and public managers, on
the other hand, exhibit substantial dispersion in their ability to select outperforming submarkets. In
the top half of the distribution, portfolio managers do appear to positively enhance returns through
submarket selection. Submarket selection ability and market timing ability, however, appear to be
5As we do not have reliable individual property returns, we cannot attribute the fraction of an individual propertyreturn that is due to the sub-market. However, prior work has shown that individual property returns are similar tothe submarket returns to a large extent. See e.g. Crane and Hartzell (2007).
3
negatively correlated, suggesting that managerial resources devoted to adding value through selection
of better-performing property submarket classes may come at the expense of market timing.
Market timing ability appears to be mildly persistent, for lags of up to two years, reverting over
a three year horizon. This persistence in timing ability is stronger when timing ability is measured
versus benchmarks with a higher level of aggregation. Persistence is weaker, however, when measured
using rank persistence measures, where we find widespread persistence only over one year. When
we examine the permanence of managers in the top quartile and the top decile of performance over
time, i.e., the extent to which managers remain in the top quartile or decile, we find that only about
one quarter of top-quartile managers remain in the top quartile over multiple years, and only around
one tenth of top-decile managers remain in the top decile for multiple years.
In contrast, we find strong positive persistence in submarket selection ability, especially in rank-
ings, over a one- and two-year horizon. Over a three-year horizon, persistence is also present when
measuring submarket selection ability versus benchmarks with a high level of aggregation. However,
as is the case for the timing measures, the degree of permanence in the top market quartile and decile
are also low for submarket selection ability. Unlike with timing, however, due to the nature of the
distribution of selection performance across managers, such permanence is not necessary in order to
identify outperforming managers ex-ante based on prior returns. These patterns of persistence are
reinforced when examining the stability and skewness of manager rank decile transition matrices.
When we examine the returns to a trading strategy that allocates capital in each year to all
portfolio managers ranked in the top decile of timing or submarket selection in the prior year, we find
that investing in managers in the top decile of timing ability produces significant negative returns. In
contrast, significant positive returns on the order of 0.24% to 2.62% per year can be obtained from
following the trading strategy of investing in the top decile of managers by sub-market selection
measure. For public REIT managers, these one-year forward returns reflect the managers’ abilities
to select property types rather than their ability to select specific geographic areas for those property
types.
Finally, we perform a cross-sectional analysis in an attempt to identify portfolio characteristics
4
that are associated with higher timing or submarket selection ability. When we regress the timing and
selection measures on manager and portfolio characteristics, it appears that smaller portfolios with
less turnover are associated with higher timing ability for public managers, while for private managers,
larger portfolios with higher turnover appear to be associated with higher timing ability. In contrast,
there is no similar clear association between most portfolio characteristics and selection performance,
suggesting that idiosyncratic ability plays a larger role in this respect. Public managers with higher
submarket selection ability are characterized by smaller portfolios with higher turnover, however
we observe no consistent patterns of portfolio characteristics associated with private managers with
higher submarket selection ability.
Despite differences in taxation and trading restrictions for private and public portfolio managers
that ex-ante may have been expected to lead to differences in investment patterns and decisions,
we note that we observe little difference between the timing and selection abilities of private and
public portfolio managers. Overall, our findings suggest that in a market such as the commercial real
estate market, in which transactions are often costly and time-consuming,6 both REIT and private
fund managers appear to generate value primarily through asset subclass selection. Furthermore, it
appears that using information on past performance, profitable portfolio allocations may be made
based on this knowledge.
Our work contributes to the large literature exploring the generation of abnormal returns by port-
folio managers. A large literature, beginning with Jensen (1968), has explored the ability of portfolio
managers to systematically pick stocks and time their investments so as to generate abnormal re-
turns and justify the fees and expenses of active money management.7 Our work contributes to an
emerging literature that attempts to generate a systematic view of how potential trading profits are
made in alternative asset markets (see e.g. Kaplan and Schoar (2005) in private equity and venture
capital markets and Bond and Mitchell (2010) in real estate). In contrast to existing studies of other
6Commercial property practitioners regard three to twelve months to complete a property sale as reasonable.7Abnormal profits (or the lack thereof) for mutual funds in the stock market have been studied extensively in the
literature (see e.g. Jensen (1968, 1969), Brown and Goetzmann (1995), Gruber (1996), Carhart (1997)). Managers’ability to select individual investments versus benchmarks and to time the market have been studied by Daniel et al.(1997) as well as e.g. Wermers (2003), Kacperczyk, Sialm and Zheng (2005).
5
alternative asset markets, which are often limited by data availability8, the real estate transaction
data we employ in this study allows us to conduct a detailed analysis of manager returns in such
markets. While contemporaneous studies such as Bond and Mitchell (2010) also use private real es-
tate fund data to assess outperformance, their work focuses on performance measures such as overall
alpha, whereas our analysis assesses the timing and submarket selection components that may drive
such alpha.
The remainder of this paper is structured as follows. Section 2 describes the data we employ for
our analysis. Section 3 details our methodology as well as our empirical findings. Section 4 discusses
and concludes.
2 Data
We focus our analysis on the institutional grade commericial real estate market, a segment valued at
$3-4 trillion in 2009. The data for our analysis are obtained from three primary data sources. Prop-
erty transaction data for REIT portfolio managers are obtained from SNL Financial, which aggregates
data from 10-K and 10-Q reports of all publicly traded REITs that are considered institutional-grade.
The SNL Financial DataSource dataset provides comprehensive coverage of corporate, market, and
financial data on institutional-grade publicly traded REITs and selected privately held REITs, and
REOCs (Real Estate Operating Companies). The SNL data contains accounting variables for each
firm, as well as a listing of properties held in each firm’s portfolio, which we use for this study. For
each property, the dataset lists a variety of property characteristics, as well as which REIT bought
and sold the property and the dates for these transactions. By aggregating across these properties
on a firm-by-firm basis in any particular time period, we can compute a REIT’s fractional exposure
to particular sets of characteristics such as property type and geographic segment. The SNL REIT
sample runs from Q2 1995 through Q4 2008.
Property transactions data for private real estate portfolio managers are obtained from the Na-
8Data limitations in VC and PE include such difficulties as being able to observe only venture-capital financed firmsthat went public, having to rely on voluntarily reported investment returns, or by being forced to use other indirectpublic-market related measures to infer information about the more inefficient private market.
6
tional Council of Real Estate Investment Fiduciaries (NCREIF), which collects transaction-level data
for private entities (primarily pension funds). While membership in NCREIF (and thus reporting
of transactions to NCREIF) is voluntary, inclusion in NCREIF’s database is considered desirable
and prestigious on the part of private managers. NCREIF’s stated policy is to only report data
on high-grade institutional-quality commercial real estate. As a result, inclusion of one’s property
transactions in NCREIF’s database and indices is viewed as confirming a level of quality on the
included investor. As a result, most eligible managers choose to become members of NCREIF, and
thus subject themselves to quarterly reporting of transactions. NCREIF membership constitutes a
long-term contract and commitment, and once included, it is not possible for an investor to report
performance only in certain quarters and not in others; the investor is contractually obligated to
report all transactions going forward. Data reported by NCREIF members to NCREIF is protected
by a strict non-disclosure agreement.9 Thus, manipulating performance numbers is viewed as inef-
fective, as it cannot help the investor signal quality beyond membership itself. As a result, NCREIF
members are both willing and able to fully and confidentially report this data to NCREIF. This
arrangement gives us the opportunity to examine trades in a large private asset market, in a more
complete and unbiased way than the data used in past studies on other alternative asset classes. The
NCREIF sample runs from Q1 1978 through Q2 2010. We note that all our analysis is robust within
sub-periods and to excluding the portions of the NCREIF sample prior to 1995:Q2 when the REIT
sample begins.
For our submarket segment returns and benchmarks, we use real estate market returns at var-
ious levels of aggregation, which are obtained from the National Property Index (NPI) series (also
compiled by NCREIF). The NCREIF NPI is considered the de-facto standard performance index for
investible US commercial real estate.10 Index series are available on a national level, as well as dis-
aggregated by region, division, state, CBSA, property type, property sub-type, and by interactions
of the property type and geographic sub-categories.11
9As academic researchers, we are given access to NCREIF’s raw data under the same non-disclosure agreement.10NCREIF gathers data on the property investments of private institutions for the purpose of constructing the NPI.11Technically, the performance of the NPI indices is based on the trading performance of all NCREIF members
7
Our data allow for many levels of disaggregation at both the geographical and property type
levels, as well as their interactions. While this creates many degrees of freedom for analysis, it
allows us a very complete view of how value may be generated by managers through market timing
or submarket selection. NCREIF subdivides property types into subtype for all types, while SNL
does not provide certain subtypes for some property types. Wherever SNL data does not provide a
property subtype, we employ the property type benchmark (i.e. one level higher of aggregation) in
place of a subtype benchmark. Table 1 describes the breakdown of property types and sub-types as
well as geographical regions and sub-regions for our REIT and private manager samples. Our data
can additionally be disaggregated to the state and Core-Based Statistical Area (CBSA) level (for
brevity, we do not detail State and CBSA-level breakdowns in the table).
There are five major property type categories in our datasets: Apartment, Hotel, Industrial,
Office and Retail, with all but hotel broken down into two to eight further subtypes (e.g. Apartment:
Garden, Apartment: High-rise and Apartment: Low-rise). Additionally, properties are classified as
belonging to one of four Regions: East, Midwest, South and West, which in turn are broken down
first into two Divisions each (e.g. East: Mideast and East: Northeast) and then further by State
and CBSA (not detailed for brevity). For each property type and subtype as well as each Region
and Division, the table details the number of unique properties transacted in by REITs and private
portfolio managers.
Table 2 presents summary statistics for the two property datasets. The table presents time-
series statistics of quarterly holdings. In the upper panel, we present distributional statistics for the
publicly traded REIT portfolios, and in the lower panel, distributional statistics for the NCREIF
private portfolio data. Our REIT portfolio sample contains 210 portfolio managers transacting in
11,674 properties. In a given quarter, the average (median) portfolio manager holds a portfolio
consisting of 21.46 (14.39) million square feet of property, with an individual average property size of
combined. Thus, upon first glance these indices may not seem to constitute a passive benchmark. However, NCREIF’suniverse covers such a large portion of privately-held institutional-grade commercial real estate that when aggregated,this trading performance essentially shows the entire market’s transactions. It is therefore reasonable to view thesebenchmarks as passive in like manner to a stock index, which ultimately also reflects the aggregate trading behaviorof the market. This view of the NPI indices as passive benchmarks is also generally held by market practitioners.
8
182,630 (114,870) square feet, spread across an average (median) of 11.37 (8) CBSAs. The average
(median) manager in a given quarter holds 1.85 (2) types of properties in their portfolio, and 2.93 (2)
sub-types. In a given quarter, our REIT portfolio managers have an average geographic concentration
(as measured by Hirschman-Herfindahl Index or HHI) ranging from 0.5 (CBSA) to 0.72 (Region)
depending on level of geographic disaggregation, and an average property type concentration (HHI)
of 0.78 by property subtype and 0.87 by property type. The average (median) manager is present in
the sample for 8.22 (8) years of the 1995 to 2008 sample period.12
In contrast, private portfolio managers hold somewhat larger portfolios by square feet, also con-
sisting of somewhat larger individual properties, and invest in a larger number of geographical areas
and property types. Our NCREIF private portfolio sample contains 136 portfolio managers transact-
ing in 18,115 properties. In a given quarter, the average (median) portfolio manager holds a portfolio
consisting of 42.06 (29.78) million square feet of property, with an individual average property size of
268,200 (173,980) square feet, spread across an average (median) of 30.94 (22) Core-Based Statistical
Areas (CBSAs). The average (median) private portfolio manager in a given quarter holds 3.35 (4)
types of properties in their portfolio, and 7.69 (6) sub-types. In a given quarter, our private portfolio
managers have an average geographic concentration (as measured by Hirschman-Herfindahl Index
(HHI)) ranging from 0.26 (CBSA) to 0.51 (Region) depending on level of geographic disaggregation,
and an average property type concentration (HHI) of 0.49 by property subtype and 0.62 by property
type. The average (median) private portfolio manager is present in the sample for 10.84 (9.25) years
of the 1978 to 2010 sample period. Our dataset thus encompasses two somewhat different sets of
players within this market, adding to the richness and breadth of our study.
While both our underlying datasets allow us to calculate submarket exposures by square footage,
we do not observe market valuations of individual properties for REIT holdings, and therefore cannot
calculate exposures by property value for these managers. For the private portfolio manager sample,
we do observe quarterly estimates of market value for each property held. We note that all the results
12Throughout this study we use the term “manager” to refer to entire organizations (REITs or private funds) incharge of portfolio management. Our data does not allow us to see the turnover of actual management teams withinorganizations.
9
reported hereafter for our private manager sample are robust to the use of value weights rather than
square footage weights. We discuss this issue further in Section 3.5.
3 Empirical Analysis
To identify managerial ability to generate abnormal profits through submarket selection and timing,
we begin by classifying the holdings and trades of portfolio managers in terms of geographic location,
property class, and size (square footage). We use this classification to characterize each portfolio
manager’s holdings as part of a submarket investment category (i.e. style) and construct measures
to evaluate the manager’s ability to generate value within- or across categories over time. Our
approach is similar in spirit to the asset-based style factors analysis of Fung and Hsieh (2002, 2004),
who show that an investment classification and performance assessment along such lines is better
suited and more economically warranted for alternative investments than an analysis using common
asset pricing factors.13
Data for privately traded commercial Real Estate suffers from well-known issues and biases related
both to the market’s low number of trades and to the use of appraisals in its analysis. As is widely
documented in the real estate literature (see e.g. Geltner (1991), Clayton, Geltner and Hamilton
(2001)), the resulting smoothing problems primarily lead to errors estimating second moments and
co-moments of returns, while leaving longer-term first moments largely unaffected in multiple period
analysis.14 The lack of reliable second moments makes parametric, regression-based analysis of
investment performance (along the lines of Carhart (1997), Treynor and Mazuy (1966) or Henriksson
and Merton (1981)) infeasible. Instead, we resort to a holdings-based, non-parametric approach,
which consists of constructing weighted sums of outperformance across a manager’s portfolio at
each point in time.15 This outperformance can occur across time, within a style choice (Market
13Further discussion of the appropriateness of various factor types for hedge funds abound in the hedge fund literature.See, for example, Titman and Tiu (2011), for a discussion of the appropriateness of different types of factor model forthis asset class, as well as a review of the literature that treats this question.
14Stale pricing issues have also been examined in other asset classes, such as the literature examining bond fundperformance (for example Chen, Ferson and Peters (2010)).
15Non-parametric, holdings-based procedures of performance evaluation are also used in other literatures on manage-rial value-added, such as the mutual fund literature, as an alternative to return-based factor models. See, for example
10
Timing), or cross-sectionally between the performance of a manager’s actual investment choice and
the performance of a larger encompassing style portfolio (Market Selection).
3.1 Market Timing
To measure managers’ ability to time entry and exit from a style, we use a Market Timing measure,
defined as follows:
MTt =
N∑j=1
(wj,t−1Rsj ,t − wj,t−5Rsj ,t−4), (1)
where wj,τ is a fund’s fractional exposure to property j at the end of quarter τ , and Rsj ,τ is the return
to a passive portfolio that mirrors the broad investment style sj to which property j belongs.16 This
measure compares the actual style-choice induced return component from a particular submarket
exposure in a particular quarter to the style-choice induced return component that was earned by
the portfolio manager’s exposure to this submarket a year earlier. The measure will be positive for
any time period in which a fund’s weighted return derived from exposure to a particular style exceeds
the weighted returns from that fund’s exposure to this style a year earlier. If the manager increases
portfolio exposure to a style in an upturn and decreases exposure in a downturn, such positive timing
ability will be captured by the MT measure.17
To implement this measure empirically, we begin by observing the properties held by each portfo-
lio manager in the dataset (NCREIF members or public REIT managers) at the end of each quarter.
Daniel et al. (1997) whose methodology resembles ours in terms of mechanics. Other studies, such as Jiang, Yao andYu (2007) also argue in favor of using non-parametric, holdings-based techniques in such a setting.
16Employing non-parametric techniques of the nature of Equation 1 has an additional advantage in our setting. Asnoted before, commercial property returns data is likely to suffer from appraisal smoothing which overstates autocor-relation in returns. Concerns arising from this excessive autocorrelation, however, should be alleviated in performancemeasures involving lagged differencing of the returns series, as the procedure should largely remove such overstatedpersistence.
17Methodologically, this approach resembles the measures constructed by Daniel, Grinblatt, Titman and Wermers(1997) to measure characteristic timing in the mutual fund market. Here, as in DGTW, we employ a weightedsum of outperformance across time which is constructed for a portfolio in each time period. However, economically,our approach differs markedly from that of Daniel et al. (1997), as their study measures performance over investmentcharacteristics, defined along the lines of asset pricing factors, while we measure outperformance of particular submarket(i.e. style) choices over time.
11
For each property, we observe size of the holding (in square feet) as well as class (i.e. type and
sub-type) and geographic location. To compute the fractional exposure weights for Equation 1, we
use the fraction of the manager’s total square footage under management in a particular quarter
which is constituted by properties in a particular submarket (i.e. the sum of individual property
square footages held by the portfolio manager in that submarket divided by total portfolio square
footage). For our submarket returns, we employ NCREIF’s National Property Index (NPI) total
return indices at the lowest levels of aggregation. For example, if a manager owned office buildings
in Chicago’s Central Business District (CBD) in the first quarter of 2006, the relevant return to the
passive portfolio that mirrors the most narrowly-defined associated submarket would be the total
return to NCREIF’s Chicago CBD Office sub-index for the quarter. Summing up weights across all
properties managed by this manager in the particular quarter and sub-market yields the manager’s
total fractional exposure to this style, which is multiplied by the return to the relevant NCREIF sub-
index to generate the weighted return for that style in that quarter. Thus, if Chicago CBD Office
property in the quarter represented 25% of the manager’s total property holdings in that quarter,
we then multiply the 25% fractional exposure by the return of the Chicago CBD Office submarket in
that quarter. Analogously, we construct the fractional exposure and sub-market return for the prior
year, and measure the weighted return the same way.
Using this approach, and the example in the preceding paragraph, a high positive value would be
generated by the manager’s decision to increase portfolio exposure to Chicago CBD Office ahead of
a rise in this market, and/or decrease exposure ahead of a slump in the Chicago CBD Office market.
This would then be considered positive timing ability with respect to the Chicago CBD Office market
overall. We proceed analogously for all other styles (property geography and class sub-markets) to
which the manager’s portfolio has exposure. The sum across all properties and thereby all sub-
markets yields the manager’s MTt measure for that quarter. We repeat this procedure for each
quarter the manager appears in our dataset, and then compute time-series statistics by manager.
As our dataset contains benchmarks for various levels of aggregation at the property geogra-
phy/class level, it affords us the ability to examine timing ability at a variety of levels of specializa-
12
tion. Continuing the prior example, while a Chicago CBD office property could, economically, be
a bet on the Chicago Office market, it could also be considered as part of a more general bet on
the overall Chicagoland Commercial Property Market, a bet on the Midwest Office Market, Midwest
Commercial Property Market, or a bet on the nationwide Office market, etc. We thus construct our
MT measures using multiple levels of aggregation in our style benchmarks. At the geography level,
we use property portfolio index returns for the CBSA level, the state level, the divisional level, the
regional level and at the whole national level. We additionally then interact each geographical level of
specialization with property type and sub-type to indicate property class. This gives us 15 variants of
the MT measure, each constructed using portfolio weights at different levels of aggregation, for each
quarter of the sample. We then additionally compute a time-series average MT for each manager
at each level of aggregation. To enable calculation of reliable t-statistics, we include all manager
observations where the manager is present in the time series data for at least 12 quarters.18 Note
that measuring MT against the National-level benchmark (which is an average performance of the
entire commercial property market) would measure the value generated by a manager’s moving funds
into and out of commercial property as a whole. Because we do not have data on non-real-estate
holdings for the entities we examine (in as far as this is even meaningful), we are unable to assess
performance along this dimension. Therefore, we do not calculate measures related to market timing
with respect to the National benchmark.
Table 3 presents distributional statistics of the time-series average MT s from the cross-section
of portfolio managers in the public and private samples. To obtain the statistics in the table, we
first compute the MTt measure described in Equation (1) for each level of aggregation for every
manager in every quarter. For each manager, within each level of aggregation, we then compute a
time-series average over the quarters for which the manager is active in the sample, to obtain a single
average MT statistic per manager. The table displays the mean, standard deviation, minimum, first
quartile, median, third quartile and maximum of these measures across managers. We compute
18In similar exercises using weekly data, Hartzell et al. (2010) restrict analysis to observations where a manager ispresent in the time series data for at least 24 months. Our results are robust to imposing this more strict requirement.If we eliminate the restriction completely, we obtain similar means, medians and standard deviations for the sample.
13
these measures relative to benchmarks at both the geographical and property type levels, as well as
interactions between the two. For ease of interpretation, these distributional statistics are illustrated
in Figure 1. The dark line at the middle of each box represents the median of the cross-section of
MT s. The boxes represent the inter-quartile spread of the distribution, while the whiskers demarcate
1.5 times the interquartile range from the edge of box. The circles represent outliers that do not fall
within the whiskers.
As is apparent both from the table and the figure, neither the public nor private portfolio man-
agers appear to exhibit particular skill at timing versus the various levels of benchmarks. If anything,
both private and public managers appear to exhibit negative timing ability with respect to our style
portfolios at essentially all levels of specialization. From the bottom of the distribution up to and
including the third quartile, we observe that the point estimates for managers’ mean MT measures
are negative. Only in the top quartile of the distribution are there managers with positive market
timing ability, with the top managers consistently able to positively time markets, measured at all
levels of specialization.
As an illustrative example, consider timing as measured against the State/Type benchmark for
private portfolio managers. The worst portfolio manager in the sample exhibits a Market Timing
measure of -3.21% per annum, with the 25th percentile of managers logging in a -0.84% per annum
versus the benchmark. The median manager earns -0.56% per annum due to timing, and the 75th
percentile manager exhibits a timing ability that accords him -0.39% per annum. The manager
most successful at timing in our sample earned a mere 1.07% per annum from timing. Depending
on the benchmark against which the ability to time individual submarkets is measured, the top
portfolio managers (public or private) earn between 0.71% per annum (CBSA/Type for REITs) to
1.4% per annum (CBSA/Type for private portfolios) due to ability to time investment styles. The
final statistic reported for each benchmark level is the fraction of the cross-section of managers who
obtain a positive point estimate for their average MT performance. This ranges from 18.6% (for
REIT managers at the Divisional/Type level) to 7.4% (for private managers at the State/Type,
CBSA/Type, and CBSA/Subtype level). In general, the fraction with positive point estimates tends
14
to be slightly higher among REITs than private portfolio managers.
To better understand whether some managers are able to successfully and significantly time
the market versus the style benchmarks, Table 4 presents distributional statistics for the t-statistic
testing the hypothesis that a manager has zero timing ability against the two-sided alternative. As
in the previous table, for each manager in each quarter, we compute MTt as in Equation 1, and
then compute a time series average to obtain a single average MT score per manager. Following e.g.
Hartzell et al. (2010), we then compute a t-statistic for the hypothesis that this time-series mean MT
is equal to zero for each manager (such t-statistics are often refereed to as an “Information Ratio”
in the mutual fund literature). The table presents the distributional statistics of these t-statistics.
For ease of interpretation, these distributional statistics are illustrated in Figure 2. The dark line at
the middle of each box represents the median of the cross-section of information ratios. The boxes
represent the inter-quartile spread of the distribution, while the whiskers demarcate 1.5 times the
interquartile range from the edge of box. The circles represent outliers that do not fall within the
whiskers.
Clearly, if fund returns are purely random in the cross-section (with the mean fund generating
zero outperformance), pure statistical chance would on average cause 2.5% of funds to appear to
have statistically positive outperformance at the 5% significance level (assuming a two-tailed test).
Therefore, some care must be taken to distinguish outcomes that may occur purely by chance from
outcomes in which an inference about positive outperformance is warranted. To distinguish true
outperformance from such “False Discoveries”, we follow the procedure suggested by Storey (2002)
and adapted to the purpose of cross-sectional performance studies by Barras, Scaillet and Wermers
(2010), and compute a False Discovery Ratio.19 This measure hinges upon the recognition that the
cross-section of p-values associated with a hypothesis test of zero outperformance should, if fund
19A second common technique applied in the recent literature examining cross-sectional distributions of outperfor-mance is the bootstrap-analysis proposed by Kosowski, Timmermann, Wermers and White (2006). Such a technique,however, is not feasible in our setting, as it is based upon the use of a parametric factor-model analysis to generatethe original cross-sectional distribution of outperformance, and therefore requires reliable second moments. A similarconcern would apply to a procedure such as Fama and French (2010). An additional impediment to this alternativeapproach in our setting is that the number of portfolios in our study is small compared to the samples in mutual-fundstudies, reducing the effectiveness of a bootstrap.
15
returns are entirely random, show a uniform distribution from zero to one. Identifying the existence
of true outperformance is then accomplished by comparing the fraction of managers that show an
apparent statistical outperformance to the fraction of managers that should show such outperfor-
mance if p-values were uniformly distributed. The False Discovery Ratio (FDR+) is calculated as
the fraction of apparent outperformance that we should see purely by chance for the specific em-
pirical distributions we encounter in the data. When the fraction of t-statistics that are above the
5% significance level (“Fraction Sig.”) is in excess of FDR+, this then indicates the existence of
managerial outperformance in excess of what could be expected by chance. We report the positive
False-Discovery Ratio (FDR+) in Table 4.
Looking at the distributional statistics in the table (and as illustrated by the figure), it is apparent
that from the median up to and beyond the third quartile, the timing abilities of private managers are
actually statistically indistinguishable from zero, while they appear to be significantly negative for
public portfolio managers with respect to certain benchmark levels. Below the median, managers’
timing abilities are significantly negative for both private and public portfolio managers.20 More
generally, the distributions of information ratios show a wider dispersion for REITs than for private
portfolios, and also vary slightly more by benchmark level for REITs than for private portfolios. It is
apparent, however, that some managers in the upper quartile of both the private and public managers
appear to possess significantly positive timing ability. This is also indicated by “Fraction Sig.”
for each benchmark level, which shows the fraction of managers who exhibit significantly positive
timing ability. This fraction ranges from zero (for REITs at the CBSA/Type and CBSA/Subtype
level, as well as private portfolios at the State/Type level) to 2.9% (for private managers at the
Divisional) level. A comparison of these statistics with FDR+, however, reveals very few style
benchmark levels for which we find more managers with positive outperformance than one could
find due to pure chance. In fact, for REIT portfolio managers, at all benchmark levels, there are
20Studies using parametric approaches in the mutual fund literature, such as Ferson and Schadt (1996), argue thatnegative timing ability does not make economic sense, because an investor could become a good “timer” by simplytrading in the opposite direction of such portfolio managers. Ferson and Schadt (1996) argue that the evidenceof negative timing can come from using inaccurate benchmarks, such as ignoring the effect of public information.However, the arguments in this literature are less applicable in the commercial real estate setting, as properties cannotbe shorted, and holdings and trades in manager portfolios cannot be readily observed.
16
fewer managers that significantly outperform than one should see purely by chance, if there were
no true outperformance. For private portfolios, we find a fraction of portfolios beyond FDR+ that
outperform for all benchmarks at the National level, for all geographic-only benchmarks except State,
and for CBSA/Subtype benchmarks. This suggests that at the very top of the distribution there exist
some private managers who can time markets, generally at least with respect to more aggregated
benchmarks.
The results are surprisingly homogeneous throughout the different levels of benchmark specializa-
tion, suggesting that the distribution of timing abilities is very similar whether we assess managers’
ability to time large national trends or trends within particular submarkets and property types.
There does seem to be, however, a very slight upward shift and a slightly larger dispersion in timing
abilities as we employ more specialized benchmarks. We note, however, that these results need not
necessarily be due to managers’ inherent (lack of) abilities to read markets, but may be caused by
the microstructure of the market in which they act, in which property transactions may involve
extremely prolonged transaction times. It is well known that the commercial property market suffers
from slow execution of transactions and very high transaction costs when compared to other asset
markets. These frictions may be an important driver behind the timing results we observe. The lim-
its of available data, however, do not allow us to ascribe a definite underlying cause to the observed
timing patterns.
3.2 Market Selection
Our second measure captures a portfolio manager’s ability to add value cross-sectionally by making
particularly good submarket investment choices within a more broadly defined style category and
holding those submarket investments for some time. To measure managers’ ability to add value along
this dimension, we use a Market Selection measure, defined as follows:
17
MSt =N∑j=1
wj,t−1(Rsj ,t −RQj ,t) (2)
In this equation, Rsj is the return to the geography and property class submarket portfolio j, while
RQj is the return to a more broadly defined style portfolio to which the submarket belongs. The
weight wj is defined by property square-footage over total portfolio square footage (as in the previous
section) for the properties held by the manager in period t in submarket j.21 This measure will be
positive if a manager invests into geographic and property class submarkets that outperform more
broadly defined markets within a similar style. 22
To continue the example from the previous section, we would attribute the return from all holdings
of Chicago CBD office buildings to the most narrowly defined geographic and class style of Chicago
CBD Office, and thus Rsj for this submarket will be the NCREIF NPI return for Chicago CBD
Office. The return on this submarket would then be compared to the return of a larger market
of which Chicago CBD Office is a subset, for example, the overall Chicago Office market, which
becomes RQj . The manager’s specific choice of Chicago CBD Office within the overall set of possible
Chicago Office properties would have proved to be successful, if the Chicago CBD Office market
outperforms the overall Chicago Office market, and this would generate a positive excess return for
that property. The weighted average of all portfolio exposures at a particular time constitute overall
Market Selection performance for that manager in that quarter.
As is the case for our timing measure, given the wide variety of benchmark aggregations available
to us, we can construct MS by using the returns to ever larger markets for RQj , while preserving
Rsj at the smallest possible style definition. Thus, after the first run described above, we would
compute a version of MS that compares the performance of the Chicago CBD Office market with
21Analogous to the case of our market timing measure, the methodological approach used to construct this submarketselection measure resembles the characteristic selectivity measure of Daniel et al. (1997), in that here too we employa measure that constitutes the weighted sum of outperformance versus a benchmark. As before our approach differseconomically from that of Daniel et al. (1997), however, in that we compare narrower style choices to more broadlydefined ones, as opposed to comparing specific asset choices to asset-pricing characteristics-based benchmarks.
22As with theMT measure, the construction ofMS should, to a large extent, alleviate concerns arising from smoothedproperty portfolio returns, as here, too, we are differencing return series.
18
the Chicagoland Office market, then the Illinois Office market, then the overall Illinois market, and
so forth, once again using all levels of aggregation or specialization, geographically and by class.23 24
Table 5 presents the distributional statistics of the time-series average Market Selection measure
for the cross-sections of public and private portfolio managers. The statistics in this table are
computed analogously to those in Table 3. That is, we first compute the MSt measure described in
Equation (2) for every manager in every quarter. We then compute a time-series average for each
manager over the quarters for which the manager is active in the sample, to obtain a single average
MS statistic per manager.
The table displays the mean, standard deviation, minimum, first quartile, median, 3rd quartile,
and maximum of these measures across managers, as well as the fraction of managers who exhibit
a positive point estimate for average Market Selection. As was the case for the MT measures, we
compute these measures relative to benchmarks at both the geographical and property type levels,
as well as interactions between the two. For ease of interpretation, the distributional statistics are
illustrated in Figure 3. The dark line at the middle of each box represents the median of the cross-
section of MS. The boxes represent the inter-quartile spread of the distribution, while the whiskers
demarcate 1.5 times the interquartile range from the edge of box. The circles represent outliers that
do not fall within the whiskers. As CBSA/Type level returns are used as the most narrowly defined
style to which the broader style benchmarks are compared, the table naturally omits any statistics
for the CBSA/Type- or CBSA/Subtype-level benchmarks.
23Note that the NPI series disaggregated by CBSA and property subtype simultaneously contains a large amountof missing values, due to insufficient portfolio size. We overcome this limitation by defining Rsj as the return to theCBSA/Type portfolio to which property j belongs. This becomes our most narrowly defined style and we constructMS to compare this style choice to larger encompassing styles.
24An alternative to our approach, which evaluates managers’ abilities to select submarkets, would be to evaluate theoutperformance of managers’ selections of individual properties within submarkets (akin to Daniel et al. (1997). Thelack of reliable price or return series for the individual properties held by the managers in our data, however, preventsus from conducting such analysis. Given that these properties are not traded while they are held in a manager’sportfolio, they are also not priced. For REITs, this limitation is absolute, as these firms also do not disclose appraisalsof the properties in their portfolio. While NCREIF members do disclose periodic appraisals, using these would preventus from conducting a clean comparison between private and public managers. However, as argued previously, thechoice of CBSA and property type constitutes the defining characteristics of a commercial property investment and soa portfolio that reflects these choices should capture the concept of narrowly defined choice of style, which is the goalof our analysis. Furthermore, Crane and Hartzell (2007) compare the use of a CBSA/Type portfolio from the NPI tousing direct returns for a particular property where available, and find that the returns derived through the index havea correlation of more than 0.96 with the actual property returns.
19
Both the table and the figure show that the mean and median Market Selection for both types of
managers is very close to zero. Thus, approximately half the managers in our sample have negative
Market Selection measures, and the other half have positive Market Selection measures. This is in
contrast to the results on timing, where positive performance existed only in upper outliers of the
distribution. Also, unlike with the case of Market Timing, the position of the means and medians of
the Market Selection distributions are fairly homogeneous across both public and private managers.
One further pattern that emerges is that at lower levels of geographic aggregation, submarket
selection ability shows more dispersion with respect to the geographic-only benchmark than with
respect to benchmarks disaggregated by both geography and property class. Consider, for example,
the case of public portfolios at the State level. The distributions of all three sets of Market Selection
measures (State, State/Type, and State/Subtype) have means that are very close to zero (about
two basis points per annum outperformance for State, three for State/Type and a half basis point
for State/Subtype). Yet the standard deviation of Market Selection by State only is about 4.3 basis
points per annum, while those with respect to the benchmarks interacted with property class are
about 2.5 basis points per annum, or 60% as high. Similar patterns can be seen by examining
the median and the inter-quartile spread. The regional, divisional, and state level results for both
types of manager illustrate this pattern to some extent. This may suggest that selection of the right
property class within geographic subdivisions is an important part of managerial value added, and
that this component of ability contains a high degree of heterogeneity.
As we did for the MT measure, we next compute t-tests for the hypothesis that a manager has
zero selection ability against the two-sided alternative, and the distributions of t-statistics from these
tests are reported in Table 6. Once again, we compute each manager’s MSt in each quarter as in
Equation (2) and compute a time series average over the entire activity of each manager. Then, we
test the hypothesis of zero selection ability over the manager’s entire period of activity and report
the cross section of t-statistics (“information ratios”). We additionally report FDR+, the positive
False-Discovery Ratio, which shows the fraction of portfolios that would exhibit statistically positive
outperformance if true returns were random, with mean zero, for each distribution. To help with
20
interpretation, these distributions are also presented in Figure 4.
As is apparent from both the table and the figure, unlike the case for the MT measure, the entire
range of MS measures between the first and the third quartile are statistically indistinguishable from
zero at the 5% significance level. However, as is apparent in Figure 4, a t-statistic of around +1.96
is still well within the range covered by the whiskers (the same can be said about a value of −1.96).
This means that, unlike with timing, one does not have to look for extreme positive outliers to find
significant outperformance (or underperformance) along the selection dimension. Comparing the
fraction of managers that shows significant outperformance to the respective FDR+, it is apparent
that in all cases, this fraction exceeds the amount of outperformance that would be expected by mere
chance. These results again are fairly homogeneous across benchmarks and types of manager.
Overall, these results suggest considerable heterogeneity among managers in submarket selection
ability, and that an appreciable fraction of managers (around 10%) have significant ability to create
value through property class selection. For timing, in contrast, most managers exhibit negative
performance, with only the extreme positive outliers showing significantly positive value-added.
In Table 7 we examine the correlation between timing and selection ability. We calculate corre-
lation coefficients over the entire panel dataset (by manager and quarter). Specifically, for manager
m, and MT or MS performance over benchmark i at time t, we report the correlation of all MTm,i,t
with MSm,i,t, together with t-statistics for the hypothesis that the true correlation is zero, against
the two-sided alternative. Overall, manager MS and MT at the quarterly level appear to be signif-
icantly negatively correlated at all levels of benchmark for public REIT managers, and for 10 out
of 12 benchmark-levels for private managers. These negative correlations range from -2% for public
managers when MS and MT are measured versus the lowest-level of aggregation (CBSA) to a high
of -23% for public managers when MS and MT are measured versus the highest level of aggregation
(National/Type). Overall, these correlations suggest that timing ability and selection ability are
negatively correlated, and that managers who exhibit the ability to generate returns through market
selection are still performing poorly at timing their property trades, or may even be selecting at the
21
expense of timing performance.25
3.3 Time-Series Persistence
If some managers display timing or selection ability, a natural question to ask is whether such
abilities are persistent. To answer this question, we examine time-series persistence of the Timing
and Selection measures by manager, examining autocorrelations, rank persistence and permanence
of managers in the top decile or quartile. We then examine the forward returns to investing in the
portfolios of managers who ranked in the top decile on timing and selection ability in the past, and
examine the rank-decile transition matrix stability and skewness.
We begin by investigating autocorrelations in MT and MS measures for each manager over
time, using panel datasets of annualized manager-by-manager MT and MS measures. Table 8
reports these results for MT measures. The table reports the mean MT over the entire panel
with respect to each level of geographic and property-class aggregation, as well as one-year two-
year and three-year autocorrelation of MT measure, for both public and private managers. For each
correlation coefficient we test the null hypothesis that the true correlation is zero against the two-sided
alternative. It is important to note that autocorrelations indicate performance persistence only with
respect to deviations from the mean. Therefore, a positive autocorrelation simply indicates persistent
above-mean or below-mean performance from year to year. Given that the mean MT is negative,
and given the distribution of MT s we report in the previous section, above-mean performance does
not indicate actual positive generation of value through timing.
The autocorrelation estimates in the table suggest significant persistence in timing ability (or lack
thereof) in the real estate market, for both publicly and privately held portfolios. In contrast, in other
literatures that examine the value-added of active money management it is common to find little or
no persistence in the performance of actively-traded portfolios. For both types of portfolio managers,
Table 8 indicates significantly positive autocorrelation over a one-year horizon, for all but the smallest
25Henriksson (1984) argues (in a parametric setting, for mutual funds) that negative correlation between timing andselection may be a sign of mis-specified benchmarks. However, other studies such as Jagannathan and Korajczyk (1986)refute this argument and show that such an outcome is possible even with correctly specified benchmarks.
22
levels of aggregation for REITs and for all levels of aggregation for private portfolios. Over a two-
year horizon, we also find significantly positive autocorrelation of MT values, for all but the smallest
levels of aggregation (State/Subtype, CBSA/Type and CBSA/Subtype for REITs). For a one-year
horizon, the timing performance of private managers tends to have a higher autocorrelation than that
of public managers (around 0.4 with a maximum of 0.461 for the National/Subtype benchmark for
private managers, compared to a maximum value of 0.2594 for REITs, at the Regional benchmark).
For a two-year horizon, private managers still tend to have somewhat higher autocorrelation in timing
performance than public portfolio managers. Overall, autocorrelation coefficients tend to be higher
(and therefore persistence tends to be stronger) when considering timing ability with respect to only
a geographic benchmark, as opposed to a benchmark which interacts geography and property class.
Over a three-year horizon, however, the autocorrelation coefficients are negative for benchmarks
at all levels of aggregation, and significant at least at the 1% level, with values ranging from −0.2961
for REITs over a National/Subtype benchmark, to −0.1118 for Private managers over a CBSA
benchmark, with most values lying between −0.2 and −0.3. This suggests that within three years,
above-average timing will turn into below-average timing and vice versa. It therefore seems that
while the timing strategies and abilities of the managers in this market do persist in the short run,
they are still short-lived and followed by reversals. REIT timing seems to be slightly more variable
over short horizons than private market timing.
Table 9 reports panel means and autocorrelations for the MS measures. First, we note that the
panel means for this measure are much closer to zero than for the MT results, and therefore above-
mean performance in most cases will mean positive Market Selection, and below-mean performance
will mean negative Market Selection. Over a one-year horizon there is strong evidence of positive
persistence; all correlation coefficients are positive and significant at the 1% level, with the lowest
coefficient value of 0.1583 for REITs with respect to a State/Subtype benchmark, the highest of
0.4362 for private managers, with respect to a National benchmark, and most coefficient values
between 0.2 and 0.4.
Over a two-year horizon, however, the evidence for persistence of selection ability is more sparse.
23
For REITs the only positively significant autocorrelation coefficients are with respect to geographic
benchmarks, not interacted with property class. In other words, the data shows that it is only a
REIT manager’s ability to select a property class investment within a certain geographic area that
persists, and not any selection made based upon finer distinctions. In fact at the State/Type and
State/Subtype level, significant reversals begin to show, even at the two-year horizon; these are
also visible at the three-year horizon. Part of these results could be due to outliers, as can be seen
when comparing these to Table 11, so these interpretations must be approached with some caution.
Except for these two coefficients, neither significant persistence nor reversals exist in REITs’ Market
Selection at the three-year horizon.
For private managers, there is strongly significant persistence at the two-year horizon, for all levels
of benchmark aggregation, even when interacted with property class. These results are consistent
with a hypothesis of private investors showing the ability to make positive selections at a more coarse
geographical level, with REIT investors making more targeted property-class plays. At the Regional,
Divisional, State and CBSA level, over a three-year horizon, private investors also show significant
reversals in selective ability.
We next examine the persistence of manager relative rankings in timing and selection ability over
time. For each year we construct a percentile rank for each manager, based on his or her realized
performance with respect to Market Timing and Market Selection over the past year. We then
compute autocorrelations of manager percentile ranks over one-, two-, and three-year horizons.
Table 10 shows the results for rank persistence with respect to Market Timing. Similar to the
persistence results for the MT measure itself, there is strong positive persistence in manager rank
at the one-year horizon, for both public and private portfolios, at the higher levels of benchmark
aggregation, with a top coefficient of 0.2943, for private managers on a National/Type benchmark.
The interpretation of these positive coefficients is that managers tend to not cross the median over
the course of the following year: above-median managers remain above-median and below-median
managers remain below-median. However, unlike for the MT measure itself, the positive autocorre-
lation becomes less prevalent at lower levels of aggregation; for public portfolios this begins at the
24
State level, with no significantly positive coefficients for State or CBSA and their interactions, while
for private portfolios there is no such effect, with positive, albeit smaller rank correlations even at
lower levels of disaggregations. By construction, rank correlations are more robust to outliers than
the direct correlations reported in Table 8. This suggests that, with few exceptions, persistence in
timing ability exists primarily with respect to coarser benchmarks.
At longer time horizons, public portfolios only show weakly positive rank persistence in timing
geographic benchmarks, but no or negative persistence in timing benchmarks that interact geography
with property class. In fact, at the two-year horizon, public portfolios show some degree of reversion
in timing rank, when combined geography-class benchmarks are used, at a wide variety of levels of
aggregation. Private portfolios show more widespread rank persistence with at least weak persistence
at the two- and three-year horizon. Furthermore, private managers show no significant reversion in
ranks at any level and some positive persistence even at this horizon for benchmarks based on only
geography.
A somewhat different picture emerges for Market Selection from Table 11. There appears to
be widespread persistence in Market Selection rank, including, with a few exceptions, at the two-
year horizon, at low levels of benchmark aggregation. This is the case for both public and private
portfolios. Even at the three-year horizon, the larger levels of aggregation (National for both types
of managers and Regional for private managers) exhibit some degree of persistence. For REIT
managers, this is in contrast to the results on persistence in MS itself, which showed little or no
persistence at horizons beyond one year. Once again, the difference here is that these correlations
are more robust to outliers. Given that the median level of MS is close to zero, these results are
encouraging for the task of portfolio formation, as past positive performance may then give some
degree of confidence of such performance in the future.
While the degree of persistence evident in both the autocorrelations and rank auto-correlations
seems encouraging for investors interested in formulating strategies for allocating capital to managers,
these correlation coefficients are primarily indicative of movements about the median, which in the
case of the MT measures is negative. As is shown in Table 4, if an investor is looking for positive
25
managerial timing ability, he will need to find a manager who is at the very top of the distribution.
We thus examine an alternative form of persistence, which we dub permanence, which captures
the extent to which a manager above a particular percentile is likely to remain at or above this
performance percentile in the future. We calculate these permanence measures for both the top
quartile and top decile. 75plusτ,t indicates, out of all managers whose performance rank was at or
above the 75th percentile at time τ , that fraction whose performance is still at or above the 75th
percentile at time t. 90plusτ,t is an analogous measure for the 90th percentile. We construct these
using MT and MS measures at all levels of benchmark aggregation.
The highest values of these measures, reported in Table 10, are for REITs at the National/Type
level, where for 75plus and 90plus, these are 0.42 and 0.36 respectively, over a one-year horizon. The
latter measure, for example, indicates that 36% of managers who were ranked in the top 10% of the
market the previous year, are ranked there in the current year as well. At a two-year and three-year
horizon, the 75plus measures drop to 0.24 and 0.27 respectively and the 90plus measures drop to
0.16 and 0.17 respectively. The overall progressions in both these measures show some degree of
reversion. Typical values are between 0.2 and 0.3 for 75plus and between 0.1 and 0.2 (or even less
than 0.1) for 90plus. This indicates a high degree of turnover in terms of which managers occupy the
top of the performance distribution. This means that at most levels of benchmarks, even if one finds
a manager that generates positive timing over a given year, this finding is unlikely to prove useful
for the purpose of future capital allocation decisions, in that a repeat performance is unlikely.26
While the autocorrelation coefficient itself is already informative for selection ability, as the
median manager has a MS of close to zero, we also examine the 75plus and 90plus measures for
Market Selection rankings. The maximum values found when examining permanence in selection
ability are 0.5 for 75plus for REITs at the National, Divisional, and CBSA level and 0.39 for 90plus for
REITs at the Divisional level. Generally, the range of values for the selection permanence measures
26Despite the fact that the vast majority of these permanence statics are substantially higher than what one mightfind by pure chance, the overall relatively low permanence in the top of the distribution still makes capital allocationsto outperforming managers difficult, due to the performance distributions’ overall low or negative means and medianslocation.
26
are slightly higher than those found in Table 10. However, there still appears to be a high level of
turnover in the top quartile or decile for selection ability as well as timing ability.
Table 12 illustrates the returns to a trading strategy that allocates capital in year t + 1 to all
portfolios ranked in the top decile of MT or MS for year t. For each year, ending at time t, we rank
managers according to annualized MT or MS performance over the previous year, with respect to
each benchmark level i and within their entity type p (i.e. public versus private). We then simulate
(for each i and p) the returns to investing equal amounts of money at time t into each portfolio
that ranked in the top decile in terms of MT or MS performance in the previous year for the same
benchmark and entity type, and holding those investments until the end of year t + 1. The table
presents an equal-weighted average MT or MS return with respect to benchmark i, obtained per
manager over the year.
The top panel of the table presents these forward investment returns for MT , and the bottom for
MS. We calculate the top-decile forward investment return with respect to each level of benchmark
aggregation. The table presents the time-series averages of equal-weighted cross-sectional mean
returns for each year, and t-statistics testing the hypothesis that the time-series average of cross-
sectional means is equal to zero against the two-sided alternative. As can be seen from the table,
investing in year t + 1 in the portfolios of REIT managers who ranked in the top decile on timing
ability in year t returns a statistically significant negative return in 9 out of the 14 levels of benchmark
aggregation; investing in year t+1 in the portfolios of private managers who ranked in the top decile
on timing ability in year t returns a statistically significant negative return in 13 out of the 14 levels
of benchmark aggregation. This negative return ranges from -1.51% when measuring MT versus
the National/Type level benchmark to -5.3% when measuring MT versus the CBSA/Type level
benchmark.
In contrast, investing in year t + 1 in the portfolios of REIT managers who ranked in the top
decile on selection ability in year t returns a statistically significant positive return in 6 out of the 14
levels of benchmark aggregation, and for private managers, in 13 out of the 14 levels of benchmark
aggregation, with positive returns ranging from 0.24% per year to 2.62% per year depending on the
27
level of benchmark aggregation. For REIT managers, at the Regional benchmark level and smaller,
the one-year forward returns also reflect the managers’ ability to select property types in general,
rather than an ability to select geographic areas in which to make these specific property-type bets.
This is in line with findings for many of our other selection results. The returns for the private
managers also show this pattern, but to a lesser extent.
Finally, in Table 13 we examine the stability and skewness of the rank-decile transition matrix
for the managers in our sample, first by MT measure and then by MS measure. For each year
ending at time t, we rank managers according to their annualized MT or MS performance over the
previous year, with respect to each benchmark level i and within their entity type p (public versus
private). We then construct 10 × 10 decile transition matrices (analogous to credit-rating transition
matrices used in fixed-income investments), one for each interaction of type of measure (MT or MS),
benchmark level i, and entity type p. The row index of these matrices represents a manager’s rank
decile for the year t− 1, dt−1, and the column index represents a manager’s rank decile for the year
t, dt. Each entry shows the fraction of managers that transitioned from a given decile dt−1 to a given
decile dt over year t, with respect to a certain measure type, benchmark level, and entity type. All
rows sum to 1.
In these matrices, the diagonal elements show the stability of deciles, that is the fraction of
managers who start and end the year in the same rank decile. The elements to the right of the
diagonal show the fractions of managers who end the year in a higher decile (upgrades), and the
elements to the left of the diagonal show the fractions of managers who end the year in lower deciles
(downgrades). For each decile in each matrix, we construct a measure of skewness, which is defined
as the ratio of upgrades over downgrades, or more specifically the ratio of the sum of elements to
the right of the diagonal, over the sum of elements to the left of the diagonal. Skewness is only
meaningful for interior deciles.27 We report stability and skewness for each decile with respect to
measure type, benchmark level, and entity type, for all years combined.
27These measures are analogous to credit-ratings transition matrix stability and skewness.
28
Table 13 reports stability and skewness from our calculated decile transition matrices.28 As an
illustrative example of the measures, consider the statistics for REIT manager MT calculated versus
the CBSA benchmark (in Panel A). The first of the two lines shows the stability statistic for each
decile. The numbers indicate the fraction of REIT managers who were in each rank decile for MT
at the start of a particular year and remained in that decile over the year. For example, for decile
1 (the worst-performing decile), the stability statistic indicates that 22.7% of managers who were in
the worst-performing decile according to MT with respect to a CBSA benchmark at the beginning
of the year remain in this decile at the end of the year. For decile 5 this fraction is 9.68% and for
decile 10 (the best-performing decile) it is 4.27%.
The second line for each matrix presents the decile skewness, or the total fraction of managers
upgraded out of that decile to a higher one from year t to t+ 1 over the total fraction of managers
downgraded out of that decile to a lower decile. Given that in decile 1, any manager that does
not remain ranked in that decile the following year must, by definition, have been upgraded and in
decile 10 every manager that does not remain in the decile the following year must, by definition,
have been downgraded, we do not report these statistics for the extreme deciles. Continuing the
example above, skewness for decile 2 of public manager MT performance measured against the
CBSA benchmark in Panel A is 7.23. This means that there were 7.23 times as many managers
who transitioned to a higher decile (upgrades) as there were managers who transitioned to a lower
decile (downgrades) in the ranks of REIT manager MT measured over the CBSA benchmark.29 For
decile 5, skewness is 1.38, indicating just a few more upgrades than downgrades, while for decile 9
skewness is .0833, indicating about 12 times (1/.0833) as many downgrades as upgrades. The larger
the number greater than 1, the stronger the upgrade bias, and the smaller the number less than 1,
the stronger the downgrade bias. A skewness measure of 1 indicates the same number of upgrades
as downgrades.
28As with all other results, the panels that contain measures based on MT do not report results with respect to theNational benchmark and the panels that contain measures based on MS do not report results based on benchmarkslower than CBSA, as these are not meaningful. In the former case, this is due to a lack of non-real-estate holdings dataand in the latter to the fact that we use CBSA/Type as the proxy for the property’s actual return.
29Of course, this fraction is conditional upon the manager having moved out of this decile.
29
Overall, Table 13 indicates patterns for all levels of benchmark aggregation that are very similar
to the one described in the example above. For stability of MT -based rankings (Panels A for REITs
and Panel C for private managers), we consistently observe a relatively high level of stability for
the lowest-performing decile (decile 1). These values are, with a few exceptions, between 0.2 and
0.3, with private portfolios showing a slightly higher stability in the bottom decile than public ones.
If movements among deciles were purely random and uniformly distributed, one should observe
values of 0.1 everywhere. Therefore, these results indicate that the worst underperforming managers
according to MT have an increased tendency to retain this attribute. For the other deciles in Panels
A and C, we observe stability values that fluctuate right around 0.1, indicating that movement out
of each of these deciles is essentially no different than what one would observe by pure chance. The
exception to this is decile 10, for private managers only, which does show slightly increased stability
relative to the interior deciles, indicating that a slightly higher percentage of managers remain in the
top portion of the distribution than what one would expect to observe by pure chance.
For skewness of MT -based rankings, Panels A and C indicate more or less monotonically de-
creasing numbers with the lowest deciles indicating a high amount of upgrade bias, while the highest
deciles indicate a high amount of downgrade bias. The transition between skewness values greater
than 1 and skewness values less than 1, in most cases is right at the center of the distribution, or
between deciles 5 and 6. This pattern indicates reversion in ranks towards the middle deciles, and
is very similar to what one might expect to arise purely by chance. Thus, the patterns we observe
imply no systematic patterns of movement between performance deciles conditional upon a manager
changing location in the distribution.
For MS-based rankings (Panels B and D), we observe consistent increased stability (most values
between .2 and .35 and some even above .4) at both ends of the performance distribution, i.e. deciles
1 and 10. Unlike the case for MT , this indicates increased consistency of performance, both for
underperforming and overperforming managers. The interior deciles in Panels B and D also tend
to have slightly higher stability than for MT -based rankings, with values more around .15 than .1.
This suggests that stability in this dimension is at least slightly higher than what one would observe
30
by chance, even at the center of the distribution. Overall, this is consistent with an increased level
of performance consistency (persistence) along this dimension, with a U-shaped pattern of highest
stability at both ends. The pattern of skewness measures in Panels B and D remains very similar to
that observed for MT in Panels A and C. The overall picture, therefore, is one of greater consistency
in performance, especially at the extremes of the distribution, but still with more or less random
movements through the distribution, conditional upon a manager’s relative performance changing.
Overall, the stability and skewness patterns for the MT and MS measures appear largely consis-
tent with the patterns of autocorrelations, rank persistence and one-year forward returns observed
above.
3.4 Timing, Selection and Portfolio Characteristics
Our final analysis examines the relationship of manager MT and MS to the set of observable portfolio
characteristics available in our dataset. We estimate panel regressions of the form
MTi,t = α+ β1geo.speci,t + β2type.speci,t + β3log(sizei,t) + β4turnoveri,t + εi,t (3)
MSi,t = α+ β1geo.speci,t + β2type.speci,t + β3log(sizei,t) + β4turnoveri,t + εi,t (4)
Here, the dependent variables in the regressions are, for each manager in each quarter, the current
MT measure and the current MS measure, respectively. The independent variables are taken as
one-year moving averages or trailing sums of each of the time series employed. log(sizei,t) is the
natural logarithm of manager i’s average portfolio size in square feet between quarter t − 3 and t.
turnoveri,t is the fractional portfolio turnover during the previous year and is defined analogously
to the turnover variable in CRSP’s Survivorship-Bias Free Mutual Fund Database. Specifically, in
our case we define this measure as
turnoveri,t =
∑3j=0 bt−j∑3
j=0 sizet−j/4(5)
31
where bτ = min(sqf.boughtτ , sqf.soldτ ) or the minimum of square footage bought and square footage
sold at time τ . The turnover measure is thus defined as the total square feet turned over during the
previous year, as a fraction of the average portfolio size during that same time period.
The variables geo.speci,t and type.speci,t measure the level of geographic and property type
specialization of a manager’s portfolio, respectively, and are computed as Hirschman-Herfindahl
Indices, defined as:
Hi,t =N∑s=1
w2s,t (6)
Here ws,t is manager i’s fractional exposure to submarket s in time period t (computed as the sum of
the fractional exposures to each property within that submarket). For geographic specialization, each
CBSA constitutes a submarket. For type specialization, each property type constitutes a submarket.
Both these measures will be equal to one if a manager held all his properties in one single sub-market
and will approach zero, the more a manager is diversified among sub-markets. In line with the other
independent variables, geo.speci,t and type.speci,t are then defined as moving averages from quarter
t− 3 to quarter t of the respective series of Hi,t.
We estimate a version of Equations 3 and 4 for performance measures computed with respect
to benchmarks at all the levels of aggregation we employ. We also estimate separate regressions for
public and private portfolios. We estimate panel regressions that include year fixed-effects. Standard
errors are robust heteroskedasticity-consistent and clustered by portfolio manager.
Tables 14 and 15 present estimates from our manager by manager panel regressions where the
dependent variable is the manager’s Market Timing performance measure. In contrast to Market Se-
lection, where there appears to be some variation in ability to select investments versus a benchmark,
managerial ability to time the market is less apparent in the univariate data. The adjusted-R2s for
the models range from 0.216 to 0.560, suggesting that they are able to capture a significant portion of
the variance in Market Timing. A consistent pattern is apparent for these regression models, for both
public and private managers. While specialization, both geographic and by property type, exhibits
32
no or only isolated significant relationships to Market Timing for REIT portfolios, there appears a
fairly consistent positive relationship between geographic specialization and timing performance in
private portfolios. We further find a widespread negative but concave relationship between portfolio
size and timing ability for REIT portfolios, as indicated by the negative and significant coefficients
on log(size). For private portfolios this relationship is slightly more isolated, but where it exists, it is
actually positive and concave. For both public and private portfolios we find a widespread negative
relationship between turnover and timing performance.
Tables 16 and 17 present estimates from our manager by manager panel regressions where the
dependent variable is the manager’s Market Selection measure. In general, the models appear to have
low explanatory power relative to the models for MT described above, suggesting a limited ability to
systematically explain a significant portion of the variation in selection ability using observable port-
folio characteristics, with the maximum R2 at only 0.034. Within this limitation, however, for REIT
managers, we observe widespread negative relationships between log(size) and selection performance
as well as positive relationships between turnover and selection performance. For private portfolios,
any significant effects found are small and isolated, suggesting a large amount of heterogeneity in
selection ability among these managers that cannot be explained by the portfolio characteristics
observable in our data.
While the limitations of the data available to us prevents us from exploring cross-sectional rela-
tionships of this nature, a natural question which arises for future research is how individual manager
characteristics–such as experience, background, education, etc.– relate to the manager’s ability to
either time property purchases and sales or to select individual property subclasses of investment
versus style-matched similar investments.
3.5 Robustness and other Comments
We conduct a number of additional robustness analyses to ensure that our results are not an artifact
of the manner in which we construct our measures.
First, as previously noted, commercial property market cycles are often thought to be longer
33
than the market cycles in other asset classes, due to the prolonged nature of the transactions. Our
timing measure, MT , compares current submarket returns for each submarket to returns in the
prior year in order to capture the extent to which managers are increasing weights in submarkets
that improve in performance and exit markets which subsequently see a drop in performance. As
a robustness check, we also consider an alternate measure of timing ability that accounts for the
lengthy nature of transactions in this market, MT2yr, in which we compare compare the most recent
submarket return and weight to submarket returns and weights lagged by two years (rather than
single year lag). In untabulated analysis, we find that this alternate measure is, overall, slightly
lower than the MT measures reported in Table 3 (by around five to ten basis points). The fraction
of managers acheiving positive timing performance under this alternative measure is lower by one to
two percentage points. When we examine the t-statistics for zero-outperformance, we find slightly
higher fractions of portfolios that yield statistically significantly positive outperformance (by about
two percentage points). Thus, overall, we find that making this alteration yields results that are
qualitatively very similar to those generated under our original measure, with the vast majority of
managers generating zero or negative timing profits, and only a very small fraction at the very top
of the distribution outperforming significantly.30
Second, while square footage is the measure that is most generally used by commercial property
practitioners to indicate levels of submarket exposure, general finance theory argues for the use
of value weights to characterize exposure levels within a portfolio. If price per square foot varies
considerably throughout the cross section of a manager’s portfolio, it could be the case that the two
types of weights would yield different inferences about performance. As commerical properties within
a portfolio are not continuously traded, observing prices to construct such value-weights, however, is
problematic. In the case of REITs, the data does not allow us to do this at all; for NCREIF members,
we do have quarterly estimates of market value for each property, which are based on transaction
values where available and otherwise on appraisals.
While appraisal-based price data has well-known limitations, for robustness, we construct MT
30For brevity, we do not tabulate these results, which are available upon request.
34
and MS measures based on value-weights constructed from these market value estimates in order
to assure that our square-footage based weights are not the major underlying driver of our findings.
In untabulated analysis, we find that using value-weights rather than square footage weights does
not materially alter our findings. As with square footage weights, positive timing profits exist only
in the very top of the performance distributions. We also find similar results to those obtained with
square fotage weights along the submarket selection ability dimension. Performance distributions
are centered at zero, and a substantial fraction of managers outperform significantly in this category,
with the fraction outperforming here slightly higher than that reported in Tables 5 and 6 by a few
percentage points.31 Thus, overall, the choice of square footage-based portfolio weights, while offering
distinct advantages in terms of parsimoniousness, does not seem to constitute an inadvertent driver
of results that yield to unwarranted inferences.
Finally, we note that it is common to partly finance commercial property investments through
non-recourse debt secured by the individual property. Default decisions are thus made at the property
level, and default occurs when an individual property value crosses a threshold, not portfolio value.
The equity value of individual property, and hence portfolio value, is therefore convex in underlying
property value when close to default threshold. This affects a manager’s portfolio’s true exposure to
a submarket, in that in a strongly declining market the manager could use strategic default to shift
some losses to debt holders. Unfortunately, our data does not allow us to observe either leverage
ratios or strategic defaults in a systematic way. In our setting, however, this data restriction is
somewhat less concerning, in that it will be the case that properties on which a default option
is used will drop out of the dataset and thereby act on our measures as reducing the manager’s
exposure to a submarket in the next quarter. Additionally, it is also unlikely that strategic defaults
are prevalent enough in the cross-section of managers’ strategies so as to have a major effect on the
overall distributions and results, as managers that pursue portfolio strategies that hinge on large
numbers of strategic defaults would likely face difficulties in raising debt capital in the long run.
31Once again, due to their similarity with what is reported in our tables, these results are not reported here explicitlyfor brevity, but are available upon request.
35
A potential caveat to the findings of our study is that we are unable to observe the transactions
costs associated with portfolio managers’ strategies, and thus cannot speak directly to whether
portfolio managers can obtain outsize profits for their investors net of such costs.
4 Conclusion
The ability of active money management to generate abnormal returns that justify their fees has
long been a subject of academic debate. In this paper, we examine the ability of portfolio managers
to generate abnormal profits in the Commercial Real Estate market through the timing of entry
and exit into real estate submarkets and through their ability to select outperforming submarkets
versus broader style-matched benchmarks. Like many alternative asset classes, the commercial real
estate market, which is characterized by transactions that primarily occur in relatively illiquid and
opaque private asset markets, may provide greater opportunities for managerial skill, value-added
and informational advantages to lead to abnormal profits.
Our analysis suggests that both public and private portfolio managers exhibit little or negative
ability to successfully time the market, on average, though in the upper tail of the distribution,
portfolio managers appear to have significant timing ability. There is substantial variation in the
ability of portfolio managers to select submarket segments, with approximately half of all managers
creating positive value in this dimension. We find evidence of persistence in submarket selection abil-
ity, a finding which may aid efficient portfolio allocation in order to profit from managers’ submarket
selection abilities. Portfolio characteristics explain a substantial fraction of variation in timing per-
formance, but not in submarket selection ability. Our findings are consistent with the commonly
held practitioner notion that in commercial property markets, where transactions are costly and
may involve prolonged transacting periods, managers may primarily create value for their investors
through selection of geographic and property-type submarkets versus a broader passive portfolio,
rather than through precise timing of sub-market entry and exit.
To the best of our knowledge, our study is the first to rigorously examine issues of market timing
36
and submarket selection ability in the real estate market. As such, it potentially may provide insights
for investors, portfolio managers and academics into how managers may earn abnormal profits in
these markets. Overall, our results suggest that in the commercial real estate market, managerial
value added, where it exists, appears to come primarily in the form of investment selection.
37
References
Barkham, R. and Geltner, D.: 1995, Price discovery in american and british property markets, RealEstate Economics 23(1), 21–44.
Barras, L., Scaillet, O. and Wermers, R.: 2010, False discoveries in mutual fund performance: Mea-suring luck in estimated alphas, The Journal of Finance 65(1), 179–216.
Bond, S. A. and Mitchell, P.: 2010, Alpha and persistence in real estate fund performance, Journalof Real Estate Finance and Economics 41, 53–79.
Brown, S. J. and Goetzmann, W. N.: 1995, Performance persistence, Journal of Finance 50, 679–698.
Carhart, M. M.: 1997, On persistence in mutual fund performance, Journal of Finance 52, 57–82.
Case, B. and Quigley, J.: 1991, Dynamics of real-estate prices, Review of Economics and Statistics73(1), 50–58.
Case, K. and Shiller, R.: 1990, Forecasting prices and excess returns in the housing market, AREUEAJournal 18(3), 253–273.
Chen, Y., Ferson, W. and Peters, H.: 2010, Measuring the timing ability and performance of bondmutual funds, Journal of Financial Economics 98(1), 72–89.
Clayton, J., Geltner, D. and Hamilton, S.: 2001, Smoothing in commercial property valuations:Evidence from individual appraisals, Real Estate Economics 29(3).
Crane, A. D. and Hartzell, J. C.: 2007, Is there a disposition effect in corporate investment decisions?evidence from real estate investment trusts, SSRN eLibrary, http://ssrn.com/paper=1031010 .
Daniel, K., Grinblatt, M., Titman, S. and Wermers, R.: 1997, Measuring mutual fund performancewith characteristic-based benchmarks, Journal of Finance 52(3), 1035–1058.
Fama, E. and French, K.: 2010, Luck versus skill in the cross-section of mutual fund returns, TheJournal of Finance 65(5), 1915–1947.
Ferson, W. and Schadt, R.: 1996, Measuring fund strategy and performance in changing economicconditions, Journal of Finance pp. 425–461.
Florance, A., Miller, N., Peng, R. and Spivey, J.: 2010, Slicing, dicing, and scoping the size of theus commercial real estate market, Journal of Real Estate Portfolio Management 16(2), 101–118.
Fung, W. and Hsieh, D. A.: 2002, Asset-based style factors for hedge funds, Financial AnalystsJournal 58(5), pp. 16–27.URL: http://www.jstor.org/stable/4480414
Fung, W. and Hsieh, D. A.: 2004, Hedge fund benchmarks: A risk-based approach, FinancialAnalysts Journal 60(5), pp. 65–80.URL: http://www.jstor.org/stable/4480604
Geltner, D.: 1991, Smoothing in appraisal-based returns, Journal of Real Estate Finance and Eco-nomics 4(3), 327–345.
38
Geltner, D. and Mei, J.: 1995, The present value model with time-varying discount rates: Implica-tions for commercial property valuation and investment decisions, Journal of Real Estate Financeand Economics 11(2), 119–135.
Gruber, M. J.: 1996, Another puzzle: The growth in actively managed mutual funds, Journal ofFinance 51, 783–810.
Hartzell, J., Muhlhofer, T. and Titman, S.: 2010, Alternative benchmarks for evaluating mutualfund performance, Real Estate Economics 38(1), 121–154.
Henriksson, R.: 1984, Market timing and mutual fund performance: An empirical investigation,Journal of Business pp. 73–96.
Henriksson, R. and Merton, R.: 1981, On market timing and investment performance. ii. statisticalprocedures for evaluating forecasting skills, Journal of Business pp. 513–533.
Jagannathan, R. and Korajczyk, R. A.: 1986, Assessing the market timing performance of managedportfolios, The Journal of Business 59(2), 217–35.URL: http://ideas.repec.org/a/ucp/jnlbus/v59y1986i2p217-35.html
Jensen, M.: 1968, The performance of mutual funds in the period 1945-1964, Journal of Finance23, 389–416.
Jensen, M.: 1969, Risk, the pricing of capital assets, and the evaluation of investment portfolios,Journal of Business 42, 167–247.
Jiang, G., Yao, T. and Yu, T.: 2007, Do mutual funds time the market? evidence from portfolioholdings, Journal of Financial Economics 86(3), 724–758.
Kacperczyk, M., Sialm, C. and Zheng, L.: 2005, On the industry concentration of actively managedequity mutual funds, The Journal of Finance 60(4), 1983–2011.
Kallberg, J. G., Liu, C. L. and Trzcinka, C.: 2000, The value added from investment managers: Anexamination of funds of reits, Journal of Financial and Quantitative Analysis 35, 387–408.
Kaplan, S. N. and Schoar, A.: 2005, Private equity performance: Returns persistence and capital,Journal of Finance 60, 1791–1823.
Kosowski, R., Timmermann, A., Wermers, R. and White, H.: 2006, Can mutual fund stars reallypick stocks? new evidence from a bootstrap analysis, The Journal of finance 61(6), 2551–2595.
Liu, C. H. and Mei, J.: 1992, The predictability of returns on equity reits and their comovementwith other assets, Journal of Real Estate Finance and Economics 5(4), 401–418.
Liu, C. H. and Mei, J.: 1994, An analysis of real estate risk using the present value model, Journalof Real Estate Finance and Economics 8, 5–20.
Muhlhofer, T.: 2012, They would if they could: Assessing the bindingness of the property holdingconstraint for reits., Real Estate Economics Forthcoming.
Plazzi, A., Torous, W. and Valkanov, R.: 2010, Expected returns and expected growth in rents ofcommercial real estate, Review of Financial Studies 23(9), 3469.
39
R Development Core Team: 2008, R: A Language and Environment for Statistical Computing, RFoundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0.URL: http://www.R-project.org
Storey, J.: 2002, A direct approach to false discovery rates, Journal of the Royal Statistical Society:Series B (Statistical Methodology) 64(3), 479–498.
Titman, S. and Tiu, C.: 2011, Do the best hedge funds hedge?, Review of Financial Studies24(1), 123–168.
Treynor, J. and Mazuy, K.: 1966, Can mutual funds outguess the market, Harvard Business Review44(4), 131–136.
Wermers, R.: 2003, Is money really ’smart’? new evidence on the relation between mutual fundflows, manager behavior, and performance persistence. Working Paper, University of Maryland.
pub.nat
pub.nat.type
pub.nat.sub
pub.reg
pub.reg.type
pub.reg.sub
pub.div
pub.div.type
pub.div.sub
pub.sta
pub.sta.type
pub.sta.sub
pub.cbsa
pub.cbsa.type
pub.cbsa.sub
priv.nat
priv.nat.type
priv.nat.sub
priv.reg
priv.reg.type
priv.reg.sub
priv.div
priv.div.type
priv.div.sub
priv.sta
priv.sta.type
priv.sta.sub
priv.cbsa
priv.cbsa.type
priv.cbsa.sub
−0.04
−0.03
−0.02
−0.01
0.00
0.01
Distribu
tion
al Plo
ts of M
T M
easures B
y Man
ager
Figure 1: This figure shows box-and-whisker diagrams comparing the cross-sectional distributions ofMT measures obtained by managers with respect to the various aggregation levels of benchmarks.The data presented includes REIT managers (starting with pub) and private managers (starting withpriv).
pub.nat
pub.nat.type
pub.nat.sub
pub.reg
pub.reg.type
pub.reg.sub
pub.div
pub.div.type
pub.div.sub
pub.sta
pub.sta.type
pub.sta.sub
pub.cbsa
pub.cbsa.type
pub.cbsa.sub
priv.nat
priv.nat.type
priv.nat.sub
priv.reg
priv.reg.type
priv.reg.sub
priv.div
priv.div.type
priv.div.sub
priv.sta
priv.sta.type
priv.sta.sub
priv.cbsa
priv.cbsa.type
priv.cbsa.sub
−10
−5 0
Distributional P
lots of MT
Measure T
−Statistics B
y Manager
Figure 2: This figure shows box-and-whisker diagrams comparing the cross-sectional distributionsof the t-statistic that a manager’s MT-measure is different from zero. The data presented includesREIT managers (starting with pub) and private managers (starting with priv).
pub.nat
pub.nat.type
pub.nat.sub
pub.reg
pub.reg.type
pub.reg.sub
pub.div
pub.div.type
pub.div.sub
pub.sta
pub.sta.type
pub.sta.sub
pub.cbsa
priv.nat
priv.nat.type
priv.nat.sub
priv.reg
priv.reg.type
priv.reg.sub
priv.div
priv.div.type
priv.div.sub
priv.sta
priv.sta.type
priv.sta.sub
priv.cbsa
−0.03
−0.02
−0.01
0.00
0.01
0.02
Distribu
tion
al Plo
ts of M
S M
easures B
y Man
ager
Figure 3: This figure shows box-and-whisker diagrams comparing the cross-sectional distributions ofMS measures obtained by managers with respect to the various aggregation levels of benchmarks.The data presented includes REIT managers (starting with pub) and private managers (starting withpriv).
pub.nat
pub.nat.type
pub.nat.sub
pub.reg
pub.reg.type
pub.reg.sub
pub.div
pub.div.type
pub.div.sub
pub.sta
pub.sta.type
pub.sta.sub
pub.cbsa
priv.nat
priv.nat.type
priv.nat.sub
priv.reg
priv.reg.type
priv.reg.sub
priv.div
priv.div.type
priv.div.sub
priv.sta
priv.sta.type
priv.sta.sub
priv.cbsa
−6
−4
−2 0 2 4 6
Distributional P
lots of MS
Measure T
−Statistics B
y Manager
Figure 4: This figure shows box-and-whisker diagrams comparing the cross-sectional distributionsof the t-statistic that a manager’s MS-measure is different from zero. The data presented includesREIT managers (starting with pub) and private managers (starting with priv).
Table 1: Subdivisions by Geography and Property Type
This table presents the numbers of properties held by private investors and REITs, organized by NCREIF Type,
Subtype, Region, and Division. NCREIF also offers organizations by state and CBSA, which we do not present here.
Private REITs
Type and Subtype
Apartment 3654 1871
Garden 2600
High-rise 532
Low-rise 410
Hotel 245 24
Industrial 6385 3144
Warehouse 4930 1068
R&D 768 130
Flex Space 725 519
Manufacturing 30 133
Showroom 14
Other 90 76
Office 4780 3731
CBD 966 3465
Suburban 3821 266
Retail 3151 2904
Community 1062
Theme/Festival 7
Fashion/Specialty 71
Neighborhood 1155
Outlet 6 71
Power Center 203 43
Regional 338 133
Super Regional 214
Single Tenant 253 466
Regions and Divisions
East 4092 3132
Mideast 2103 1446
Northeast 1989 1686
Midwest 2927 2183
East North Central 2118 1642
West North Central 809 541
South 5170 3392
Southeast 2839 2002
Southwest 2331 1390
West 5950 2967
Mountain 1466 770
Pacific 4484 2197
Table 2: Summary Statistics
This table presents summary statistics for the sets of properties held by both private investors and Real Estate
Investment Trusts.
Mean Std. Dev. 1st Quartile Median 3rd Quartile
REITs
Property Sizes (1000 Sq. ft.) 182.63 235.44 52.87 114.87 228
Portfolio Sizes (1000 Sq. ft.) 21, 460.15 26, 077.87 6, 263.01 14, 386.73 24, 868.9
Portfolio Presence in Number of CBSAs 11.37 12.27 2.25 8 16
Number of Property Types in Portfolios 1.85 1.01 1 2 2
Number of Property Subtypes in Portfolios 2.93 2.1 1 2 4
Property Holding Periods (years) 6.03 4.88 2.68 4.87 8.21
Manager Number of Years in Sample 8.22 4.59 4 8 12.25
Manager HHI, by Region 0.72 0.25 0.5 0.71 1
Manager HHI, by Division 0.64 0.28 0.41 0.59 0.98
Manager HHI, by State 0.57 0.3 0.32 0.5 0.9
Manager HHI, by CBSA 0.5 0.31 0.24 0.4 0.79
Manager HHI, by Type 0.87 0.18 0.76 0.99 1
Manager HHI, by Subtype 0.78 0.23 0.6 0.84 1
Number of Properties: 11,674 Number of Managers: 210
Private Portfolios
Property Sizes (1000 Sq. ft.) 268.02 432 92.02 173.98 316.82
Portfolio Sizes (1000 Sq. ft.) 42, 060.43 44, 311.33 10, 917.88 29, 775.05 52, 475.24
Portfolio Presence in Number of CBSAs 30.94 28.15 10.75 22 41.25
Number of Property Types in Portfolios 3.35 1.3 3 4 4
Number of Property Subtypes in Portfolios 7.69 4.82 4 6 11
Property Holding Periods (years) 4.65 3.55 2.25 3.75 6.25
Manager Number of Years in Sample 10.84 8.29 3.75 9.25 15.38
Manager HHI, by Region 0.51 0.23 0.34 0.42 0.62
Manager HHI, by Division 0.41 0.24 0.23 0.32 0.51
Manager HHI, by State 0.33 0.24 0.15 0.25 0.41
Manager HHI, by CBSA 0.26 0.22 0.1 0.18 0.35
Manager HHI, by Type 0.62 0.23 0.45 0.57 0.8
Manager HHI, by Subtype 0.49 0.21 0.33 0.43 0.59
Number of Properties: 18,115 Number of Managers: 136
Table 3: Market Timing Measures: Means.
This table presents distributional characteristics across managers, for time-series means of quarterly Market Timing
(MT ) measures. The measures are computed relative to style benchmarks at the National, Regional, Divisional, State,
and CBSA level. At each geographic level of disaggregation, we use the benchmark for all property types combined, as
well as benchmarks separated and matched by type and subtype. We include only managers for whom we can compute
a MT measure over 12 quarters or more.
REITs NCREIF Private
National/Type National/Subtype National/Type National/Subtype
Mean −0.004598 −0.00466 −0.005521 −0.005396
StdDev 0.006139 0.006111 0.005768 0.005918
Min. −0.029997 −0.029478 −0.02984 −0.029021
1st Qu. −0.007991 −0.00775 −0.00767 −0.007999
Median −0.004059 −0.003985 −0.005375 −0.005229
3rd Qu. −0.000865 −0.000997 −0.002919 −0.002957
Max. 0.010026 0.009789 0.011138 0.012389
Fraction > 0 0.161905 0.161905 0.095588 0.088235
Regional Regional/Type Regional/Subtype Regional Regional/Type Regional/Subtype
Mean −0.004756 −0.004482 −0.004428 −0.005423 −0.00563 −0.005427
StdDev 0.005661 0.006164 0.006218 0.005611 0.006034 0.006188
Min. −0.030541 −0.032973 −0.033849 −0.029509 −0.032463 −0.030963
1st Qu. −0.007282 −0.007403 −0.00734 −0.008363 −0.008466 −0.008226
Median −0.004009 −0.004035 −0.003882 −0.005022 −0.005332 −0.00558
3rd Qu. −0.000832 −0.000397 −0.000336 −0.003231 −0.002907 −0.002871
Max. 0.008067 0.011174 0.010465 0.007705 0.00855 0.009552
Fraction > 0 0.152381 0.17619 0.17619 0.088235 0.088235 0.095588
Divisional Divisional/Type Divisional/Subtype Divisional Divisional/Type Divisional/Subtype
Mean −0.004759 −0.00456 −0.004737 −0.005565 −0.005938 −0.005741
StdDev 0.005714 0.006236 0.006462 0.005682 0.00616 0.00605
Min. −0.027595 −0.032525 −0.034198 −0.031078 −0.035284 −0.030842
1st Qu. −0.007239 −0.007404 −0.007843 −0.008082 −0.008589 −0.008184
Median −0.004102 −0.004234 −0.004162 −0.005057 −0.005385 −0.005276
3rd Qu. −0.001459 −0.000455 −0.001003 −0.003116 −0.003385 −0.003215
Max. 0.00957 0.012797 0.011377 0.007377 0.009152 0.008797
Fraction > 0 0.147619 0.185714 0.166667 0.088235 0.080882 0.088235
State State/Type State/Subtype State State/Type State/Subtype
Mean −0.004972 −0.004849 −0.004937 −0.005455 −0.005936 −0.005898
StdDev 0.005739 0.006491 0.007033 0.005777 0.005806 0.006071
Min. −0.029287 −0.031255 −0.038958 −0.027539 −0.032138 −0.03085
1st Qu. −0.007886 −0.007692 −0.007843 −0.008276 −0.008408 −0.00866
Median −0.00454 −0.004499 −0.004189 −0.005097 −0.005596 −0.005583
3rd Qu. −0.001239 −0.00091 −0.000884 −0.003113 −0.003878 −0.003453
Max. 0.010206 0.012984 0.012984 0.008753 0.010676 0.010998
Fraction > 0 0.119048 0.157143 0.133333 0.102941 0.073529 0.088235
CBSA CBSA/Type CBSA/Subtype CBSA CBSA/Type CBSA/Subtype
Mean −0.005255 −0.005293 −0.004583 −0.00567 −0.006156 −0.006125
StdDev 0.005609 0.006593 0.007804 0.0059 0.005995 0.006235
Min. −0.028819 −0.037491 −0.046658 −0.027463 −0.029777 −0.027018
1st Qu. −0.007991 −0.007918 −0.007288 −0.007821 −0.009138 −0.009885
Median −0.004722 −0.004424 −0.003504 −0.005203 −0.005483 −0.006133
3rd Qu. −0.001648 −0.00147 −0.000102 −0.003209 −0.003329 −0.003374
Max. 0.008393 0.007059 0.010474 0.011471 0.014188 0.016744
Fraction > 0 0.104762 0.119048 0.147619 0.088235 0.073529 0.073529
Table 4: Market Timing Measures: t-statistics.
This table presents distributional characteristics across managers, for t-statistics testing the hypothesis Ho : MT = 0
against the two-sided alternative, over a manager’s time series of quarterly MT measure observations. The MT
measures are computed relative to benchmarks at the National, Regional, Divisional, State, and CBSA level. At each
geographic level of disaggregation, we use the benchmark for all property types combined, as well as benchmarks
separated and matched by type and subtype. We include only managers for whom we can compute a MT measure over
12 quarters or more. Fraction Sig. is the fraction of portfolios that show a significantly positive t-statistic, at the 5%
level (two-tailed test). FDR+ is the positive-side False-Discovery Ratio, or the fraction of portfolios with significantly
positive performance that would be observed if returns were purely determined by luck.
REITs NCREIF Private
National/Type National/Subtype National/Type National/Subtype
Mean −1.450189 −1.411691 −1.76214 −1.719102
Min. −5.628049 −5.744656 −6.12599 −5.652484
1st Qu. −2.687202 −2.56827 −2.794327 −2.844073
Median −1.836674 −1.681195 −1.820428 −1.746882
3rd Qu. −0.247335 −0.445283 −1.070069 −1.023229
Max. 3.379433 3.51457 3.441438 3.990084
Fraction Sig. 0.014286 0.009524 0.022059 0.022059
FDR+ 0.01994 0.020536 0.017004 0.017923
Regional Regional/Type Regional/Subtype Regional Regional/Type Regional/Subtype
Mean −1.600414 −1.432858 −1.300095 −1.665149 −1.697443 −1.664536
Min. −4.988081 −5.790907 −5.280524 −5.196902 −5.070494 −4.822903
1st Qu. −2.510735 −2.669551 −2.395451 −2.58113 −2.639516 −2.753458
Median −1.939086 −1.645071 −1.427114 −1.761968 −1.87808 −1.780817
3rd Qu. −0.382176 −0.164844 −0.149508 −1.081622 −0.947389 −1.016944
Max. 4.073819 3.156732 3.303863 3.841806 3.549419 3.942267
Fraction Sig. 0.009524 0.004762 0.004762 0.022059 0.014706 0.014706
FDR+ 0.021131 0.022321 0.025 0.017463 0.017923 0.017923
Divisional Divisional/Type Divisional/Subtype Divisional Divisional/Type Divisional/Subtype
Mean −1.594406 −1.389694 −1.334713 −1.713678 −1.771529 −1.701093
Min. −5.829871 −5.540403 −5.096477 −7.917501 −6.892256 −6.409009
1st Qu. −2.447641 −2.557836 −2.364666 −2.635298 −2.69865 −2.592254
Median −1.915493 −1.575252 −1.420866 −1.780566 −1.823243 −1.818251
3rd Qu. −0.477872 −0.136487 −0.341782 −1.012937 −1.037759 −0.967833
Max. 4.273815 3.181013 3.123875 3.444613 3.74164 3.141749
Fraction Sig. 0.009524 0.009524 0.009524 0.029412 0.014706 0.014706
FDR+ 0.020833 0.021131 0.024702 0.017463 0.017463 0.017923
State State/Type State/Subtype State State/Type State/Subtype
Mean −1.745335 −1.438056 −1.303139 −1.670242 −1.712724 −1.619501
Min. −10.69233 −5.43857 −6.052973 −5.858985 −6.253351 −6.299009
1st Qu. −2.702192 −2.610382 −2.378467 −2.680802 −2.560097 −2.616981
Median −2.023049 −1.610232 −1.292217 −1.704782 −1.783884 −1.761259
3rd Qu. −0.520679 −0.258782 −0.233132 −1.137348 −0.903273 −0.866664
Max. 4.071906 2.779545 2.91737 3.302668 2.051281 3.508843
Fraction Sig. 0.009524 0.009524 0.004762 0.014706 0 0.014706
FDR+ 0.021131 0.022024 0.027083 0.018382 0.017004 0.019761
CBSA CBSA/Type CBSA/Subtype CBSA CBSA/Type CBSA/Subtype
Mean −1.725321 −1.451822 −1.238115 −1.620218 −1.585274 −1.517362
Min. −5.285499 −5.078405 −6.600948 −6.080619 −6.607411 −6.333351
1st Qu. −2.625622 −2.447465 −2.378021 −2.47084 −2.404432 −2.33023
Median −1.880332 −1.301726 −1.146668 −1.687081 −1.722097 −1.703037
3rd Qu. −0.716154 −0.466703 −0.021334 −0.90028 −0.907327 −0.845984
Max. 2.671161 2.105177 1.931608 3.444613 3.511602 3.03063
Fraction Sig. 0.004762 0 0 0.022059 0.014706 0.022059
FDR+ 0.022024 0.02619 0.033333 0.017463 0.020221 0.02068
Table 5: Market Selection Measures: Means.
This table presents distributional characteristics across managers, for time-series means of quarterly Market Selection
(MS) measures. The measures are computed relative to benchmarks at the National, Regional, Divisional, State, and
CBSA level. At each geographic level of disaggregation, we use the benchmark for all property types combined, as well
as benchmarks separated and matched by type and subtype. We include only managers for whom we can compute an
MS measure over 12 quarters or more.
REITs NCREIF Private
National National/Type National/Subtype National National/Type National/Subtype
Mean −0.00022 −0.000146 −0.000859 −0.000192 −0.000215 −0.000187
StdDev 0.004523 0.004013 0.004205 0.003532 0.002621 0.002439
Min. −0.019699 −0.015679 −0.019044 −0.013187 −0.009116 −0.008667
1st Qu. −0.002593 −0.002252 −0.003346 −0.001773 −0.001264 −0.001052
Median −5e− 04 −0.000472 −0.000887 −7.6e− 05 −0.000228 −8.2e− 05
3rd Qu. 0.002045 0.001788 0.001482 0.001681 0.001197 0.001089
Max. 0.020836 0.01626 0.015667 0.012988 0.00695 0.007643
Fraction > 0 0.366667 0.333333 0.3 0.397059 0.367647 0.382353
Regional Regional/Type Regional/Subtype Regional Regional/Type Regional/Subtype
Mean 0.000194 0.000192 −0.000444 8.4e− 05 7.1e− 05 9.4e− 05
StdDev 0.00443 0.003325 0.003654 0.003147 0.00212 0.002047
Min. −0.021777 −0.014922 −0.014332 −0.009163 −0.007357 −0.007432
1st Qu. −0.001591 −0.001379 −0.002004 −0.001415 −0.00073 −0.000688
Median −2.5e− 05 −9e− 06 −0.000375 0.000335 0.000155 0.00017
3rd Qu. 0.001972 0.001668 0.001166 0.001685 0.001152 0.001021
Max. 0.016648 0.013952 0.012119 0.011523 0.009032 0.008708
Fraction > 0 0.390476 0.395238 0.338095 0.477941 0.455882 0.433824
Divisional Divisional/Type Divisional/Subtype Divisional Divisional/Type Divisional/Subtype
Mean 0.000189 0.000229 −0.000432 4.8e− 05 −3.4e− 05 2.5e− 05
StdDev 0.004484 0.003143 0.003366 0.002934 0.001798 0.001822
Min. −0.022997 −0.016104 −0.015745 −0.009131 −0.007509 −0.007106
1st Qu. −0.001346 −0.000908 −0.001638 −0.00116 −0.000672 −0.000575
Median 0.000176 0.000204 −0.000206 9.1e− 05 3.4e− 05 5.9e− 05
3rd Qu. 0.001503 0.001534 0.001055 0.001343 0.000758 0.000776
Max. 0.016489 0.013453 0.012204 0.011177 0.007313 0.007319
Fraction > 0 0.428571 0.438095 0.352381 0.426471 0.426471 0.411765
State State/Type State/Subtype State State/Type State/Subtype
Mean 0.000156 0.000308 −5.1e− 05 7.6e− 05 0.00018 0.000163
StdDev 0.004344 0.00253 0.002505 0.002897 0.001566 0.001622
Min. −0.023742 −0.015998 −0.015572 −0.008915 −0.007159 −0.00643
1st Qu. −0.001256 −0.000328 −0.000533 −0.000994 −0.000455 −0.000561
Median 0.000126 0.00023 2.5e− 05 0.000315 2.3e− 05 −2.3e− 05
3rd Qu. 0.00142 0.001008 0.000807 0.001114 0.000599 0.00065
Max. 0.016086 0.012239 0.011195 0.01031 0.007938 0.007784
Fraction > 0 0.442857 0.471429 0.37619 0.463235 0.419118 0.389706
CBSA CBSA
Mean 3.1e− 05 3.6e− 05
StdDev 0.003996 0.002412
Min. −0.028883 −0.008159
1st Qu. −0.00131 −0.000728
Median 0.000153 8.3e− 05
3rd Qu. 0.001405 0.000955
Max. 0.01266 0.009639
Fraction > 0 0.42381 0.433824
Table 6: Market Selection Measures: t-statistics.
This table presents distributional characteristics across managers, for t-statistics testing the hypothesis Ho : MS = 0
against the two-sided alternative, over a manager’s time series of quarterly MS measure observations. The MS
measures are computed relative to benchmarks at the National, Regional, Divisional, State, and CBSA level. At each
geographic level of disaggregation, we use the benchmark for all property types combined, as well as benchmarks
separated and matched by type and subtype. We include only managers for whom we can compute a MS measure over
12 quarters or more. Fraction Sig. is the fraction of portfolios that show a significantly positive t-statistic, at the 5%
level (two-tailed test). FDR+ is the positive-side False-Discovery Ratio, or the fraction of portfolios with significantly
positive performance that would be observed if returns were purely determined by luck.
REITs NCREIF Private
National National/Type National/Subtype National National/Type National/Subtype
Mean −0.375851 −0.336299 −0.68265 −0.090703 −0.126823 −0.062521
Min. −6.103733 −6.152487 −7.342043 −4.806187 −4.382483 −4.230968
1st Qu. −1.71303 −1.491534 −2.126945 −1.418639 −1.464125 −0.972616
Median −0.395455 −0.369881 −0.658239 −0.045002 −0.225546 −0.080836
3rd Qu. 0.885615 1.07333 0.847101 1.201405 1.313817 0.96297
Max. 6.178993 4.760577 5.126133 4.603225 3.478985 3.41616
Fraction Sig. 0.180952 0.12381 0.128571 0.169118 0.176471 0.183824
FDR+ 0.01994 0.01994 0.020536 0.017923 0.017004 0.017923
Regional Regional/Type Regional/Subtype Regional Regional/Type Regional/Subtype
Mean 0.003664 0.067785 −0.236078 0.205065 0.15181 0.175524
Min. −6.39094 −6.981408 −6.981408 −3.523223 −2.984255 −3.244873
1st Qu. −1.001047 −1.151088 −1.406943 −1.203038 −0.763916 −0.776307
Median −0.012537 −0.017465 −0.332384 0.324941 0.202219 0.212406
3rd Qu. 1.152849 1.23596 0.861898 1.449287 1.17493 1.061433
Max. 5.625015 4.752053 4.752053 4.909625 3.289341 3.961995
Fraction Sig. 0.161905 0.142857 0.138095 0.213235 0.198529 0.154412
FDR+ 0.021131 0.022321 0.025 0.017463 0.017923 0.017923
Divisional Divisional/Type Divisional/Subtype Divisional Divisional/Type Divisional/Subtype
Mean 0.020859 0.161611 −0.17038 0.135817 0.101373 0.110553
Min. −5.622909 −6.757757 −6.757757 −3.298095 −3.020899 −3.410074
1st Qu. −0.999397 −0.996447 −1.274746 −1.059857 −0.79793 −0.684182
Median 0.129395 0.132275 −0.175734 0.094459 0.028535 0.080927
3rd Qu. 1.138629 1.175503 0.898637 1.289605 0.991962 1.0491
Max. 6.747472 4.622942 4.622942 3.959653 3.582001 4.134138
Fraction Sig. 0.185714 0.138095 0.119048 0.154412 0.161765 0.147059
FDR+ 0.020833 0.021131 0.024702 0.017463 0.017463 0.017923
State State/Type State/Subtype State State/Type State/Subtype
Mean 0.016491 0.304491 0.041045 0.168665 0.227929 0.109475
Min. −5.698834 −3.737185 −3.612207 −3.893137 −3.179769 −4.653705
1st Qu. −0.861276 −0.547847 −0.721977 −1.063348 −0.716312 −0.809837
Median 0.088806 0.327738 0.049302 0.258182 0.065522 −0.015105
3rd Qu. 0.936132 1.067133 0.819504 1.32766 1.198209 1.171855
Max. 5.943529 3.699163 3.897693 3.539257 3.61891 3.706276
Fraction Sig. 0.2 0.171429 0.109524 0.176471 0.147059 0.139706
FDR+ 0.021131 0.022024 0.027083 0.018382 0.017004 0.019761
CBSA CBSA
Mean −0.063453 0.155002
Min. −5.722754 −4.775663
1st Qu. −0.89255 −0.906985
Median 0.1407 0.14987
3rd Qu. 1.067773 1.165479
Max. 3.805435 4.181431
Fraction Sig. 0.185714 0.183824
FDR+ 0.022024 0.017463
Table 7: Correlations Between MT and MS
This table shows correlation coefficients between Market Timing and Market Selection performance. The table shows
these over the entire panel dataset by manager and quarter. T-statistcs testing a hypothesis that the true correlation
is zero, against the two-sided alternative are included.
REITs NCREIF Private
Panel Correlation t-statistic Correlation t-statistic
National/Type −0.1484 −11.34∗∗∗ −0.1066 −7.25∗∗∗
National/Subtype −0.2319 −17.87∗∗∗ −0.0669 −4.53∗∗∗
Regional −0.1171 −8.91∗∗∗ −0.0849 −5.77∗∗∗
Regional/Type −0.1256 −9.56∗∗∗ −0.0567 −3.84∗∗∗
Regional/Subtype −0.1888 −14.25∗∗∗ −0.0399 −2.7∗∗∗
Divisional −0.0954 −7.24∗∗∗ −0.0607 −4.12∗∗∗
Divisional/Type −0.0987 −7.49∗∗∗ −0.0365 −2.47∗∗
Divisional/Subtype −0.16 −11.91∗∗∗ −0.0521 −3.5∗∗∗
State −0.0513 −3.88∗∗∗ −0.024 −1.62
State/Type −0.041 −3.04∗∗∗ 0.0085 0.57
State/Subtype −0.1198 −8.48∗∗∗ −0.0459 −3.06∗∗∗
CBSA −0.0219 −1.65∗ −0.0402 −2.72∗∗∗
∗ : significance level ≤ 10%. ∗∗: significance level ≤ 5%. ∗∗∗: significance level ≤ 1%.
Table 8: Market Timing Measures: Persistence
This table illustrates persistence of managers’ MT performance through time. From a panel dataset, we annualize the
quarterly MT measure for each manager and then we compute the overall mean, as well as one-year, two-year, and
three-year autocorrelations of annual MT .
REITs NCREIF Private
National/Type National/Subtype National/Type National/Subtype
Mean −0.01752 −0.01797 −0.02821 −0.02802
ρt,t−1 0.2543∗∗∗ 0.2255∗∗∗ 0.4582∗∗∗ 0.461∗∗∗
ρt,t−2 0.1267∗∗∗ 0.1126∗∗∗ 0.19∗∗∗ 0.183∗∗∗
ρt,t−3 −0.2812∗∗∗ −0.2961∗∗∗ −0.2083∗∗∗ −0.2011∗∗∗
Regional Regional/Type Regional/Subtype Regional Regional/Type Regional/Subtype
Mean −0.01775 −0.01725 −0.01734 −0.02711 −0.028 −0.02782
ρt,t−1 0.2594∗∗∗ 0.2243∗∗∗ 0.1766∗∗∗ 0.4412∗∗∗ 0.4344∗∗∗ 0.4342∗∗∗
ρt,t−2 0.1622∗∗∗ 0.0922∗∗∗ 0.0693∗∗ 0.1899∗∗∗ 0.1748∗∗∗ 0.1625∗∗∗
ρt,t−3 −0.3333∗∗∗ −0.2523∗∗∗ −0.2788∗∗∗ −0.1536∗∗∗ −0.1831∗∗∗ −0.1739∗∗∗
Divisional Divisional/Type Divisional/Subtype Divisional Divisional/Type Divisional/Subtype
Mean −0.01771 −0.01721 −0.01818 −0.02719 −0.02832 −0.0283
ρt,t−1 0.2583∗∗∗ 0.2035∗∗∗ 0.1492∗∗∗ 0.4281∗∗∗ 0.4224∗∗∗ 0.4251∗∗∗
ρt,t−2 0.1554∗∗∗ 0.0771∗∗ 0.0304 0.2014∗∗∗ 0.1772∗∗∗ 0.1815∗∗∗
ρt,t−3 −0.3225∗∗∗ −0.236∗∗∗ −0.2132∗∗∗ −0.1514∗∗∗ −0.1785∗∗∗ −0.1748∗∗∗
State State/Type State/Subtype State State/Type State/Subtype
Mean −0.01836 −0.01762 −0.01752 −0.027 −0.02869 −0.02918
ρt,t−1 0.1896∗∗∗ 0.123∗∗∗ 0.0519∗ 0.4221∗∗∗ 0.3742∗∗∗ 0.3457∗∗∗
ρt,t−2 0.146∗∗∗ 0.0806∗∗∗ 0.0282 0.1797∗∗∗ 0.1469∗∗∗ 0.1171∗∗∗
ρt,t−3 −0.2214∗∗∗ −0.1843∗∗∗ −0.1576∗∗∗ −0.1246∗∗∗ −0.1857∗∗∗ −0.2046∗∗∗
CBSA CBSA/Type CBSA/Subtype CBSA CBSA/Type CBSA/Subtype
Mean −0.01939 −0.01858 −0.01692 −0.02777 −0.02894 −0.02979
ρt,t−1 0.1475∗∗∗ 0.0485 0.0247 0.3751∗∗∗ 0.3145∗∗∗ 0.3179∗∗∗
ρt,t−2 0.1169∗∗∗ 0.0247 0.0052 0.1365∗∗∗ 0.143∗∗∗ 0.1428∗∗∗
ρt,t−3 −0.1903∗∗∗ −0.2114∗∗∗ −0.1474∗∗∗ −0.1118∗∗∗ −0.1996∗∗∗ −0.1947∗∗∗
∗ : significance level ≤ 10%. ∗∗: significance level ≤ 5%. ∗∗∗: significance level ≤ 1%.
Table 9: Market Selection Measures: Persistence
This table illustrates persistence of managers’ MS performance through time. From a panel dataset, we annualize the
quarterly MS measure for each manager and then we compute the overall mean, as well as one-year, two-year, and
three-year autocorrelations of annual MS.
REITs NCREIF Private
National National/Type National/Subtype National National/Type National/Subtype
Mean −0.00093 −0.00116 −0.00349 −0.00035 −0.00076 −0.00068
ρt,t−1 0.3707∗∗∗ 0.255∗∗∗ 0.2288∗∗∗ 0.4362∗∗∗ 0.4087∗∗∗ 0.4018∗∗∗
ρt,t−2 0.0784∗∗∗ 0.0452 0.0453 0.1561∗∗∗ 0.1891∗∗∗ 0.197∗∗∗
ρt,t−3 0.004 0.0457 0.0554∗ 0.0015 0.0245 0.0141
Regional Regional/Type Regional/Subtype Regional Regional/Type Regional/Subtype
Mean 0.001 0.00033 −0.0019 0.00084 0.00019 0.00016
ρt,t−1 0.3908∗∗∗ 0.2132∗∗∗ 0.2236∗∗∗ 0.4325∗∗∗ 0.3606∗∗∗ 0.3394∗∗∗
ρt,t−2 0.1179∗∗∗ 0.0291 0.0435 0.1575∗∗∗ 0.1736∗∗∗ 0.1667∗∗∗
ρt,t−3 0.0153 0.0044 0.0255 −0.0605∗ −0.0133 −0.0066
Divisional Divisional/Type Divisional/Subtype Divisional Divisional/Type Divisional/Subtype
Mean 0.001 0.0006 −0.00186 0.00074 0.00013 −0.00002
ρt,t−1 0.4157∗∗∗ 0.244∗∗∗ 0.2394∗∗∗ 0.4136∗∗∗ 0.2935∗∗∗ 0.3197∗∗∗
ρt,t−2 0.1384∗∗∗ 3e− 04 0.0162 0.1627∗∗∗ 0.1332∗∗∗ 0.1122∗∗∗
ρt,t−3 0.019 −0.0148 0.0524 −0.0681∗ −0.0187 7e− 04
State State/Type State/Subtype State State/Type State/Subtype
Mean 0.00087 0.00111 −0.00023 0.00118 0.00072 0.00027
ρt,t−1 0.4002∗∗∗ 0.185∗∗∗ 0.1583∗∗∗ 0.402∗∗∗ 0.2868∗∗∗ 0.2998∗∗∗
ρt,t−2 0.1142∗∗∗ −0.1129∗∗∗ −0.1436∗∗∗ 0.1374∗∗∗ 0.1195∗∗∗ 0.0917∗∗∗
ρt,t−3 −0.0072 −0.1151∗∗∗ −0.1014∗∗∗ −0.0794∗∗ −0.0091 −0.0078
CBSA CBSA
Mean 0.00054 0.00084
ρt,t−1 0.4906∗∗∗ 0.3519∗∗∗
ρt,t−2 0.2151∗∗∗ 0.0845∗∗
ρt,t−3 0.0316 −0.0945∗∗∗
∗ : significance level ≤ 10%. ∗∗: significance level ≤ 5%. ∗∗∗: significance level ≤ 1%.
Table 10: Market Timing Measures: Rank Persistence
This table shows the persistence of relative rankings of managers, according to their MT measure. Each year, we rank managers
according to their realized Market Timing performance. ρt,τ shows the correlation between a manager’s rank at time τ and at
time t. 75plus and 90plus show permanence above the 75th and 90th percentile. Specifically, 75plusτ,t shows the fraction of
all managers whose performance was at or above the 75th percentile at time τ , whose performance is still at or above the 75th
percentile at time t. 90plus shows the analogous statistic for the 90th percentile.
REITs NCREIF Private
National/Type National/Subtype National/Type National/Subtype
ρt,t−1 0.2829∗∗∗ 0.2043∗∗∗ 0.2943∗∗∗ 0.282∗∗∗
ρt,t−2 −0.0246 0.0045 0.1036∗∗∗ 0.0945∗∗∗
ρt,t−3 0.0071 −0.0288 0.0602∗ 0.089∗∗∗
75plust−1,t 0.42 0.34 0.42 0.38
75plust−2,t 0.24 0.27 0.32 0.28
75plust−3,t 0.27 0.24 0.27 0.26
90plust−1,t 0.36 0.1 0.2 0.26
90plust−2,t 0.16 0.17 0.2 0.1
90plust−3,t 0.17 0.09 0.06 0.06
Regional Regional/Type Regional/Subtype Regional Regional/Type Regional/Subtype
ρt,t−1 0.1669∗∗∗ 0.1594∗∗∗ 0.0797∗∗∗ 0.3074∗∗∗ 0.2297∗∗∗ 0.2294∗∗∗
ρt,t−2 0.1419∗∗∗ −0.0349 −0.0297 0.0953∗∗∗ 0.0694∗∗ 0.0701∗∗
ρt,t−3 0.1465∗∗∗ −0.0029 −0.0636∗ 0.0953∗∗∗ 0.0277 0.0773∗∗
75plust−1,t 0.28 0.34 0.3 0.4 0.34 0.34
75plust−2,t 0.34 0.27 0.25 0.32 0.28 0.27
75plust−3,t 0.36 0.31 0.26 0.27 0.23 0.23
90plust−1,t 0.07 0.11 0.13 0.27 0.18 0.16
90plust−2,t 0.23 0.08 0.17 0.18 0.14 0.12
90plust−3,t 0.28 0.19 0.07 0.13 0.08 0.08
Divisional Divisional/Type Divisional/Subtype Divisional Divisional/Type Divisional/Subtype
ρt,t−1 0.1171∗∗∗ 0.1172∗∗∗ 0.0715∗∗ 0.2801∗∗∗ 0.1972∗∗∗ 0.1923∗∗∗
ρt,t−2 0.1011∗∗∗ −0.0468 −0.0606∗ 0.0963∗∗∗ 0.0834∗∗∗ 0.0524∗
ρt,t−3 0.1538∗∗∗ −0.0315 −0.0194 0.0731∗∗ 0.043 0.0502
75plust−1,t 0.26 0.3 0.28 0.36 0.33 0.33
75plust−2,t 0.29 0.21 0.24 0.29 0.24 0.27
75plust−3,t 0.36 0.22 0.21 0.25 0.26 0.26
90plust−1,t 0.14 0.09 0.12 0.28 0.17 0.16
90plust−2,t 0.09 0.09 0.14 0.15 0.13 0.12
90plust−3,t 0.33 0.15 0.1 0.13 0.09 0.08
State State/Type State/Subtype State State/Type State/Subtype
ρt,t−1 0.0417 −0.0038 −0.024 0.1565∗∗∗ 0.0789∗∗∗ 0.1337∗∗∗
ρt,t−2 0.0939∗∗∗ 0.0299 −0.0233 0.083∗∗∗ 0.0564∗ 0.0596∗
ρt,t−3 0.083∗∗ −0.019 −0.0109 0.0819∗∗ 0.0337 0.0366
75plust−1,t 0.24 0.25 0.24 0.31 0.27 0.29
75plust−2,t 0.35 0.25 0.24 0.31 0.27 0.27
75plust−3,t 0.34 0.25 0.28 0.27 0.22 0.22
90plust−1,t 0.08 0.1 0.12 0.14 0.14 0.16
90plust−2,t 0.1 0.13 0.11 0.13 0.13 0.11
90plust−3,t 0.21 0.16 0.2 0.06 0.05 0.09
CBSA CBSA/Type CBSA/Subtype CBSA CBSA/Type CBSA/Subtype
ρt,t−1 −0.0494∗ −0.0265 −0.0196 0.116∗∗∗ 0.0564∗ 0.1068∗∗∗
ρt,t−2 0.088∗∗∗ −0.0316 −0.0429 0.0229 0.0458 0.063∗
ρt,t−3 0.0851∗∗ 0.0031 −0.0472 0.0874∗∗∗ 0.0139 0.017
75plust−1,t 0.2 0.22 0.16 0.28 0.28 0.3
75plust−2,t 0.31 0.25 0.25 0.22 0.24 0.29
75plust−3,t 0.33 0.23 0.23 0.26 0.28 0.22
90plust−1,t 0.04 0.11 0.07 0.14 0.09 0.18
90plust−2,t 0.12 0.12 0.12 0.09 0.07 0.12
90plust−3,t 0.13 0.16 0.15 0.12 0.12 0.14
∗ : significance level ≤ 10%. ∗∗: significance level ≤ 5%. ∗∗∗: significance level ≤ 1%.
Table 11: Market Selection Measures: Rank Persistence
This table shows the persistence of relative rankings of managers, according to their MS measure. Each year, we rank
managers according to their realized Market Selection performance. ρt,τ shows the correlation between a manager’s
rank at time τ and at time t. 75plus and 90plus show permanence above the 75th and 90th percentile. Specifically,
75plusτ,t shows the fraction of all managers whose performance was at or above the 75th percentile at time τ , whose
performance is still at or above the 75th percentile at time t. 90plus shows the analogous statistic for the 90th
percentile.
REITs NCREIF Private
National National/Type National/Subtype National National/Type National/Subtype
ρt,t−1 0.4324∗∗∗ 0.3736∗∗∗ 0.3609∗∗∗ 0.4499∗∗∗ 0.4101∗∗∗ 0.3703∗∗∗
ρt,t−2 0.1927∗∗∗ 0.2254∗∗∗ 0.241∗∗∗ 0.2264∗∗∗ 0.2295∗∗∗ 0.2324∗∗∗
ρt,t−3 0.0506 0.1624∗∗∗ 0.1705∗∗∗ 0.0949∗∗∗ 0.1135∗∗∗ 0.074∗∗
75plust−1,t 0.5 0.45 0.44 0.46 0.44 0.38
75plust−2,t 0.34 0.38 0.41 0.38 0.33 0.35
75plust−3,t 0.28 0.32 0.37 0.31 0.26 0.27
90plust−1,t 0.3 0.26 0.27 0.31 0.23 0.23
90plust−2,t 0.21 0.21 0.25 0.16 0.19 0.14
90plust−3,t 0.2 0.17 0.18 0.06 0.12 0.12
Regional Regional/Type Regional/Subtype Regional Regional/Type Regional/Subtype
ρt,t−1 0.3873∗∗∗ 0.2666∗∗∗ 0.2623∗∗∗ 0.3934∗∗∗ 0.329∗∗∗ 0.3075∗∗∗
ρt,t−2 0.1806∗∗∗ 0.1138∗∗∗ 0.1287∗∗∗ 0.1874∗∗∗ 0.1739∗∗∗ 0.1947∗∗∗
ρt,t−3 0.0376 0.0484 0.0446 0.04 0.0839∗∗ 0.0547
75plust−1,t 0.49 0.4 0.41 0.44 0.42 0.4
75plust−2,t 0.29 0.33 0.32 0.36 0.34 0.33
75plust−3,t 0.24 0.26 0.26 0.3 0.3 0.28
90plust−1,t 0.33 0.22 0.26 0.34 0.28 0.19
90plust−2,t 0.2 0.18 0.21 0.23 0.22 0.18
90plust−3,t 0.22 0.21 0.14 0.03 0.13 0.07
Divisional Divisional/Type Divisional/Subtype Divisional Divisional/Type Divisional/Subtype
ρt,t−1 0.3843∗∗∗ 0.2746∗∗∗ 0.2358∗∗∗ 0.3468∗∗∗ 0.2732∗∗∗ 0.306∗∗∗
ρt,t−2 0.166∗∗∗ 0.0965∗∗∗ 0.1072∗∗∗ 0.1788∗∗∗ 0.1515∗∗∗ 0.1414∗∗∗
ρt,t−3 0.0091 −0.0057 0.0339 0.0231 0.0833∗∗ 0.0364
75plust−1,t 0.5 0.42 0.38 0.45 0.4 0.38
75plust−2,t 0.32 0.32 0.3 0.38 0.32 0.3
75plust−3,t 0.24 0.26 0.25 0.28 0.29 0.26
90plust−1,t 0.39 0.26 0.31 0.34 0.26 0.25
90plust−2,t 0.19 0.22 0.15 0.21 0.18 0.18
90plust−3,t 0.2 0.19 0.13 0.07 0.1 0.12
State State/Type State/Subtype State State/Type State/Subtype
ρt,t−1 0.359∗∗∗ 0.1909∗∗∗ 0.1609∗∗∗ 0.3295∗∗∗ 0.2752∗∗∗ 0.2891∗∗∗
ρt,t−2 0.1179∗∗∗ 0.0431 −0.0165 0.1501∗∗∗ 0.1187∗∗∗ 0.1036∗∗∗
ρt,t−3 −0.0328 −0.0942∗∗∗ −0.0843∗∗ −0.018 0.0656∗ 0.0013
75plust−1,t 0.45 0.38 0.34 0.46 0.44 0.42
75plust−2,t 0.3 0.31 0.24 0.34 0.31 0.32
75plust−3,t 0.22 0.24 0.24 0.28 0.26 0.27
90plust−1,t 0.33 0.27 0.28 0.27 0.39 0.31
90plust−2,t 0.19 0.11 0.15 0.19 0.23 0.22
90plust−3,t 0.19 0.08 0.11 0.1 0.11 0.15
CBSA CBSA
ρt,t−1 0.4066∗∗∗ 0.2805∗∗∗
ρt,t−2 0.1336∗∗∗ 0.0741∗∗
ρt,t−3 −0.0528 −0.037
75plust−1,t 0.5 0.38
75plust−2,t 0.32 0.33
75plust−3,t 0.23 0.26
90plust−1,t 0.34 0.22
90plust−2,t 0.19 0.15
90plust−3,t 0.14 0.06
∗ : significance level ≤ 10%. ∗∗: significance level ≤ 5%. ∗∗∗: significance level ≤ 1%.
Table 12: Top-Decile Forward Investment Returns
This table shows Market Timing- and Market Selection-based returns with respect to a particular benchmark level,
obtained by investing during year t + 1 into all portfolios ranked in the top decile for year t, with respect to the
same measure and benchmark level. The figures shown are time-series averages of equal-weighted cross-sectional mean
returns for each year, and t-statistics testing the hypothesis that the time-series average of cross-sectional means is
equal to zero, against the two-sided alternative.
REITs NCREIF Private
MT Mean t-statistic Mean t-statistic
National/Type −0.0151 −0.69 −0.0284 −1.62
National/Subtype −0.0211 −1.17 −0.0331 −1.95∗
Regional −0.0299 −1.75 −0.0316 −1.95∗
Regional/Type −0.0344 −1.80∗ −0.0335 −1.97∗
Regional/Subtype −0.0359 −2.35∗∗ −0.0373 −2.22∗∗
Divisional −0.0296 −1.70 −0.0319 −1.94∗
Divisional/Type −0.0328 −1.90∗ −0.0350 −2.09∗
Divisional/Subtype −0.0364 −2.21∗∗ −0.0399 −2.29∗∗
State −0.0368 −2.47∗∗ −0.0397 −2.69∗∗
State/Type −0.0423 −2.14∗ −0.0384 −2.22∗∗
State/Subtype −0.0480 −1.94∗ −0.0401 −2.51∗∗
CBSA −0.0470 −2.72∗∗ −0.0408 −3.03∗∗
CBSA/Type −0.0530 −2.45∗∗ −0.0372 −2.35∗∗
CBSA/Subtype −0.0462 −1.61 −0.0373 −2.44∗∗
MS Mean t-statistic Mean t-statistic
National 0.0224 3.41∗∗∗ 0.0130 3.16∗∗∗
National/Type 0.0139 2.81∗∗ 0.0080 2.91∗∗
National/Subtype 0.0102 1.67 0.0046 1.64
Regional 0.0259 3.43∗∗∗ 0.0169 3.90∗∗∗
Regional/Type 0.0049 1.02 0.0078 2.61∗∗
Regional/Subtype 0.0052 0.93 0.0069 2.65∗∗
Divisional 0.0262 3.39∗∗∗ 0.0139 3.21∗∗∗
Divisional/Type 0.0057 0.98 0.0056 2.40∗∗
Divisional/Subtype 0.0087 1.70 0.0073 2.46∗∗
State 0.0240 3.34∗∗∗ 0.0137 3.49∗∗∗
State/Type 0.0049 1.08 0.0063 2.88∗∗
State/Subtype 0.0024 0.52 0.0060 2.39∗∗
CBSA 0.0197 3.22∗∗∗ 0.0108 2.81∗∗
∗ : significance level ≤ 10%. ∗∗: significance level ≤ 5%. ∗∗∗: significance level ≤ 1%.
Table 13: Rank-Decile Transition Stability and Skewness
This table shows stability and skewness of rank-decile transition matrices by investor type, measure type, and bench-
mark. For each benchmark level, the first line shows stability for each decile. This is the fraction of managers that were
in decile d in year t− 1 and are still in decile d in year t. The second line for each benchmark level shows skewness for
each decile. This is the ratio of the fraction of managers who were in decile d in year t− 1 and were in a decile > d in
year t (upgrades), over the fraction of managers who were in decile d in year t− 1 and were in a decile < d in year t
(downgrades).
Panel A: REITs, MT .
Decile 1 2 3 4 5 6 7 8 9 10
National/Type 0.2479 0.1765 0.2061 0.0635 0.1157 0.2033 0.2114 0.1057 0.0977 0.3562
3.2609 2.4667 2.3714 0.7541 0.661 0.7963 0.4865 0.2632
National/Subtype 0.2743 0.1679 0.152 0.0656 0.1181 0.1769 0.0726 0.123 0.1382 0.0963
4.45 2.3125 1.7143 1 0.8136 0.5132 0.321 0.4324
Regional 0.25 0.1579 0.1328 0.1679 0.0976 0.0846 0.1154 0.1176 0.1221 0.0735
1.8235 2.5806 2.8929 2.0833 1.3333 0.5333 0.2069 0.1979
Regional/Type 0.2231 0.189 0.1417 0.1181 0.1603 0.072 0.121 0.1048 0.1679 0.1057
3.9048 3.037 1.6047 1.3404 0.871 0.3625 0.5634 0.1515
Regional/Subtype 0.1966 0.2177 0.1138 0.0909 0.1024 0.0551 0.1405 0.128 0.1395 0.1282
3.4091 3.36 1.973 1.85 0.6438 0.4247 0.2824 0.0777
Divisional 0.3115 0.2034 0.119 0.0887 0.0698 0.063 0.097 0.0873 0.0692 0.1385
5.2667 3.1111 2.5312 2.0769 0.8889 0.3444 0.1058 0.1101
Divisional/Type 0.2295 0.1613 0.0787 0.127 0.1318 0.1102 0.1102 0.1484 0.0977 0.0894
4.4737 3.3333 1.6829 1.383 0.9825 0.4868 0.3625 0.1429
Divisional/Subtype 0.181 0.1453 0.1545 0.0847 0.1048 0.122 0.1406 0.1513 0.1311 0.1217
3.7619 3.5217 1.9189 1.1765 0.6364 0.3415 0.2169 0.1042
State 0.2288 0.127 0.144 0.1077 0.1371 0.0846 0.1061 0.0887 0.1128 0.0813
6.3333 2.2424 2.1351 1.8158 0.8594 0.5325 0.2989 0.1132
State/Type 0.181 0.1032 0.041 0.1074 0.1885 0.1395 0.1008 0.1318 0.152 0.0917
5.2778 3.5 1.5714 0.8679 1.1765 0.45 0.3176 0.0291
State/Subtype 0.1887 0.1038 0.1261 0.0943 0.125 0.0893 0.1593 0.1607 0.1339 0.1143
4.2778 3.85 2 1.2791 0.7288 0.6102 0.2368 0.0319
CBSA 0.2269 0.1371 0.0833 0.0709 0.0968 0.1032 0.0703 0.0866 0.0859 0.0427
7.2308 3.5833 2.9333 1.383 0.8525 0.5867 0.234 0.0833
CBSA/Type 0.1373 0.1028 0.1593 0.1101 0.1518 0.0965 0.1356 0.0885 0.0855 0.1143
3.3636 2.9583 2.129 1.3171 0.9808 0.5 0.2875 0.0808
CBSA/Subtype 0.2727 0.1064 0.0941 0.125 0.1277 0.1809 0.1158 0.1429 0.0444 0.0732
8.3333 2.3478 1.4062 1.4848 1.2 0.75 0.3 0.0488
Panel B: REITs, MS.
Decile 1 2 3 4 5 6 7 8 9 10
National 0.2857 0.1825 0.209 0.1496 0.1538 0.156 0.197 0.2016 0.2422 0.3248
3.1481 2.3125 1.5116 0.9298 1.0517 0.5362 0.6885 0.2763
National/Type 0.3433 0.1628 0.1085 0.0889 0.2016 0.1971 0.1825 0.2 0.1716 0.2895
7.3077 2.3824 1.2364 1.0196 0.9643 0.8361 0.3867 0.2759
National/Subtype 0.3333 0.1719 0.136 0.1128 0.1855 0.1429 0.1642 0.2 0.1894 0.3036
4.5789 3.32 1.8095 0.9057 0.7812 0.7231 0.6 0.2299
Regional 0.3358 0.1729 0.1298 0.1615 0.1756 0.1176 0.1756 0.1705 0.203 0.3504
5.1111 2.2571 1.6585 1.5714 0.8462 0.3333 0.5286 0.3086
Regional/Type 0.3411 0.1866 0.1429 0.1085 0.145 0.156 0.1128 0.1579 0.2 0.2435
6.7857 2.5625 2.0263 1.7317 0.9194 0.6389 0.3333 0.1818
Regional/Subtype 0.3279 0.16 0.088 0.1562 0.125 0.112 0.1278 0.1417 0.152 0.2719
6.5 3.75 1.9189 1.383 0.7903 0.45 0.4156 0.2471
Divisional 0.3478 0.1769 0.1111 0.1667 0.1705 0.1037 0.1231 0.1783 0.2462 0.4083
4.6316 3.4444 1.4444 1.6098 0.5316 0.4805 0.493 0.3067
Divisional/Type 0.3507 0.1742 0.1719 0.1417 0.1194 0.1387 0.1752 0.1374 0.188 0.287
7.3846 2.1176 2.5161 1.5106 0.7612 0.5067 0.4487 0.2273
Divisional/Subtype 0.312 0.152 0.1393 0.15 0.1475 0.1562 0.1406 0.119 0.1048 0.3304
5.625 2.5 2.7778 1.5366 0.8621 0.6418 0.3537 0.1443
State 0.3404 0.1439 0.1212 0.1008 0.1429 0.0677 0.1163 0.2188 0.2385 0.3388
5.2778 1.9744 1.6977 1.2353 0.6757 0.5 0.4286 0.32
State/Type 0.3876 0.1382 0.127 0.136 0.1339 0.1406 0.12 0.1769 0.0873 0.2832
7.8333 2.5484 1.9189 1.8947 0.6923 0.4865 0.4459 0.2105
State/Subtype 0.3036 0.1364 0.1441 0.1391 0.1321 0.181 0.1429 0.1197 0.0862 0.297
4.2778 3.75 2.1935 1.5556 0.8269 0.4571 0.3553 0.1648
CBSA 0.2929 0.1374 0.1194 0.1615 0.16 0.1364 0.1719 0.1797 0.2481 0.35
3.0357 1.4583 1.5349 1.234 1.3265 0.5143 0.4189 0.2436
Panel C: NCREIF Private, MT .
Decile 1 2 3 4 5 6 7 8 9 10
National/Type 0.2909 0.1111 0.1579 0.1532 0.1327 0.1667 0.1121 0.1322 0.2328 0.1827
2.84 2.4286 1.6111 0.9216 0.9444 0.3377 0.3816 0.2714
National/Subtype 0.3208 0.1509 0.1652 0.1504 0.1858 0.1129 0.1083 0.1271 0.1624 0.2376
5.9231 1.7429 1.5263 1.359 0.8644 0.5286 0.2561 0.1264
Regional 0.3056 0.1712 0.1404 0.1217 0.0957 0.0826 0.0957 0.087 0.1349 0.2626
2.8333 2.3793 1.0612 1.8108 0.6818 0.5758 0.2805 0.1848
Regional/Type 0.2885 0.1217 0.1092 0.1171 0.0893 0.104 0.1525 0.1864 0.1681 0.1735
4.05 3.6087 1.3333 1.3182 0.5135 0.5625 0.3714 0.1928
Regional/Subtype 0.2963 0.1308 0.0847 0.1121 0.1062 0.1157 0.1121 0.1417 0.181 0.1531
4.4706 1.7 1.8611 0.8704 1.1837 0.5606 0.2118 0.1585
Divisional 0.3148 0.2091 0.1282 0.1327 0.0948 0.1167 0.1478 0.1197 0.1624 0.2547
4.8 2.5172 1.2273 1.1 0.963 0.6333 0.1977 0.1807
Divisional/Type 0.2667 0.1339 0.1513 0.1062 0.0763 0.1271 0.1034 0.1356 0.15 0.16
4.7059 3.8095 1.9706 0.9818 0.7167 0.6 0.2911 0.2143
Divisional/Subtype 0.2941 0.1491 0.1652 0.1404 0.0734 0.1026 0.1892 0.1026 0.1552 0.1515
3.619 3.3636 1.7222 0.9804 1.0588 0.5254 0.3462 0.1011
State 0.2593 0.0909 0.1379 0.1193 0.0932 0.15 0.0833 0.161 0.1197 0.1373
5.6667 3.3478 1.5263 0.9815 0.9245 0.746 0.32 0.1839
State/Type 0.2336 0.0982 0.0973 0.1304 0.1754 0.1057 0.1217 0.1293 0.1368 0.1368
3.2083 2.9231 2.7037 1.1364 0.8966 0.4028 0.2625 0.0978
State/Subtype 0.2828 0.1545 0.1071 0.1351 0.0943 0.1102 0.0609 0.114 0.1239 0.163
4.4706 3.3478 1.7429 1.4 0.6935 0.6119 0.2317 0.1124
CBSA 0.2212 0.08 0.1228 0.0926 0.1429 0.0984 0.1121 0.1239 0.1138 0.1368
3.381 2.7037 1.45 0.9592 1.1569 0.4714 0.2532 0.09
CBSA/Type 0.2 0.1215 0.0943 0.1579 0.0762 0.1102 0.1034 0.027 0.1466 0.086
5.7143 3.1739 1.7429 1.1556 0.6406 0.5072 0.2857 0.1647
CBSA/Subtype 0.2381 0.0808 0.1415 0.1048 0.1176 0.0849 0.1009 0.1359 0.1698 0.1758
5.5 4.3529 1.4737 0.8367 0.6441 0.7193 0.2535 0.0864
Panel D: NCREIF Private, MS.
Decile 1 2 3 4 5 6 7 8 9 10
National 0.4272 0.2039 0.11 0.1489 0.1515 0.1238 0.1881 0.1935 0.243 0.3553
4.125 1.9667 1.5 1.1538 1.0909 0.6078 0.5306 0.2857
National/Type 0.4571 0.1717 0.1961 0.134 0.1212 0.13 0.1373 0.117 0.1869 0.2632
3.5556 2.4167 1.4 1.4167 0.8125 0.6604 0.431 0.2609
National/Subtype 0.4151 0.1633 0.101 0.1373 0.1196 0.0865 0.1569 0.0879 0.1765 0.2597
2.9048 2.2963 1.75 0.9756 1.2093 0.6538 0.2969 0.1667
Regional 0.3558 0.1748 0.1837 0.2308 0.0968 0.1485 0.0909 0.1075 0.1604 0.375
3.0476 3.7059 1.5 1.0488 0.9111 0.6981 0.431 0.2535
Regional/Type 0.3714 0.1313 0.1456 0.1111 0.043 0.1415 0.1515 0.1458 0.13 0.3086
5.1429 2.52 2.0345 1.4054 0.569 0.5 0.4909 0.2254
Regional/Subtype 0.3148 0.1735 0.1429 0.1485 0.1489 0.1875 0.1165 0.1398 0.1386 0.2
3.7647 3.4211 1.3889 1.2222 0.8571 0.5167 0.4286 0.2794
Divisional 0.3654 0.1538 0.11 0.2041 0.125 0.0882 0.17 0.129 0.1981 0.3846
6.3333 2.56 1.4375 1.2703 0.898 0.431 0.4727 0.2319
Divisional/Type 0.2621 0.1731 0.1667 0.1078 0.1053 0.1143 0.0928 0.1099 0.181 0.2987
3.5263 4.3125 1.5278 0.8478 0.7885 0.4915 0.4211 0.2464
Divisional/Subtype 0.2547 0.1236 0.1524 0.1522 0.0737 0.1532 0.1489 0.1489 0.2258 0.2561
2.5455 2.069 1.6 1.2 0.9184 0.6667 0.3333 0.1429
State 0.3302 0.1327 0.1277 0.1828 0.1635 0.1509 0.15 0.134 0.2039 0.3
4.6667 3.1 1.303 1.4167 0.6364 0.5741 0.3548 0.4138
State/Type 0.2736 0.0722 0.1562 0.1111 0.1237 0.1165 0.1895 0.2174 0.2574 0.3953
5.4286 2.6818 1.6667 1.125 0.5167 0.7111 0.3091 0.1905
State/Subtype 0.3619 0.2211 0.134 0.1489 0.0538 0.0833 0.0962 0.1556 0.1856 0.3375
4.6923 3 2.6364 1.1463 0.913 0.4242 0.2459 0.2742
CBSA 0.3235 0.2143 0.1717 0.102 0.1099 0.1019 0.0926 0.125 0.0926 0.2778
4.9231 2.9048 1.8387 1.7931 0.7321 0.6333 0.4237 0.2727
Table 14: Regression Results, MT Measures, Public Portfolios
Dependent variable: Quarterly MT measure by manager, for various aggregation levels of benchmark. This table
presents results from panel regressions, by time and manager, for public portfolios. The independent variables are
geographic portfolio specialization, property-type portfolio specialization, log of portfolio size and fractional turnover
of the portfolio over time. All regressions contain year fixed-effects and standard errors clustered by manager.
National/Type National/Subtype
Coefficient t-statistic Coefficient t-statistic
(Intercept) 0.007913 2.97∗∗∗ 0.005655 1.95∗
geo.spec −0.001282 −1.28 −0.000511 −0.5
type.spec 0.002303 2.14∗∗ 0.001479 1.33
log(size) −0.000569 −3.52∗∗∗ −0.000417 −2.41∗∗
turnover −0.001810 −2.86∗∗∗ −0.001748 −2.77∗∗∗
R2 0.459 0.412
F 311.116 254.641
Regional Regional/Type Regional/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) 0.004792 1.95∗ 0.008530 3.14∗∗∗ 0.007593 2.53∗∗
geo.spec 0.000945 1.01 −0.001509 −1.41 −0.000855 −0.75
type.spec −0.000005 −0.01 0.002300 2.26∗∗ 0.001282 1.22
log(size) −0.000171 −1.14 −0.000597 −3.68∗∗∗ −0.000525 −2.87∗∗∗
turnover −0.001353 −2.7∗∗∗ −0.001832 −2.83∗∗∗ −0.001852 −2.67∗∗∗
R2 0.508 0.411 0.343
F 378.345 256.207 187.227
Divisional Divisional/Type Divisional/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) 0.005960 2.33∗∗ 0.010498 3.72∗∗∗ 0.009303 2.83∗∗∗
geo.spec 0.000488 0.5 −0.002057 −1.91∗ −0.001782 −1.48
type.spec −0.000058 −0.06 0.002095 1.93∗ 0.001700 1.44
log(size) −0.000252 −1.62 −0.000697 −4.23∗∗∗ −0.000667 −3.41∗∗∗
turnover −0.001326 −2.69∗∗∗ −0.001612 −2.71∗∗∗ −0.001639 −2.73∗∗∗
R2 0.482 0.377 0.299
F 341.483 221.699 150.315
State State/Type State/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) 0.004582 1.79∗ 0.011113 3.63∗∗∗ 0.008035 2.27∗∗
geo.spec 0.000040 0.04 −0.002158 −1.87∗ −0.000831 −0.63
type.spec −0.000188 −0.19 0.002104 1.73∗ 0.001201 0.88
log(size) −0.000276 −1.74∗ −0.000780 −4.27∗∗∗ −0.000541 −2.46∗∗
turnover −0.001414 −3.17∗∗∗ −0.001417 −2.87∗∗∗ −0.002149 −2.86∗∗∗
R2 0.418 0.286 0.245
F 262.503 144.597 104.692
CBSA CBSA/Type CBSA/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) 0.003035 1.19 0.010545 3.23∗∗∗ 0.009679 2.53∗∗
geo.spec 0.000292 0.28 −0.001716 −1.27 −0.001361 −0.89
type.spec −0.000489 −0.46 0.001574 1.2 0.000726 0.49
log(size) −0.000244 −1.63 −0.000736 −3.83∗∗∗ −0.000683 −2.95∗∗∗
turnover −0.001486 −3.3∗∗∗ −0.001848 −2.87∗∗∗ −0.002055 −2.61∗∗∗
R2 0.346 0.232 0.216
F 190.882 99.724 75.911
∗ : significance level ≤ 10%. ∗∗: significance level ≤ 5%. ∗∗∗: significance level ≤ 1%.
Table 15: Regression Results, MT Measures, Private Portfolios
Dependent variable: Quarterly MT measure by manager, for various aggregation levels of benchmark. This table
presents results from panel regressions, by time and manager, for private portfolios. The independent variables are
geographic portfolio specialization, property-type portfolio specialization, log of portfolio size and fractional turnover
of the portfolio over time. All regressions contain year fixed-effects and standard errors clustered by manager.
National/Type National/Subtype
Coefficient t-statistic Coefficient t-statistic
(Intercept) −0.007710 −2.01∗∗ −0.006309 −1.48
geo.spec 0.002269 1.9∗ 0.002644 2.03∗∗
type.spec −0.001367 −1.34 −0.001646 −1.52
log(size) 0.000390 2.11∗∗ 0.000315 1.53
turnover −0.000921 −2.96∗∗∗ −0.000795 −2.49∗∗
R2 0.56 0.548
F 195.593 185.88
Regional Regional/Type Regional/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) −0.010816 −2.59∗∗∗ −0.008428 −2.13∗∗ −0.006272 −1.34
geo.spec 0.002895 2.44∗∗ 0.001947 1.43 0.002827 1.76∗
type.spec −0.000769 −0.77 −0.001821 −1.64 −0.002160 −1.81∗
log(size) 0.000497 2.61∗∗∗ 0.000377 1.92∗ 0.000276 1.29
turnover −0.000774 −2.81∗∗∗ −0.000820 −2.74∗∗∗ −0.000743 −2.25∗∗
R2 0.557 0.536 0.503
F 193.063 177.683 155.196
Divisional Divisional/Type Divisional/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) −0.007972 −1.47 −0.009355 −2.14∗∗ −0.004774 −0.95
geo.spec 0.002994 1.98∗∗ 0.001773 1.05 0.003646 1.76∗
type.spec −0.000964 −0.93 −0.001432 −1.22 −0.002421 −1.89∗
log(size) 0.000504 2.42∗∗ 0.000432 2∗∗ 0.000342 1.43
turnover −0.000648 −2.32∗∗ −0.000694 −2.21∗∗ −0.000639 −1.89∗
R2 0.547 0.518 0.494
F 185.742 165.44 147.501
State State/Type State/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) −0.006485 −1.24 −0.009941 −1.94∗ −0.001818 −0.31
geo.spec 0.002285 1.19 0.003686 2.08∗∗ 0.004750 1.78∗
type.spec −0.001567 −1.47 −0.001717 −1.42 −0.001448 −1.03
log(size) 0.000362 1.53 0.000512 2.1∗∗ 0.000582 2.16∗∗
turnover −0.000786 −2.7∗∗∗ −0.000819 −2.35∗∗ −0.000796 −2.28∗∗
R2 0.519 0.462 0.442
F 165.778 131.366 117.464
CBSA CBSA/Type CBSA/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) −0.006814 −1.2 −0.005732 −1.02 −0.000977 −0.14
geo.spec 0.002767 1.38 0.001823 0.92 0.002664 0.8
type.spec −0.001659 −1.51 −0.001255 −0.95 −0.001162 −0.74
log(size) 0.000436 1.85∗ 0.000505 1.91∗ 0.000490 1.5
turnover −0.000756 −2.1∗∗ −0.000912 −2.19∗∗ −0.000912 −2.09∗∗
R2 0.494 0.424 0.417
F 147.33 109.84 101.723
∗ : significance level ≤ 10%. ∗∗: significance level ≤ 5%. ∗∗∗: significance level ≤ 1%.
Table 16: Regression Results, MS Measures, Public Portfolios
Dependent variable: Quarterly MS measure by manager, for various aggregation levels of benchmark. This table
presents results from panel regressions, by time and manager, for public portfolios. The independent variables are
geographic portfolio specialization, property-type portfolio specialization, log of portfolio size and fractional turnover
of the portfolio over time. All regressions contain year fixed-effects and standard errors clustered by manager.
National National/Type National/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) 0.011762 3.58∗∗∗ 0.009184 3.08∗∗∗ 0.008714 2.82∗∗∗
geo.spec −0.002169 −1.56 −0.002085 −1.72∗ −0.002065 −1.62
type.spec −0.001453 −1.21 0.000121 0.11 −0.000361 −0.32
log(size) −0.000555 −3.06∗∗∗ −0.000528 −3.15∗∗∗ −0.000512 −3.03∗∗∗
turnover 0.001116 1.75∗ 0.000642 1.27 0.000869 1.86∗
R2 0.022 0.02 0.033
F 8.225 7.54 11.899
Regional Regional/Type Regional/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) 0.010335 3.31∗∗∗ 0.007512 2.79∗∗∗ 0.008322 2.73∗∗∗
geo.spec −0.002026 −1.45 −0.001915 −1.69∗ −0.002269 −1.7∗
type.spec −0.001559 −1.61 −0.000138 −0.18 −0.000064 −0.07
log(size) −0.000464 −2.39∗∗ −0.000414 −2.65∗∗∗ −0.000396 −2.21∗∗
turnover 0.001174 1.76∗ 0.000842 1.78∗ 0.000806 1.93∗
R2 0.016 0.015 0.023
F 6.206 6.067 8.588
Divisional Divisional/Type Divisional/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) 0.009374 3.09∗∗∗ 0.006999 2.81∗∗∗ 0.007050 2.27∗∗
geo.spec −0.001660 −1.2 −0.001525 −1.39 −0.002181 −1.61
type.spec −0.001247 −1.32 −0.000161 −0.23 0.000019 0.02
log(size) −0.000435 −2.3∗∗ −0.000387 −2.82∗∗∗ −0.000422 −2.27∗∗
turnover 0.001080 1.67∗ 0.000824 2.04∗∗ 0.001029 2.54∗∗
R2 0.012 0.014 0.016
F 4.814 5.569 6.168
State State/Type State/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) 0.007897 2.76∗∗∗ 0.003927 2.34∗∗ 0.004496 2.45∗∗
geo.spec −0.001875 −1.4 −0.001347 −1.52 −0.001290 −1.25
type.spec −0.001056 −1.16 −0.000304 −0.58 0.000010 0.02
log(size) −0.000320 −1.78∗ −0.000181 −2.02∗∗ −0.000227 −2.45∗∗
turnover 0.001055 1.65∗ 0.000451 1.25 0.000579 2.09∗∗
R2 0.009 0.006 0.005
F 3.952 2.976 2.328
CBSA
Coefficient t-statistic
(Intercept) 0.003760 1.44
geo.spec −0.001113 −1.02
type.spec −0.001043 −1.11
log(size) −0.000196 −1.25
turnover 0.000927 1.64
R2 0.004
F 2.331
∗ : significance level ≤ 10%. ∗∗: significance level ≤ 5%. ∗∗∗: significance level ≤ 1%.
Table 17: Regression Results, MS Measures, Private Portfolios
Dependent variable: Quarterly MS measure by manager, for various aggregation levels of benchmark. This table
presents results from panel regressions, by time and manager, for private portfolios. The independent variables are
geographic portfolio specialization, property-type portfolio specialization, log of portfolio size and fractional turnover
of the portfolio over time. All regressions contain year fixed-effects and standard errors clustered by manager.
National National/Type National/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) −0.004853 −1.07 0.003248 0.74 0.008126 1.6
geo.spec −0.003353 −1.57 −0.003472 −1.64 −0.004740 −1.97∗∗
type.spec 0.002276 1.79∗ 0.000473 0.53 −0.000266 −0.31
log(size) 0.000207 0.79 0.000008 0.03 −0.000270 −0.98
turnover −0.000004 −0.02 −0.000183 −0.96 −0.000136 −0.72
R2 0.034 0.021 0.019
F 5.536 3.732 3.49
Regional Regional/Type Regional/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) −0.001697 −0.43 −0.001412 −0.39 0.000285 0.08
geo.spec −0.004178 −2.18∗∗ −0.002991 −1.71∗ −0.003058 −1.7∗
type.spec 0.002451 2.09∗∗ 0.000484 0.67 −0.000322 −0.45
log(size) 0.000171 0.73 0.000054 0.28 −0.000018 −0.09
turnover −0.000106 −0.47 −0.000302 −1.94∗ −0.000312 −2.14∗∗
R2 0.026 0.008 0.01
F 4.42 2.054 2.369
Divisional Divisional/Type Divisional/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) −0.001308 −0.37 −0.000456 −0.16 −0.001721 −0.6
geo.spec −0.003496 −1.98∗∗ −0.001860 −1.32 −0.001045 −0.61
type.spec 0.002113 1.86∗ 0.000134 0.22 −0.000603 −0.95
log(size) 0.000116 0.56 −0.000013 −0.09 0.000086 0.59
turnover 0.000067 0.32 −0.000049 −0.33 −0.000059 −0.38
R2 0.023 0.004 0.009
F 4.006 1.513 2.101
State State/Type State/Subtype
Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic
(Intercept) −0.004183 −1.24 −0.004010 −1.84∗ −0.004032 −1.57
geo.spec −0.002317 −1.18 0.000359 0.29 0.000321 0.19
type.spec 0.001667 1.49 0.000141 0.3 −0.000105 −0.2
log(size) 0.000152 0.73 0.000071 0.59 0.000091 0.71
turnover 0.000086 0.43 −0.000087 −0.71 0.000009 0.06
R2 0.019 0.001 0.005
F 3.548 1.192 1.581
CBSA
Coefficient t-statistic
(Intercept) −0.002053 −0.71
geo.spec −0.002882 −1.97∗∗
type.spec 0.001715 1.85∗
log(size) 0.000066 0.37
turnover 0.000056 0.31
R2 0.018
F 3.411
∗ : significance level ≤ 10%. ∗∗: significance level ≤ 5%. ∗∗∗: significance level ≤ 1%.