revisiting `stylized facts` about hedge funds - Workspace

REVISITING ‘STYLIZED FACTS’ ABOUT HEDGE
FUNDS
JUHA JOENVÄÄRÄ, ROBERT KOSOWSKI
AND PEKKA TOLONEN
Centre for Hedge Fund Research, Risk Management Laboratory
Imperial College Business School
Working Paper 10
Date: 3/2012
Revisiting ‘stylized facts’ about hedge funds*
Juha Joenvääräa, Robert Kosowskib, and Pekka Tolonen c
a
b
University of Oulu
Imperial College Business School, Imperial College, London
c
University of Oulu and GSF
The First Draft; May 2011
This Draft; February 2012
_______________________________________________________________________
Abstract
This paper presents new stylized facts about hedge fund performance and data biases
based on a novel database aggregation. By highlighting the effect of data base differences on
previously documented results we aim to improve the ability of researchers in this literature to
compare results across different studies. We document economically important positive riskadjusted performance of the average fund while differences in its magnitude are due to
differences in fund size, domicile and data biases, but not differences in fund risk exposures.
Measures of misreporting and return smoothing by funds are similar across different data bases.
Performance persistence is sensitive to share restrictions, rebalancing frequency, fund size and
weighting scheme as well as more pronounced biases in certain databases. Hedge funds with
greater managerial incentives, smaller funds and younger funds outperform while hedge funds
with strict share restrictions are not associated with higher risk-adjusted returns. Since several
stylized facts are sensitive to the choice of the database it is important to use a high quality
consolidated database such as the one used in this paper.
JEL Classifications: G11, G12, G23
Keywords: hedge fund performance, persistence, sample selection bias, managerial skill
_______________________________________________________________________
*We would like to thank participants at the CREST Paris 4th Annual Hedge Fund conference 2012 and
GSF summer workshop in Helsinki, in particular, Arjen Siegmann, the discussant, and Petri Jylhä. We
would like to thank Pertrac for providing us access to with the Pertrac Analytical Platform. We are grateful
for the support of OP-Pohjola Group Research Foundation. Contact address: Juha Joenväärä, University of
Oulu, juha.joenvaara@oulu.fi. Robert Kosowski, Imperial College Business School, Imperial College,
London, UK, r.kosowski@imperial.ac.uk. Pekka Tolonen, University of Oulu and GSF,
pekka.tolonen@oulu.fi. Online appendix is available at the web page:
https://workspace.imperial.ac.uk/business-school/Public/research/f_agroup/JKT_Appendix.pdf.
The usual disclaimer applies.
Electronic copy available at: http://ssrn.com/abstract=1989410
1.
Introduction
This paper presents new stylized facts about hedge fund performance and data biases
based on a novel database aggregation and a comprehensive analysis of differences between the
main commercial hedge fund databases. By highlighting the effect of data base differences on
previously documented results we aim to improve the ability of researchers in this literature to
compare results across different studies. While several hedge fund studies build and use a large
consolidated database containing multiple databases, there is no standard merging methodology
described in the literature that could be used as a benchmark to help gauge the sensitivity of
findings to the databases employed. Another contribution of this paper is that our results can be
viewed as laying the foundations for an industry standard for matching hedge fund databases so
that the consolidated data is designed to be as close as possible to the true unobserved population.
We demonstrate the properties of our consolidated database by revisiting stylized facts about
average hedge fund performance, performance persistence, data biases, management company
domicile and fund-specific characteristics explaining cross-sectional differences in hedge fund
performance. First, we find that the results inferred from the aggregate database qualitatively
differ from those found in the previous literature. Importantly, the stylized facts obtained using
one database are often in contrast to results inferred from the consolidated database. Our aim is
not to produce ‘back-tests’ of earlier studies and our results should not be interpreted as
questioning earlier findings. The reason is that differences in our findings compared to previous
studies may also be due to revisions the same database over time, an issue recently documented
by Patton, Ramadorai and Streatfield (2011). Second, our overall findings show the importance of
using an aggregate database in hedge fund research and also when allocating capital to hedge
funds in practice, since the results based on a single database are often not representative and may
even be misleading compared to findings based on the aggregate database.
While the hedge fund literature is continuing to grow significantly, to the best of our
knowledge there is no comprehensive study that would compare the stylized facts in the literature
based on individual and consolidated databases. Such a comparison would be very useful for
academic researchers and practitioners; for mutual funds, for example, Elton, Gruber and Blake
(2001) find systematic differences in returns between the popular Morningstar and CRSP mutual
fund databases. They show that these differences are important, since they may change the
conclusion about individual mutual funds or a group of mutual funds.
Our paper fills the
equivalent research gap for hedge funds and aims to assist hedge fund researchers in evaluating
Electronic copy available at: http://ssrn.com/abstract=1989410
their database choice and the understanding of differences in results between databases. We argue
that the database selection is even more important in hedge fund studies than in the mutual fund
literature for three reasons. First there are 5-10 commercially available hedge fund databases
while there are only two main databases used in the majority of mutual fund studies (CRSP and
Morningstar). Second, hedge fund databases are also non-overlapping - we find that almost 70%
of funds in our consolidated database report only to one of the used major databases. Third,
existing research documents a larger number of data biases in hedge fund databases than in
mutual funds which highlights the importance of comparing the quality of individual databases.
To understand why the performance results differ across databases, we start by
highlighting how the return, the asset under management (AuM) time-series, and the attrition
rates differ between commercial and consolidated databases based for a data download that we
carried out in Q3 of 2011. Overall, we find that the aggregate data set consists of 24,749 unique
hedge funds that report at least 12 return observations. For these hedge funds, 8,512 are active,
while 16,237 are not providing any data to vendors, thus, we classify them as defunct. The
number of hedge funds in our database is close to that reported by the UBS’ proprietary AIS
database consisting of about 20,000 hedge funds and 45,000 share classes 1, while the PerTrac
2010 hedge fund database study finds that hedge fund industry contains about 23,600 funds.
Therefore, we believe that our aggregate database containing the union of five major databases is
close to the true unobservable population of hedge funds.
We document that the number of hedge funds ranges across data vendors from 6,772 for
Morningstar to 9,719 for BarclayHedge. EurekaHedge is the largest data vendor in terms of 4,452
active hedge funds. Importantly, the proportion of alive and defunct funds varies significantly
across databases. BarclayHedge, HFR and TASS (EurekaHedge and Morningstar) contain
relatively more (fewer) defunct funds than alive funds. The attrition rates are also remarkably
different across data vendors and time periods. We find that EurekaHedge and Morningstar have
very limited information about defunct funds before 2004. In contrast, BarclayHedge, HFR and
TASS do not seem to suffer from the same lack of data. Hence, the survivorship and backfilling
biases are much higher in EurekaHedge and Morningstar than in the BarclayHedge, HFR and
TASS databases. Therefore, we test the hypotheses that EurekaHedge and Morningstar should
have higher average returns, but weaker performance persistence than the other databases due to
1
See, Güner, Rachev, Edelman, and Fabozzi (2010). The AIS database includes funds that UBS allocated
capital to, but that do not report to commercial databases.
Electronic copy available at: http://ssrn.com/abstract=1989410
these biases. The rationale is that the EurekaHedge’s and Morningstar’s bottom deciles may not
contain a significant number of liquidated funds that deliver poor performance suggesting the
spread portfolio between the top and bottom deciles may be indistinguishable from zero.
Overall, we find that about 28 percent of AuM observations are missing from our
aggregate database. The amount of missing AuM observations varies significantly across data
vendors, being lowest for BarclayHedge (11%) and HFR (19%), while significantly higher for
EurekaHedge (37%), TASS (34%), and Morningstar (32%). Since the performance of the overall
hedge-fund industry is evaluated using value-weighted portfolios, we test the hypothesis that
these differences in the AuM time-series properties lead to different average performance results
based on different databases. To evaluate hedge-fund industry’s average performance, we
examine equal-weighted (EW) and value-weighted (VW) portfolios formed using the aggregate
and single databases. We find clear evidence that hedge funds deliver, on average, economically
and statistically significant abnormal performance on an equal- and value-weighted basis, as well
as across investment strategies, domiciles, size categories, and time-periods, a finding that is
consistent with previous studies such as Kosowski, Naik and Teo (2007).
Specifically, for the aggregate EW portfolio, we find an annualized average excess return
of 8.5 percent and an annualized alpha of 5.9 percent. The average excess returns (alpha) is lower
for the aggregate VW portfolio, at 7.4 (4.8) percent per year. The conclusion that the average
hedge fund (EW or VW) delivers risk-adjusted returns for investors differs significantly across
databases. For example, the annualized equal-weighted average return ranges from 7.9 percent for
TASS to 10.3 percent for EurekaHedge. In contrast, when returns are measured on a valueweighted basis, TASS shows the highest value-weighted average returns of 8.6 percent, being
almost a quarter higher than the lowest respective return for BarclayHedge.
Since we find that Sharpe ratios, risk loadings and R2’s of the Fung and Hsieh (2004)
seven-factor model are very similar across individual and the aggregate databases, this suggests
that return differences cannot be explained by risk exposures. Consistent with our hypothesis we
find that the equal-weighted return differences across databases are driven by the different levels
of survivorship and backfilling biases in the commercial databases. The equal-weight returns are
the highest (lowest) for EurekaHedge and Morningstar (BarclayHedge, HFR and TASS), which
has the lowest (highest) amount of defunct funds. The missing AuM observations and a relatively
high performance of large funds drive the differences in value-weighted portfolios. We find that
high value-weighted returns for TASS are due to the outstanding performance of large funds and
non-randomly missing AuM observations. Importantly, only in the TASS data, a significant
number of funds have a missing AuM observation exactly at the same time when a fund exhibits
a poor return. A low value-weighted performance in the BarclayHedge data is associated with the
fact that it is able to gather significantly more AuM observations for poor performing funds
compared to other databases. We conclude that the return differences between databases are
smaller after 2004 when the survivorship and backfilling biases decrease in the EurekaHedge and
Morningstar databases suggesting that the properties of databases seems to converge over time.
In the paper, we classify hedge funds into twelve main strategies as shown in Online Appendix 2
and show that there are significant differences in the average performance across strategies. Our
findings show that the fund size is strongly related to the levels of average performance. In the
aggregate database, we find that for ten of the twelve style groups small funds outperform large
funds. We also find the onshore based management firms and hedge funds to outperform offshore
vehicles suggesting that domicile also explains differences in the average performance.
Next, we address the economically important question of whether hedge funds are able to
add value after fees by delivering superior performance persistence consistently through time. We
focus on an annual horizon, a realistic frequency given practical redemption restrictions. We
tackle this issue by sorting funds on the t-statistic of alpha obtained from the Fung and Hsieh
(2004) model. The main results obtained using the aggregate database show that there is
economically and statistically significant performance persistence at annual horizon for equalweight portfolios. However, for value-weight portfolios, we document significant performance
persistence only at monthly horizons, which implies that performance persistence vanishes rather
quickly as the holding period increases. This suggests that performance persistence may be
driven by small funds.
We demonstrate that the conclusions about performance persistence are sensitive to the
choice of the data vendor. To highlight the issue, we find that BarclayHedge, HFR and TASS
show economically significant performance persistence for the equal-weighted portfolios at semiannual horizons, whereas using EurekaHedge and Morningstar databases we do not find any
evidence about performance persistence. Consistent with our hypothesis, we find that the average
risk-adjusted returns of EurekaHedge’s and Morningstar’s bottom deciles are significantly higher
2
Online appendix is available at the web page:
https://workspace.imperial.ac.uk/businessschool/Public/research/f_agroup/JKT_Appendix.pdf.
than the respective returns in the BarclayHedge, HFR and TASS databases. In addition, the
dropout rates show that bottom deciles in BarclayHedge, HFR, and TASS databases contain a
significantly higher number of defunct funds compared to the respective deciles for EurekaHedge
and Morningstar. Hence, one plausible reason driving the results is a larger survivorship and
backfilling bias in EurekaHedge and Morningstar. An alternative explanation would be that
BarclayHedge, HFR and TASS contain a set of ‘high quality’ funds that do not report to
EurekaHedge and Morningstar. However, the returns of top decile portfolios are very similar
across data vendors ruling out the possibility that high quality funds are only present in the
BarclayHedge, HFR, and TASS databases. Finally, since none of the databases show significant
persistence at annual horizon as the results for the aggregate database suggest, we interpret this as
evidence supporting the use of a consolidated database in making the conclusion about hedge
fund performance persistence.
We perform also a performance persistence test for the aggregate database in order to
measure whether a real-time investor is able to exploit the short-term performance persistence of
hedge funds. We implement feasible rebalancing strategies by taking into account fund-specific
share restrictions within each rebalancing horizon. Specifically, we exclude funds having lockup,
notice, and redemption periods longer than the rebalancing horizon in question. Based on the
length of notice period, we use lagged information in persistence tests to mitigate look-ahead
bias. For the aggregate database, the results suggest persistence only at quarterly horizon. These
results indicate that one should take share restrictions into account when implementing
persistence tests.
Finally, we examine the cross-sectional relationship between fund characteristics and
hedge fund performance. The existing literature has documented that managerial incentives, share
restrictions and capacity constraints are associated with cross-sectional differences in hedge fund
performance. Using portfolio sorts and the Fama and MacBeth (1973) regressions, we
demonstrate that smaller and younger funds, and funds having greater capital flows deliver better
future returns than their peers. The result is very robust and homogenous across databases. Our
conclusion is in line with the previous literature (e.g., Teo, (2010) and Aggarwal and Jorion
(2010)). In contrast to the existing literature, we find, however, that fund characteristics related to
managerial discretion or illiquidity do not consistently explain hedge fund cross-sectional returns.
In fact, we find very little evidence that share restrictions in the form of lockup, notice and
redemption periods are related to higher risk-adjusted returns when we control for the role of
other characteristics in multivariate regression. Like Aragon (2007), we find the strongest
evidence using the TASS database, whereas using other commercial and aggregate databases the
conclusion is much weaker. For example, we find a significantly positive relation between notice
period and performance only using TASS. Using our aggregate database, we find a significantly
positive cross-sectional relationship between managerial incentives proxied by incentive fees and
high-water mark provision and hedge fund return. Importantly, the signs and significance of highmark provision change wildly across databases suggesting that the conclusion about the impact of
managerial incentives on hedge fund performance varies based on the used data vendor.
However, the overall results that are obtained using the aggregate database are consistent with
Agarwal, Daniel and Naik (2009) showing that hedge funds with greater managerial incentives
deliver higher performance than their peers. We stress that these results should not be interpreted
as ‘back-tests’ of earlier studies since differences between our findings and previous studies may
also be due to revisions of the same data base over time (Patton, Ramadorai and Streatfield
(2011)).
Our paper is related to four streams of performance evaluation literature. First, Elton,
Gruber and Blake (2001) document systematic return differences in CRSP and Morningstar
mutual fund databases. Harris, Jenkinson and Kaplan (2012) show that there are economically
important performance differences among private equity fund databases. Liang (2000) compares
hedge fund survivorship rates between HFR and TASS databases. We add to this literature by
showing that the hedge fund performance results are sensitive to the choice of database or sample.
Importantly, the stylized facts do not only differ between relatively young databases such as
EurekaHedge and Morningstar, but we also document significant heterogeneity among mature
databases such as BarclayHedge, HFR and TASS.
Second, our paper contributes to the hedge fund literature providing an economically and
statistically well-motivated merging approach that other researchers can easily follow. Indeed, it
is not a trivial task to remove duplicate funds from the aggregate database due to the fact that
there is no common identifier for the same hedge fund in the different vendors’ databases.
Therefore, few of the existing papers provide transparent and detailed explanations of how their
database is constructed. Notable exceptions are Ramadorai and Patton (2011) and Aggarwal and
Jorion (2010). We add to this literature by using additional variables such domicile and legal
structure to match share classes, and by providing a formal statistical algorithm in identifying
unique hedge funds.
Third, the paper relates to the literature examining hedge fund data biases, misreporting,
and strategic reporting behaviour. Due to the voluntary reporting, it is well known that hedge
fund databases are associated with many data biases (e.g., Fung and Hsieh (2000, 2009), Liang
(2000), and Getmansky, Lo, and Makarov (2004)), while the recent studies (e.g., Bollen and Pool
(2008, 2009), Patton, Ramadorai, and Streatfield (2011), and Aragon and Nanda (2011)) show
that hedge funds misreport, revisit, and strategically delay their returns when reporting in
commercial databases. We add to this literature by showing that a database selection bias may
arise when a study relies only on one of the hedge fund databases making a conclusion about
hedge fund performance.
Fourth, we contribute to the literature documenting new stylized facts about hedge fund
performance. The recent literature (e.g., Kosowski, Naik, and Teo (2007) and Jagannathan,
Malakhov, and Novikov (2010)) has shown using the sophisticated econometric methods that
hedge fund performance persists at annual horizons, while earlier studies (e.g., Brown,
Goetzmann, and Ibotson (1999), Agarwal and Naik (2000), and Liang (2000)) document that
hedge fund performance persists only at quarterly horizons. Using the consolidated database, we
confirm that hedge fund performance persists at annual horizon even when just using
unsophisticated econometric methods. However, there is persistence only at monthly horizon
when it is evaluated using value-weighted portfolios. When the fund-level share restrictions and
the effect of the look-ahead bias are controlled in persistence tests, we document equal-weight
performance persistence at quarterly horizon.
Finally, the hedge fund literature has documented cross-sectional performance
differences among hedge funds by showing that funds with greater managerial incentives (e.g.
Agarwal, Daniel, and Naik (2009), Aggarwal and Jorion (2010)), strict share restrictions (e.g.,
Aragon (2007)), and less binding capacity constraints (e.g., Teo (2010)), on average, outperform
their peers on a risk-adjusted basis. We contribute to this literature by confirming that smaller
and younger funds and funds having greater capital flows and managerial incentives deliver
higher future returns than their peers, while our results suggest that strict share restrictions are not
associated with the higher risk-adjusted returns after we control the role of other fund
characteristics using multivariate regressions.
The paper is structured as follows. Section 2 describes the data and methodology. Section
3 summarizes the stylized facts about average fund performance based on different databases.
Section 4 reports stylized facts about performance persistence. Section 5 describes stylized facts
about hedge fund performance and cross-sectional characteristics. Section 6 concludes.
2.
Data and merging approach
In this section, we propose an industry standard in constructing an aggregate hedge fund
database by merging multiple commercial databases. Our merging approach is based on the two
transparent main steps that can be easily replicated on a regular basis. Due to the fact that these
steps are nearly automatic, we update our database each month making data real time applicable.
Easily replicable steps also imply that other researchers can follow them in constructing their own
data set or even use the same aggregate data.
2.1. Merging procedure of hedge fund databases
We combine five major hedge fund databases (BarclayHedge, EurekaHedge, Hedge Fund
Research (HFR), Morningstar and TASS) to the aggregate data set. It is not a trivial task to merge
several commercial hedge fund databases and to separate unique hedge funds from the share
classes. The main reason is that all the commercial data vendors only provide an identifier to
unique share classes, but there are no identifiers for unique hedge funds. The problem is serious
even for the studies that are conducted using only one of the commercial databases, since the
individual databases contain significant numbers of multiple share classes that cannot be captured
only by excluding different currency classes. Thus, it is important to remove duplicate share
classes even if a study is conducted using only one of the databases.
Figure 1 highlights the
issue by showing the histograms of pairwise correlation coefficients of share classes estimated
within management companies for each database. All share class pairs are required to have at
least 12 monthly return observations. Figure 1 shows a significant spike at the 0.99 correlation
level for each database. The results suggest that multiple share class structures exist within
management companies. The histogram of the aggregate database suggests that highly correlated
share classes exist within management companies, which can be due to (i) multiple share class
structures within management companies or (ii) duplicate share classes between databases.
We address these issues by combining five major hedge fund databases to form the
aggregate data set using a novel merging approach that is based on two main steps. First, we
develop a matching algorithm of hedge fund share classes that aims to identify exactly the same
share classes with each other across databases. Second, we propose a formal statistical algorithm
that combines sufficiently similar share classes within management companies as a group
allowing us to obtain the longest possible time-series for each unique hedge fund. In contrast to
other studies that often very loosely describe that they ‘carefully’ remove duplicate hedge funds
without describing in detail their merging process, we report and rationalize clearly the variables
that are used in share class matching and open up the details of the statistical algorithm that
identifies the unique hedge funds.3 Specifically, along with the share class and company names,
we opt to use information about the legal structure, domicile and currency base. We do not use
compensation structure and share restriction information due to different reporting standards
across databases, and the changing nature of these variables.4 Furthermore, applying the statistical
algorithm, we can select the individual share class along with the longest track record. This also
allows us to define a unique fund by taking either the equal-weighted or the value-weighted
cross-sectional average return of share classes belonging to the same group within the
management company. Hence, compared to other studies, our innovation is aimed at combining
share class information so that data biases are mitigated. Finally, we emphasize that there is no
reason to believe that previous papers do not use rigorous merging procedures. Due to the fact
that the merging procedures are not discussed in these papers, there is a high demand for the
proposed standardized merging approach that could be exploited by researchers, investment
professionals and even regulators. Appendix discusses the details of the merging algorithm.
The baseline version of our database contains monthly net-of-fees returns, assets under
management (AUM), and other characteristics, such as manager compensation (management fee,
performance-based fee, and high-watermark provision), share restrictions (lockup period, advance
notification period, and redemption frequency), domicile, currency code, style category, and
inception. The database can be easily extended to contain other interesting fund characteristics
that are provided only by some of the databases. For example, EurekaHedge provides detailed
information about hedge fund soft and hard lockups as well as the locations of hedge funds’ head
and research offices. This feature makes the merged aggregate database more attractive for
3
Notable exceptions are Ramadorai and Patton (2011) and Aggarwal and Jorion (2010) who provide the
details about the used merging or cleaning procedures. However, we add to this literature by using
additional variables such as domicile and legal structure to match share classes and by providing a formal
algorithm in cleaning databases from the multiple share classes.
4
For example, Agarwal and Ray (2011) document that hedge fund fees are changing. These changes are
not updated to databases; therefore, we believe that fees do not provide reliable information that could be
used in merging. Anecdotal evidence suggests that hedge funds’ share restrictions are also changed due to
investors’ higher demand for liquidity.
research and investment purposes, since using it one can exploit all the novel properties of
individual databases and while benefiting from the advantages of a large aggregate database.
The two main strengths of our database choice are quality and scope. TASS and HFR
databases form a natural cornerstone of our database, since they are widely used among
researchers and their properties are well-documented in literature. BarclayHedge, EurekaHedge
and Morningstar are not as frequently used; however, we argue that a comprehensive coverage
depends crucially on these less used data vendors. BarcleyHedge provides the most
comprehensive set of asset under management observations, whereas EurekaHedge covers the
largest number of active funds among these databases. The large cross-section of funds that we
obtain as a result has the advantage of making it possible to include important variables from
other sources such as hedge fund equity holdings and proxies of an operational risk.5 Specifically,
BarclayHedge provides information on hedge fund equity holdings obtained from SEC’s 13-F
fillings, while Morningstar collected variables related to hedge funds’ operational risk from
SEC’s Form ADV disclosures. These additional unique features of BarclayHedge, EurekaHedge
and Morningstar suggest that they provide additional useful information for our aggregate
database that the most used databases - HFR and TASS - cannot provide at the time. Finally, we
opt not to use the CISDM database, since its data seem not to be of the same commercial quality
as that of others. The main reason is that CISDM does not provide frequent updates suggesting
that one cannot update the data on a monthly or quarterly basis, which is a very attractive feature
of our consolidated database.
2.2.
Properties of commercial databases
This section compares the properties of commercial and aggregated databases by focusing
on return and AuM time-series as well as misreporting and attrition rates. We find that there are
significant differences between commercial databases that may be important in explaining
performance differences across commercial databases. Specifically, we find that EurekaHedge
and Morningstar have significantly lower proportion of defunct funds compared to BarclayHedge,
HFR and TASS. The result is driven by the periods before 2004 when EurekaHedge and
Morningstar did not gather information about defunct funds. It is important to note that
5
Aragon and Martin (2010) show that hedge fund equity holdings can be used to predict future stock
returns. Goetzmann, Brown, Liang, and Schwart (2008, 2009) measure operational risk using data obtained
from ADV disclosures containing information on the business, ownership, and disciplinary events.
BarclayHedge, HFR and TASS do not suffer from such a problem. In addition, we find that the
amount of AuM observations varies significantly across databases. BarclayHedge is superior in
the terms of AuM coverage, since it has the lowest number of missing AuM observations and the
longest AuM time-series suggesting different behavior when aggregate returns are calculated on a
value-weighted basis.
Our sample covers a time period from January 1994 to June 2011. We focus on the post1994 period to mitigate a potential survivorship bias, as most databases started to collect defunct
funds only after 1994. Our sample restricts each fund to have at least 12 monthly non-missing
return observations. We find that returns and assets under management of share classes are
reported in up to 25 different currencies and approximately 77.2% of share classes are reported in
US dollars. We use end-of-month spot exchange rates to convert all non-USD observations of
currency share classes to US dollars. After all restrictions, the total number of share classes in
sample is 48,121. Obviously, the number of unique share classes and hedge funds is considerably
smaller because of (i) multiple share class structures within management companies and (ii)
duplicate share classes between databases.
After merging five major databases using our merging procedure, we find 24,749 unique
hedge funds. Our estimate of unique hedge funds is consistent with other estimates about the
number of funds in hedge fund industry. For example, Güner, Rachev, Edelman, and Fabozzi
(2010) report that the UBS’ proprietary AIS database has information on over 20,000 hedge funds
and 45,000 share classes, while PerTrac 2010 hedge fund database study finds that hedge fund
industry contains approximately 23,603 hedge funds. The descriptive statistics presenting our
aggregate and commercial databases suggest that an empirical hedge fund study that is conducted
using only one of the databases may infer biased results. The sample of a single commercial
database may not be representative, since even the largest database solely represents less than half
of the hedge funds that exist in our aggregate database. Indeed, the aggregate database has
24,749 hedge funds, while BarclayHedge is the largest database having 9,719 funds (39.27%
relative to the aggregate data) and Morningstar being the smallest 6772 funds (27.36% relative to
the aggregate data).
Figure 2 shows a Venn diagram describing the percentage amounts of unique and duplicate
share classes in our union of five major databases. Figure shows that the total amount of unique
funds in the union is 67.7%. BarclayHedge has the largest percentage share (15.6%), and even the
EurekaHedge database, which has the lowest share (10.4%), has a considerable number of unique
share classes. Hence, all of the databases have a large number of share classes that are not in other
databases suggesting the importance of the use of multiple databases. The numbers shown above
indicate that the properties of the databases differ significantly from each other suggesting that
the sample based on a single database may not be representative.
Share classes in commercial databases are classified into two groups, namely active and
defunct. Active funds report to commercial data vendors, while defunct funds no longer report to
the most recent snapshots. Hedge funds stop reporting because they liquidate, merge with other
funds, cannot be reached by the vendor, stop accepting new investments, are dormant or stop
reporting for unknown reasons. It is important to note that funds belonging to the defunct
category are not necessarily liquidated. However, recent studies document that defunct funds
typically have lower performance measures relative to alive funds.
Among the first, Ackermann, McEnally, and Ravenscraft, (1999) and Liang (1999), and
Fung and Hsieh (2000) note that hedge fund performance measures are affected by the
survivorship bias when the database contains only alive funds. Furthermore, hedge funds often
report return data prior to their listing date in the database, thereby creating a backfill bias.
Clearly, due to the incentives, the backfilled returns are usually higher than the nonbackfilled
returns. To confirm that our findings are robust to incubation and backfill biases, we repeat our
analysis by excluding the first 12 months of data. We also ensure that serial correlation does not
drive our findings. We unsmooth hedge fund returns using the Getmansky, Lo, and Makarov
(2004) specification. We then repeat performance tests to confirm our conclusions. 6
Table 1 presents the summary statistics of hedge fund database returns and AUMs
categorized by all, alive, and defunct funds. The sample period extends from January 1994 to
June 2011 and contains funds with at least 12 non-missing monthly returns. Single commercial
databases are cleaned and merged using the proposed statistical algorithm. The statistics of
returns are annualized and reported in percentages. According to Panel A, our aggregate database
has 37,592 unique share classes after databases are merged at the share class level using the name
matching. The aggregate database has 8,353 unique management firms and 24,749 unique hedge
funds (of which 34.39% are alive) after databases are merged using the statistical algorithm.
Annualized performance measures are higher for alive funds if compared to defunct funds
6
The results are available upon request.
(12.15% / year vs 8.35% / year). The aggregate data has 1.36% (27.56%) of missing return
(AuM) observations.7 The results suggest that alive hedge funds are larger than defunct hedge
funds. The average AuM for alive (defunct) funds is 160.2 (83.82) million US dollars.
Overall, for each of the databases, a typical alive fund has a larger annualized return than a
typical defunct fund. Morningstar database has the largest amount of missing returns, 2.58%.
Alive funds are also larger based on a statistics of the average AUM. EurekaHedge and TASS
have the largest amounts of missing AUM observations, namely, 37.12% and 34.4%. Based on
the coverage of AuM reporting, BarclayHedge is superior to its peers, since it has only 11.49% of
missing AuM observations. In each database, the coverage of AUM observations is important if
performance is measured on a value-weighted basis.
Table 1 (Panel A) presents the summary statistics of the length of the time series of both
the return and assets under management. The results are reported separately for all hedge funds
and separately for alive, and defunct funds. The aggregate data shows that the average length of
fund-return time series is 62 months. On average, the longest return time series are in the
Morningstar database (70 months). Defunct funds seem to have shorter return time series
compared to alive funds. BarclayHedge has the lowest amount of missing AUM observations and
the longest fund-level AUM time series on average (57.22 months).
Panel A shows that the largest commercial database in term of the number of hedge funds
is BarclayHedge having 9,719 funds from which 36.4% are alive. The proportion of alive funds
in TASS and HFR are close to our aggregate data (37.83% and 39.61%, respectively). The largest
proportions of alive funds are in EurekaHedge (62.14%) and Morningstar (52.54%). These results
suggest that a survivorship bias may have the largest effects in the EurekaHedge and Morningstar
databases.
Panel B in Table 1 describes the statistical properties of hedge fund returns categorized by
alive and defunct funds. In the aggregate database, over 51.36% of funds have non-normal returns
based on the Jarque-Bera test of normality (5% level of significance) and 21.44% of funds exhibit
significant serial correlation in returns for the first six lags based on the Ljung-Box test (5% level
7
We calculate missing AuM observations conditional on fund-level return time series that have at least 12
monthly returns for each fund.
of significance. Results are similar across databases. It is well-known from the prior literature
(e.g., Bollen and Pool (2008, 2009)) that hedge funds misreport or ‘manage’ their returns. Panel
B summarizes hedge fund misreporting using a measure proposed by Jylhä (2011). The measure
formalizes the fact that the number of small gains exceeds the number of small losses
documented by Bollen and Pool (2009). Table 1 presents the measure of discontinuity (DC(%))
and z statistic (z(DC)) for testing the existence of discontinuity across databases. We follow Jylhä
(2011) and use only USD-based funds. Our findings suggest that misreporting is quite similar
across databases. Alive hedge funds seem to misreport less than defunct ones. However, the
difference is small. To examine the number of funds showing higher serial correlation when
returns are low, we document the estimates of conditional smoothing using the methodology
proposed by Bollen and Pool (2006). Consistent with Bollen and Pool (2006), we find that 5.77%
of funds exhibit conditional smoothing in our aggregate database. Alive funds in the Morningstar
database show the highest amount of funds showing conditional smoothing (7.07%). Agarwal,
Daniel and Naik (2011) show that hedge funds with high performance-based incentives, low
lockup and restriction periods, high volatility, and low liquidity exhibit significant December
spikes in returns. We test the existence of a December spike and apply the Fung and Hsieh (2004)
residual returns approach to measure the difference between the average December values and the
average of January-November values. After correcting for clustering at the fund-level, consistent
with Agarwal, Daniel and Naik (2011), we find significant differences between the average
December and January-November values across all databases. To summary, our measures of
statistical properties of hedge fund returns are consistent with recent literature.
Table 2 presents the attrition rates across databases and for the aggregate data. Attrition
rates are calculated as the ratio of the number of dissolved funds to the number that existed at the
start of the year. Attrition rates are remarkably different across commercial databases. Attrition
rates for EurekaHedge and Morningstar show an interesting pattern. Specifically, Table 2 shows
that these two databases have started to gather information on defunct funds after 2004, since
their attrition rates are close to zero before 2004. For EurekaHedge, the results is in a line with
Teo (2009), who examines using the EurekaHedge database whether hedge funds’ access to local
information is associated with superior performance. In contrast, during these early years attrition
rates are considerably higher for BarclayHedge, HFR and TASS as well as for the aggregate
database. The results of attrition rates suggest that EurekaHedge and Morningstar databases are
subject to higher survivorship bias if compared to TASS, HFR, and Barclay. According to recent
literature, survivorship bias imparts an upward bias to performance measures (Liang (2000)).
3.
Average hedge fund performance
In this section, we investigate hedge funds’ average performance on an equal- and value
weight basis as well as across time-periods, investment strategies and fund domiciles. An
important question in performance evaluation is whether active managers add value on average.
A common approach to evaluate hedge fund abnormal performance is to estimate the alpha,
which is the return that cannot be explained by exposure to common risk factors. For investors, a
positive abnormal performance is crucial since it tells how much an average hedge fund adds
positive value, if any, after management and incentive fees. The major interest is in the results of
our novel aggregate database that serves a proxy of the unobservable whole hedge-fund industry.
Also, we identify possible differences in performance estimates between databases that may
suggest existence of sampling biases in hedge funds databases occurring from the fact that all
hedge fund databases are not drawn from the same population of hedge funds. Consistent with
previous studies (e.g., Kosowski, Naik, and Teo (2007)), we find that hedge funds add value even
after fees. Specifically, our results show positive and significant aggregate abnormal returns for
the aggregate and commercial databases. The databases that are associated with more pronounce
biases deliver the highest average performance.
3.1. Equally-weighted and value-weighted portfolios
We construct equal-weight (EW) and value-weight (VW) portfolios of hedge funds. EW
portfolio measures how an average hedge fund performs, while VW portfolio summarizes the
performance of aggregate wealth invested in hedge-fund industry as a whole. We expect to find
that the aggregate EW portfolio has a higher average return than the VW portfolio, since the
evidence suggests (e.g., Teo (2010)) that smaller hedge funds, on average, outperform the larger
ones. In addition, the performance of a VW portfolio can also give some hints how AuM
reporting affects performance. It is important to note that we control for the potential impact of a
duplicate bias, since all commercial databases are merged and cleaned from multiple share classes
using a statistical algorithm proposed in the Appendix of the paper.
Figure 5 shows the cumulative excess returns of EW and VW portfolios. EW portfolios
show that the cumulative excess return of our aggregate database is very close to the respective
return for TASS, HFR, and Barclay. However, EurekaHedge and Morningstar clearly outperform
other databases suggesting that there are differences across databases. VW portfolios show the
highest cumulative return pattern for TASS. Figure 5 suggests that the aggregate returns of the
databases are highly correlated with each other, but the average levels of returns vary between
databases. This suggests that the same common factors may drive the hedge fund return across
databases.
Table 3 shows summary statistics for excess returns of EW and VW portfolios from
January 1994 to December 2010. Our aggregate data indicates that hedge funds add value on
average. The annualized average EW (VW) excess return for the aggregate data is 8.45% per year
(7.36% per year). EurekaHedge and Morningstar databases show the highest annualized average
excess returns (10.28% and 9.75%). TASS has the lowest average excess return being 7.92% per
year. To highlight differences between databases, we find that EurekaHedge has approximately
23 percent higher average excess return than TASS. Average excess returns show much more
variation between databases than annualized standard deviations. Hence, risk levels do not differ
significantly across databases. The results are consistent with our hypothesis suggesting that
EurekaHedge and Morningstar have the highest returns due to the fact that these two databases
have the lowest attrition rates and proportions of defunct funds.
In contrast to the results of EW portfolios, Table 3 shows the highest VW return for
TASS (8.56%). Except for TASS, all databases show lower value-weight average returns if
compared to equal-weight average returns. This is consistent with Teo (2010), who documents
that smaller funds have higher average returns. We find that TASS has a large amount of missing
AuM observations that equals to 34.4%. The results may potentially explain why TASS’s VW
returns are so high. Perhaps, relatively large and high-performing funds report AuM data to
TASS, but poor performing funds skip reporting. We address this issue rigorously in section 3.2.
It is interesting to note that BarclayHedge has the lowest VW return of 6.91% per year.
BarclayHedge provides the best AuM coverage with the longest time-series and the smallest
amount of missing observations. The relatively low average return of BarclayHedge may be
explained by the fact that it is able to gather AuM information from poor performing hedge funds.
The value-weighted return for BarclayHedge is close to the VW return that is obtained for our
consolidated database, being 7.36% per year. Therefore, we believe that BarclayHedge’s VW
portfolio is the closest proxy among the commercial databases for the overall performance of
hedge-fund industry.
Next, we investigate whether an average hedge fund is capable to deliver positive
abnormal performance. Kosowski, Naik, and Teo (2007) conclude that the average abnormal
return across hedge funds from 1994 to 2002 is 0.42% per month (5.04% per year), while Fung,
Hsieh, Naik, and Ramadorai (2008) document that hedge fund performance and risk-taking differ
across time-periods. As a benchmark model, we use the Fung and Hsieh (2004) seven-factor
model that is the standard workhorse in hedge fund performance evaluation studies. 8 We regress
the net-of-fee monthly excess returns (in excess of risk-free rate) of a hedge fund against
traditional buy-and-hold and primitive trend-following factors
Ri,t = a + b1(SP - Rf) + b2(RL - SP) + b3(TY - Rf) + b 4(BAA - TY) + b5(PTFSBD - Rf) +
b6(PTFSFX - Rf) + b7(PTFSCOM - Rf) + e,
(1)
where the risk factors are defined as the excess return of the S&P 500 index (SP-RF), the return
of the Russell 2000 index minus the return of the S&P 500 index (RL-SP), the excess return of
ten-year Treasuries (TY-RF), the return of Moody's BAA corporate bonds minus ten-year
Treasuries (BAA-TY), the excess returns of look-back straddles on bonds (PTFSBD-RF),
currencies (PTFSFX-RF), and commodities (PTFSCOM-RF). Fung and Hsieh (2004) show that
their seven-factor model considerably explains time series variation in hedge fund returns.
Table 3 presents the estimates of time series regression (1) for EW (panel A) and VW
(panel B) portfolios of hedge funds in our sample from January 1994 to December 2010. The
intercepts of the Fung and Hsieh (2004) model summarize the average abnormal performance of
hedge funds (EW portfolios) and the abnormal performance of aggregate wealth invested in
hedge funds (VW portfolios). The alpha terms for net-of-fees returns show us whether hedge
funds have sufficient private information to cover costs that they impose on investors. Based on
the annualized intercept estimates of the Fung and Hsieh (2004) model, all hedge fund databases
show positive abnormal performance. The result is consistent with previous literature like
Kosowski, Naik, and Teo (2007).
The results in Table 3 show that the aggregate hedge fund EW and VW alphas equal to
5.87% and 4.88% per year, respectively. The results for abnormal performance suggest that hedge
funds add value on average after adjusting for common systematic risk factors. Table 3 shows
8
In an unreported robustness test, we augment the model with liquidity, currency and carry-trade risk
factors. The main inference remains unchanged.
substantial variation in abnormal returns between databases. The highest EW abnormal return
estimate is found for EurekaHedge and Morningstar databases (7.61% and 7.28%). TASS has the
lowest EW alpha estimate equaling to 5.31% per annum. Again, the EW returns are the highest
for the databases with the lowest attrition rates. As for the excess returns, BarclayHedge has the
lowest VW alpha, while the respective alpha is relatively high for TASS. The differences between
the average EW and VW aggregate alphas can be seen from Figure 6 that shows EW and VW
cumulative abnormal returns of hedge funds for January 1994 – December 2010. Figure 6 shows
clearly that TASS database underperforms the other databases in terms of EW returns but
outperforms in terms of VW returns.
Since the factor loadings and R-squares of the Fung and Hsieh (2004) model in Table 3
do not show significant variation between aggregate and across databases, it implies that
differences in risk taking do not explain significant variation in alphas. We also include in
unreported robustness checks additional factors such as liquidity, carry, and currency risk factors.
We find that the levels of alphas are insignificantly lower, but the t-statistics of alphas are slightly
higher since the risk factors explain better the residual variance. Therefore, we argue that
differences in the survivorship bias and AuM coverage across commercial databases are driving
the alpha differences between databases.
3.2. Backfilling- and smoothing-adjustment, size categories, and missing
AuM observations
To address rigorously our arguments about the determinants that are driving performance
difference across databases, we examine the impact of backfilling bias, serial correlation, fund
size and missing AuM observations. We find that small hedge funds, which size is below 10
million dollars, deliver extremely high performance. It is the most important reason why we
document so high average hedge fund performance earlier using EW and VW portfolios.
Differences in survivorship bias remain as an explanation why equally-weighted average
performance varies significantly across commercial databases. We demonstrate two main reasons
why value-weighted performance differs considerably across data vendors. TASS’s superior VW
performance is explained by the fact that TASS contains large hedge funds having relatively high
performance. More importantly, TASS has a significant number of missing AuM observations
exactly at the same time when respective hedge funds deliver poor returns. Other commercial
databases do not share this property. Therefore, we argue that it is important determinant in
explaining VW performance differences among data vendors.
Panel C in Table 3 presents the average performance results that address the impact of
backfilling bias. We demonstrate that after excluding the first 12 return observations the average
performance is significantly lower.9 To compare performance with the baseline results presented
in Panel A of Table 3, we conclude that the backfilled average performance is significantly higher
across databases. Using an aggregate database, we find backfilled alpha (t-statistic of alpha) is
over 20% (25%) higher than the respective non-backfilled measures. Among commercial
databases, the impact of backfilling bias is more pronounce for mature databases like TASS,
HFR, and BarclayHedge. The difference between backfilled and non-backfilled returns is not so
remarkable for EurekaHedge and Morningstar. This may be due to the fact that these databases
suffer from significant survivorship bias, which is difficult to disentangle from backfilling bias.
Panel D in Table 3 presents the EW average performance results that address the impact
of the performance smoothing that is a consequence of serial correlation in hedge fund returns.
Recent papers in the related literature argue that serial correlation in hedge fund returns is either
due to (i) holdings of illiquid assets (Getmansky, Lo and Makarov (2004) or (ii) misreporting
(Bollen and Pool 2008, 2009). First, to measure the impact of serial correlation on average
performance, we add a MA(2) process to the Fund and Hsieh (2004) model’s error term. Second,
we estimate the econometric model proposed by Getmansky, Lo and Makarov (2004) and adjust
the standard deviations of excess returns and Sharpe ratios for serial correlation. Our results in
Panel D suggest that the adjustment leads to lower alpha t-values in Fung and Hsieh (2004)
model. However, the t-values remain statistically significant at the 5% level. EurekaHedge and
Morningstar databases show the highest average alphas (7.68% per year and 7.35% per year).
Thus, the database ranking based on the average performance remains the same as shown in
Panels A and B. The results suggest also that the adjustment for serial correlation leads to lower
Sharpe ratios and higher standard deviation of excess returns. These conclusions are consistent
with Getmansky, Lo and Makarov (2004).
9
Another possibility to address backfilling bias would be to use information from the ‘Date added to
database’ –field. However, all of the databases do not provide such data field. Hence, for consistence, we
opt not to adjust backfilling bias using it.
To examine whether hedge funds are capable to deliver superior performance
consistently through time, we divide our aggregate database into subperiods following Hsieh,
Naik and Ramadorai (2008). Panel E in Table 3 presents the results showing that hedge fund risk
taking is changing through time, but hedge funds are able to deliver alpha during all the
subperiods expect during the period from January 1997 to September 1998. Finally, Table A1 in
Appendix (Panel E) shows that the return differences between databases are smaller after 2004
when the survivorship and backfilling biases are not so pronounced in EurekaHedge and
Morningstar. It implies that the properties of databases seem to converge over the time.
Panel F in Table 3 presenting the average performance across size categories show that
the average performance is strongly related to fund size. Small hedge funds deliver an extremely
high performance, while the large ones are not able to deliver statistically significant alpha.
Indeed, the hedge funds having AuM below 10 million dollars deliver outstanding performance
with alpha (t-statistic of alpha) of 7.25% per year (6.80). The magnitude of the average alpha and
its statistical significance are implausible high. Therefore, we argue that small hedge funds are
associated with pronounce data biases. In addition, large hedge funds, which size is above 250
million dollars, are not able to deliver significant alpha. There also seems to be monotonic
relationship between hedge fund size and performance. It is interesting to note that hedge funds,
which do not report even a single AuM observation, provide superior performance. We see it as a
one of the reasons to explain why EW returns are higher than VW returns, since also the hedge
funds having solely missing AuM observations are included in EW portfolios. To sum up, these
results shed further light why we document so high performance for equally- and value-weighted
portfolios.
Table 4 presents convincing evidence why the average performance of EW and VW
portfolios differs among data vendors. Panel A in Table 4 shows that large hedge funds’ average
returns are the highest for TASS when compared to two other mature databases, HFR and
BarclayHedge. Specifically, average returns are consistently the highest for TASS when fund size
is above 500 million dollars. BarclayHedge delivers the smallest average returns among large
funds, respectively. This is one of the reasons why TASS has the highest VW performance,
whereas BarclayHedge delivers the lowest VW returns. Panel B in Table 4 shows another
important reason in explaining performance differences. We demonstrate that missing AuM
observations show different behavior for TASS if compared to other mature databases. Since the
AuM coverage of TASS is poor if especially compared to BarclayHedge, we estimate its impact
on average returns. Only for TASS, but not for other databases, we find that when AuM
observations are missing from the middle or the end of AuM time-series, then the respective
returns tend to be extremely poor. Specifically, the associated mean return is 3.48% per annum in
TASS, while the mean return is 6.28% in BarclayHedge being almost as twice high. The
magnitude is also important, since TASS contains 2,972 hedge funds that have missing AuM
observations at the middle or at the end of AuM time-series. Clearly, missing AuM observations
also shed some light why the TASS’s VW performance is superior over its EW performance,
since EW portfolio contains those returns that do not have the respective AuM observations. We
conclude that differences in VW returns among data vendors can be explained by non-randomly
missing AuM observations.
3.3. Investment objectives
Next, we turn on the average performance of hedge fund strategies by examining whether
hedge funds grouped by investment objective add value on a net-of-fees basis. Table 5 reports the
results for the consolidated database, while Table A2 in Internet Appendix displays the results for
each of the commercial databases. We classify hedge funds into 12 categories: CTA, Emerging
Markets, Event Driven, Global Macro, Long/Short, Long Only, Market Neutral, Multi-Strategy,
Relative Value, Short Bias, Sector and Others. The Internet Appendix provides the classification
of hedge fund strategies. Figure 3 presents the proportions of hedge funds by investment
objective. The proportions are similar across data vendors. This implies that our merging
approach delivers an accurate strategy matching and multiple share class deletion approach.
Table 5 presents the average performance of hedge fund strategies for the aggregate
database. Overall, we find that hedge fund strategies are capable to deliver significantly positive
risk-adjusted returns. The only exception is Long Only strategy, which has a positive but
insignificant alpha (t-statistic=1.76). The results in Table A2 show that the strategy-level average
performance varies significantly between commercial databases. It seems that for the most of the
hedge fund strategies, EurekaHedge and Morningstar (BarclayHedge, HFR and TASS) are among
top (bottom) performing databases. For example, the annual average alpha of Emerging Markets
strategy equals to 12.63 (5.23) percent for Morningstar (TASS). Therefore, the strategy-level
results give us further evidence that differences in survivorship bias are driving the performance
differences between commercial databases. Importantly, we also demonstrate significant
differences in the average performance between TASS, HFR and BarclayHedge. For example,
Emerging Markets strategy shows the annual average alpha of 9.03 % per year for HFR, while
TASS’s and BarclayHedge’s respective alphas are remarkable lower being 5.23% per year and
6.92 % per year. Hence, not only the average hedge fund performance differs across databases, as
we documented earlier, but the strategy-level average alphas differ even across mature
commercial databases.
Panel B in Table 5 address the effect of the fund size on the strategy-level average
performance using the aggregate database. The results indicate that for ten of the twelve indices
(all except sector and short bias) small funds outperform large funds. For example, in Emerging
Markets category, small (large) funds exhibit the average alpha of 11.88% per year (0.86% per
year). Hedge funds are sorted into terciles each December based on fund-level monthly nonmissing AuM observations. Portfolio returns are calculated for equal-weight portfolios monthly
using 12-month holding period. Results in Panel B support the stylized facts shown in Table 3
and Table 4. Panel B in Table A2 describes the effect of the fund size on the strategy-level
performance using the five individual databases. The results are consistent with results shown in
Table 5 (Panel B). For the most of the strategies and databases, small funds outperform large
funds in terms of the annual average performance.
3.4. Domiciles
We investigate whether hedge funds add value, on average, across domiciles. The prior
literature such as Aragon, Liang, and Park (2011) documents that onshore hedge funds registered
in USA deliver higher performance than the offshore hedge funds. We extend their work by
examining the hedge fund average performance around the whole world. The domicile regions of
hedge funds and management firms are divided to two groups: (1) onshore; and (2) offshore.
United States and Canada are classified as onshore regions. Other domicile regions are classified
into four groups: (1) Asia and Pacific; (2) Caribbean; (3) Europe; (4) Rest of world. Figure 4
shows the pie charts of the proportions of funds grouped by fund-level domicile region. In
BarclayHedge database, 46% of funds are onshore funds. In other databases, most of the funds
are domiciled in Caribbean (37% in the aggregate database). Overall, the proportion of funds
between onshore and offshore seems to be similar across all databases.
Table 6 shows results of the number of unique funds and firms in domicile region groups
as well as the average performance results of funds grouped by domiciles. Panel A in Table 6
shows the number of unique hedge funds in domicile regions. Results are obtained using Hedge
Fund Research, BarclayHedge, and EurekaHedge, since these data vendors provide information
on domicile of the management firms. Most of the firms are established in ‘onshore regions’
(64%) and most of the funds are domiciled as onshore vehicles (38%). Hedge funds that are
established in the Caribbean (Europe) account for 35% (16%) of all hedge funds (15,805). The
number of unique funds suggests that the domicile region of both firms and funds provide the
same type of classification of offshore and onshore domicile.
Panel B in Table 6 presents the results of the average performance grouped by fund
domicile. All portfolios are equal-weighted monthly. The results show that there are significant
differences in average performance across domicile groups. On average, we find that onshore
based funds outperform offshore based funds. Onshore (offshore) category has the annualized
Fung and Hsieh (2004) alpha of 7.45% per year (4.74% per year). Indeed, Europe based hedge
funds deliver the poorest performance (alpha equals to 3.21% per year). The aggregate database
shows the highest annual average alpha for Asia-Pacific group, 8.16 % with the t-statistic of 3.99.
All domicile groups show statistically significant average alpha with t-statistics above the 5%
level. Finally, the results in Table A3 in Appendix show that EurekaHedge and Morningstar seem
to outperform mature databases that do not suffer pronounce data biases. Specifically, in the
following groups: (1) all funds; (2) onshore; (3) all offshore funds ; (4) Caribbean; (5) rest of
world, EurekaHedge and Morningstar outperform BarclayHedge, HFR and TASS. This is
consistent with the fact that EurekaHedge and Morningstar have smaller attrition rates and higher
average performance if compared to BarclayHedge, HFR and TASS databases. Panel C in Table
6 provides the results of the average performance grouped by the firm-level domicile region. On
average, onshore funds outperform offshore funds in terms of the average alphas (6.88% per year
and 5.14% per year). We document also the average performance grouped by cities where
management firms are legally established. New York and London based management companies
generate similar levels of the average alphas (6.33% per year and 6.16% per year).
4.
Hedge fund performance persistence
In this section, we examine hedge fund performance persistence. When the abnormal
performance of hedge funds is due to manager skills, then the same top performing hedge funds
should have a high return year after year. If some hedge fund managers have access to superior
information, sorting funds on past performance should indicate whether past winners are on
average future winners and past losers are future losers. For investors, the performance
persistence is crucial since hedge funds typically restrict capital withdrawals by imposing lockup,
advance notification, and redemption periods.10 All these restrictions indicate that new investors
are not able to withdraw from hedge funds in a timely fashion. Therefore, hedge funds that are
able to add value after fees consistently through time is a rewarding feature for investors. The
pioneering literature (e.g., Agarwal and Naik (2000), Brown, Goetzmann, and Ibbotson (1999),
and Liang (1999)) documents that hedge fund performance only persists for short periods and it
disappears at the annual horizons. However, recently, using a sophisticated econometric
approaches, Jagannathan, Malakhov, and Novikov (2010), and Kosowski, Naik, and Teo (2007)
show that top abnormal performance of hedge funds persists even at annual horizons. The
stylized fact is that hedge funds are able to deliver economically significant performance
persistence.
The overall results documented in Table 7 show that hedge fund performance persists,
but the persistence is driven by small funds. Using an aggregate database, we document that postranking Fung and Hsieh (2004) alpha estimates are positive and statistically significant for the top
decile across different holding periods and size terciles. Hence, hedge fund investors that invest in
top decile hedge funds are able to earn continuously statistically and economically significant
alpha even at annual horizons. However, the post-ranking alpha estimates for spread portfolio,
calculated as top decile minus bottom decile, are not always significantly positive. For the
aggregate database, equal-weight spread portfolios show positive persistence even at annual
horizon, while value-weighted spread portfolios only indicate significant persistence on a
monthly basis. Hence, hedge fund performance persistence seems to be driven by small funds. By
dividing hedge funds into three size terciles, we confirm the issue by showing that performance
persistence is the strongest for small funds and weakest for the large funds. The conclusion about
the performance persistence changes when one draws the inference using only one of the
commercial databases. Using mature databases (BarclayHedge, HFR and TASS), we find
significant evidence about performance persistence. In contrast, using EurekaHedge and
Morningstar, we are not able to document significant performance persistence. Differences in
10
Hedge funds can impose a lockup provision specifying a time period during which new investors are not
able to withdraw their shares. Investors can withdraw their shares at the end of the lockup period by giving
an advance notice. When the notice is given, investors have to wait until the pre-specified redemption
interval is at hand. About 25% of hedge funds apply one year lockup, while a typical hedge fund imposes a
30-day’s notice and allows quarterly redemptions.
performance persistence can be explained by the fact that BarclayHedge, HFR and TASS are not
associated with pronounce data biases, while EurekaHedge and Morningstar suffer significantly
from survivorship and backfilling biases.
We start by cleaning commercial and aggregate databases from multiple share classes. To
investigate hedge fund performance persistence, we follow the standard way in performance
evaluation literature (e.g. Carhart (1997)). We sort hedge funds based on the t-statistic of Fung
and Hsieh (2004) alpha. Kosowski Timmermann, Wermers and White (2006) describe several
attractive features of t-statistic. Although alpha measures the size of the abnormal performance,
those estimates can be sensitive to outliers. Due to short time series, the alphas are also estimated
imprecisely. The t-statistic of alpha provides a correction, since it is measured as a ratio of alpha
estimate and the estimated precision of alpha estimate. Specifically, we sort hedge funds into
decile portfolios based on their t-statistics of alpha obtained from the Fung and Hsieh (2004)
model that is estimated over the previous two years. We report persistence results using four
different holding periods: (i) monthly, (ii) quarterly, (iii) semiannual, and (iv) annual over the
period from January 1994 to December 2010. For example, if a quarterly holding period is used,
we sort hedge funds on March, June, September and December of each year into decile portfolios.
Following Kosowski, Naik, and Teo (2007), the post-ranking portfolio excess returns are created
monthly, so the weights are readjusted whenever a fund disappears. Hedge funds with the highest
past two-year alpha’s t-statistic comprise decile 10 (top decile), while funds with the lowest past
two-year alpha’s t-statistic form decile 1 (bottom decile). We estimate also the abnormal return of
the spread portfolio (decile 10 minus decile 1). A significant difference in abnormal performance
between top and bottom decile provides evidence of performance persistence during the selected
holding period. As a baseline, we form equally-weighted (EW) post-ranking portfolios. To study
whether the performance persistence is driven by small funds, we also construct value-weighted
(VW) portfolios.
Importantly, to understand the impact of data biases between commercial databases, we
calculate dropout rates for each of the decile portfolios used in persistence tests. The dropout rate
is the percentage of hedge funds dropping out from the underlying decile portfolio during the
holding period. We expect that dropout rates are relatively low (high) for the top (bottom) decile.
This is due to the fact that poor performing hedge funds close their operations, while top ones
continue to operate. In addition, since the attrition rates in Table 2 are considerable lower for
EurekaHedge and Morningstar, we hypothesize that across decile portfolios dropout rates vary
significantly more to BarclayHedge, HFR and TASS than to EurekaHedge and Morningstar. We
also expect that the spread between the top and the bottom deciles is wider for BarclayHedge,
HFR and TASS than for EurekaHedge and Morningstar. Hence, BarclayHedge, HFR and TASS
should show more significant performance persistence than EurekaHedge and Morningstar.
Table 7 shows the results of EW, VW, and size tercile hedge fund performance
persistence tests. Table reports the annualized Fung and Hsieh (2004) alphas, t-statistics of
alphas, and dropout rates for bottom, top, and spread portfolios. Panel A in Table 7 presents the
persistence tests for the equally-weighted portfolios. Using the aggregate database, we find a
significantly positive alpha for spread portfolios across different holding periods. In particular,
using the monthly rebalancing, the spread portfolio delivers an annual post-rank alpha of 5.80%
with the t-statistic of 4.10. For a yearly holding period, the respective post-rank alpha is 2.84%
per year, being statistically significant with the t-statistic of 1.97. Hence, the results suggest that
hedge fund performance persists even at annual horizons.
To examine whether small hedge funds are driving the performance persistence, we
construct post-ranking portfolios on value-weighted basis. According to Panel B in Table 7, the tstatistics of VW post-rank spread portfolio alpha reveal performance persistence only at monthly
horizon for the aggregate database. Specifically, the spread portfolio between the top and the
bottom deciles of the aggregate database shows the annual alpha of 4.62% with t-statistic
equaling to 2.89. The aggregate database do not show statistically significant spread portfolio
alpha for quarterly, semiannual, and yearly holding periods. The results suggest that performance
persistence is driven by relatively small hedge funds that have smaller weight in VW portfolios
relative to large funds. Overall, results of persistence suggest that the results depend on weighting
schema and performance persistence of VW portfolios disappear faster compared to EW
portfolios.
To confirm the performance persistence is driven by small funds, we investigate hedge
fund performance persistence among different size categories including small, median, and large
funds.11 It would be important to find performance persistence among large hedge funds, since
investors could exploit it in practice. In addition, the data biases are more pronounce among small
11
Table A5 in Appendix presents the number of hedge funds and AuM cut points in these three terciles.
The first size group contains hedge funds that are, on average, smaller than 10 million dollars, while the
second size group includes funds, which average size is between 10 and 50 million dollars. The third
group’s average hedge fund size is above 50 million dollars.
funds than large funds. Panel C in Table 7 presents the persistence test results for aggregate
database across three size groups. The results show that small funds are driving the performance
persistence. Specifically, the spread portfolio alphas are almost consistently the highest for small
funds and the lowest for large funds. Indeed, using only the group of small funds, we document
significant performance persistence even at annual horizons. For the median and large hedge
funds, we find significant performance persistence at semiannual horizons, but not at annual
horizons. However, the top decile alphas are still economically and statistically significant even
for large funds at annual horizon. This implies that hedge fund investors may be able to earn
significant alpha by investing in funds based on the t-statistic of alpha.
According to Table 7, the conclusion about hedge fund performance persistence varies
significantly across commercial and aggregate databases. We cannot document significant
performance persistence at annual horizons for any of the commercial databases. This suggests
that the conclusion changes if one relies only on one of the commercial databases. On an equallyweighted basis, Morningstar and EurekaHedge do not show any performance persistence. Indeed,
we find that their spread portfolio’s post-rank alphas are indistinguishable from zero even with
monthly rebalancing. In contrast, for TASS, HFR, and BarclayHedge, the results show
performance persistence at monthly, quarterly and semiannual horizons. Even these mature
databases do not show performance persistence at annual horizons, as our aggregate database
does. On a value-weighted basis, only the HFR database shows also persistence at monthly
horizon. Hence, using commercial databases, performance persistence seems to vanish very
quickly.
To investigate why performance persistence results differ across databases, we calculate
dropout rates. First, we find that dropout rates are remarkably wider for mature databases than to
younger databases. At annual holding period, Table 7 presents that the difference in dropout rates
between top and bottom portfolios is 12.61% (BarclayHedge), 13.80% (HFR), and 15.27%
(TASS) for mature databases, whereas they are significantly lower 4.19% (Morningstar), and
6.97% (EurekaHedge) for younger databases. This is consistent with the findings that
EurekaHedge and Morningstar databases have small attrition rates for 1994-2004 periods if
compared to other databases. Second, the alphas of bottom portfolios vary significantly across
commercial databases being the highest for EurekaHedge and Morningstar. This is shown clearly
in Figure 7 that plots the EW annualized Fung and Hsieh (2004) alphas for persistence portfolios
using four different holding periods. Based on the graph, EurekaHedge and Morningstar have
considerably higher levels of average alphas for bottom portfolios if compared to other databases.
The levels of the average alphas of the other databases are close to each other. In contrast, the top
portfolios’ alphas are remarkable similar across databases. Based on these two reasons, we argue
that relatively low attrition rates in Morningstar and EurekaHedge databases are driving the
differences in persistence test among databases.
In our robustness checks, we rule out the possibility that differences in performance
persistence are driven by heterogeneity in risk exposures.
Table A4 in Appendix presents
persistence tests at annual level and reports annualized average excess returns and standard
deviations as well as the alpha, risk-loadings, R-square with respect to Fung and Hsieh (2004)
model, and dropout rate. Table A4 shows that Sharpe ratios, risk loadings and R2:s of the sevenfactor Fund and Hsieh (2004) model are very similar across individual databases and the
aggregate databases. Therefore, we argue that performance persistence differences cannot be
explained by risk exposures among databases.
Finally, we perform a performance persistence test for the aggregate database to measure
whether a real-time investor is able to exploit the short-term performance persistence of hedge
funds. We build feasible rebalancing strategies by taking into account fund-specific share
restrictions within each rebalancing horizon. We exclude funds having lockup, notice, or
redemption periods longer than the rebalancing period in question. For the aggregate database,
results suggest EW persistence only at quarterly horizon when funds are restricted to have share
restrictions smaller than six months. Without these constraints, our baseline result (Table 7, Panel
A) shows EW persistence at semiannual horizon.
We sort hedge funds into portfolios based on their t-statistics of alphas using the recent
24-months of returns preceding the evaluation period. Portfolios are based on four rebalancing
horizons: (1) monthly; (2) quarterly; (3) semiannually; and (4) yearly. Within each rebalancing
horizon, decile portfolios are formed using only the feasible information taking into account fundspecific share restrictions. For instance, for the feasible quarterly rebalancing strategy, we
exclude funds that have lockup, redemption, or notice periods longer than three months. This
implies that, for this strategy, we use 3-month lagged information to estimate persistence in order
to mitigate effects on look-ahead bias. Table 8 shows the annualized EW post-rank alphas for
persistence portfolios. Portfolio returns are calculated for EW portfolios monthly, so the weights
are rebalanced whenever a fund disappears.
Results in Table 8 suggest that the length of share restrictions affects to the estimates of
performance persistence. Panel A in Table 7 shows performance persistence (Equal-weight) for
the aggregate database at semiannual level (t-statistic=3.48). According to Table 8, the feasible
strategy for the semiannual holding horizon does not suggest significant performance persistence.
Thus, the results indicate that funds with notice period longer than six month show performance
persistence, and the real-time investor cannot exploit this predictability in semiannual rebalancing
strategies.
5.
Hedge fund performance and fund characteristics
In this section, we examine the cross-sectional relationship between hedge fund
characteristics and performance. Specifically, we explain the cross-sectional variation in hedge
fund returns and alphas by fund-specific characteristics related to managerial incentives, liquidity,
and capacity constraints. Several academic papers have documented that hedge fund-specific
characteristics explain cross-sectional differences in fund performance. First, Ackermann,
McEnally, and Ravenscraft, (1999) and Liang (1999) find a positive relation between incentive
fees and Sharpe ratio. Using a comprehensive database (a union of CISDM, HFR, MSCI, and
TASS), Agarwal, Daniel, and Naik (2009) document that hedge funds with greater manager’s
option delta deliver superior performance. The findings suggest that managerial incentives are
associated with superior performance. Second, Aragon (2007) argues that share restrictions allow
hedge funds to manage illiquid assets and earn an illiquidity premium. Aragon uses the TASS
database for January 1994 through December 2001 and documents that hedge funds with a lockup
period deliver approximately 4% higher annual returns than their peers. Finally, using a union of
TASS and HFR databases, Teo (2010) shows that small hedge funds outperform large ones by
2.75% per year after adjusting for risk. Using the TASS database, Aggarwal and Jorion (2010)
show that due to particularly strong financial incentives, emerging hedge funds are able to
outperform their peers.
To understand how fund-specific characteristics differ across databases, Panel A in Table
9 presents the descriptive statistics of hedge fund characteristics. The cross-sectional averages
and standard deviations of fund characteristics are quite similar across commercial databases. The
average incentives fee between databases ranges from 17 to 19%, while management fee is
consistently about 2% across data vendors. The high-water mark indicator varies significantly
among databases. Using HFR, we find that 88% of hedge funds impose a high-water provision,
while BarclayHedge suggests that only 61% of hedge funds use high-water mark. The number of
hedge funds imposing a lockup varies significantly across databases. According to EurekaHedge,
only 19% of hedge funds use a lockup, but based on Morningstar, we find that 45% of hedge
funds use lockups. Table also reports the proportion of hedge funds having a missing value for a
specific variable. The results indicate that the coverage of hedge fund characteristics differ
between databases. Morningstar has the highest amount of missing observations for both
compensation and share restrictions variables. The other four databases have almost the similar
amount of missing observations. BarclayHedge has relatively large number of missing
redemption and lockup period observations, while all of the four databases have almost the
perfect coverage of compensation variables. Overall, these results imply that the properties of
compensation and share restriction variables differ remarkably among commercial databases.
Therefore, it is interesting to investigate cross-sectional performance differences across
commercial databases and compare results to ones obtained using the aggregate database.
We first address this issue by forming decile portfolios based on fund age, size, and flow,
inventive fee as well as lockup, notice and redemption periods across databases and size
categories. We rebalance portfolios on a yearly basis, and estimate the spread between top and
bottom deciles in order to examine whether the average performance differs between the extreme
realizations of a specific fund characteristic. Panel B in Table 9 presents the Fung and Hsieh
(2004) alphas and the associated t-statistics across fund characteristics and databases. The
portfolio sorts across databases show that younger and smaller hedge funds outperform, while
hedge funds with significant short term flows do not deliver superior future performance.
Managerial incentives seem to be important, since hedge funds with great incentive fees
outperform the funds with low fees across databases. Most interestingly, we find that share
restrictions seem not to explain consistently hedge fund performance across database. As Aragon
(2007), we document that hedge funds imposing strict lockups are associated with significant
outperformance. However, notice and redemption periods seem not to explain consistently crosssectional differences in hedge fund risk-adjusted returns. Using TASS and HFR, we document a
significant and positive relation between notice period and risk-adjusted average return, while
other commercial databases and even the aggregate database show insignificant relation between
notice period and risk-adjusted returns. The relationship is extremely strong in TASS, since the
average spread between the top and bottom deciles is 4.30% per year with the t-statistics of 4.66.
Hence, the conclusion about the impact of notice period on hedge fund performance varies
significantly across databases. The unreported tests suggest that results are robust even after we
exclude CTAs from our sample.
Using the Fama-McBeth (1973) approach, we examine which of the fund-specific
characteristics are the most important variables in explaining the cross-section of hedge fund
performance. Formally, the Fama-McBeth (1973) procedure can be expressed as
Ri,t =
where Ri,t
0
+
1
Yi,t +
2
Zi + u
(2)
refers to excess return (alpha) of a hedge fund i at the time t,
representing the slope coefficients for time-variant characteristics, and
2
1
is a vector
is a vector representing
the slope coefficients for time-invariant characteristics. The vector of time-variant characteristics
(Yi,t) includes hedge fund size, flow, and age, while the vector of time-invariant characteristics
(Zi) contains management and incentive fees, high-water mark provision and share restrictions in
the form of lockup, notice, and redemption periods. We control for the strategy and domicile
fixed effects, and adjust standard errors for autocorrelation and heteroskedasticity following
Newey and West (1987).12
Panel A in Table 10 reports Fama-McBeth regression results using hedge fund excess
returns from January 1994 to December 2010, while Panel B of Table 10 presents the results for
the Fung and Hsieh (2004) alphas. For the aggregate database, we find evidence that small, young
and the funds with greater managerial incentives outperform their peers. We cannot document
that strict share restrictions are associated with superior performance. The only share restriction
variable that shows weak significance is lockup, but after controlling for the role of common risk
factors, the relation between the lockup and future alphas is also insignificant.
Table 10 demonstrates that the results differ significantly across databases suggesting that
the conclusion about which fund characteristics explain cross-sectional differences between fund
returns may be database specific. Specifically, Table 10 shows that the results for fund size, age
12
It is important to control for domicile fixed effects, since Aragon, Liang, and Park (2011) document that
the impact of share restriction on hedge fund performance varies across domiciles. We find very similar
results when we adjust standard errors for within-cluster correlation, heteroskedasticity, and
autocorrelation.
and flow are very robust across databases as the previous literature (e.g., Teo, (2010, 2011) and
Aggarwal and Jorion (2010)) suggests. However, the variables that are related to managerial
incentives do not explain hedge fund cross-sectional returns consistently across databases. The
sign and significance for the high-water mark is changing wildly. However, all of the variables
related to incentives are significant for aggregate database. Therefore, we conclude that hedge
funds with greater managerial incentives outperform. Finally, we find very little evidence that
share restrictions in the form of lockup, notice and redemption periods are related to higher riskadjusted returns when we control for the role of other characteristics in multivariate regression.
As Aragon (2007), we find the strongest evidence using the TASS database, whereas using other
databases and even the aggregate database the conclusion is much weaker. For example, we find
a significantly positive coefficient for notice period only using TASS, but none of the other
databases.
Panel C of Table 10 presents results for aggregate database across three size groups. The
size groups are the same as in persistence tests. The results show that the relationship between
size and age is not significant consistently within size groups. None of the coefficients for fund
size is significant implying that cross-sectional performance difference may not be monotonic.
The fund’s age seems to be important only for small funds. This may be explained by more
pronounced backfilling bias.13 The relationship between the fund flow and performance is
consistent across size groups. A positive (negative) past month flow is associated with a higher
(lower) next’s month performance. All of the coefficients related to compensation structure are
positive among size groups. However, managerial incentives seem to be more important for small
funds, since the coefficient is the highest for them. Finally, share restrictions are weakly
associated with hedge fund performance also across size groups. The coefficients for lockup are
significantly positive for median and large funds. However, none of the other share restriction
variables are significantly related to performance. This suggests that share restrictions are not able
to explain cross-sectional differences in hedge fund performance when the role of other
characteristics is controlled for. Overall, the results for size categories are in the line with the
findings that especially managerial incentives are important for emerging hedge funds.
13
We define fund age as a length of return time-series. Aggarwal and Jorion (2010) examine the
relationship between fund age and performance using a novel event study methodology. We do not follow
this methodology, since only some of the databases provide ‘date added to database’ –variable that is
needed in order to follow their methodology.
6. Conclusion
This paper provides new stylized facts about the hedge fund industry using a large
consolidated hedge fund database. We document the sensitivity of hedge fund performance and
data biases to different data bases.
We propose a novel, easily repeatable methodology that can be used to identify unique
hedge funds from multiple share class structures containing more than two units linked to the
same risky investment portfolio. The major advantage of the approach is that we obtain as long as
possible time-series for unique hedge funds. Therefore, we can measure hedge fund performance
more accurately and mitigate the impact on several biases more efficiently. We highlight the
importance of using multiple databases in order to achieve more reliable proxy of the
unobservable population of hedge funds. After merging commercial databases, one of the striking
observations is that 67.7% of all share classes are covered exclusively by only one database with
BarclayHedge having the largest share of unique share classes (15.6%).
We demonstrate importance of using a large consolidated database by investigating
hedge fund average performance, performance persistence, and the cross-sectional relation
between fund characteristics and hedge fund performance. First, our aggregate database shows
that hedge funds add value after fees: average excess returns and alphas are economically
significant on both equal- and value-weight basis across databases. We document differences in
average performance measures between databases suggesting that hedge fund performance results
are subject to database selection.
Second, using our aggregate database, TASS, HFR, and
BarclayHedge we find that hedge fund performance persists at annual horizon, if portfolios are
created on an equal-weight basis. Based on value-weighted portfolios, the aggregate data and
HFR show persistence at monthly horizon. Hence, the results suggest that performance
persistence vanishes faster on a value-weight basis, a result that is likely to be driven by small
funds. For EurekaHedge and Morningstar, we do not find persistence at all. This is due to the fact
that EurekaHedge and Morningstar have lower attrition rates and higher a survivorship bias
creating an upward bias in results.
The stylized facts proposed in the paper suggest that one should be careful in interpreting
the results of hedge fund performance if the analysis is conducted using only one of the
commercial databases. Our study of databases shows differences in coverage of returns, assets
under management, and number of funds in the graveyard module.
The comparison of commercial databases to our aggregate database allows us to evaluate
whether all hedge fund databases contain the same level of information and whether differences
between databases induce biased inference. The major finding is that the conclusions based on the
consolidated database are qualitatively different from those based on the individual databases. To
avoid major biases, we argue that it is important to use a consolidated database because stylized
facts inferred from individual databases may differ from the true population.
References
Ackermann, C., McEnally, R., Ravenscraft, D., 1999. The performance of hedge funds: risk,
return and incentives. Journal of Finance 54, 833-874.
Agarwal, V., Daniel N. D., and Naik N. Y., 2009. Role of managerial incentives and discretion in
hedge fund performance. The Journal of Finance 64(5), 2221-2256
Agarwal, V., and Naik, N.Y., 2000. Multi-period performance persistence analysis of hedge
funds. Journal of Financial and Quantitative Analysis 53, 327-342.
Agarwal, V., and Ray S., 2011. Determinants and implications of fee changes in the hedge fund
industry. Georgia State University.
Aggarwal, R., K. and Jorion, P., 2010. The performance of emerging hedge funds and managers.
Journal of financial economics. 96, 238-256.
Aragon, G. O., 2007. Share restrictions and asset pricing: Evidence from the hedge fund industry.
Journal of Financial Economics 83(1), 33.58.
Aragon, G. O. Liang, B., and Park H., 2011. Onshore and offshore hedge funds: Are they twins?
Arizona State University.
Aragon, G. O. and Martin, J. S., 2010. A unique view of hedge fund derivatives usage: Safeguard
or speculation? forthcoming at the Journal of Financial Economics.
Aragon, G. O. and Nanda V., 2011. Strategic delays and clustering in hedge fund reported
returns. Arizona State University and Georgia Institute of Technology.
Bollen, N. P.B., and Pool, V. K., 2009. Do hedge fund managers misreport
returns? Evidence from the pooled distribution , Journal of Finance 64, 2257-2288.
Bollen, N. P.B., and Pool, V. K., 2008. Conditional return smoothing in the
hedge fund industry , Journal of Financial and Quantitative Analysis 43, 267-298.
Brown, S., Goetzmann, W., Ibbotson, R., 1999. Offshore hedge funds: Survival and performance,
1989-95. Journal of Business 72, 91-117.
Carhart, M., 1997. On persistence of mutual fund performance. Journal of Finance 52, 57-82.
Elton, E. J., Gruber M. J., and Blake C. R., 2001. A first look at the accuracy of the CRSP mutual
fund database and a comparison of the CRSP and Morningstar mutual fund databases.
Journal of Finance 56(6), 2415-2430.
Fama, E. F., and MacBeth J.D., 1973. Risk, return, and equilibrium: Empirical tests. Journal of
Political Economy 81, 607-636.
Fung, W., Hsieh, D., 2000. Performance characteristics of hedge funds and CTA funds: Natural
versus spurious biases. Journal of Financial and Quantitative Analysis 35, 291-307.
Fung, W., Hsieh, D., 2004. Hedge fund benchmarks: a risk based approach. Financial Analyst
Journal 60, 65-80.
Fung, W. and Hsieh D., 2009, Measurement biases in hedge fund performance data: An
update, Financial Analysts Journal 65, 36-38.
Fung, W., Hsieh, D., Naik, N., Ramadorai, T., 2008. Hedge funds: performance, risk, and capital
formation. Journal of Finance, 63 (4), 1777-1803.
Getmansky, M., Lo, A., Makarov, I., 2004. An econometric model of serial correlation and
illiquidity of hedge fund returns. Journal of Financial Economics 74, 529-610.
Goetzmann, W. N., Brown, S., Liang, B., and Schwarz C., 2009. Estimating Operational Risk for
Hedge Funds: The Omega-Score, Financial Analysts Journal, 5(1), 43-53,
Goetzmann, W.N., Brown, S., Liang, B., and Schwarz C., 2008. Mandatory Disclosure and
Operational Risk: Evidence from Hedge Fund Registration, Journal of Finance, 63(6),
2785-2815.
Güner, B., Rachev, S. T., Edelman, B., and Fabozzi, F. J., 2010. Bayesian inference for hedge
funds with stable distribution of returns. Ch.4 in Rethinking Risk Measurement and
Reporting: Volume II, Klaus Böcker, eds.
Jagannathan, R., Malakhov A., and Novikov D., 2010. Do hot hands exist among hedge fund
managers? An empirical evaluation. The Journal of Finance 65(1), 217.255.
Jylhä P., (2011). Hedge fund return misreporting: Incentives and effects. Aalto University.
Liang, B., 1999. On the performance of hedge funds. Financial Analysts Journal 55(4), 72.85.
Liang, B., 2000. Hedge funds: The living and the dead. Journal of Financial and Quantitative
Analysis 35(3), 309.326.
Kosowski, R., Naik N. Y., and Teo M., 2007. Do hedge funds deliver alpha? A Bayesian and
bootstrap analysis. Journal of Financial Economics 84(1), 229.264.
Kosowski, R., Timmermann, A., Wermers, R., White, H., 2006. Can mutual fund “stars” really
pick stocks? New evidence from a bootstrap analysis. Journal of Finance, 61(6).
Newey, W., West, K., 1987. A simple, positive semi-definite, heteroscedasticity and
autocorrelation consistent covariance matrix. Econometrica 55, 703-708.
Patton, A. J. and Ramadorai, T., 2011. On the Dynamics of Hedge Fund Risk Exposures.
Forthcoming in Journal of Finance.
Patton, A. J., Ramadorai, T, and Streatfield M., 2011. The reliability of voluntary disclosures:
Evidence from hedge funds. Duke University and University of Oxford.
Teo, M., 2009. Geography of hedge funds. Review of Financial Studies, 22, 3531-3561
Teo, M., 2010. Does size matter in hedge fund industry? Singapore Management University School of Business.
Teo, M., 2011. The Liquidity Risk of Liquid Hedge Funds, Journal of Financial Economics,
100, 24–44.
Table 1
Summary statistics of hedge fund returns and asset under management [January 1994 - June 2011]
Table provides summary statistics of hedge funds with at least 12 monthly return observations. Results are reported separately for each database. Aggregate database is a
merged database of single databases including TASS, Hedge Fund Research, BarclayHedge, EurekaHedge, and Morningstar. Databases are merged using a novel statistical
algorithm. First, all databases are grouped based on the management company name that is cleaned from errors and abbreviations. Second, correlation coefficients are
estimated for each share class pair within each management company. Each share class pair in correlation analysis is required to have at least 12 non-missing return
observations. Third, multiple share classes within management companies are grouped based on 0.99 correlation coefficient limit. Within groups, the share class having the
longest return time series represents a unique hedge fund. If a single group has more than one share class left with the same length of returns, the largest share class is selected
based on the asset under management to represent a unique hedge fund. If more than one share class has the same amount of returns and assets, the US dollar currency share
class is selected. The row "aggregate" contains results of the merged database that applies all five databases as a source data. The aggregate database is removed from multiple
share classes than can exist either between databases or within management companies. Single hedge fund databases are removed from multiple share classes using the same
statistical procedure.
Panel A shows summary statistics of returns and asset under management. Panel shows the number of unique share classes (column "# share classes"). The name matching
algorithm (presented in Appendix) is used to merge databases at the share class level in order to obtain the number of unique share classes in the aggregate data. Next column
shows the number of unique management firms in each database. For simple returns, Panel reports the number of funds (column 4), annualized average return (column 5), and
annualized standard deviation (column 6). Percentage amount of missing simple returns are reported in column 7. The next two columns consider the mean and standard
deviation of the length of simple return time series. Summary statistics are also reported for asset under management (in millions of US dollars). Panel B shows the statistical
properties of hedge fund returns. The first column describes the amount of funds in each database. The next three columns show the statistics of normality including skewness,
kurtosis and the number of funds for which the Jarque-Bera test of normality is rejected at the 5% level. The next three columns show the first two order serial correlations and
the number of funds for which the Ljung-Box test of serial correlation is rejected at the 5% level. The results of conditional smoothing are based on the regression proposed by
Bollen and Pool (2006):
Rt
a b1 R t
1
b1 (1 I t 1 )R t
1
t
,
1 if the return in month t-1 is greater than its mean and zero otherwise. The next four columns show the number of funds having statistically significant
where I t 1
positive and negative estimates of smoothing profiles based on the 5% level of significance. The next two columns show the results of misreporting based on a measure proposed
by Jylhä (2011). The measure formalizes the fact that the number of small gains exceeds the number of small losses documented by Bollen and Pool (2009). Panel presents the
measure of discontinuity (DC(%)) and z statistic (z(DC)) for testing the existence of discontinuity across databases. The final columns show the difference between the average
December values and the average of January-November values and the p-value for the test that this difference equals zero after correcting standard errors for clustering at the
fund-level. These measures test the result that on average, the December values are higher than January-November values as documented by Agarwal, Daniel and Naik (2011).
A. Summary statistics
All funds
Returns
Asset Under Management
Lenght of time series
Database
Lenght of time series
# share classes
# firms
# funds
Mean % pa
Std % pa
Missing (%)
Mean
Std
# funds
Mean
Std
Missing (%)
Mean
Std
Lipper TASS
10073
3338
8220
8.98
4.19
0.68
64.45
44.49
6719
101.92
312.29
34.40
42.57
43.76
Hedge Fund Research
11423
3710
9648
10.05
4.08
0.98
65.15
45.36
7929
107.22
635.96
19.43
53.01
48.38
BarclayHedge
11358
4224
9719
10.47
4.52
0.94
64.04
45.15
9308
112.24
502.23
11.49
57.22
45.17
EurekaHedge
8632
3040
7164
10.37
3.91
0.29
66.81
45.41
6399
141.45
397.82
37.12
42.14
38.91
Morningstar
8274
2817
6772
9.31
3.69
2.58
70.33
48.38
5570
122.41
418.65
32.23
48.92
48.79
Aggregate
37592
8353
24749
9.66
4.46
1.36
62.03
44.37
21308
110.08
519.81
27.56
45.56
43.62
Alive
Returns
Asset Under Management
Lenght of time series
Database
Lenght of time series
# share classes
# firms
# funds
Mean % pa
Std % pa
Missing (%)
Mean
Std
# funds
Mean
Std
Missing (%)
Mean
Std
Lipper TASS
4144
1386
3110
11.95
3.20
0.35
74.69
50.33
2413
123.17
356.34
37.75
46.66
50.81
Hedge Fund Research
4674
1716
3822
11.81
3.27
0.39
78.31
52.65
2943
170.94
1007.65
19.36
63.39
58.20
BarclayHedge
4256
1755
3536
12.44
3.44
0.34
78.15
52.80
3394
167.62
679.50
10.73
70.00
53.79
EurekaHedge
5441
2133
4452
12.39
3.62
0.27
73.15
48.78
4003
174.27
482.06
35.60
47.24
42.49
Morningstar
4369
1698
3558
10.99
3.33
2.14
77.58
52.25
2838
141.46
505.08
32.31
53.66
53.34
Aggregate
12929
4191
8512
12.15
3.71
0.35
72.45
50.61
7328
160.20
751.83
26.45
53.48
51.12
Defunct
Returns
Asset Under Management
Lenght of time series
Database
Lipper TASS
Lenght of time series
# share classes
# firms
# funds
Mean % pa
Std % pa
Missing (%)
Mean
Std
# funds
Mean
Std
Missing (%)
Mean
Std
5929
1952
5110
7.18
4.61
0.94
58.23
39.25
4306
90.01
283.99
31.80
40.09
38.64
Hedge Fund Research
6749
1994
5826
8.90
4.50
1.52
56.52
37.43
4986
69.61
200.45
19.49
46.20
39.22
BarclayHedge
7102
2469
6183
9.34
5.01
1.42
55.98
37.84
5914
80.45
359.58
12.09
49.91
37.49
EurekaHedge
3191
907
2712
7.05
4.17
0.34
56.40
36.98
2396
86.62
172.25
40.34
33.76
30.39
Morningstar
3905
1119
3214
7.44
3.98
3.19
62.29
42.27
2732
102.62
302.68
32.13
43.67
42.60
Aggregate
24663
4162
16237
8.35
4.77
2.03
56.57
39.64
13980
83.82
336.99
28.29
41.40
38.48
B. Statistical properties of hedge fund returns
All funds
Normality
Database
# funds
Skew
Kurt
Serial Correlation
JB
1
2
(% Rejection)
TASS
8072
-0.22
3.56
54.98
0.12
0.04
Discontinuity
Conditional Smoothing
LB
b+
DC (%)
b-
(% rejection)
Pos (%)
Neg (%)
Pos (%)
Neg (%)
23.27
12.57
87.43
5.93
94.07
December Spike
z(DC)
Avg (Dec minus
Diff
Jan-Nov) %/mth (p-value)
9.21
14.70
0.28
< 0.01
Hedge Fund Research
9508
-0.16
3.26
54.29
0.12
0.03
24.61
14.20
85.80
5.04
94.96
7.92
14.39
0.20
<0.01
Barclay
9588
-0.07
3.09
53.90
0.10
0.02
22.53
12.99
87.01
5.09
94.91
7.92
14.57
0.27
<0.01
Eureka
7009
-0.15
2.92
53.09
0.12
0.04
24.17
12.53
87.47
5.68
94.32
6.86
10.53
0.32
<0.01
Morningstar
6681
-0.22
3.28
54.42
0.13
0.03
25.92
13.73
86.27
7.07
92.93
7.87
12.04
0.36
<0.01
Aggregate
24768
-0.16
3.02
51.36
0.10
0.03
21.44
11.34
88.66
5.77
94.23
7.91
20.05
0.28
<0.01
Alive
Normality
Database
# funds
Skew
Kurt
2963
-0.19
3.28
Serial Correlation
JB
1
2
0.13
0.06
(% Rejection)
TASS
57.75
Discontinuity
Conditional Smoothing
LB
b+
b-
(% rejection)
Pos (%)
Neg (%)
Pos (%)
Neg (%)
27.07
12.69
87.31
6.41
93.59
December Spike
DC (%)
z(DC)
7.22
7.89
Avg (Dec minus
Diff
Jan-Nov) %/mth (p-value)
0.29
<0.01
Hedge Fund Research
3682
-0.15
3.33
58.17
0.13
0.04
29.88
15.92
84.08
5.57
94.43
8.14
10.61
0.22
<0.01
Barclay
3400
-0.07
3.15
57.47
0.11
0.03
27.15
14.26
85.74
5.88
94.12
7.38
9.40
0.27
<0.01
Eureka
4296
-0.12
3.02
55.49
0.12
0.04
26.05
13.08
86.92
5.61
94.39
6.86
8.87
0.30
<0.01
Morningstar
3463
-0.16
3.11
54.87
0.12
0.02
27.40
13.84
86.16
7.48
92.52
7.52
9.17
0.38
<0.01
Aggregate
10644
-0.14
2.83
52.02
0.10
0.03
23.97
11.34
88.66
6.28
93.72
7.21
13.17
0.30
<0.01
# funds
Skew
Kurt
DC (%)
z(DC)
Avg (Dec minus
TASS
5109
-0.24
3.72
Hedge Fund Research
5826
-0.18
3.22
Defunct
Normality
Database
Serial Correlation
JB
Discontinuity
Conditional Smoothing
LB
b-
b+
December Spike
Diff
1
2
(% rejection)
Pos (%)
Neg (%)
Pos (%)
Neg (%)
53.38
0.12
0.03
21.06
12.50
87.50
5.64
94.36
10.48
13.67
0.28
<0.01
51.84
0.12
0.03
21.28
13.11
86.89
4.70
95.30
9.23
13.31
0.19
<0.01
(% Rejection)
Jan-Nov) %/mth (p-value)
Barclay
6188
-0.07
3.06
51.94
0.09
0.02
19.99
12.28
87.72
4.65
95.35
8.02
12.01
0.27
<0.01
Eureka
2713
-0.18
2.77
49.28
0.12
0.04
21.19
11.65
88.35
5.79
94.21
7.54
7.47
0.35
<0.01
Morningstar
3218
-0.29
3.47
53.95
0.15
0.03
24.33
13.62
86.38
6.62
93.38
9.18
10.01
0.33
<0.01
Aggregate
14124
-0.17
3.16
50.86
0.11
0.02
19.53
11.34
88.66
5.39
94.61
8.72
17.42
0.26
<0.01
Table 2
Attrition rates
Table presents attrition rates for each hedge fund database. Aggregate database is a merged database of single databases including Lipper TASS, Hedge Fund Research,
BarclayHedge, EurekaHedge, and Morningstar. Databases are merged with a novel statistical procedure that is described in Appendex of the paper. Single databases are
removed from multiple share classes using the same statistical procedure. Attrition rate is calculated as the ratio of the number of dissolved funds to the number that existed
at the start of the year. Table includes year, number of funds that existed at the start of the year (Start), number of new defunct funds, and attrition rate in percentage (AR
%).
Year
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Start
.
772
941
1112
1336
1488
1770
1985
2244
2548
2965
3471
3892
4251
4283
3935
3724
TASS
Exit
9
48
97
72
108
145
167
189
191
190
242
334
406
631
898
616
580
AR %
.
6.22
10.31
6.47
8.08
9.74
9.44
9.52
8.51
7.46
8.16
9.62
10.43
14.84
20.97
15.65
15.57
Hedge Fund Research
Start
Exit
AR %
.
4
.
1100
38
3.45
1398
130
9.30
1616
152
9.41
1786
261
14.61
1882
180
9.56
2112
245
11.60
2314
213
9.20
2594
212
8.17
2974
250
8.41
3406
264
7.75
3998
389
9.73
4499
477
10.60
4859
671
13.81
4963
986
19.87
4526
677
14.96
4321
610
14.12
Start
.
1085
1331
1501
1721
1965
2319
2552
2944
2997
3480
3939
4380
4669
4745
4278
4160
BarclayHedge
Exit
9
138
132
108
112
95
236
191
537
250
326
413
534
660
1058
594
631
AR %
.
12.72
9.92
7.20
6.51
4.83
10.18
7.48
18.24
8.34
9.37
10.48
12.19
14.14
22.30
13.88
15.17
Start
.
234
308
418
546
715
1037
1321
1715
2243
2885
3454
3737
4143
4434
4354
4394
EurekaHedge
Exit
.
.
.
.
.
1
.
.
9
35
178
398
374
401
670
498
615
AR %
.
.
.
.
.
0.14
.
.
0.52
1.56
6.17
11.52
10.01
9.68
15.11
11.44
14.00
Start
.
283
384
517
668
849
1115
1391
1739
2217
2775
3471
4130
4537
4619
4240
3951
Morningstar
Exit
.
.
.
.
.
.
.
.
.
1
27
113
359
576
927
672
805
AR %
.
.
.
.
.
.
.
.
.
0.05
0.97
3.26
8.69
12.70
20.07
15.85
20.37
Start
.
2420
2979
3407
3838
4240
4895
5356
6148
7042
8347
9833
11072
12132
12648
11907
11624
Aggregate
Exit
22
207
302
289
405
365
552
424
609
499
662
982
1266
1633
2520
1807
1729
AR %
.
8.55
10.14
8.48
10.55
8.61
11.28
7.92
9.91
7.09
7.93
9.99
11.43
13.46
19.92
15.18
14.87
Table 3
Hedge fund average performance [January 1994 - December 2010].
Table presents results of the average performance of hedge funds for all databases. Aggregate database is a merged database of single databases including TASS, Hedge Fund
Research, BarclayHedge, EurekaHedge, and Morningstar. Databases are merged using a novel statistical procedure that is described in Appendix of the paper. Single databases are
removed from multiple share classes using the same statistical procedure. Panel A presents results of the average performance for equal-weight portfolios and Panel B presents
results for value-weight portfolios. Panel C shows the backfilling-adjusted results of the average performance for equal-weight portfolios. Panel D shows performance smoothingadjusted results of the average performance for equal-weight portfolios. The annualized Fung and Hsieh (2004) alpha is adjusted for serial correlation by adding the MA(2) process to
the model's error term. The annualized standard deviations of the excess returns and the Sharpe ratios are adjusted for serial correlation using the methodology proposed by
Getmansky, Lo and Makarov (2004). Panel E shows results of the equal-weight average performance of the aggregate database in sub periods that are selected based on Fung, Hsieh,
Naik and Ramadorai (2008). The first sub period is from January 1994 to December 1996. The second sub period is January 1997 - September 1998. The third sub period is October
1998 - March 2000. The fourth sub period is April 2000 - December 2004. The last sub period is January 2005 - December 2010. Panel F shows results of the equal-weight
performance of the aggregate database in size groups. Funds are sorted into size groups every December based on monthly fund-level non-missing AUM observations (reported in
millions of US dollars). "Missing AUM" portfolio consists of sorting of funds into portfolio every December based on the fund-level monthly missing AUM observations. This group
includes funds that report at least one missing AUM observation. Equal-weight returns are calculated for each portfolio monthly, so the portfolio weights are adjusted monthly.
In each Panel, the first column describes the name of the database. The second column presents the number of funds in each portfolio. The third column presents the percentage
amount of funds that are defunct. Annualized mean and standard deviation are reported in the next two columns. Values are reported in percentages. Sharpe ratio is annualized and
defined as the average excess return divided by the standard deviation of return. Alpha is the measure of abnormal return estimated from the seven factor model proposed by Fung
and Hsieh (2004). Alpha is annualized and reported in percentage. Risk factors are: the excess return of the S&P 500 index (SP-RF), the return of the Russell 2000 index minus the
return of the S&P 500 index (RL-SP), the excess return of ten-year Treasuries (TY-RF), the return of Moody's BAA corporate bonds minus ten-year Treasuries (BAA-TY), the excess
returns of look-back straddles on bonds (PTFSBD-RF), currencies (PTFSFX-RF), and commodities (PTFSCOM-RF). RSQ is the R-square of the model. Values of t-statistics are reported in
parentheses.
A. Equal-weight
Dataset
TASS
No. Of Funds
8072
% of Dead
63.29
Mean ER % pa
7.92
Std ER % pa
7.07
Hedge Fund Research
9508
61.27
8.72
7.01
BarclayHedge
9588
64.54
9.02
6.51
EurekaHedge
7009
38.71
10.28
7.57
Morningstar
6681
48.17
9.75
7.01
Aggregate
24768
57.03
8.45
7.11
No. Of Funds
6644
% of Dead
65.32
Mean ER % pa
8.56
Std ER % pa
7.24
Hedge Fund Research
7844
63.64
7.73
6.19
BarclayHedge
9193
64.68
6.91
5.68
EurekaHedge
6310
38.92
7.72
6.86
Morningstar
5695
49.53
8.07
6.14
Aggregate
21587
58.39
7.36
6.61
Sharpe (pa) Alpha % pa
1.12
5.31
(5.13)
1.24
6.20
(6.53)
1.38
6.87
(7.08)
1.35
7.61
(6.85)
1.38
7.28
(7.14)
1.19
5.87
(5.83)
SP-RF
0.28
(13.32)
0.30
(15.60)
0.26
(13.37)
0.31
(13.66)
0.28
(13.86)
0.29
(14.32)
RL-SP
0.16
(6.65)
0.18
(8.02)
0.14
(6.04)
0.16
(6.18)
0.16
(6.62)
0.16
(6.56)
TY-RF
0.10
(2.59)
0.08
(2.07)
0.08
(2.20)
0.10
(2.25)
0.09
(2.26)
0.10
(2.48)
BAA-TY
0.28
(5.77)
0.22
(5.04)
0.25
(5.52)
0.29
(5.65)
0.26
(5.44)
0.28
(6.01)
PTFSBD-RF
0.00
-(0.55)
0.00
-(0.29)
0.01
(1.00)
0.00
(0.15)
0.00
(0.05)
0.00
(0.04)
PTFSFX-RF
0.01
(2.61)
0.01
(2.46)
0.02
(3.60)
0.01
(2.80)
0.01
(2.54)
0.01
(2.76)
PTFSCOM-RF
0.01
(1.74)
0.01
(1.56)
0.01
(2.36)
0.01
(1.95)
0.01
(2.00)
0.01
(1.88)
RSQ
0.66
Sharpe (pa) Alpha % pa
1.18
5.97
(4.60)
1.24
5.59
(5.48)
1.21
5.01
(5.07)
1.12
5.48
(4.62)
1.31
6.04
(5.73)
1.11
4.88
(4.61)
SP-RF
0.25
(9.55)
0.23
(11.44)
0.20
(10.14)
0.23
(9.82)
0.21
(10.09)
0.25
(11.78)
RL-SP
0.16
(5.23)
0.15
(5.96)
0.11
(4.71)
0.11
(3.79)
0.15
(5.76)
0.13
(5.21)
TY-RF
0.15
(3.01)
0.09
(2.20)
0.12
(3.14)
0.11
(2.30)
0.12
(2.93)
0.12
(2.96)
BAA-TY
0.18
(3.06)
0.18
(3.73)
0.20
(4.34)
0.30
(5.37)
0.20
(4.11)
0.24
(4.79)
PTFSBD-RF
-0.02
-(2.13)
-0.01
-(1.17)
0.00
-(0.16)
0.00
-(0.11)
0.00
-(0.50)
-0.01
-(1.40)
PTFSFX-RF
0.01
(1.70)
0.01
(1.97)
0.01
(2.51)
0.01
(1.56)
0.01
(1.45)
0.01
(2.07)
PTFSCOM-RF
0.02
(1.79)
0.01
(1.76)
0.02
(2.59)
0.02
(2.86)
0.02
(3.00)
0.01
(2.03)
RSQ
0.50
0.71
0.65
0.66
0.67
0.69
B. Value-weight
Dataset
TASS
0.58
0.53
0.53
0.54
0.60
C. Backfill-adjusted results of the average performance (Equal-weight)
Dataset
TASS
No. Of Funds
7963
% of Dead
63.23
Mean ER % pa
7.40
Std ER % pa
7.30
Hedge Fund Research
9384
61.11
8.08
7.37
BarclayHedge
9464
64.41
7.91
6.80
EurekaHedge
6924
38.65
9.93
7.91
Morningstar
6635
48.21
9.23
7.24
Aggregate
24420
56.91
7.74
7.45
Sharpe (pa) Alpha % pa
1.01
4.32
(3.96)
1.09
5.06
(4.95)
1.16
5.13
(5.10)
1.25
6.73
(5.65)
1.27
6.28
(5.79)
1.04
4.57
(4.29)
SP-RF
0.28
(13.35)
0.31
(15.55)
0.27
(13.99)
0.31
(13.52)
0.29
(13.61)
0.30
(14.52)
RL-SP
0.17
(6.66)
0.19
(7.99)
0.15
(6.38)
0.17
(6.19)
0.17
(6.62)
0.16
(6.67)
TY-RF
0.10
(2.38)
0.08
(1.98)
0.10
(2.66)
0.10
(2.20)
0.09
(2.18)
0.11
(2.58)
BAA-TY
0.28
(5.82)
0.23
(5.15)
0.26
(5.85)
0.30
(5.72)
0.27
(5.58)
0.29
(6.19)
PTFSBD-RF
0.00
-(0.61)
0.00
-(0.41)
0.00
(0.50)
0.00
(0.11)
0.00
-(0.14)
0.00
-(0.29)
PTFSFX-RF
0.01
(2.46)
0.01
(2.39)
0.02
(3.75)
0.01
(2.59)
0.01
(2.54)
0.01
(2.74)
PTFSCOM-RF
0.01
(1.54)
0.01
(1.36)
0.01
(2.22)
0.01
(1.58)
0.01
(1.65)
0.01
(1.69)
RSQ
0.68
Sharpe (pa) Alpha % pa
0.85
5.42
(4.01)
0.95
6.27
(5.19)
1.12
6.89
(6.08)
1.04
7.68
(5.41)
1.08
7.35
(5.89)
0.92
5.93
(4.71)
SP-RF
0.29
(12.92)
0.31
(16.88)
0.27
(12.80)
0.31
(13.43)
0.29
(13.51)
0.30
(13.87)
RL-SP
0.15
(7.21)
0.17
(9.12)
0.13
(6.43)
0.15
(6.25)
0.15
(6.74)
0.15
(6.87)
TY-RF
0.11
(2.22)
0.09
(1.96)
0.09
(1.95)
0.10
(1.95)
0.09
(2.04)
0.10
(2.18)
BAA-TY
0.24
(6.46)
0.19
(5.30)
0.23
(6.18)
0.26
(6.13)
0.23
(6.19)
0.25
(6.81)
PTFSBD-RF
0.00
-(0.36)
0.00
-(0.24)
0.01
(0.87)
0.00
(0.10)
0.00
(0.11)
0.00
(0.13)
PTFSFX-RF
0.01
(2.46)
0.01
(2.12)
0.01
(3.59)
0.01
(2.67)
0.01
(2.43)
0.01
(2.67)
PTFSCOM-RF
0.01
(2.39)
0.01
(2.37)
0.02
(2.83)
0.02
(2.87)
0.02
(2.69)
0.02
(2.53)
RSQ
0.68
0.73
0.69
0.68
0.68
0.71
D. Smoothing-adjusted results of the average performance (Equal-weight)
Database
TASS
No. Of Funds
8072
% of Dead
63.29
Mean ER % pa
7.92
Std ER % pa
9.31
Hedge Fund Research
9508
61.27
8.72
9.10
BarclayHedge
9588
64.54
9.02
8.04
EurekaHedge
7009
38.71
10.28
9.87
Morningstar
6681
48.17
9.75
8.95
Aggregate
24768
57.03
8.45
9.13
0.73
0.66
0.68
0.69
0.70
E. Performance of the aggregate database in sub periods (Equal-weight)
Sub period
Jan 1994 - Dec 1996
No. Of Funds
% of Dead
Mean ER % pa
Std ER % pa
3771
82.84
9.70
4.38
2.16
Jan 1997 - Sep 1998
4635
79.03
4.89
7.57
0.64
Oct 1998 - Mar 2000
Apr 2000 - Dec 2004
Jan 2005 - Dec 2010
5627
12319
20432
73.75
21.87
68.71
7.81
47.94
6.03
6.73
5.64
8.85
Sharpe (pa) Alpha % pa
3.23
1.42
0.68
SP-RF
RL-SP
TY-RF
PTFSFX-RF
PTFSCOM-RF
RSQ
8.59
0.25
0.16
0.04
BAA-TY PTFSBD-RF
0.04
0.00
0.01
0.04
0.62
(3.61)
(3.56)
(2.26)
(0.34)
(0.13)
(0.12)
(1.58)
(3.11)
2.97
0.39
0.18
-0.30
0.25
0.02
0.01
0.04
(1.00)
(6.77)
(2.67)
-(1.32)
(0.66)
(0.85)
(0.47)
(1.28)
15.82
0.29
0.26
0.41
0.56
0.05
0.00
-0.02
(4.35)
(4.23)
(5.85)
(1.77)
(2.12)
(1.78)
(0.15)
-(1.05)
0.88
0.88
5.71
0.28
0.17
0.23
0.16
0.00
0.03
0.01
(5.23)
(12.22)
(6.65)
(6.16)
(2.14)
(0.47)
(5.15)
(1.14)
0.88
5.66
0.37
-0.09
-0.02
0.25
0.01
0.00
0.02
(2.50)
(8.05)
-(1.24)
-(0.23)
(3.51)
(0.43)
-(0.14)
(1.48)
SP-RF
RL-SP
TY-RF
PTFSFX-RF
PTFSCOM-RF
RSQ
5.91
0.32
0.17
0.12
0.32
-0.01
0.01
0.01
0.67
(4.72)
(13.02)
(5.93)
(2.35)
(5.80)
-(0.70)
(1.68)
(1.27)
7.25
0.28
0.17
0.10
0.24
0.01
0.02
0.02
(6.80)
(13.65)
(6.80)
(2.37)
(5.02)
(1.38)
(3.87)
(2.32)
4.71
0.31
0.18
0.09
0.25
0.00
0.01
0.01
(4.63)
(15.54)
(7.78)
(2.25)
(5.54)
-(0.55)
(2.88)
(1.83)
0.73
F. Performance of the aggregate database in size groups (Equal-weight)
Size group
No. Of Funds
% of Dead
Mean ER % pa
Std ER % pa
Missing AUM observations
9110
54.31
9.37
8.22
1.14
$0M <= AUM <= $10M
10152
65.57
9.89
7.03
1.41
$10M < AUM <= $50M
10532
59.60
7.81
7.37
1.06
$50M < AUM <= $250M
$250M < AUM <= $500M
$500M < AUM <= $1000M
AUM > $1000M
7750
2521
1313
626
52.58
44.03
41.20
42.81
5.73
4.84
5.84
5.45
7.35
7.32
6.83
8.75
Sharpe (pa) Alpha % pa
0.78
0.66
0.85
0.62
BAA-TY PTFSBD-RF
2.48
0.29
0.15
0.11
0.31
-0.01
0.01
0.01
(2.32)
(13.86)
(6.05)
(2.68)
(6.45)
-(1.13)
(1.98)
(1.90)
1.59
0.26
0.13
0.13
0.34
-0.01
0.01
0.01
(1.34)
(11.21)
(4.63)
(2.90)
(6.49)
-(1.33)
(1.50)
(1.37)
2.80
0.24
0.09
0.12
0.29
-0.01
0.01
0.01
(2.39)
(10.66)
(3.25)
(2.61)
(5.48)
-(1.91)
(1.92)
(1.18)
1.58
0.30
0.10
0.18
0.30
-0.02
0.01
0.01
(0.97)
(9.47)
(2.56)
(2.75)
(4.16)
-(2.35)
(1.20)
(1.27)
0.67
0.73
0.70
0.63
0.58
0.50
Table 4
Average performance in size groups [January 1994 - December 2010]
Table shows results of the average performance in size groups. Results are reported for each database. Aggregate database is a merged database including TASS, Hedge Fund
Research, BarclayHedge, EurekaHedge, and Morningstar. Databases are merged using a novel statistical procedure that is described in Appendix of the paper. Single
databases are removed from multiple share classes using the statistical algorithm. All funds are required to have at least 12 monthly return observations. In panel A, funds are
sorted into size groups every December based on monthly fund-level non-missing AUM observations (reported in millions of US dollars). "Missing AUM" portfolio consists of
sorting of funds into portfolio every December based on the fund-level monthly missing AUM observations. This group includes funds that report at least one missing AUM
observation. In panel B, three subsamples are created for each database. The first subsample consists of funds that do not report AUM observations. The second subsample
consists of all fund-level AUM observations that are missing in the beginning of the time series (conditional on non-missing returns) until the first reported non-missing AUM
observation appears. The third subperiod consists of all fund-level missing AUM observations that are reported after the first reported AUM observation. All portfolios are
equal-weighted monthly.
In Panel A and B, column "# funds" is the number of funds that are included in each portfolio. "Dead %" column tells the amount of dead funds in percentage. Columns "Mean
ER" and "Std ER" are the annualized mean and standard deviation of excess returns (in percentages). "Alpha" is the measure of abnormal return estimated from the seven
factor model proposed by Fung and Hsieh (2004). Alpha is annualized and reported in percentage. Risk factors are: the excess return of the S&P 500 index (SP-RF), the return
of the Russell 2000 index minus the return of the S&P 500 index (RL-SP), the excess return of ten-year Treasuries (TY-RF), the return of Moody's BAA corporate bonds minus
ten-year Treasuries (BAA-TY), the excess returns of look-back straddles on bonds (PTFSBD-RF), currencies (PTFSFX-RF), and commodities (PTFSCOM-RF). RSQ is the R-square of
the model. Values of t-statistics are reported in parentheses. In Panel B, "Obs" is the number of excess returns in each portfolio and Sharpe ratio (Column "Sharpe") is
annualized and defined as the average excess return divided by the standard deviation of return.
A. Average performance in size groups
TASS
Size group
Hedge Fund Research
BarclayHedge
# funds
Dead %
Mean ER
Std ER
Alpha
# funds
Dead %
Mean ER
Std ER
Alpha
# funds
Dead %
Mean ER
Std ER
Alpha
Missing AUM observations
3804
60.17
9.08
7.87
5.87
2485
63.06
9.80
7.83
6.49
1558
62.20
9.77
8.48
6.26
$0M <= AUM <= $10M
3138
70.94
9.35
7.48
4016
67.06
11.25
7.68
4947
69.90
10.53
6.39
$10M < AUM <= $50M
3507
65.10
7.49
7.24
4723
62.95
8.07
7.16
4710
62.97
8.60
6.87
(4.49)
6.53
(5.30)
(5.70)
4.55
2683
60.23
5.76
7.05
2.82
851
53.47
4.59
7.65
1.52
3419
55.60
5.95
7.24
458
53.93
6.64
8.32
3.20
1116
43.55
5.02
7.03
205
60.49
6.27
11.43
2.14
579
37.82
5.14
7.00
3602
56.91
5.93
6.92
2.25
269
31.97
5.99
7.84
2.81
1246
49.28
5.45
6.88
2.34
(1.92)
680
45.74
5.10
6.44
2.59
(2.16)
290
43.10
4.40
6.07
(1.91)
EurekaHedge
2.85
(2.75)
(1.81)
(0.84)
Size group
2.04
5.82
(5.96)
(1.80)
(2.03)
AUM > $1000M
2.86
8.46
(7.83)
(2.73)
(1.19)
$500M < AUM <= $1000M
5.25
(5.42)
(2.64)
$250M < AUM <= $500M
(5.30)
(7.79)
(4.35)
$50M < AUM <= $250M
8.41
1.88
(1.66)
Morningstar
Aggregate
# funds
Dead %
Mean ER
Std ER
Alpha
# funds
Dead %
Mean ER
Std ER
Alpha
# funds
Dead %
Mean ER
Std ER
Alpha
Missing AUM observations
3780
38.78
10.87
8.03
7.55
3260
49.33
8.84
7.34
5.69
9110
54.31
9.37
8.22
5.91
$0M <= AUM <= $10M
2149
42.86
12.98
7.95
2581
51.38
13.28
7.86
10152
65.57
9.89
7.03
(6.29)
10.28
(4.66)
(7.59)
$10M < AUM <= $50M
2931
40.09
9.50
9.08
6.50
2488
36.70
8.21
7.76
4.92
2999
49.68
10.01
7.36
926
29.81
7.52
8.73
4.52
2261
45.78
7.36
7.47
542
29.34
5.85
7.54
AUM > $1000M
249
23.69
3.11
8.28
2.91
761
41.52
6.56
7.48
(0.83)
59.60
7.81
7.37
4.18
4.01
404
37.87
4.80
7.57
197
42.64
5.66
7.31
2.08
7750
52.58
5.73
7.35
(2.08)
2.48
(2.32)
2521
44.03
4.84
7.32
1.59
(1.34)
1313
41.20
5.84
6.83
626
42.81
5.45
8.75
(1.34)
2.98
4.71
(4.63)
(2.74)
(1.92)
1.36
10532
(3.68)
(2.56)
$500M < AUM <= $1000M
7.08
7.25
(6.80)
(6.73)
(3.71)
$250M < AUM <= $500M
(4.72)
(9.00)
(4.11)
$50M < AUM <= $250M
10.52
2.80
(2.39)
1.58
(0.97)
B. Average performance conditional on reporting of missing AUM observations
TASS
Hedge Fund Research
Category
Obs
# funds
Dead %
Mean ER
Std ER
Sharpe
Alpha
Obs
# funds
Dead %
Mean ER
Std ER
Sharpe
Alpha
No AUM reported
204
1435
53.59
9.44
8.76
1.07
6.63
204
1666
50.06
9.93
8.04
1.23
6.96
Missing AUM in the beginning
203
2181
59.70
10.93
8.04
1.35
201
927
85.87
10.79
5.70
1.88
(4.00)
8.07
(5.66)
(6.06)
Missing AUM in the middle-end
203
2972
72.07
3.48
7.58
0.46
0.93
8.93
(9.93)
199
220
95.00
9.49
12.74
0.74
(0.82)
8.29
(2.61)
BarclayHedge
EurekaHedge
Category
Obs
# funds
Dead %
Mean ER
Std ER
Sharpe
No AUM reported
204
427
57.14
9.44
8.41
1.12
Alpha
Obs
# funds
Dead %
Mean ER
Std ER
Sharpe
6.64
204
784
33.16
9.94
7.06
1.40
(5.34)
Missing AUM in the beginning
203
1939
72.98
11.71
8.14
1.44
Missing AUM in the middle-end
199
410
79.51
6.28
9.50
0.66
8.93
7.33
(6.75)
203
3474
41.02
11.62
7.72
1.50
187
576
43.58
5.27
16.93
0.31
(7.77)
2.47
Alpha
9.09
(8.07)
(1.54)
2.99
(0.80)
Morningstar
Aggregate
Category
Obs
# funds
Dead %
Mean ER
Std ER
Sharpe
Alpha
Obs
# funds
Dead %
Mean ER
Std ER
Sharpe
Alpha
No AUM reported
204
1036
38.42
8.43
7.58
1.11
5.75
204
3399
44.81
8.95
7.79
1.14
6.11
(4.31)
Missing AUM in the beginning
203
1705
52.73
9.65
6.88
1.40
7.31
(5.03)
203
6472
58.27
10.95
8.04
1.36
(6.27)
Missing AUM in the middle-end
201
2697
51.72
9.00
8.05
1.11
6.70
(4.69)
8.07
(6.79)
203
4189
63.98
4.66
7.76
0.60
2.02
(1.65)
Table 5
Average performance of the aggregate database by investment objective [January 1994 - December 2010]
Table shows results of the average performance grouped by the investment objective. Results are reported for the aggregate database including TASS, Hedge Fund Research,
BarclayHedge, EurekaHedge, and Morningstar. Databases are merged using a statistical procedure that is described in Appendix of the paper. All funds are required to have at
least 12 monthly return observations. Panel A shows results of the average performance where hedge funds are categorized based on the investment objectives. Classification
of strategies is provided in Appendix of the paper. In Panel B, hedge funds are sorted into terciles each December based on monthly AuM observations and results of the
average performance are reported for each size tercile and investment objective. In Table, the second column describes the number of funds in each portfolio and Column "% of
Dead" tells the percentage amount of defunct funds if compared to all funds. The next two columns include the annualized average excess return and the standard deviation of
excess returns. Sharpe ratio is annualized and defined as the average excess return divided by the standard deviation of return. Alpha is the measure of abnormal return
estimated from the seven factor model proposed by Fung and Hsieh (2004). Alpha is annualized and reported in percentage. Risk factors are: the excess return of the S&P 500
index (SP-RF), the return of the Russell 2000 index minus the return of the S&P 500 index (RL-SP), the excess return of ten-year Treasuries (TY-RF), the return of Moody's BAA
corporate bonds minus ten-year Treasuries (BAA-TY), the excess returns of look-back straddles on bonds (PTFSBD-RF), currencies (PTFSFX-RF), and commodities (PTFSCOM-RF).
RSQ is the R-square of the model. Values of t-statistics are reported in parentheses.
A. Performance of the aggregate database by the investment objective (Equal-weight)
Investment objective
All
CTA
No. Of Funds % of Dead
24768
2413
57.03
59.22
Mean ER % pa
8.45
7.59
Std ER % pa Sharpe (pa)
7.11
6.87
1.19
1.11
Emerging Markets
2724
33.11
11.79
15.72
0.75
Event Driven
1363
63.98
8.15
6.37
1.28
Global Macro
1601
61.02
6.82
5.30
1.28
Long Only
473
31.08
7.28
12.96
0.56
Long/Short
6991
59.30
9.61
9.90
0.97
Market Neutral
1354
63.88
5.80
4.20
1.36
Multi-Strategy
3480
55.57
8.04
7.27
1.10
Others
1084
82.38
7.96
6.62
1.20
Relative Value
2522
58.29
6.27
4.92
1.28
Sector
648
62.19
12.65
12.96
0.97
Short Bias
115
74.78
2.14
11.86
0.18
Alpha % pa
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
PTFSFX-RF
PTFSCOM-RF
RSQ
5.87
0.29
0.16
0.10
0.28
0.00
0.01
0.01
0.69
(5.83)
(14.32)
(6.56)
(2.48)
(6.01)
(0.04)
(2.76)
(1.88)
6.74
0.09
0.03
0.21
0.17
0.02
0.04
0.03
(4.78)
(3.13)
(0.74)
(3.76)
(2.65)
(1.87)
(5.70)
(3.13)
7.06
0.53
0.23
-0.01
0.46
-0.02
0.00
0.00
(2.50)
(9.30)
(3.42)
-(0.08)
(3.52)
-(1.32)
(0.30)
-(0.07)
5.60
0.23
0.14
0.02
0.31
-0.02
0.00
0.00
(6.98)
(14.03)
(7.54)
(0.77)
(8.25)
-(3.42)
(1.27)
-(0.41)
5.10
0.19
0.07
0.16
0.15
0.00
0.01
0.02
(5.18)
(9.39)
(3.17)
(4.20)
(3.25)
(0.23)
(2.84)
(2.38)
2.64
0.60
0.30
0.05
0.39
0.00
0.01
0.00
(1.76)
(19.78)
(8.44)
(0.90)
(5.53)
(0.15)
(0.88)
(0.34)
6.20
0.45
0.30
0.06
0.18
0.00
0.01
0.01
(5.02)
(18.24)
(10.22)
(1.29)
(3.10)
-(0.47)
(1.24)
(0.94)
4.28
0.12
0.04
0.11
0.17
-0.01
0.01
0.00
(5.25)
(7.57)
(1.98)
(3.43)
(4.42)
-(1.47)
(2.15)
(0.84)
6.77
0.13
0.06
0.18
0.32
0.02
0.02
0.04
(4.54)
(4.32)
(1.55)
(3.18)
(4.56)
(2.43)
(3.44)
(3.69)
5.51
0.26
0.16
0.11
0.20
-0.01
0.01
0.01
(5.37)
(12.67)
(6.64)
(2.81)
(4.21)
-(0.98)
(1.62)
(1.08)
4.07
0.12
0.05
0.14
0.38
-0.01
0.00
0.00
(5.51)
(8.26)
(2.97)
(4.95)
(11.04)
-(1.83)
(0.57)
(0.30)
8.74
0.56
0.44
0.05
0.10
0.00
0.01
0.00
(4.65)
(14.66)
(9.75)
(0.70)
(1.17)
-(0.28)
(0.61)
(0.34)
4.93
-0.52
-0.40
-0.03
0.26
0.00
0.01
0.01
(2.53)
-(13.17)
-(8.54)
-(0.44)
(2.83)
-(0.08)
(0.68)
(0.82)
0.34
0.49
0.75
0.46
0.79
0.76
0.41
0.34
0.63
0.65
0.67
0.58
B. Average performance of the investment objectives in size groups (aggregate data)
CTA
Group
No. Of Funds
% of Dead
Mean ER % pa
Small
1464
63.39
8.73
Std ER % pa Sharpe (pa) Alpha % pa
7.00
1.25
Med
911
59.60
5.28
6.87
0.77
Large
528
47.73
3.50
7.13
0.49
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
PTFSFX-RF PTFSCOM-RF
7.56
0.08
0.02
0.20
0.16
0.02
0.04
0.02
(4.91)
(2.82)
(0.62)
(3.30)
(2.34)
(2.29)
(5.31)
(2.11)
3.95
0.08
0.04
0.21
0.13
0.01
0.04
0.03
(2.72)
(2.67)
(1.30)
(3.70)
(2.05)
(1.44)
(6.36)
(2.74)
2.12
0.07
0.01
0.21
0.14
0.01
0.04
0.02
(1.31)
(2.24)
(0.32)
(3.33)
(1.91)
(0.78)
(4.98)
(2.43)
RSQ
0.31
0.36
0.27
Emerging markets
Group
No. Of Funds
% of Dead
Mean ER % pa
Small
998
39.08
16.40
Med
Large
1340
849
37.01
30.86
11.86
6.50
Std ER % pa Sharpe (pa) Alpha % pa
18.65
15.96
16.14
0.88
0.74
0.40
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
11.88
0.58
0.25
-0.07
0.59
0.02
PTFSFX-RF PTFSCOM-RF
0.01
0.01
(3.22)
(8.09)
(2.94)
-(0.51)
(3.58)
(0.80)
(0.38)
(0.55)
6.55
0.53
0.23
0.01
0.46
-0.03
0.00
0.00
(2.18)
(8.97)
(3.34)
(0.09)
(3.42)
-(1.57)
(0.28)
(0.03)
0.86
0.47
0.22
0.03
0.56
-0.04
0.00
0.01
(0.28)
(7.84)
(3.07)
(0.23)
(4.10)
-(2.37)
(0.04)
(0.29)
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
8.55
0.30
0.22
0.04
0.20
-0.01
0.00
-0.01
(5.79)
(10.37)
(6.58)
(0.75)
(3.06)
-(1.08)
(0.42)
-(0.88)
5.23
0.25
0.16
0.01
0.34
-0.02
0.01
-0.01
(5.65)
(14.04)
(7.64)
(0.37)
(8.17)
-(3.42)
(1.53)
-(1.09)
3.91
0.18
0.12
0.01
0.34
-0.02
0.00
0.00
(4.30)
(10.41)
(5.75)
(0.15)
(8.32)
-(4.18)
(0.55)
(0.28)
RSQ
0.44
0.50
0.48
Event driven
Group
No. Of Funds
% of Dead
Mean ER % pa
Std ER % pa Sharpe (pa) Alpha % pa
Small
470
69.79
11.54
8.57
1.35
Med
687
62.88
8.32
7.25
1.15
Large
634
60.88
6.58
6.33
1.04
PTFSFX-RF PTFSCOM-RF
RSQ
0.58
0.77
0.70
Global macro
Group
No. Of Funds
% of Dead
Mean ER % pa
Small
858
63.64
6.65
Med
Large
567
408
60.85
54.41
4.97
5.40
Std ER % pa Sharpe (pa) Alpha % pa
5.58
5.87
5.88
1.19
0.85
0.91
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
PTFSFX-RF PTFSCOM-RF
5.14
0.16
0.07
0.08
0.17
0.01
0.00
0.03
(4.41)
(7.17)
(2.52)
(1.75)
(3.27)
(1.09)
(0.44)
(3.48)
2.71
0.16
0.09
0.20
0.20
0.00
0.02
0.01
(2.29)
(7.00)
(3.33)
(4.29)
(3.78)
(0.26)
(4.45)
(1.99)
2.67
0.18
0.09
0.23
0.14
-0.01
0.02
0.01
(2.24)
(7.65)
(3.43)
(4.86)
(2.66)
-(1.98)
(3.14)
(1.55)
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
RSQ
0.38
0.42
0.41
Long only
Group
No. Of Funds
% of Dead
Mean ER % pa
Small
144
36.81
11.35
Med
Large
197
186
32.49
25.81
8.48
7.04
Std ER % pa Sharpe (pa) Alpha % pa
13.77
14.58
13.30
0.82
0.58
0.53
PTFSFX-RF PTFSCOM-RF
8.14
0.58
0.19
-0.03
0.38
0.03
0.01
-0.01
(3.81)
(14.00)
(3.94)
-(0.36)
(3.98)
(2.04)
(0.98)
-(0.85)
2.77
0.66
0.32
0.09
0.39
0.00
0.01
0.00
(1.51)
(18.56)
(7.56)
(1.19)
(4.74)
-(0.22)
(0.67)
(0.40)
1.48
0.59
0.27
0.11
0.41
-0.01
0.00
0.01
(0.88)
(17.94)
(7.04)
(1.66)
(5.53)
-(1.20)
(0.22)
(0.51)
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
RSQ
0.67
0.77
0.77
Long/Short
Group
No. Of Funds
% of Dead
Mean ER % pa
Small
3008
62.77
11.44
Med
Large
3245
2183
59.11
54.10
9.27
6.78
Std ER % pa Sharpe (pa) Alpha % pa
11.29
10.10
10.34
1.01
0.92
0.65
PTFSFX-RF PTFSCOM-RF
7.46
0.52
0.35
0.01
0.14
0.00
0.01
0.00
(5.41)
(19.52)
(11.09)
(0.16)
(2.34)
(0.07)
(1.46)
(0.33)
5.57
0.46
0.32
0.03
0.16
0.00
0.01
0.00
(4.60)
(19.69)
(11.35)
(0.72)
(2.98)
(0.05)
(1.17)
(0.32)
2.81
0.46
0.29
0.06
0.17
-0.01
0.01
0.01
(1.98)
(16.61)
(9.00)
(1.08)
(2.63)
-(1.07)
(0.90)
(1.10)
RSQ
0.79
0.80
0.73
Market neutral
Group
No. Of Funds
% of Dead
Mean ER % pa
Small
581
68.85
8.39
5.46
1.51
Med
643
69.05
4.71
4.11
1.12
Large
450
60.22
3.60
Std ER % pa Sharpe (pa) Alpha % pa
4.02
0.87
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
7.07
0.08
0.04
0.09
0.11
-0.01
PTFSFX-RF PTFSCOM-RF
0.01
0.01
(5.27)
(3.23)
(1.18)
(1.69)
(1.83)
-(0.95)
(1.76)
(1.76)
3.15
0.11
0.03
0.08
0.17
0.00
0.01
0.00
(3.68)
(6.78)
(1.47)
(2.50)
(4.48)
-(0.81)
(2.40)
(0.58)
2.12
0.14
-0.02
0.08
0.10
-0.01
0.00
0.01
(2.54)
(8.29)
-(1.24)
(2.43)
(2.82)
-(1.24)
(1.27)
(1.19)
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
RSQ
0.14
0.38
0.38
Multi-strategy
Group
No. Of Funds
% of Dead
Mean ER % pa
Small
1673
61.98
8.69
Std ER % pa Sharpe (pa) Alpha % pa
8.78
0.99
Med
1413
55.56
7.30
7.85
0.93
Large
965
48.70
6.27
7.21
0.86
PTFSFX-RF PTFSCOM-RF
7.30
0.10
0.07
0.23
0.22
0.03
0.04
0.05
(3.87)
(2.77)
(1.70)
(3.08)
(2.65)
(2.26)
(4.31)
(4.34)
5.49
0.15
0.08
0.24
0.22
0.02
0.03
0.04
(3.32)
(4.73)
(2.10)
(3.65)
(3.04)
(2.28)
(3.94)
(3.98)
4.58
0.13
0.04
0.22
0.29
0.02
0.02
0.03
(2.92)
(4.18)
(1.05)
(3.51)
(4.17)
(2.55)
(3.07)
(2.85)
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
RSQ
0.34
0.37
0.33
Others
Group
No. Of Funds
% of Dead
Mean ER % pa
Small
387
84.24
9.63
Std ER % pa Sharpe (pa) Alpha % pa
8.27
1.17
Med
414
78.99
6.77
8.24
0.82
Large
270
82.96
5.66
5.90
0.95
PTFSFX-RF PTFSCOM-RF
6.32
0.30
0.20
0.11
0.22
-0.01
0.02
0.00
(4.38)
(10.70)
(5.89)
(1.99)
(3.46)
-(1.22)
(2.35)
-(0.11)
3.62
0.34
0.20
0.07
0.21
0.00
0.01
0.01
(2.81)
(13.38)
(6.67)
(1.45)
(3.72)
-(0.52)
(1.29)
(0.87)
3.32
0.18
0.16
0.10
0.19
-0.01
0.00
0.01
(3.15)
(8.96)
(6.47)
(2.41)
(4.14)
-(1.50)
-(0.33)
(1.68)
RSQ
0.56
0.65
0.54
Relative value
Group
No. Of Funds
% of Dead
Mean ER % pa
Small
942
59.45
8.38
5.04
1.67
Med
1153
59.58
5.57
5.02
1.11
Large
986
57.30
4.57
Std ER % pa Sharpe (pa) Alpha % pa
5.68
0.80
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
6.08
0.13
0.07
0.16
0.35
0.00
PTFSFX-RF PTFSCOM-RF
0.00
0.00
(7.17)
(7.71)
(3.74)
(4.96)
(9.21)
-(0.09)
(0.85)
(0.45)
3.23
0.12
0.06
0.11
0.38
-0.01
0.00
0.00
(4.15)
(8.05)
(3.10)
(3.54)
(10.83)
-(2.12)
(0.46)
(0.10)
2.05
0.11
0.03
0.12
0.46
-0.01
0.00
0.00
(2.22)
(6.04)
(1.22)
(3.30)
(11.09)
-(2.37)
-(0.94)
(0.26)
RSQ
0.60
0.66
0.62
Sector
Group
No. Of Funds
% of Dead
Mean ER % pa
Small
321
67.60
11.91
16.66
0.71
Med
361
60.94
12.68
14.01
0.90
Large
218
48.62
7.90
Std ER % pa Sharpe (pa) Alpha % pa
13.01
0.61
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
6.77
0.62
0.51
0.04
0.15
-0.01
PTFSFX-RF PTFSCOM-RF
0.00
0.00
(2.35)
(11.12)
(7.62)
(0.37)
(1.18)
-(0.58)
-(0.27)
(0.04)
8.51
0.63
0.45
0.01
0.01
0.01
0.00
0.00
(4.11)
(15.60)
(9.43)
(0.08)
(0.12)
(0.80)
(0.39)
-(0.10)
3.64
0.51
0.45
0.07
0.16
0.00
0.00
0.01
(1.78)
(12.81)
(9.53)
(0.85)
(1.73)
-(0.06)
(0.34)
(0.62)
RSQ
0.57
0.69
0.65
Short bias
Group
No. Of Funds
% of Dead
Mean ER % pa
Small
60
70.00
0.43
16.01
0.03
Med
74
70.27
6.58
17.43
0.38
Large
33
72.73
-3.53
Std ER % pa Sharpe (pa) Alpha % pa
14.70
-0.24
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
3.58
-0.72
-0.39
0.18
0.35
-0.01
PTFSFX-RF PTFSCOM-RF
-0.01
0.02
(1.27)
-(13.20)
-(5.96)
(1.62)
(2.79)
-(0.67)
-(0.66)
(0.90)
11.30
-0.64
-0.59
-0.10
0.04
-0.01
0.01
0.02
(3.59)
-(10.45)
-(8.09)
-(0.80)
(0.27)
-(0.32)
(0.61)
(0.77)
-0.73
-0.70
-0.45
0.18
0.49
-0.01
0.00
0.00
-(0.31)
-(15.19)
-(8.28)
(1.96)
(4.66)
-(0.37)
(0.36)
-(0.24)
RSQ
0.56
0.53
0.64
Table 6
Average performance of the aggregate database by domicile region [January 1994 - December 2010]
Table shows results of the average performance grouped by the domicile region. Panel A shows the number of
unique hedge funds divided by the manager firm domicile. Panel B provides results of the average performance
where funds are grouped based on the fund-level domicile region. Panel C shows the results of the average
performance where funds are grouped based on the firm domicile region. The domicile regions are divided to two
groups: (1) onshore; (2) offshore. United States and Canada are classified as onshore regions. Other domicile
regions are classified as offshore and they are further divided to four sub categories: (1) Asia and Pacific; (2)
Caribbean; (3) Europe; (4) Rest of World. In Panel A, "# of firms" is the number of unique management firms in
each firm domicile group. Numbers in Panel A are calculated using Hedge Fund Research, BarclayHedge, and
EurekaHedge databases, since only these databases report information on the domicile of the management firm.
In Panel A, within each firm domicile, the number of unique hedge funds are reported in each fund domicile group.
In Panel B, the second column describes the number of funds in each portfolio and Column "% of Dead" tells the
percentage amount of defunct funds if compared to all funds. The next two columns include the annualized
average excess return and the standard deviation of excess returns. Sharpe ratio is annualized and defined as the
average excess return divided by the standard deviation of return. Alpha is the measure of abnormal return
estimated from the seven factor model proposed by Fung and Hsieh (2004). Alpha is annualized and reported in
percentage. Risk factors are: the excess return of the S&P 500 index (SP-RF), the return of the Russell 2000 index
minus the return of the S&P 500 index (RL-SP), the excess return of ten-year Treasuries (TY-RF), the return of
Moody's BAA corporate bonds minus ten-year Treasuries (BAA-TY), the excess returns of look-back straddles on
bonds (PTFSBD-RF), currencies (PTFSFX-RF), and commodities (PTFSCOM-RF). RSQ is the R-square of the model.
Values of t-statistics are reported in parentheses. In Panel B, results are reported for the aggregate database
including TASS, Hedge Fund Research, BarclayHedge, EurekaHedge, and Morningstar. Panel C shows results of the
average performance grouped by (i) the management company domicile region; (ii) the firm city. BarclayHedge is
the only database that provides information on the city of the management company.
A. Number of hedge funds in management firm and fund domicile groups
Fund domicile (total number of funds = 15,805)
Firm domicile
# of firms
Onshore
Asia and Pacific
Caribbean
Europe
Rest of world
Onshore
Region
4229
5846
4
1923
101
854
Asia and Pacific
565
24
240
760
43
44
Caribbean
212
39
.
370
9
11
Europe
1317
54
5
2237
2274
251
Rest of world
325
38
.
268
36
374
Sum
6648
6001
249
5558
2463
1534
B. Average performance of the aggregate database by fund domicile (Equal-weight)
No. Of Funds
% Of Dead
Mean ER % pa
Std ER % pa
Sharpe (pa)
Alpha % pa
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
PTFSFX-RF
PTFSCOM-RF
RSQ
All
Domicile
24768
57.03
8.45
7.11
1.19
5.87
0.29
0.16
0.10
0.28
0.00
0.01
0.01
0.69
(5.83)
(14.32)
(6.56)
(2.48)
(6.01)
(0.04)
(2.76)
(1.88)
Onshore
8202
64.61
9.52
6.56
1.44
7.45
0.29
0.19
0.03
0.17
0.00
0.01
0.01
(8.68)
(16.87)
(9.03)
(0.89)
(4.26)
(0.86)
(3.47)
(2.31)
Offshore (all)
Caribbean
16566
9304
53.27
59.18
7.58
7.29
7.53
7.25
1.01
1.01
Asia and Pacific
730
29.59
11.19
10.47
1.07
Europe
4150
41.35
6.18
8.56
0.72
Rest of world
2382
58.23
9.49
7.99
1.19
4.74
0.28
0.14
0.13
0.34
0.00
0.01
0.01
(4.12)
(12.20)
(5.16)
(2.80)
(6.29)
-(0.62)
(2.31)
(1.55)
4.61
0.26
0.15
0.10
0.30
-0.01
0.01
0.01
(4.05)
(11.48)
(5.64)
(2.27)
(5.68)
-(1.21)
(1.94)
(1.27)
8.16
0.28
0.06
0.15
0.55
0.01
0.01
0.01
(3.99)
(6.69)
(1.31)
(1.89)
(5.79)
(0.81)
(1.11)
(0.97)
3.21
0.30
0.12
0.19
0.41
0.01
0.02
0.02
(2.22)
(10.26)
(3.43)
(3.33)
(6.10)
(0.80)
(2.71)
(1.81)
6.57
0.30
0.14
0.09
0.35
-0.01
0.01
0.01
(5.29)
(11.88)
(4.58)
(1.95)
(6.15)
-(0.77)
(2.09)
(1.11)
0.73
0.63
0.61
0.40
0.55
0.62
C. Average performance of the aggregate database by firm domicile (Equal-weight)
Firm domicile region (Aggregate database):
Domicile
All
Onshore
Offshore (all)
No. Of Funds
% Of Dead
Mean ER % pa
Std ER % pa
Sharpe (pa)
Alpha % pa
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
PTFSFX-RF
PTFSCOM-RF
RSQ
15935
58.17
8.80
6.97
1.26
6.33
0.29
0.16
0.09
0.26
0.00
0.01
0.01
0.69
(6.50)
(14.80)
(6.77)
(2.25)
(5.82)
(0.37)
(2.98)
(1.92)
8794
7141
67.05
47.25
8.84
8.12
6.08
8.39
1.45
0.97
Caribbean
435
62.53
11.05
8.59
1.28
Asia and Pacific
1124
46.44
8.43
10.03
0.84
Europe
4861
47.07
6.88
7.98
0.86
Rest of world
721
40.50
11.75
12.82
0.92
6.88
0.26
0.17
0.04
0.16
0.00
0.01
0.01
(8.29)
(15.85)
(8.42)
(1.27)
(4.24)
(0.58)
(3.45)
(2.07)
5.14
0.31
0.14
0.11
0.38
0.00
0.01
0.01
(3.84)
(11.46)
(4.35)
(2.16)
(6.06)
-(0.21)
(2.20)
(1.38)
8.57
0.28
0.13
0.06
0.25
-0.01
0.00
0.01
(5.21)
(8.34)
(3.37)
(0.96)
(3.32)
-(0.64)
(0.64)
(1.19)
5.67
0.36
0.16
0.07
0.32
0.01
0.01
0.01
(3.14)
(9.97)
(3.67)
(1.04)
(3.77)
(1.38)
(1.20)
(0.77)
4.07
0.28
0.13
0.15
0.38
0.00
0.02
0.01
(3.07)
(10.40)
(4.15)
(2.87)
(6.16)
(0.17)
(3.06)
(1.57)
0.71
0.60
0.42
0.49
0.57
7.44
0.42
0.16
0.02
0.51
-0.03
0.00
0.00
(3.43)
(9.64)
(3.11)
(0.24)
(5.03)
-(2.07)
-(0.16)
(0.31)
0.55
Alpha % pa
SP-RF
RL-SP
TY-RF
BAA-TY
PTFSBD-RF
PTFSFX-RF
PTFSCOM-RF
RSQ
0.78
Firm City (BarclayHedge):
Domicile
No. Of Funds
% Of Dead
Mean ER % pa
Std ER % pa
Sharpe (pa)
New York
1650
66.85
8.71
6.65
1.30
London
San Franciso
1435
197
59.30
70.56
8.78
10.24
8.30
8.87
1.06
1.15
6.33
0.30
0.19
0.02
0.16
-0.01
0.01
0.00
(8.09)
(19.04)
(10.28)
(0.61)
(4.51)
-(1.29)
(2.58)
(0.69)
6.16
0.27
0.12
0.08
0.35
0.00
0.02
0.01
(4.14)
(9.16)
(3.45)
(1.43)
(5.09)
-(0.11)
(2.22)
(1.07)
8.06
0.35
0.30
-0.06
0.10
0.00
0.01
0.01
(5.96)
(13.01)
(9.18)
-(1.16)
(1.53)
(0.52)
(0.83)
(0.73)
0.50
0.64
Table 7
Results of persistence for holding periods [January 1994 - December 2010]
Hedge funds are sorted into portfolios on (1) monthly; (2) quarterly; (3) Semiannually; and (4) december each year
based on their t-statistics of alphas estimated using the seven factor model of Fung and Hsieh (2004). Databases are
merged and cleaned from multiple share classes using a novel statistical procedure (see Appendix). The t-statistics of
alphas are estimated using the most recent 24 months of returns preceding the evaluation period. Portfolio returns
are calculated for equal- and value-weight portfolios monthly, so the weights are readjusted whenever a fund
disappears. Results of persistence are reported for bottom, top and spread portfolios. Table shows annualized alphas
in percentages and t-values in parentheses. The column "DO (%)" shows the drop out rate (in percentage) for each
portfolio describing the average number of funds that drop from each portfolio during the specified holding period.
For spread portfolios, the column "Do (%) (diff) contains the difference in drop out rates between the bottom and the
top portfolio. The positive number suggests the higher drop out rate for the bottom portfolio if compared to the top
portfolio. Risk factors are: the excess return of the S&P 500 index (SP-RF), the return of the Russell 2000 index minus
the return of the S&P 500 index (RL-SP), the excess return of ten-year Treasuries (TY-RF), the return of Moody's BAA
corporate bonds minus ten-year Treasuries (BAA-TY), the excess returns of look-back straddles on bonds (PTFSBD-RF),
currencies (PTFSFX-RF), and commodities (PTFSCOM-RF). Panel A shows the results of the equal-weight persistence
portfolios and Panel B shows the results of the value-weight portfolios. Panel C shows the results of the equal-weight
persistence portfolios for the aggregate data base in the size terciles. First, funds are sorted into size terciles based on
monthly AUM observations. Second, within size terciles, funds are sorted into deciles based on t-statistics of alphas.
A. Equal-weight
Bottom
1-month
Database
TASS
3-month
DO (%)
Alpha % pa
DO (%)
Alpha % pa
DO (%)
Alpha % pa
DO (%)
1.91
3.99
2.17
10.97
2.07
19.12
2.79
31.87
(1.38)
1.43
3.81
(0.98)
BarclayHedge
2.78
3.68
6.87
3.15
3.99
5.54
2.71
10.15
3.75
8.41
8.74
(0.90)
3.52
17.78
7.72
7.96
15.54
(1.44)
1.85
4.49
29.08
9.54
27.03
(5.21)
(3.80)
10.44
30.45
(3.37)
15.45
6.40
4.17
(3.09)
(4.16)
7.17
2.14
(1.78)
18.45
(2.51)
(3.97)
1.23
2.84
(1.98)
(4.41)
(3.38)
Aggregate
(1.36)
10.70
(2.78)
(3.66)
Morningstar
2.83
(1.84)
(2.14)
EurekaHedge
12-month
Alpha % pa
(1.26)
Hedge Fund Research
6-month
7.58
27.02
(4.71)
18.49
(1.32)
3.05
31.13
(2.24)
Top
1-month
Database
TASS
3-month
Alpha % pa
DO (%)
6.19
1.38
(6.22)
Hedge Fund Research
7.27
1.54
7.12
1.49
1.94
3.96
6.88
4.38
7.13
2.07
7.03
1.58
7.08
5.67
8.18
6.40
8.76
8.59
4.48
16.60
5.55
16.66
5.82
16.47
5.74
20.07
(4.17)
11.96
(7.14)
(7.31)
5.22
(5.46)
10.55
6.42
DO (%)
(6.47)
6.94
7.20
Alpha % pa
(5.20)
(5.30)
6.09
(8.08)
(7.15)
DO (%)
(6.83)
5.43
7.02
Alpha % pa
(7.31)
4.31
7.55
12-month
(5.58)
(6.22)
7.18
(7.91)
Aggregate
6.00
(7.36)
7.56
(5.99)
Morningstar
DO (%)
(7.83)
(7.27)
EurekaHedge
Alpha % pa
(6.21)
(7.89)
BarclayHedge
6-month
5.57
22.83
(6.13)
6.84
9.06
(6.74)
5.89
17.51
(5.84)
Spread
1-month
Database
TASS
Alpha % pa
4.28
DO (%) (diff)
2.61
(2.78)
Hedge Fund Research
5.84
4.34
0.69
1.64
1.21
5.80
(4.10)
7.01
4.04
3.14
-0.86
0.64
-0.15
4.93
(3.27)
10.94
3.56
3.42
-0.52
0.02
4.90
4.99
(3.48)
13.80
1.33
12.61
-3.81
6.97
-(1.97)
3.58
(0.01)
5.96
1.38
(0.84)
-(0.27)
1.87
15.27
(0.93)
9.19
(2.27)
3.31
2.43
(1.51)
9.69
(2.32)
5.84
-(0.08)
2.16
3.60
(2.29)
6.32
-(0.45)
(0.94)
Aggregate
3.83
(2.01)
(0.35)
Morningstar
12-month
Alpha % pa DO (%) (diff)
(2.47)
2.19
(2.99)
EurekaHedge
6-month
Alpha % pa DO (%) (diff)
(2.44)
2.27
(3.62)
BarclayHedge
3-month
Alpha % pa DO (%) (diff)
-2.01
4.19
-(1.18)
9.44
2.84
(1.97)
13.63
B. Value-weight
Bottom
1-month
Database
TASS
3-month
DO (%)
Alpha % pa
DO (%)
Alpha % pa
DO (%)
Alpha % pa
DO (%)
4.52
6.11
5.61
13.68
6.55
21.95
4.48
34.91
(2.65)
2.41
3.97
(1.62)
BarclayHedge
4.02
3.95
6.02
4.43
5.41
10.81
8.11
8.10
4.19
4.73
10.17
(1.01)
2.99
5.76
18.80
6.88
18.33
(1.88)
4.66
4.85
30.52
7.42
35.10
(4.30)
(3.68)
12.25
31.48
(3.15)
21.65
7.02
5.88
(4.24)
(4.01)
(4.00)
1.60
(2.00)
19.56
(3.47)
12.47
8.15
4.20
(2.77)
(4.55)
(3.90)
Aggregate
(2.75)
11.24
(3.20)
(3.43)
Morningstar
3.82
(2.72)
(2.50)
EurekaHedge
12-month
Alpha % pa
(2.01)
Hedge Fund Research
6-month
7.52
29.64
(4.13)
21.04
(2.68)
5.02
33.94
(3.38)
Top
1-month
Database
TASS
3-month
Alpha % pa
DO (%)
5.84
4.28
(5.40)
Hedge Fund Research
6.01
1.74
5.94
1.73
6.12
5.87
6.21
6.80
5.68
4.97
5.89
2.60
5.68
4.73
12.37
5.49
9.83
9.24
5.93
5.49
19.54
4.54
18.51
4.94
17.37
4.16
27.90
(3.34)
14.35
(5.56)
(6.05)
4.40
(4.51)
15.20
5.41
DO (%)
(5.37)
6.37
4.78
Alpha % pa
(4.54)
(3.74)
7.74
(6.23)
(6.33)
DO (%)
(6.39)
7.86
5.81
Alpha % pa
(6.97)
4.63
5.64
12-month
(4.69)
(4.57)
3.28
(6.31)
Aggregate
4.96
(5.74)
2.87
(4.82)
Morningstar
DO (%)
(6.62)
(5.41)
EurekaHedge
Alpha % pa
(4.74)
(6.59)
BarclayHedge
6-month
4.92
25.42
(4.47)
11.33
(6.01)
4.48
19.82
(4.90)
Spread
Database
TASS
1-month
3-month
6-month
12-month
Alpha % pa DO (%) (diff)
Alpha % pa DO (%) (diff)
Alpha % pa DO (%) (diff)
Alpha % pa DO (%) (diff)
1.32
1.83
(0.59)
Hedge Fund Research
3.61
1.92
0.10
-2.23
1.55
4.62
(2.89)
-2.47
0.92
-2.34
2.69
(1.65)
1.29
0.61
4.60
-2.10
-1.61
0.83
(0.48)
-1.34
12.96
0.09
13.16
(0.05)
6.45
-3.27
7.20
-(1.66)
3.99
-(0.84)
6.32
15.37
-(0.91)
9.56
-(1.07)
2.43
-0.08
-(0.03)
9.73
(0.34)
-(1.11)
2.13
9.58
(0.85)
6.17
-(1.24)
-(1.09)
Aggregate
0.48
-1.81
-(0.75)
6.27
(0.26)
(0.05)
Morningstar
1.87
(1.28)
2.22
(1.07)
EurekaHedge
6.88
-(0.30)
2.23
(2.29)
BarclayHedge
-0.65
-2.59
4.22
-(1.35)
9.71
-0.54
-(0.34)
14.12
C. Equal-weight persistence of the aggregate data in size groups
Bottom
1-month
Size group
Small
3-month
DO (%)
Alpha % pa
DO (%)
Alpha % pa
DO (%)
Alpha % pa
DO (%)
1.36
5.09
2.04
14.27
1.72
24.97
2.37
40.41
(1.23)
1.12
3.24
(0.77)
Large
(1.04)
2.41
9.44
(1.50)
-0.09
2.52
-(0.06)
All
12-month
Alpha % pa
(0.90)
Med
6-month
1.24
3.70
7.32
2.31
(1.45)
16.99
(1.43)
1.55
(1.01)
(0.91)
2.30
1.70
10.34
1.98
29.98
(2.87)
12.83
(1.16)
(1.55)
4.27
2.13
22.88
(1.51)
18.31
(1.40)
3.15
30.79
(2.28)
Top
1-month
Size group
Small
3-month
DO (%)
Alpha % pa
DO (%)
Alpha % pa
DO (%)
Alpha % pa
DO (%)
7.14
2.15
7.98
6.04
9.13
11.76
9.37
22.31
7.25
(5.76)
1.49
(6.84)
Large
5.97
1.52
6.97
7.16
(5.82)
4.40
(6.83)
(6.54)
All
12-month
Alpha % pa
(5.58)
Med
6-month
5.77
1.52
4.41
7.03
8.93
(6.32)
(6.15)
(7.07)
(5.98)
6.63
4.31
18.72
(5.06)
5.04
8.45
(5.34)
(7.20)
5.38
3.80
16.61
(3.82)
6.73
8.66
(6.61)
5.82
16.75
(5.76)
Spread
1-month
Size group
Small
3-month
Alpha % pa DO (%) (diff)
5.79
2.94
(3.21)
Med
6.12
6.06
5.73
(4.00)
8.23
4.75
1.00
4.22
4.72
(3.09)
Alpha % pa DO (%) (diff)
Alpha % pa DO (%) (diff)
7.41
13.21
4.33
2.91
3.34
4.75
(3.29)
18.11
1.11
11.26
(0.67)
4.38
(2.21)
6.03
7.00
(3.37)
8.06
(2.60)
(2.68)
2.18
12-month
(3.49)
5.04
(2.83)
(4.03)
All
5.94
(3.01)
1.75
(3.81)
Large
Alpha % pa DO (%) (diff)
6-month
1.67
6.27
(1.08)
9.65
2.67
(1.84)
14.04
Table 8
Feasible portfolio rebalancing and performance persistence [January 1994 - December
2010]
Table shows the annualized Fung and Hsieh (2004) alphas for portfolios based on (1) monthly; (2);
quarterly; (3) semiannually; and (4) yearly rebalancing horizons. Hedge funds are sorted into
portfolios based on their t-statistics of alphas estimated using the seven factor model proposed by
Fung and Hsieh (2004). The t-statistics of alphas are estimated using the most recent 24 months of
returns preceding the evaluation period. Portfolio returns are calculated for equal-weight portfolios
monthly, so the weights are readjusted whenever a fund disappears. Within each rebalancing horizon
decile portfolios are formed using only the feasible information taking into account fund-specific
share restrictions. For instance, for the feasible semiannual (annual) rebalancing strategy, we exclude
the funds that have redemption and lockup periods longer than 6 (12) months. Also, for semiannual
(annual) rebalancing strategies, we exclude funds having notice periods longer than 6 (12) months.
For the 1-month rebalancing strategy, we exclude funds having notice period longer than 1-month.
This implies that we use at least 1-month lagged information to estimate persistence in order to
mitigate look-ahead bias. Notice period determines the lag length. The annualized alphas of feasible
rebalancing strategies are reported in Column "Feasible". Columm "Baseline" shows the results of
persistence without restrictions for lockup and redemption periods. Results of persistence are
reported for bottom, top and spread portfolios. Databases are merged and cleaned from multiple
share classes using a novel statistical procedure that is described in Appendix of the paper.
Bottom
Redemption 1 m
Redemption 3 m
Redemption 6 m
Redemption 12 m
Lockup 1 m
Lockup 3 m
Lockup 6 m
Lockup 12 m
Feasible
Baseline
Feasible
1m
Notice 3 m
Notice 6 m
Notice 12 m
Top
Notice 6 m
Notice 12 m
Spread
Alpha % pa
Notice 6 m
Notice 12 m
12m
Alpha % pa
Alpha % pa
2.14
1.53
1.85
2.85
3.05
(0.56)
(0.90)
(0.44)
(1.44)
(0.98)
(1.32)
(1.86)
(2.24)
.
.
2.40
.
1.98
.
3.05
.
.
.
(1.51)
.
(1.15)
.
(1.60)
.
.
.
.
.
3.44
.
3.31
.
.
.
.
.
(2.19)
.
(2.06)
.
.
.
.
.
.
.
4.64
.
.
.
.
.
.
.
(2.97)
.
Baseline
Feasible
Baseline
Feasible
Baseline
Feasible
Feasible
3m
6m
Alpha % pa
Baseline
12m
Alpha % pa
Alpha % pa
5.44
7.03
5.43
7.08
4.83
6.84
4.80
5.89
(4.21)
(7.15)
(4.27)
(7.31)
(3.70)
(6.74)
(4.10)
(5.84)
.
.
5.95
.
5.12
.
5.01
.
.
.
(5.68)
.
(5.09)
.
(5.25)
.
.
.
.
.
4.36
.
3.41
.
.
.
.
.
(4.23)
.
(3.35)
.
.
.
.
.
.
.
3.68
.
.
.
.
.
.
.
(3.53)
.
Baseline
Feasible
Baseline
Feasible
Baseline
Feasible
Feasible
3m
Alpha % pa
Notice 3 m
6m
Baseline
0.70
1m
Notice 1 m
Feasible
1.23
Alpha % pa
Notice 3 m
Baseline
0.89
1m
Notice 1 m
Feasible
3m
Alpha % pa
Notice 1 m
Baseline
6m
Alpha % pa
Baseline
12m
Alpha % pa
Alpha % pa
4.55
5.80
4.73
4.93
3.31
4.99
1.95
2.84
(2.61)
(4.10)
(2.75)
(3.27)
(1.93)
(3.48)
(1.18)
(1.97)
.
.
3.55
.
3.14
.
1.96
.
.
.
(2.28)
.
(1.85)
.
(1.03)
.
.
.
.
.
0.92
.
0.10
.
.
.
.
.
(0.54)
.
(0.06)
.
.
.
.
.
.
.
-0.96
.
.
.
.
.
.
.
(-0.62)
.
Table 9
Summary statistics of characteristics and performance results of univariate sorts
Table describes statistics of fund-level characteristics and the results of performance based on univariate sorts. Results are reported for each database. Aggregate database is a merged
database of single databases including TASS, Hedge Fund Research, BarclayHedge, EurekaHedge, Morningstar. Databases are merged using a novel statistical algorithm that is
described in Appendix of the paper. All hedge funds are required to have at least 12 monthly non-missing returns. Single databases are removed from multiple share classes using the
same statistical procedure. In Panel A, all values of characteristics are based on hedge fund database snapshot variables that include one observation for each fund. Table contains
variables of fees including management and incentive fee (in percentages) and high-water mark binary variable (one for funds having high-water mark and zero otherwise). Second,
table includes variables of share restrictions (reported in years) including lockup period, advance notification period and redemption frequency. Columns include a total number of
hedge funds in question (# of Funds) and a number of hedge funds that have missing values for each variable (Missing %). Statistics include cross-sectional mean and standard
deviation. In Panel B, hedge funds are sorted into five portfolios on December each year based on one of the time-varying variables (age, size (AuM), or Flow). Equal-weight and valueweight portfolio returns are calculated for each month, so the weights are readjusted whenever a fund disapprears. Also, hedge funds are sorted into portfolios based on timeinvariant characteristics including incentive fee, lockup period, notification period, and redemption period. Within each group, hedge funds in the aggregate data are sorted into
terciles using monthly AuM observations. In Panel B, annualized Fung and Hsieh (2004) alphas and t-values of alphas are reported for each portfolio.
A. Summary statistics
Incentive Fee
Database
TASS
Lockup period (in years)
# of Funds
Missing (%)
Mean
Std
Database
8072
3.47
0.17
0.07
TASS
# of Funds
Missing (%)
Mean
Std
8072
0.00
0.24
0.53
Hedge Fund Research
9508
1.89
0.19
0.05
Hedge Fund Research
9508
7.43
0.29
0.55
BarclayHedge
9588
0.01
0.19
0.06
BarclayHedge
9588
17.48
0.27
0.54
EurekaHedge
7009
1.95
0.19
0.05
EurekaHedge
7009
2.18
0.19
0.48
Morningstar
6681
12.72
0.19
0.05
Morningstar
6681
46.79
0.45
0.62
Aggregate
24768
5.20
0.18
0.06
Aggregate
24768
17.10
0.25
0.53
Mean
Std
Management Fee
Database
# of Funds
Missing (%)
Notification period (in years)
Mean
Std
Database
# of Funds
Missing (%)
TASS
8072
0.94
0.02
0.01
TASS
8072
0.00
0.08
0.09
Hedge Fund Research
9508
1.31
0.02
0.01
Hedge Fund Research
9508
5.55
0.10
0.08
BarclayHedge
9588
0.01
0.02
0.01
BarclayHedge
9588
0.01
0.07
0.10
EurekaHedge
7009
1.73
0.02
0.01
EurekaHedge
7009
6.15
0.10
0.09
Morningstar
6681
13.29
0.02
0.01
Morningstar
6681
39.60
0.10
0.08
Aggregate
24768
3.66
0.02
0.01
Aggregate
24768
10.57
0.08
0.09
Database
# of Funds
Mean
Std
Database
# of Funds
Mean
Std
High-Water mark (1=yes, 0=no)
Missing (%)
Redemption frequency (in years)
Missing (%)
TASS
8072
0.94
0.63
0.48
TASS
8072
9.32
0.17
0.19
Hedge Fund Research
9508
0.00
0.88
0.32
Hedge Fund Research
9508
2.77
0.17
0.22
BarclayHedge
9588
0.67
0.61
0.49
BarclayHedge
9588
24.52
0.19
0.22
EurekaHedge
7009
0.49
0.87
0.34
EurekaHedge
7009
2.01
0.13
0.17
Morningstar
6681
25.25
0.86
0.35
Morningstar
6681
33.59
0.18
0.21
Aggregate
24768
5.98
0.70
0.46
Aggregate
24768
19.02
0.16
0.21
B. Univariate sorts
Fund age
Portfolio
TASS
EW
Young
Mature
Spread
HFR
VW
EW
Barclay
VW
EW
Eureka
VW
EW
Morningstar
VW
EW
VW
Aggregate
EW
VW
Small
Med
Large
7.63
9.82
9.39
9.92
9.58
6.16
9.63
11.72
9.45
9.78
8.23
7.45
9.77
6.50
5.85
(6.17)
(6.75)
(8.24)
(7.14)
(8.75)
(3.71)
(5.38)
(5.02)
(7.82)
(5.85)
(7.31)
(5.34)
(8.11)
(5.56)
(4.94)
5.28
6.92
5.47
4.96
6.79
6.10
6.44
4.16
8.61
5.99
5.49
5.19
8.76
5.95
6.09
(4.42)
(4.73)
(5.53)
(4.74)
(4.21)
(6.15)
(4.64)
(3.74)
(2.70)
(6.57)
(4.46)
(4.85)
(6.97)
(5.38)
(6.14)
4.79
5.90
4.84
6.12
5.53
4.94
8.98
7.52
7.51
7.70
4.99
6.17
6.48
6.27
5.59
(4.46)
(5.55)
(4.64)
(4.89)
(5.45)
(4.98)
(6.09)
(4.89)
(6.62)
(7.17)
(4.77)
(5.53)
(5.22)
(5.36)
(5.08)
4.47
7.40
5.03
5.03
4.96
5.38
7.25
6.64
7.54
6.95
4.39
5.30
4.96
6.39
4.90
(4.15)
(4.12)
(4.98)
(4.12)
(4.78)
(4.82)
(5.37)
(4.85)
(6.98)
(5.93)
(4.19)
(3.97)
(3.70)
(5.88)
(4.36)
5.52
4.41
5.52
5.22
(4.61)
(2.86)
(4.67)
(5.02)
4.16
5.97
4.70
6.21
4.13
5.78
6.55
6.16
6.21
7.48
3.96
(3.65)
(3.55)
(4.63)
(5.74)
(3.46)
(4.97)
(4.49)
(4.67)
(5.13)
(5.68)
(3.63)
3.46
3.85
4.69
3.71
5.45
0.38
3.08
5.56
3.24
2.31
4.27
1.92
5.36
0.98
0.63
(4.12)
(3.09)
(5.98)
(3.71)
(6.09)
(0.29)
(2.20)
(3.02)
(3.52)
(1.60)
(5.65)
(1.82)
(3.72)
(0.86)
(0.81)
Fund size
Portfolio
Small
Medium
Large
Spread
TASS
HFR
Barclay
Eureka
Morningstar
Aggregate
EW
VW
EW
VW
EW
VW
EW
VW
EW
VW
EW
VW
Small
Med
Large
7.32
12.60
9.19
13.33
9.35
10.67
10.64
12.52
11.64
13.43
8.28
11.22
11.84
12.77
9.99
(5.80)
(9.16)
(8.15)
(7.91)
(7.99)
(8.00)
(7.22)
(7.89)
(9.56)
(9.47)
(7.21)
(8.20)
(9.32)
(10.38)
(8.67)
11.38
8.38
8.13
7.33
(8.49)
(6.41)
(7.75)
(6.46)
7.22
12.46
7.33
11.53
7.94
12.51
11.21
15.01
9.56
13.66
7.14
(6.31)
(8.85)
(6.88)
(8.83)
(6.73)
(8.61)
(8.03)
(9.87)
(7.74)
(9.37)
(6.42)
5.22
9.29
6.08
9.10
6.74
9.62
5.95
10.68
7.80
10.18
5.57
9.02
6.54
6.12
5.33
(4.78)
(7.54)
(5.98)
(8.07)
(6.68)
(8.82)
(3.48)
(6.27)
(7.24)
(8.89)
(5.35)
(7.68)
(4.83)
(5.60)
(4.98)
3.99
7.66
4.19
7.27
5.01
7.37
6.44
8.66
5.90
9.06
4.10
7.30
4.70
3.67
3.70
(6.26)
(3.58)
(6.41)
(4.01)
(6.54)
(4.80)
(6.68)
(4.24)
(6.33)
(4.99)
(7.71)
(3.75)
(3.92)
(3.27)
(3.56)
3.30
6.28
3.13
5.41
3.04
4.87
5.20
5.81
4.53
6.48
2.88
5.12
-0.43
-0.49
3.03
(2.93)
(4.20)
(2.93)
(5.00)
(2.80)
(4.63)
(3.62)
(4.58)
(3.99)
(5.80)
(2.67)
(4.49)
-(0.34)
-(0.41)
(2.69)
4.02
6.31
6.07
7.92
6.31
5.80
5.43
6.70
7.11
6.96
5.39
6.11
12.28
13.27
6.96
(4.17)
(5.09)
(7.83)
(6.09)
(7.20)
(5.59)
(4.10)
(4.57)
(8.94)
(6.28)
(6.82)
(6.27)
(9.89)
(11.57)
(8.88)
Fund Flow
TASS
Portfolio
Low
Great
Spread
HFR
Barclay
Eureka
Morningstar
Aggregate
EW
VW
EW
VW
EW
VW
EW
VW
EW
VW
EW
VW
Small
Med
Large
5.78
8.80
6.65
5.71
6.32
5.37
8.20
6.36
8.42
5.45
6.06
6.41
8.98
7.94
7.04
(4.94)
(3.77)
(5.65)
(4.09)
(5.28)
(4.44)
(4.77)
(3.45)
(6.50)
(3.91)
(5.25)
(4.75)
(5.81)
(6.87)
(5.67)
4.73
6.27
5.10
5.56
6.34
5.93
7.50
6.52
7.52
5.77
5.24
5.59
7.71
6.39
5.22
(4.52)
(4.53)
(5.64)
(5.32)
(6.11)
(5.20)
(4.81)
(4.48)
(6.50)
(4.92)
(5.20)
(4.60)
(6.23)
(6.10)
(4.88)
5.62
5.27
6.23
5.30
(4.72)
(5.13)
(6.25)
(5.33)
4.56
5.99
5.61
5.66
6.08
5.61
7.77
7.50
7.32
8.71
4.86
(4.29)
(4.94)
(5.65)
(5.34)
(6.82)
(4.12)
(6.57)
(4.74)
(6.88)
(7.24)
(5.06)
5.44
7.50
5.88
6.67
6.55
5.92
7.75
7.58
7.83
8.04
5.53
5.85
5.93
6.13
5.67
(4.75)
(5.23)
(5.73)
(5.39)
(5.92)
(5.44)
(5.75)
(5.46)
(6.69)
(6.51)
(5.23)
(4.88)
(4.99)
(5.75)
(5.30)
6.45
5.27
6.51
6.30
6.34
4.88
8.06
5.68
8.41
7.82
6.08
5.17
7.69
4.32
4.70
(4.37)
(5.13)
(3.44)
(5.42)
(5.10)
(4.68)
(3.83)
(5.10)
(3.79)
(6.86)
(5.65)
(4.73)
(5.33)
(3.19)
(4.03)
-0.67
3.53
0.14
-0.59
-0.02
0.48
0.14
0.68
0.00
-2.37
-0.02
1.24
1.29
3.62
2.35
-(0.86)
(1.79)
(0.25)
-(0.63)
-(0.02)
(0.44)
(0.12)
(0.38)
(0.00)
-(1.95)
-(0.03)
(1.28)
(1.14)
(3.93)
(2.80)
Incentive Fee (IF)
Portfolio
IF = 0
0 < IF< 20
IF = 20
IF > 20
Spread
TASS
HFR
Barclay
Eureka
Morningstar
Aggregate
EW
VW
EW
VW
EW
VW
EW
VW
EW
VW
EW
VW
Small
Med
Large
2.46
4.81
3.69
3.36
3.55
3.28
5.16
5.63
4.48
3.53
3.26
3.60
0.69
3.50
4.80
(1.97)
(2.68)
(3.58)
(4.00)
(3.28)
(2.48)
(2.95)
(2.45)
(2.82)
(2.54)
(2.79)
(2.90)
(0.47)
(3.11)
(4.25)
4.21
6.56
5.38
5.49
5.41
4.20
5.96
8.99
5.62
5.98
4.52
4.88
3.46
4.44
7.79
(3.37)
(4.20)
(4.25)
(4.39)
(4.25)
(3.51)
(3.71)
(4.13)
(4.53)
(3.90)
(3.54)
(3.96)
(2.63)
(3.23)
(5.72)
6.79
7.06
7.25
6.67
7.87
6.08
8.85
6.31
8.70
6.99
7.29
6.18
7.31
7.95
8.13
(6.81)
(5.33)
(7.67)
(5.89)
(8.51)
(5.93)
(8.35)
(5.49)
(8.93)
(6.64)
(7.59)
(5.65)
(7.04)
(8.19)
(8.33)
8.72
5.96
10.36
7.52
12.30
8.80
13.64
4.94
12.20
9.65
11.47
7.27
13.97
12.40
10.30
(5.05)
(5.98)
(7.43)
(4.89)
(7.67)
(5.62)
(6.17)
(3.35)
(6.35)
(5.50)
(8.13)
(5.05)
(8.34)
(7.92)
(6.10)
6.26
1.16
6.66
4.16
8.74
5.52
8.49
-0.70
7.73
6.12
8.22
3.67
13.28
8.90
5.50
(4.44)
(0.70)
(5.30)
(2.73)
(6.21)
(3.39)
(3.43)
-(0.30)
(3.57)
(3.54)
(6.90)
(2.77)
(7.54)
(6.38)
(3.86)
Lockup Period (LP)
Portfolio
TASS
EW
LP= 0
0 < LP < 1
LP = 1
LP > 1
Spread
HFR
VW
EW
Barclay
VW
EW
Eureka
VW
EW
Morningstar
VW
EW
VW
Aggregate
EW
VW
Small
Med
Large
5.47
6.06
6.38
5.62
7.44
6.09
8.26
6.36
8.03
5.81
6.23
5.55
6.34
6.77
7.65
(4.98)
(4.66)
(6.42)
(5.77)
(7.38)
(5.81)
(6.95)
(4.97)
(7.99)
(5.43)
(5.91)
(5.18)
(5.89)
(6.31)
(7.27)
7.79
8.68
8.54
5.18
9.33
6.67
11.26
6.97
9.73
6.66
9.03
7.18
8.34
10.09
8.07
(7.48)
(4.57)
(9.77)
(4.30)
(8.38)
(3.01)
(9.11)
(4.38)
(8.41)
(5.01)
(8.84)
(4.15)
(5.51)
(8.84)
(6.62)
8.65
7.07
8.08
7.82
8.55
6.66
8.93
7.45
8.66
7.11
8.07
6.65
6.49
8.67
8.97
(8.66)
(4.08)
(7.31)
(5.77)
(8.76)
(5.17)
(8.65)
(7.29)
(8.68)
(6.19)
(8.05)
(4.65)
(5.20)
(8.36)
(8.71)
11.88
10.05
12.11
16.36
10.23
5.92
10.92
8.19
11.58
8.89
11.24
10.89
11.56
11.93
11.22
(8.24)
(5.94)
(8.29)
(8.62)
(6.20)
(2.84)
(7.19)
(4.61)
(7.18)
(5.69)
(8.12)
(5.83)
(3.83)
(6.34)
(6.88)
6.41
4.00
5.73
10.74
2.80
-0.17
2.65
1.82
3.55
3.08
5.01
5.33
5.11
5.16
3.57
(6.80)
(2.82)
(5.09)
(6.16)
(2.26)
-(0.09)
(2.36)
(1.10)
(3.04)
(2.20)
(5.16)
(3.09)
(1.86)
(3.45)
(2.92)
Med
Large
Notice Period (NP)
TASS
Portfolio
EW
NP < 1
1 NP < 30
NP = 30
30 < NP < 60
NP = 60
NP > 60
Spread
HFR
VW
EW
Barclay
VW
EW
Eureka
VW
EW
Morningstar
VW
EW
VW
Aggregate
EW
VW
Small
4.13
4.15
5.88
5.26
7.88
5.56
7.36
5.88
9.03
9.16
6.57
5.46
8.88
7.55
6.99
(3.04)
(2.50)
(4.80)
(2.64)
(6.28)
(4.41)
(4.28)
(1.60)
(4.16)
(3.88)
(5.25)
(4.19)
(6.86)
(5.58)
(5.16)
5.59
6.58
6.82
6.11
6.00
6.24
7.98
6.05
8.65
4.80
6.11
5.99
5.10
6.33
8.16
(4.06)
(3.11)
(5.92)
(5.80)
(4.81)
(4.38)
(5.23)
(3.61)
(6.29)
(3.83)
(5.10)
(4.25)
(3.84)
(4.94)
(6.63)
7.00
6.34
6.37
5.91
7.20
5.02
8.68
5.98
7.70
6.09
6.73
5.26
6.93
7.51
7.46
(6.49)
(4.84)
(6.30)
(4.81)
(6.94)
(3.87)
(7.97)
(4.63)
(7.43)
(5.64)
(6.55)
(4.04)
(6.01)
(7.32)
(6.78)
7.59
7.48
7.69
6.11
8.82
6.13
8.78
5.85
7.96
6.93
7.72
6.12
8.08
8.95
7.60
(7.37)
(6.27)
(6.99)
(4.56)
(8.50)
(5.85)
(7.81)
(4.40)
(8.20)
(7.80)
(7.09)
(5.14)
(4.90)
(7.31)
(6.41)
8.09
8.23
7.54
6.19
8.61
5.69
9.85
8.12
8.43
7.51
7.76
5.70
6.42
7.26
9.09
(8.08)
(5.61)
(7.34)
(5.02)
(8.72)
(5.01)
(7.48)
(4.74)
(8.36)
(5.81)
(7.77)
(5.48)
(5.43)
(6.80)
(8.09)
8.44
8.18
8.89
10.45
8.57
7.13
9.16
9.90
8.49
6.55
8.42
8.54
8.90
9.17
8.86
(8.32)
(5.20)
(9.07)
(4.63)
(8.46)
(5.39)
(9.48)
(5.51)
(8.33)
(7.55)
(8.80)
(5.62)
(5.69)
(8.84)
(8.60)
4.30
4.03
3.01
5.19
0.69
1.57
1.80
4.03
-0.54
-2.61
1.85
3.08
0.02
1.62
1.87
(4.66)
(2.27)
(2.95)
(2.04)
(0.51)
(1.05)
(1.24)
(1.18)
-(0.27)
-(1.15)
(1.71)
(1.92)
(0.01)
(1.23)
(1.50)
Redemption Period (RP)
TASS
Portfolio
EW
RP 1
1 < RP 3
RP > 3
Spread
HFR
VW
EW
Barclay
VW
EW
Eureka
VW
EW
Morningstar
VW
EW
VW
Aggregate
EW
VW
Small
Med
Large
5.23
5.07
6.40
5.32
6.27
4.70
8.14
6.46
8.41
5.75
5.84
4.65
6.00
6.40
6.95
(4.60)
(3.87)
(6.14)
(5.20)
(5.70)
(4.15)
(6.26)
(4.42)
(7.52)
(5.03)
(5.29)
(4.14)
(5.27)
(5.53)
(6.53)
7.24
8.24
7.58
7.77
8.31
6.52
9.07
6.41
8.18
7.26
7.64
7.63
6.40
8.22
8.87
(7.33)
(5.24)
(7.76)
(5.79)
(8.58)
(6.09)
(8.58)
(5.71)
(8.41)
(6.54)
(7.76)
(5.73)
(5.92)
(8.37)
(8.00)
8.02
7.63
7.65
7.79
7.87
6.13
9.02
9.45
7.89
8.32
7.62
6.38
7.32
7.37
8.43
(8.24)
(6.11)
(7.83)
(5.37)
(8.77)
(4.80)
(10.14)
(7.14)
(9.14)
(5.27)
(8.38)
(5.47)
(5.68)
(6.93)
(8.80)
2.80
2.55
1.26
2.47
1.61
1.44
0.88
3.00
-0.52
2.57
1.78
1.73
1.32
0.97
1.48
(2.90)
(2.12)
(1.50)
(2.00)
(2.08)
(1.21)
(0.77)
(1.95)
-(0.55)
(1.71)
(2.13)
(1.60)
(1.08)
(1.02)
(1.80)
Table 10
Cross-sectional regressions [January 1994 - December 2010]
Table shows results of the cross-sectional regressions based on Fama-McBeth (1973) procedure. Results are reported for each database. Aggregate database is a merged
database of single databases including TASS, Hedge Fund Research (HFR), BarclayHedge, EurekaHedge, and Morningstar. Databases are merged using a novel statistical
procedure that is described in Appendix of the paper. Single databases are removed from multiple share classes using the same statistical procedure. In panel A, excess returns
are regressed against fund-level characteristics including fund size (AUM in a log scale), age, flow, compensation structure and share restrictions (measured in years). Highwater mark is a binary variable that equals one for funds that apply high-water mark and zero otherwise. Fund-level size, age, and flow are lagged one month. Age is measured
as a length of the return time series. In panel B, fund-level alphas are regressed against fund-level characteristics. Alpha is calculated as a sum of the intercept and the time
series of the residual estimated from the Fund and Hsieh (2004) model. In Panel C, hedge funds are sorted into size terciles based on the AUM observations each December and
fund-level excess returns are regressed against fund-level characteristics within size terciles. In panel D, fund-level alphas are regressed against characteristics within size
terciles. Time period is from January 1994 - December 2010. All parameter estimates are multiplied by 100.
A. Excess returns
Variable
Intercept
Sizet
1
Aget
1
Flowt
1
Management Fee
Incentive Fee
High-Water Mark
Lockup period
Notice period
Redemption
TASS
1.13
(3.20)
-0.06
-(3.57)
-0.13
-(3.19)
1.62
(5.19)
10.06
(2.97)
0.95
(2.60)
0.15
(3.64)
0.05
(1.11)
1.22
(4.49)
0.13
(1.58)
HFR
1.83
(3.41)
-0.07
-(4.70)
-0.17
-(3.38)
1.62
(5.80)
8.10
(2.54)
0.68
(2.00)
0.04
(1.35)
0.09
(2.67)
0.23
(0.78)
0.06
(0.95)
BarclayHedge EurekaHedge Morningstar
1.45
1.25
1.64
(4.30)
(2.35)
(3.99)
-0.10
-0.03
-0.12
-(5.46)
-(1.42)
-(6.58)
-0.07
-0.14
-0.16
-(1.39)
-(1.42)
-(3.65)
1.28
1.74
1.60
(4.36)
(2.69)
(4.19)
12.59
15.42
5.25
(3.81)
(2.78)
(0.80)
0.56
0.70
1.38
(1.55)
(0.90)
(3.35)
0.12
-0.18
-0.01
(2.45)
-(0.88)
-(0.08)
0.09
0.10
0.02
(2.30)
(1.25)
(0.43)
0.00
0.49
0.70
-(0.02)
(1.06)
(2.00)
0.06
0.11
0.08
(0.71)
(0.86)
(0.96)
B. Fung and Hsieh (2004) alphas
Aggregate
1.37
(3.84)
-0.06
-(4.27)
-0.12
-(3.49)
1.52
(5.50)
6.14
(2.58)
0.75
(2.61)
0.10
(3.81)
0.08
(2.38)
0.27
(1.25)
0.10
(1.75)
TASS
0.79
(3.41)
-0.05
-(4.40)
-0.18
-(4.51)
1.10
(4.96)
10.02
(3.44)
1.44
(5.40)
0.14
(3.66)
0.05
(1.23)
0.80
(2.99)
-0.01
-(0.17)
HFR
1.39
(3.37)
-0.06
-(5.43)
-0.24
-(4.04)
1.18
(7.22)
8.73
(3.16)
1.37
(7.12)
0.05
(2.15)
0.05
(2.25)
0.24
(0.92)
0.01
(0.21)
BarclayHedge EurekaHedge Morningstar
0.84
0.63
1.19
(3.82)
(1.43)
(3.64)
-0.08
-0.03
-0.10
-(6.84)
-(1.83)
-(6.47)
-0.09
-0.12
-0.19
-(1.35)
-(1.48)
-(3.97)
1.00
1.35
1.23
(6.17)
(2.61)
(5.04)
13.13
21.19
6.84
(4.17)
(4.06)
(1.19)
0.93
1.24
1.26
(3.67)
(1.98)
(3.42)
0.11
-0.24
0.00
(2.66)
-(1.18)
-(0.05)
0.06
0.15
0.01
(2.10)
(1.82)
(0.16)
-0.09
0.36
0.60
-(0.41)
(0.83)
(1.84)
0.00
0.08
0.05
-(0.02)
(0.63)
(0.69)
Aggregate
0.84
(4.27)
-0.05
-(5.59)
-0.17
-(5.18)
1.10
(6.91)
8.06
(3.94)
1.26
(6.39)
0.08
(3.37)
0.04
(1.67)
0.24
(1.19)
0.02
(0.35)
C. Excess return
Variable
Intercept
Size t
1
Age t
1
Flow r
1
Management Fee
Incentive Fee
High-Water Mark
Lockup period
Notice period
Redemption
[1]
1.45
(3.02)
-0.04
-(1.11)
-0.19
-(2.39)
1.49
(4.29)
11.75
(2.07)
1.36
(2.77)
0.15
(2.28)
-0.02
-(0.20)
0.37
(1.01)
0.06
(0.50)
AUM terciles
[2]
0.72
(1.58)
-0.04
-(1.02)
0.01
(0.18)
1.69
(5.03)
7.97
(2.24)
0.67
(1.75)
0.13
(3.21)
0.13
(3.24)
0.24
(0.86)
0.03
(0.38)
D. Fung and Hsieh (2004) alphas
[3]
1.07
(2.40)
-0.02
-(0.59)
0.01
(0.13)
2.22
(4.75)
2.97
(1.06)
0.11
(0.38)
0.04
(1.14)
0.07
(2.22)
-0.04
-(0.17)
0.18
(2.13)
[1]
0.92
(2.76)
-0.06
-(1.83)
-0.33
-(4.03)
0.97
(4.09)
22.51
(4.62)
2.18
(5.73)
0.08
(1.47)
-0.02
-(0.34)
0.53
(1.49)
-0.02
-(0.21)
AUM terciles
[2]
0.53
(1.95)
-0.04
-(1.32)
-0.10
-(2.33)
1.26
(6.44)
3.21
(0.93)
1.12
(3.74)
0.08
(2.41)
0.09
(2.79)
0.20
(0.84)
-0.02
-(0.26)
[3]
0.82
(3.31)
-0.02
-(0.90)
-0.10
-(1.65)
1.64
(5.36)
3.46
(1.40)
0.43
(2.13)
0.06
(1.94)
0.05
(1.85)
-0.12
-(0.58)
0.08
(1.41)
Figure 1
Histograms of pairwise correlation coefficients of hedge funds
This figure shows histograms of pairwise correlation coefficients of share classes estimated within management companies for each database.
Databases are merged using a novel statistical algorithm. First, all databases are grouped based on a management company name that is
cleaned from errors and abbreviations. Second, correlation coefficients are estimated for each share class pair within each management
company. Each share class pair in correlation analysis is required to have at least 12 non-missing monthly return observations. Share classes
that do not have enough return observations for pairwise correlation analysis are automatically excluded from analysis. Share classes have
return time series between January 1994 – June 2011.
Figure 2
Venn diagram of overlapping and unique share classes
Figure presents Venn diagram showing overlapping between databases and the percentage amount of unique share classes. Databases are merged
using a statistical algorithm presented in Appendix of the paper. First, all databases including all share classes are classified based on management
company names. Second, pairwise correlations between all possible share class pairs are estimated within management companies. Multiple share
classes are identified using a 0.99 correlation limit and correlated share classes are classified in groups. Each group gives information on the amount
of databases each share class is reported. In Figure, all presented numbers are in percentages.
Figure 3
Proportion of hedge funds by investment objective
Figure presents the number of hedge funds by investment objective. Results are reported for each database. All funds are required to have at least 12
monthly return observations. Aggregate database is constructed using a statistical procedure that is described in Appendix. All databases are cleaned from
multiple share classes using the same statistical procedure. Numbers in the pie charts are percentage amounts of hedge funds by investment objective.
Pie chart of aggregate data includes all main investment objectives. Appendix provides details of the classification of the investment objectives.
Hedge Fund Research
TASS
7%
1%
BarclayHedge
CTA
10 %
5% 1% 5%
CTA
4%
EMERGING MARKETS
9%
LONG/SHORT
MARKET NEUTRAL
7%
6%
5%
18 %
7%
EurekaHedge
MARKET NEUTRAL
14 %
3%
SHORT BIAS
22 %
Aggregate
11 %
1%
CTA
10 %
7%
GLOBAL MACRO
16 %
4%
5%
6%
11 %
8%
6%
LONG/SHORT
5%
6%
EVENT DRIVEN
11 %
4%
OTHERS
RELATIVE VALUE
RELATIVE VALUE
35 %
SHORT BIAS
GLOBAL MACRO
LONG ONLY
6%
14 %
6%
MULTI-STRATEGY
OTHERS
SHORT BIAS
EMERGING MARKETS
10 %
MARKET NEUTRAL
MARKET NEUTRAL
MULTI-STRATEGY
EVENT DRIVEN
3%
<1%
GLOBAL MACRO
LONG ONLY
LONG/SHORT
10 %
EMERGING MARKETS
EVENT DRIVEN
7%
RELATIVE VALUE
CTA
EMERGING MARKETS
5%
OTHERS
SECTOR
4%
Morningstar
12 %
MARKET NEUTRAL
MULTI-STRATEGY
8%
CTA
3%
31 %
5%
SHORT BIAS
30 %
LONG ONLY
LONG/SHORT
6%
SECTOR
SHORT BIAS
<1%
LONG/SHORT
RELATIVE VALUE
RELATIVE VALUE
11 %
GLOBAL MACRO
8%
MULTI-STRATEGY
MULTI-STRATEGY
OTHERS
36 %
EVENT DRIVEN
GLOBAL MACRO
8%
GLOBAL MACRO
7%
EMERGING MARKETS
12 %
12 %
EVENT DRIVEN
EVENT DRIVEN
13 %
<1%
EMERGING MARKETS
10 %
11 %
6%
CTA
2%
6%
LONG/SHORT
MARKET NEUTRAL
MULTI-STRATEGY
OTHERS
RELATIVE VALUE
SECTOR
28 %
SHORT BIAS
Figure 4
Proportion of hedge funds by domicile region
Figure presents the number of hedge funds by fund-level domicile regions. Results are reported for each database. All funds are required to have at least
12 monthly return observations. Aggregate database is constructed using a statistical procedure (see Appendix for details). All databases are cleaned from
multiple share classes using the same statistical procedure. Numbers in the pie charts are percentage amounts of hedge funds by domicile regions that are
categorized as (1) Onshore; (2) Asia and Pacific; (3) Caribbean; (4) Europe; and (5) Rest of World. Funds that are legally established in either United States
or Canada are classified as onshore funds. Other categories of fund domiciles are classified as offshore funds.
Hedge Fund Research
TASS
9%
7%
12 %
29 %
12 %
Onshore
10 %
37 %
Asia and Pacific
1%
Onshore
Caribbean
Europe
Europe
Rest of World
1%
40 %
49 %
14 %
46 %
Onshore
Aggregate
10 %
28 %
17 %
Onshore
33 %
17 %
Onshore
Asia and Pacific
Asia and Pacific
Caribbean
Caribbean
Caribbean
Europe
Europe
2%
Europe
5%
Rest of World
31 %
8%
33 %
Caribbean
Europe
EurekaHedge
5%
Asia and Pacific
2%
Morningstar
16 %
Onshore
Asia and Pacific
Caribbean
Rest of World
41 %
BarclayHedge
Rest of World
Asia and Pacific
Rest of World
45 %
3%
37 %
Rest of World
Figure 5
Equal- and value-weight cumulative excess returns of hedge funds [January 1994 – December 2010]
Figure presents equal- and value-weight cumulative excess returns of hedge funds for each database. Databases are removed from multiple share
classes that exist either between databases or within management companies. Procedure to exclude multiple share classes is performed using a
statistical algorithm (presented in Appendix). Equal- and value-weight portfolio excess returns are calculated using single hedge funds that have at
least 12 monthly return observations. Aggregate data is the merged database using all five databases as source datasets.
6
Equal-weight
Value-weight
4.3
3.8
5
3.3
4
2.8
3
2.3
2
1.8
1
0
Jan-94
1.3
Jul-96
Jan-99
Jul-01
Jan-04
Jul-06
Jan-09
0.8
Jan-94
Jul-96
Jan-99
Jul-01
Jan-04
Jul-06
Jan-09
TASS
Hedge Fund Research
TASS
Hedge Fund Research
BarclayHedge
EurekaHedge
BarclayHedge
EurekaHedge
Morningstar
Aggregate
Morningstar
Aggregate
Figure 6
Equal- and value-weight cumulative abnormal returns of hedge funds [January 1994 – December 2010]
Figure presents equal- and value-weight cumulative abnormal returns of hedge funds for each database. Time series of abnormal returns are measured
as the model’s intercept (alpha) plus the time series of residual. All hedge funds are required to have at least 12 monthly non-missing return
observations. Aggregate data is constructed using a statistical procedure (see Appendix for details). Abnormal return is based on the seven factor
model proposed by Fung and Hsieh (2004). Risk factors are: the excess return of the S&P 500 index (SP-RF), the return of the Russell 2000 index minus
the return of the S&P 500 index (RL-SP), the excess return of ten-year Treasuries (TY-RF), the return of Moody's BAA corporate bonds minus ten-year
Treasuries (BAA-TY), the excess returns of look-back straddles on bonds (PTFSBD-RF), currencies (PTFSFX-RF), and commodities (PTFSCOM-RF).
Equal-weight
Value-weight
2.7
3.4
2.5
2.3
2.9
2.1
1.9
2.4
1.7
1.9
1.5
1.3
1.4
1.1
0.9
Jan-94
Jul-96
Jan-99
Jul-01
Jan-04
Jul-06
Jan-09
0.9
Jan-94
Jul-96
Jan-99
Jul-01
Jan-04
Jul-06
Jan-09
Lipper TASS
Hedge Fund Research
TASS
Hedge Fund Research
BarclayHedge
EurekaHedge
BarclayHedge
EurekaHedge
Morningstar
Aggregate
Morningstar
Aggregate
Figure 7
Equal-weight annual alphas of persistence portflios (January 1994 – December 2010)
Figure shows the annualized Fung and Hsieh (2004) alphas for persistence portfolios using four different holding periods. Top (bottom)
portfolios are displayed as single (dashed) lines. Results are reported for the five databases that are used in the paper and all hedge funds are
required to have at least 12 monthly return observations. The left y-axis shows the annualized Fung and Hsieh (2004) alphas in perentages
and x-axis displays the holding period in months. Returns of persistence portfolios are equal-weighted monthly, so the weights are adjusted
every month if there is funds that drop from the portfolio during the holding period.
10.00
9.00
8.00
7.00
6.00
5.00
4.00
3.00
2.00
1.00
1-month
3-month
6-month
12-month
TASS(Bottom)
TASS (Top)
HFR(Bottom)
HFR (Top)
Barclay (Bottom)
Barclay (top)
Eureka (Bottom)
Eureka (Top)
Morningstar (Bottom)
Morningstar (Top)
Aggregate (Bottom)
Aggregate (Top)
Download PDF