Identifying Performance

Persistence in Hedge
Funds Using a Measure
of Manager Skill
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

Arik Ben Dor, Jingling Guan, and Xiaming Zeng

A rik Ben Dor

is a managing director in KEY FINDINGS
the Quantitative Portfolio • The authors introduce a novel measure of skill and demonstrate it is capable of differen-
Strategy Group at Barclays tiating among hedge funds based on their past performances.
in New York, NY. • Funds with better skill generate subsequently both higher returns and improved risk
arik.bendor@barclays.com properties.
• The results cannot be explained by differences in fund characteristics or market exposures
Jingling Guan and are robust to the choice of rebalancing frequency and the existence of backfill bias.
is a director in the
Quantitative Portfolio
Strategy Group at Barclays
in New York, NY. ABSTRACT: In this article, the authors discuss $3.32 trillion as of the end of 2019.1 The
jingling.guan@barclays.com the challenges in identifying performance persis- number of hedge funds available to inves-
tence among hedge funds and introduce a novel tors and the large potential dispersion in
X iaming Zeng
is a vice president in the measure of manager skill, termed MMS, to address their performance makes identifying future
Quantitative Portfolio them. Funds with higher MMS scores deliver top performers challenging. One commonly
Strategy Group at Barclays larger returns with lower volatilities, better tail- used approach is to rely on funds’ past track
in London, United risk behavior, and less liquidation risk in subse- records and compute various performance
Kingdom. quent periods than funds with lower MMS scores. statistics, such as average returns and Sharpe
xiaming.zeng@barclays.com
The results cannot be explained by differences in ratios. Funds will then be evaluated based on
fund characteristics or market exposures and are these statistics, with the implicit assumption
robust to the choice of rebalancing frequency and that these statistics are predictive of the funds’
the existence of backfill bias. future performance. This, in turn, relies on
the concept of performance persistence—the idea
TOPICS: Manager selection, performance
that when managers’ successful track record
measurement*
ref lects skill (as opposed to sheer luck), their
future performance is also likely to be supe-

O
rior to peers’.
ver the past three decades, the
Extensive research has investigated the
hedge fund industry has expe-
existence of performance persistence among
rienced exponential growth,
hedge funds, using different return measures
with more than 9,400 funds and
*All articles are now assets under management (AUM) exceeding
categorized by topics 1
Numbers are from “HFR Global Hedge
and subtopics. View at Fund Industry Report—Year End 2019” (released on
PM-Research.com. January 20, 2020; www.HedgeFundResearch.com).

Manager Fund Selection 2020 The Journal of Portfolio Management   1

 and various statistical techniques (Brown, Goetzmann, since 1997. In particular, an equally weighted portfolio
and Ibbotson 1999; Agarwal and Naik 2000; Bares, of winners delivered a Sharpe ratio of 1.23 between
Gibson, and Gyger 2003; Kosowski, Naik, and Teo 1997 and 2019, compared with a Sharpe ratio of 0.42
2007; Jagannathan, Malakhov, and Novikov 2010; for losers over the same period. Additional tests con-
Ammann, Huber, and Schmid 2013). Most studies firm that the superior performance of winners is not
did not find evidence supporting performance per- driven by differences in characteristics (compared with
sistence, or they documented persistence patterns that loser funds) or a larger exposure to market risk and is
were highly sensitive to the choice of the database robust to different rebalancing frequencies.
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

and sample period (see, e.g., Joenvaara et al. 2019).

In addition, the methodologies used in these studies PERFORMANCE PERSISTENCE: DISCUSSION
were often complex, required multiple assumptions, AND ILLUSTRATION
and were not easily adaptable for use by investors.
The inability to detect performance persistence When selecting a hedge fund for investment, is the
among hedge funds ref lects the difficulty in identifying fund manager’s prior track record useful? If past perfor-
investment skill. Distilling skill from random success mance is indicative of future results, this information is
requires both a methodology that can properly account valuable. If not, investors may be better off basing their
for risk (both systematic and idiosyncratic) and a suffi- selection on other considerations, such as the manager’s
ciently long track record. The unique characteristics of reputation, investment style, and fee structure.
hedge funds present serious challenges on both fronts. Before introducing MMS, our measure of man-
Although their investment mandates typically allow the ager skill, we illustrate the challenges in detecting
use of leverage, short selling, derivatives, and highly persistence using two common performance statistics:
illiquid securities—with few and often no limits on mean return and Sharpe ratio. We construct portfo-
geographical and issuer-specific concentrations—hedge lios of hedge funds based on rankings generated by
funds do not follow any designated benchmarks and are these statistics and analyze their subsequent perfor-
not required to provide detailed information on their mance. Our primary data source is the Hedge Fund
holdings or the type of strategies they use. Furthermore, Research (HFR) database, which classifies funds into
many hedge funds have relatively short histories that seven broad strategy groups (equity hedge, event-
limit the accuracy of any statistical inference. driven, macro, relative value, risk parity, blockchain,
We present a new measure for evaluating the per- and fund of funds) and more than two dozen substrat-
formance of hedge funds, which we term measure of man- egies. Our sample spans 24 years from January 1996
ager skill (MMS). It controls for risk and is invariant to to December 2019 and consists of 2,467 active funds
leverage. Furthermore, the measure is model free and and 8,825 defunct funds, with monthly net returns
requires only limited data. MMS decomposes a fund’s reported in US dollars. Funds of funds are excluded
returns into manager-specific and systematic compo- to avoid double counting. Risk parity and blockchain
nents and assigns higher scores to funds that had high are also removed owing to their limited history and
manager-specific performance. After discussing the relatively small universe.
pitfalls of traditional performance measures and con-
firming that they are not able to detect performance per- Identifying Performance Persistence
sistence, we analyze the performance of various hedge
fund portfolios formed using MMS scores. Portfolios are formed by ranking all funds annu-
We f ind very strong evidence of persistence. ally based on their average returns or Sharpe ratios
Specifically, funds with the top 25% of MMS scores calculated over the previous 12-month period. Funds
(termed winners) have a 36% chance of maintaining are then assigned to four quartile portfolios, Q1–Q4,
their ranking in the subsequent year, whereas losers based on their rankings.2 Funds that receive the highest
(those with the bottom 25% of MMS scores) have a
similar probability (34%) of remaining losers. These 2
Specifically, the ranking is done as of December 31 of each
results are present not only in the aggregate but year. Funds with fewer than 12 monthly consecutive returns are
across all hedge fund styles and in almost every year not ranked.

2   Identifying Performance Persistence in Hedge Funds Using a Measure of Manager Skill Manager Fund Selection 2020
 ranking (winners) are assigned to portfolio Q4, whereas Exhibit 1
those with the lowest scores (losers) are included in Q1. Postformation Performance of Quartile Portfolios
Postformation portfolio returns are computed monthly 1997–2019
based on equally weighting all constituents, which
remain unchanged until the next ranking date, unless Q1 Q4
they are removed because they no longer report their (Loser) Q2 Q3 (Winner)
performances.3 Panel A: Ranking by Average Return
If higher values for a certain performance sta- Mean Return (%/mo) 0.62 0.54 0.64 0.72
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

Volatility (%/mo) 2.57 1.52 1.73 2.94

tistic (e.g., the average return or Sharpe ratio) are
5th Percentile (%) –3.03 –1.92 –2.17 –3.97
indeed associated with better skill, then the funds with
Worst Month (%) –7.87 –6.86 –6.12 –12.66
higher past average returns or Sharpe ratios would be Maximum Drawdown (%) –28.40 –18.37 –16.58 –28.01
expected to outperform funds with lower values in the Sharpe Ratio (Ann.) 0.56 0.77 0.88 0.61
future, on average. Specifically, in the context of our Panel B: Ranking by Sharpe Ratio
exercise, the portfolios’ postformation performance Mean Return (%/mo) 0.53 0.67 0.67 0.65
should exhibit a clear monotone ordering with the Volatility (%/mo) 2.26 2.26 2.19 1.78
Q4 and Q1 portfolios achieving the best and worst 5th Percentile (%) –2.90 –2.86 –2.90 –2.23
performance, respectively. Worst Month (%) –8.11 –7.36 –8.16 –7.78
Exhibit 1 reports various performance statistics for Maximum Drawdown (%) –26.84 –20.28 –21.93 –21.92
the four portfolios based on their postformation returns Sharpe Ratio (Ann.) 0.51 0.72 0.73 0.87
using either average return (Panel A) or Sharpe ratio
Notes: Funds are ranked annually in December based on mean returns or
(Panel B) as the ranking statistic. Overall, the results Sharpe ratios over the previous 12 months and then assigned into quartile
do not indicate that selecting funds based on either past portfolios Q1–Q4. Funds in the winner and loser portfolios are ranked in
average return or Sharpe ratio would result in superior the top and bottom 25% of all funds, respectively. Performance statistics
are based on postformation returns. Sharpe ratios are computed based on
future performance. Panel A reveals that although Q4, the one-month LIBOR rate.
the winner portfolio, boasts the highest average return Source: HFR, Barclays Research.
among all portfolios, it also has the worst risk proper-
ties, with almost twice the monthly volatility and draw-
down of the Q3 portfolio (2.94% and -28.01% compared Why is the use of Sharpe ratios unable to gen-
with 1.73% and -16.58%, respectively). Consequently, erate performance persistence? Indeed, unlike absolute
the winner portfolio actually realizes the second lowest return, the Sharpe ratio is a measure of risk-adjusted
Sharpe ratio, consistent with the notion that higher past performance. However, it does not distinguish between
absolute returns are associated with taking more risk and the drivers of performance (i.e., manager specific versus
not with having better skill. systematic). Thus, a fund with a large exposure to a
The results in Panel B suggest a positive, albeit certain systematic factor but no manager-specific per-
weak, relation between rankings and subsequent perfor- formance can be termed a winner if the factor realiza-
mance. However, the winner portfolio, Q4, generates tions are sufficiently high during the ranking period. As
slightly worse returns than Q2 and Q3. The difference long as the factor realizations enjoy positive momentum,
in Sharpe ratios among the portfolios is also statistically the fund will continue to generate strong returns post
and economically insignificant (0.87, 0.73, and 0.72 for ranking. However, if this is not the case, the fund will
Q4, Q3, and Q2, respectively). In addition, the tail risk not outperform and may even underperform if the factor
properties of the quartile portfolios are very similar. realizations exhibit mean reversion.4
In fact, the drawdown and worst month return of the One episode that illustrates this issue occurred in
Q2 portfolio are actually better than those of Q4. 2007–2009, when the financial markets experienced a
4
The fund may continue to outperform even if the factor
3
For example, the portfolios’ returns for 2009 are based realizations revert as long as it readjusts its market exposures appro-
on compositions formed by ranking 2008 performance data. priately in a timely fashion, a process also known as market timing.
The majority of funds that stopped reporting became defunct. For a discussion on the difference between market timing and alpha,
See Ambastha and Ben Dor (2010) for more details. see Ben Dor, Dynkin, and Gould (2006).

Manager Fund Selection 2020 The Journal of Portfolio Management   3

 Exhibit 2
Annual Performance of Q4 and Q1 Portfolios in 2007–2009
Portfolios Ranked by Avg. Return Portfolios Ranked by Sharpe Ratio

60% 60%
Annual Cum. Ret.

Annual Cum. Ret.

30% 30%
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

0% 0%

–30% –30%
2007 2008 2009 2007 2008 2009

Q1 (Loser) Q4 (Winner)

Notes: Funds are assigned into quartile portfolios Q1–Q4 each December based on their mean returns or Sharpe ratios computed over the previous
12 months. Funds in the winner and loser portfolios are ranked in the top and bottom 25% of funds, respectively. Annual performance is based on
postformation returns.
Source: HFR, Barclays Research.

sharp downturn starting in late 2007 and then a strong take large risks ex ante. For example, consider the case
reversal from March 2009 onward. Exhibit 2 plots the of Funds A and B, which implement the same exact
postformation performance of the winner and loser strategy, except that Fund A is twice leveraged. By
portfolios in these years based on ranking by average construction, then, the average return of Fund A would
return or Sharpe ratio. Hence, funds comprising the be twice that of Fund B. Not adjusting for leverage
loser portfolio in 2009 were those that experienced the would lead us to mistakenly conclude that Fund A is
worst performance (as measured by the appropriate per- better or worse than Fund B.
formance statistic) in 2008, likely as a result of having Although conceptually simple, measuring the risk
more leverage and systematic market exposure. It is taken by hedge funds is particularly challenging. Unlike
not surprising then that these funds outperformed the mutual funds, hedge funds operate with a much greater
winner significantly in 2009; their characteristics ben- f lexibility and use leverage and derivatives extensively.
efited them once the market rebounded strongly. The Because of the f lexibility they enjoy, hedge funds exhibit
winner portfolio based on ranking by average return significant variations in risk profile cross sectionally and
and Sharpe ratio underperformed the loser portfolio by over time, even for the same fund.
a striking 51% and 35%, respectively. Thus, not only do Separation of manager-specific and system-
we not observe persistence in fund performance, we see atic components. The second challenge in finding
quite the opposite, with the ordering of funds’ perfor- persistence is the lack of an accepted, well-defined
mances completely reversed. benchmark for hedge funds. Such a benchmark would
The results in Exhibits 1 and 2 highlight some of enable decomposition of returns into manager-specific
the challenges that need to be addressed by any perfor- and systematic components as in Equation 1, in which
mance statistic used to identify performance persistence. the former represents a manager’s skill, and the latter
Two aspects of hedge fund performance that deserve refers to exposure to any systematic factor, either tra-
special attention are risk adjustment and separation of ditional (i.e., market indexes or asset classes) or exotic
manager-specific and systematic contributions. (e.g., volatility), directly or indirectly. Funds with strong
Risk-adjusted performance. To measure per- past performance owing to high manager-specific per-
formance properly, one has to account for all risks a formance are more likely to be future top performers
fund is taking. Otherwise, many of the funds with irrespective of market conditions, unlike cases in which
higher performance ex post would also be those that

4   Identifying Performance Persistence in Hedge Funds Using a Measure of Manager Skill Manager Fund Selection 2020
 performance is driven by the systematic exposures The limited data available for hedge funds make it
(Exhibit 2). almost infeasible to estimate nonlinear exposures. Third,
even with sufficient data, the dynamics of the hedge
Fund return = Manager-specific contribution fund industry may require constantly updating the set of
+ Systematic contribution (1) factors used, contradicting the backward-looking nature
of these models.6
Because of the variety and complexity of trading To separate the strategy-specific dynamics from the
strategies used by hedge funds, benchmarking their fund returns, we compute MMS in three steps. In the
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

returns is extremely difficult. Although some factor- first step, we use synthetic hedge fund portfolios that
based models have been proposed (e.g., the seven-factor are representative of different hedge fund strategies as
model from Fung and Hsieh 2004), such model-based benchmarks. We use variable selection techniques to
benchmarks are quite data intensive and cannot be choose which set of strategies go into the benchmark.
precisely estimated given the relatively short history of On one hand, a fund may use more than one strategy
most funds. at the same time. Using a prespecified strategy, such as
the self-reported one, might be insufficient and even
MEASURE OF MANAGER SKILL inaccurate because funds may misreport. On the other
hand, many strategies might not be relevant to a specific
To address the challenges discussed in the pre- fund. Including all of them agnostically undermines the
vious section, we propose a new measure for evaluating estimation precision. We adopt a stepwise regression
the performance of individual funds: MMS. Unlike approach to identify strategies to which a fund has signif-
the traditional factor-based approach that models the icant exposures and include only those in the benchmark.
systematic component in returns via risk factor load- In the second step, we account for funds’ leverages
ings, MMS uses the systematic component to capture on different types of strategies. Given that hedge funds
strategy-specific dynamics. As a result, the manager- are likely to have extreme returns, we remove potential
specific contribution becomes the idiosyncratic effects outliers using the Cook’s distance statistics (Cook 1977).
caused by a manager’s decisions that deviate from those The leverages are then estimated with the remaining
of an average manager in the industry. Most importantly, return sample.
now the manager-specific part can still be a manager’s In the third step, we control for the idiosyncratic
clever bet on the market or any other type of risk factors. risk of funds by accounting for the volatility in the
Such skills, by construction, cannot be ref lected by the manager-specific component. After taking out the sys-
factor-based approach because they would be taken out tematic contribution from the fund performance, we
by the factor loadings.
Factor regressions also have several other important (i.e., unrelated to the broad market), whereas in down markets, the
limitations. First, the number and nature of factors for probability that a deal will fail increases the larger the decline. This
hedge funds are largely unknown. It is therefore difficult is because the target would like to maintain the original deal price,
to decide which factors should be included, given that whereas the acquirer would like to adjust the price downward to
so many factors are out there (see, e.g., Harvey, Liu, and ref lect the new state of the market. For more details see Mitchell
and Pulvino (2001).
Zhu 2016 for a zoo of factors just in the equity space). 6
In fact, factor regressions are usually more suited for building
Second, factor exposures can be highly nonlinear for hedge fund replicators (Ben Dor, Jagannathan, and Meier 2003)
hedge funds and hence require a large amount of data to rather than evaluating an individual hedge fund’s performance.
estimate. For example, hedge funds have few restrictions When tracking a large universe of hedge funds, an overall evalua-
in trading derivatives or exotic products. Furthermore, tion at the aggregate level would be sufficient. Issues discussed in
the main text then become less of a problem. More importantly,
strategies such as merger arbitrage may exhibit a payoff
factor analysis is almost inevitable for replication because finding
profile similar to an out-of-the-money put on the S&P out what managers have been doing is critical so that these activi-
500 index even when they do not use any options.5 ties can be replicated. When it comes to distinguishing high skill
versus low skill funds, however, a precise assessment of performance
5
The reason for the option-like profile is that the strategy is needed at the cross-sectional level; as long as a relative ranking
payoff depends on whether the merger is completed or not. In up among managers can be formulated, it is not a necessity to know
or f lat markets, an unsuccessful acquisition is an idiosyncratic event what managers have been doing.

Manager Fund Selection 2020 The Journal of Portfolio Management   5

 Exhibit 3 the rest would be included in one of the three remaining
Transition Frequencies among Quartile Portfolios portfolios. Because the same argument would also apply
Based on MMS Rankings to the Q1–Q3 portfolios in year t, the winner portfolio
in year t + 1 would comprise an equal number of funds
Ranking at Year t + 1 (50) from the Q1–Q4 portfolios in year t.
Q1 (Loser) Q2 Q3 Q4 (Winner) The results in Exhibit 3 indicate, however, that
Q1 (Loser) 34% 27% 23% 16% 72 of the funds (the equivalent of 36% in our example)
Ranking
at Year t

Q2 29% 26% 25% 20% ranked in the top quartile (winner) in year t maintain
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

Q3 24% 26% 26% 24% their top ranking in the subsequent year, and only 34
Q4 (Winner) 17% 22% 26% 36% (i.e., 17%) are reclassified as losers. Although funds in
the Q3 portfolio (and to a slightly lesser extent Q2)
Notes: Funds are assigned into quartile portfolios Q1–Q4 every
December between 1997 and 2019, based on their MMS scores computed
are distributed almost uniformly in year t + 1, the
over the previous 12 months. Funds termed winner and loser are ranked worst performing ones (losers) exhibit strong persis-
in the top and bottom 25% of funds, respectively. The exhibit reports the tence as well, with a 34% probability of staying in the
average transition frequencies among quartile portfolios between any two bottom quartile in the following year. Consequently,
consecutive years using only funds with complete 12-month histories.
the probability a winner fund would be ranked above
Source: HFR, Barclays Research.
the median fund in year t + 1 is 62%, much higher
than that of losers (39%), or 124 funds compared with
adjust the average manager-specific component by its just 78 out of the 200 funds in each of the portfolios
time-series standard deviation and use that as the MMS in our example.
score. A fund that can deliver manager-specific perfor- The results in Exhibit 3 hold not just for the aggre-
mance in a consistent manner would therefore be valued gate but also for different hedge fund styles and indi-
more favorably. vidual years. Exhibit 4 shows the percentage of funds
If MMS is an effective measure of a manager’s skill, with above-median ranking separately for funds that
funds with high scores should deliver better risk-adjusted are ranked as winner or loser in the prior year. Panel A
performance consistently, and there should be a positive shows that the percentage of winner funds that outper-
relation between past rankings and future performance. form the median is higher than 50% in all years, with
We first present evidence of performance persistence the exception of 2017. In contrast, the percentage of
based on using MMS and then analyze the performance loser funds that outperform the median is consistently
of hedge funds that received high MMS scores. lower than 50%. Similarly, winner funds outperform the
median with more than 50% likelihood across all styles,
Performance Persistence with relative value fixed-income funds exhibiting the
strongest persistence (72%). Loser funds again exhibit
Exhibit 3 presents the average annual transition the mirror image, with only 38% outperforming the
probabilities based on MMS rankings between any two median on average.
consecutive years over the sample period. The rows and Another important aspect of the ability to identify
columns represent funds’ rankings in year t and t + 1, performance persistence is attrition rates (see, e.g., Ben
respectively, and the diagonal cells therefore represent Dor et al. 2012). Hedge funds as a group experience a
the percentage of funds with unchanged ranking. Thus, rate of attrition considerably higher than other types of
performance persistence requires that the value in the asset managers. For example, Ambastha and Ben Dor
bottom diagonal (i.e., for Q4) be significantly higher (2010) found that in 2008, more than one quarter of
than 25%, which would represent a random draw. funds in the HFR database stopped reporting, and most
For example, if the population of funds in year of them were liquidated.
t and t + 1 is 800, the top-quartile portfolio (winner) Exhibit 5 plots the postranking attrition rates for
in year t would include 200 funds. Lack of persistence winner and loser funds. Consistent with the evidence
implies that these funds would be equally likely to end in Exhibit 3, loser funds are much more likely to stop
up in any of the quartile portfolios in year t + 1. Thus, of reporting during the year post ranking. Their attrition
the 200 funds, only 50 on average would maintain their rate is 18% on average, more than twice that of winner
winner status (based on their rankings in year t + 1), and funds. In addition, the attrition rate of loser funds varies

6   Identifying Performance Persistence in Hedge Funds Using a Measure of Manager Skill Manager Fund Selection 2020
 Exhibit 4
Percentage of Winner and Loser Funds with above Median Ranking in the Following Year
Panel A: By Year (aggregated over styles)
100%

75%
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

50%

25%

0%
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010

2012
2013
2014
2015
2016
2017
2018
2019
2011
Panel B: By Style (aggregated over time)
100%

75%

50%

25%

0%
Market
Neutral

Fundamental
Growth

Fundamental
Value

Special

Discretionary
Thematic
Quantitative
Directional

Technology/

Restructuring

Situations
Healthcare

Distressed/

Fixed
Income

Multi-Strategy
Systematic
Diversified

Loser (Q1) Winner (Q4)

Notes: Funds are assigned into quartile portfolios Q1–Q4 every December based on their MMS scores computed over the previous 12 months.
Funds termed winner and loser are ranked in the top and bottom 25% of funds, respectively. Probabilities are computed only for funds with rankings
in year t and t + 1. Style categories are based on HFR classifications.
Source: HFR, Barclays Research.

considerably with the overall market environment. In to maintain their respective rankings beyond what is
2008, at the peak of the financial crisis, the attrition implied by random chance. In this section, we provide
rate for loser funds reached a record of 32% whereas for direct evidence on the relation between MMS scores and
winner funds it was only 11%, slightly above its long- future absolute and risk-adjusted returns. To examine
term average. this relation, Exhibit 6 presents various performance
statistics for the quartile portfolios Q1–Q4 constructed
Portfolio Performance based on MMS rankings over the entire sample and for
two separate subperiods.7 The statistics ref lect returns
The last section introduced MMS as a ranking sta-
tistic and showed that funds with MMS rankings in the 7
Recall that the constituent funds in each portfolio are
top and bottom quartiles of the population are likely determined annually.

Manager Fund Selection 2020 The Journal of Portfolio Management   7

 Exhibit 5
Yearly Attrition Rate for Winner and Loser Funds
35%

30%

25%
Yearly Attrition Rate
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

20%

15%

10%

5%

0%
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

‘Loser’ Funds ‘Winner’ Funds

Notes: Funds are assigned into quartile portfolios every December based on their MMS scores computed over the previous 12 months. Funds termed winner
and loser are ranked in the top and bottom 25% of funds, respectively. Attrition rates are computed as the number of winner and loser funds that stop
reporting in a certain year divided by the total number of funds at the beginning of the year. Because funds are delisted from the HFR dataset only three
months after they stop reporting, the data for 2019 are not final yet and are therefore excluded.
Source: HFR, Barclays Research.

during the 12-month postformation until the next rebal- and loser portfolios (represented by the vertical distance
ancing date. between the two lines in Exhibit 7) is not constant. For
Despite significant variation in absolute return over example, it declined substantially between 2006 and early
time, the ordering of performance based on past MMS 2008 as loser funds outperformed winners in absolute
rankings is preserved. Furthermore, there is a clear nega- terms during the bull market. This highlights the fact that
tive relation between rankings and risk, irrespective of although persistence should result in higher-ranked funds
the measurement techniques. In particular, the winner outperforming lower-ranked ones on average, outperfor-
portfolio experienced return volatility and drawdown mance is not necessarily the case in the short run. This is
of 1.53% and -14.56%, respectively—almost half that of because funds with low rankings (i.e., no manager-specific
the portfolio with the lowest ranked funds, Q1 (2.49% performance) but large market exposures may still gen-
and -31.62%, respectively). As a result, the Sharpe ratio erate higher returns than better-ranked funds if the market
increases consistently across portfolios by a factor of rallies. However, such an outcome may only hold for a
three, from a low of 0.42 (loser) to 1.23 (winner). These limited period. Eventually, as market conditions change,
results hold equally well in both subperiods. the benefit of superior skill will materialize, and highly
Exhibit 7 plots the rolling five-year Sharpe ratios ranked funds (i.e., with high manager-specific perfor-
of the winner portfolio alongside those of the loser port- mance) will outperform lower ranked funds.
folio. The graph indicates that the risk-adjusted perfor- To measure more accurately the magnitude of the
mance of the winner portfolio is superior to that of the difference in manager-specific performance between the
loser portfolio in every five-year period since 1997. In winner and loser portfolios, we analyze the performance
particular, the Sharpe ratio of the winner portfolio is of a theoretical portfolio representing long and short
always positive and is higher than that of the loser port- positions in the winner and loser portfolios, respectively.
folio by at least 0.3 throughout the sample. To ensure the systematic risk of the long–short portfolio
Notice, however, that the magnitude of the differ- is eliminated, the position in the winner portfolio is
ence in risk-adjusted performance between the winner scaled such that its risk ex ante is equalized to that of

8   Identifying Performance Persistence in Hedge Funds Using a Measure of Manager Skill Manager Fund Selection 2020
 Exhibit 6 and S&P 500 Indices since January 2000. Both charts
Performance Statistics for Portfolios Formed Based illustrate that the performance differential between the
on MMS Rankings winner and loser portfolios does not depend on the
overall hedge fund market (the correlation is 0.01) or
Q1 Q4 the state of the overall market as represented by the
(Loser) Q2 Q3 (Winner) S&P 500 (the beta to the S&P 500 is only -0.06, both
Panel A: Full Sample (1997–2019) statistically and economically insignificant).
Mean Return (%/mo) 0.50 0.61 0.66 0.74 Exhibit 9 plots the cumulative return of the
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

Volatility (%/mo) 2.49 2.11 1.92 1.53

long–short portfolio along with that of HFRI and S&P
5th Percentile (%) –3.58 –2.75 –2.39 –1.67
500 indices. Over the sample period 2000–2019, the
Worst Month (%) –9.33 –8.11 –7.80 –5.46
Maximum Drawdown (%) –31.62 –24.80 –19.85 –14.56
long–short portfolio delivered a monthly average return
Sharpe Ratio (Ann.) 0.42 0.66 0.82 1.23 of 0.76%, which translates into an annualized manager-
Panel B: Subperiod (1997–2009) specific performance difference between winner and
Mean Return (%/mo) 0.63 0.77 0.84 0.92 loser funds of 9.12%. Furthermore, the long–short port-
Volatility (%/mo) 2.84 2.41 2.21 1.68 folio had an annualized volatility of only 3.90% with a
5th Percentile (%) –3.60 –2.52 –2.10 –1.45 hit ratio of 80%. As a result, the long–short portfolio
Worst Month (%) –9.33 –8.11 –7.80 –5.46 boasted an information ratio of 2.34 and generated more
Maximum Drawdown (%) –31.62 –24.80 –19.85 –14.56 than 500% returns from 2000 to 2019, exceeding both
Sharpe Ratio (Ann.) 0.38 0.66 0.83 1.25 the HFRI and S&P 500 indices by a great margin.
Panel C: Subperiod (2010–2019) Perhaps even more impressive is the difference
Mean Return (%/mo) 0.34 0.39 0.42 0.51 between the results in Exhibit 2 and the performance
Volatility (%/mo) 1.93 1.63 1.44 1.27
of the long–short portfolio based on MMS rankings in
5th Percentile (%) –3.35 –2.91 –2.46 –1.90
Worst Month (%) –5.24
2008–2009 shown in Exhibit 10. The annual returns of
–4.63 –4.06 –3.76
Maximum Drawdown (%) –13.61 –10.31 –9.10 –7.34
the long–short portfolio in 2008 and 2009 were both
Sharpe Ratio (Ann.) 0.50 0.71 0.85 1.23 positive. The long–short portfolio earned 8.81% and
3.88% in 2008 and 2009, respectively, whereas the S&P
Notes: Funds are assigned into quartile portfolios Q1–Q4 every 500 was down 37% in 2008 and up 26% in 2009. In
December based on their MMS scores computed over the previous 12 addition, the long–short portfolio experienced just six
months. Funds termed winner and loser were ranked in the top and
bottom 25% of funds, respectively. Performance statistics are based on down months during the period, with December 2008
postformation returns. Sharpe ratio was computed based on the one-month representing the worst month with a loss of only 0.49%.
LIBOR rate. The quartile portfolios Q1–Q4 are well diver-
Source: HFR, Barclays Research. sified and consist of a few hundred funds each. How
will the portfolio performance change if the number
the loser portfolio.8 Consequently, the performance of of constituent funds in the portfolio ranked by MMS
the long–short portfolio is a direct measure of relative is smaller? Will the relation between past ranking and
manager-specific contribution—the skill difference future performance be equally strong?
between winner and loser funds. Though the long–short This question is of particular importance because
portfolio consists of a leg shorting the loser funds, we the number of hedge funds held in investors’ portfolios
are not suggesting shorting hedge funds. The portfolio is typically limited to several dozen or less. To address
is simply an illustration to demonstrate the outperfor- the question, we form 10 equally populated portfolios
mance of winner funds over loser funds. (i.e., deciles), based on MMS rankings, as well as the two
Exhibit 8 displays monthly returns of the long– ranking statistics we used previously: average returns and
short portfolio against those of the HFRI composite Sharpe ratios. Exhibit 11 displays the Sharpe ratios of the
different portfolios based on their postformation returns.
8
The chart indicates that the relation between MMS
Because both the winner and loser portfolios include hun-
rankings and future performance is still monotonic,
dreds of funds, their idiosyncratic risk will be negligible. The scaling
factor is calculated monthly based on the ratio of volatilities of the despite the fact that the number of constituent funds
loser to the winner portfolio over the previous 36 months with no in each portfolio has declined by more than half. The
hindsight. highest decile portfolio delivered a Sharpe ratio of 1.43

Manager Fund Selection 2020 The Journal of Portfolio Management   9

 Exhibit 7
Rolling 60-Months Sharpe Ratios of Winner and Loser Portfolios
4.0

3.0
Sharpe Ratio

2.0
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

1.0

0.0

–1.0
Dec 2001 Dec 2005 Dec 2009 Dec 2013 Dec 2017

Loser (Q1) Winner (Q4)

Notes: Funds were assigned into quartile portfolios Q1–Q4 every December based on their MMS scores computed over the previous 12 months.
Funds termed winner and loser were ranked in the top and bottom 25% of funds, respectively. Sharpe ratios were computed over rolling 60-months
periods based on the portfolios’ postformation returns and using the one-month LIBOR rate.
Source: HFR, Barclays Research.

Exhibit 8
Performance of Long Winner–Short Loser Portfolio vs. HFRI Composite and S&P 500 Indices
Long–Short Long–Short
8 8
Portfolio Ret. Portfolio Ret.
(%/month) (%/month)

4 4

0 0

–4 –4
–10 –5 0 5 10 –20 –10 0 10 20
HFRI Monthly Return (%) S&P 500 Monthly Return (%)

Notes: The long–short portfolio represents long and short positions in the winner and loser” portfolios, respectively, formed based on MMS rankings.
The position in the winner portfolio is scaled such that its risk ex ante is equalized to that of the loser portfolio. For additional details, see footnote 8.
Source: Bloomberg, HFR, Barclays Research.

over the 23-year period from 1997, almost five times Sources of Outperformance
that of the portfolio with the lowest MMS-ranked funds
(0.32). In contrast, the performance of the portfolios The results so far indicate a clear monotone rela-
ranked on past average returns exhibits no relation at all, tion between MMS scores and subsequent perfor-
and the pattern for those based on Sharpe ratio is very mance. Funds with higher MMS scores not only earn
weak and inconclusive. higher returns but also exhibit substantially lower risk

10   Identifying Performance Persistence in Hedge Funds Using a Measure of Manager Skill Manager Fund Selection 2020
 Exhibit 9
Cumulative Return of Long Winner–Short Loser Portfolio
700

600
Cumulative Ret. (%)

500

400
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

300

200

100

0
Jan 2000 Jan 2002 Jan 2004 Jan 2006 Jan 2008 Jan 2010 Jan 2012 Jan 2014 Jan 2016 Jan 2018

Long–Short Portfolio (‘winner’ less ‘loser’) HFRI Index S&P 500

Notes: The long–short portfolio represents a long and short positions in the winner and loser portfolios, respectively, formed based on MMS rankings.
The position in the winner portfolio is scaled such that its risk ex ante is equalized to that of the loser portfolio. For additional details, see footnote 8.
Source: Bloomberg, HFR, Barclays Research.

Exhibit 10
Performance of Long Winner–Short Loser Portfolio in 2008–2009



0RQWKO\5HW 





±
2FW

2FW
-DQ

-DQ
$SU

$SU
-XO

-XO

Notes: The long–short portfolio represents long and short positions in the winner and loser portfolios, respectively, formed based on MMS rankings.
The position in the winner portfolio is scaled such that its risk ex ante is equalized to that of the loser portfolio. For additional details, see footnote 8.
Source: HFR, Barclays Research.

and attrition rate. Although we argued that the results is simply more likely to pick on such funds. Another,
ref lect the ability of MMS to effectively identify high less likely, possibility is that funds with higher MMS
manager-specific performance, there are several alterna- scores have more exposure to systematic risk, which
tive explanations. leads to better performance. A third explanation is that
One possible explanation is that the outperfor- our results ref lect the rebalancing frequency and are
mance is driven by fund characteristics known to be not robust to other specifications. A fourth explanation
correlated with performance, such as size, and MMS is that results are driven by data sample, specifically by

Manager Fund Selection 2020 The Journal of Portfolio Management   11

 Exhibit 11
Sharpe Ratios of Decile Portfolios by Ranking Statistics
2.0

1.5
Sharpe Ratio
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

1.0

0.5

0.0
1-Low 2 3 4 5 6 7 8 9 10-High
Ranking Decile Portfolio Ranking

Avg. Ret. Ranking Sharpe Ratio Ranking MMS Ranking

Notes: Funds are assigned into decile portfolios every December between 1997 and 2019 based on one of the ranking statistics (MMS, Sharpe ratio,
or average return) computed over the previous 12 months. Sharpe ratios are computed based on the decile portfolios postformation return time series using
the one-month LIBOR rate.
Source: HFR, Bloomberg, Barclays Research.

Exhibit 12 the backfilled return data. We examine each of these

Fund Characteristics of Portfolios Formed by MMS issues in turn.
Fund characteristics. Exhibit 12 displays the
4 4 average characteristics of funds in each quartile port-
/RVHU 4 4 :LQQHU folio. These characteristics, such as fund size, age, and fee
1XPEHURI)XQGV     structures, have been found to be correlated with hedge
$YJ$80 PLOOLRQ     fund performance (Liang 1999; Edwards and Caglayan
$YJ+LVWRU\ PRQWKV     2001; Teo 2009).
$YJ0DQDJHPHQW)HH     
The exhibit reveals several differences among the
$YJ,QFHQWLYH)HH     
%HWD 0NW5I    
portfolios, but none of them can account for the per-
%HWD 60%     formance patterns we found. First, funds with higher
%HWD +0/    ± rankings are actually larger, with winner funds having an
%HWD 50: ± ± ± ± average AUM of $271 million, about $80 million more
%HWD &0$ ± ± ± ± than those in the loser portfolio. This is likely a result of
positive inf lows during the period used for ranking, in
Notes: Funds are assigned into quartile portfolios Q1–Q4 every which higher ranked funds exhibited better performance.
December based on their MMS scores computed over the previous 12
months. Funds termed winner and loser are ranked in the top and bottom Hence, the size effect is not the driver of winner port-
25% of funds, respectively. The characteristics are equally averaged across folios’ superior performance. Second, funds with higher
funds within each quartile portfolio based on previous year-end values. rankings are typically younger (the average age of funds
They are further averaged across all years from 1997 to 2019. Regres-
sion betas to Fama–French five factors are computed using an expanding
in the winner portfolio is 5.89 years compared with 6.81
window starting from the first available month since 1996 with a min- years for those in the loser portfolio), but the difference is
imum of 12 months of data. Regression betas are computed for each fund statistically insignificant—in particular among the Q1–
individually. Their values are averaged within each quartile portfolio and Q3 portfolios, despite the performance difference among
are then averaged across different years.
them. Third, funds with higher rankings also tend to
Source: HFR, Ken French Data Library, Barclays Research.
charge higher fees, consistent with the idea that they
possess higher skill, but not sufficiently high to equate

12   Identifying Performance Persistence in Hedge Funds Using a Measure of Manager Skill Manager Fund Selection 2020
 Exhibit 13
Correlations of Winner/Loser Portfolio with Major Markets

&RUUHODWLRQRI0RQWKO\5HWXUQV




Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.





±

±
+)5, +('* (TXLWLHV &RPPRGLWLHV 5DWHV &UHGLW

4 /RVHU 4 :LQQHU

Notes: HFRI and HEDG are the HFRI Composite Index and Credit Suisse Hedge Fund Index, respectively; equities is the S&P 500 Total Return
Index; commodities is the Bloomberg Commodity Index; rates and credit are represented by the 10-year Treasury future and the Bloomberg Barclays US
High Yield Index total returns. Funds are assigned to quartile portfolios Q1–Q4 every December based on their MMS scores computed over the previous
12 months. Funds termed winner and loser are ranked in the top and bottom 25% of funds, respectively. The exhibit shows the correlation between each
of the portfolios postformation monthly returns and those of the respective index using data between January 1997 and December 2019.
Source: Bloomberg, HFR, Barclays Research.

the net-of-fee performance across portfolios in light of Rebalancing frequency. The results presented
the lack of variation in the industry fee structure. Finally, so far were based entirely on a 12-month portfolio
funds in different quartile portfolios show little variation rebalancing frequency. The choice of rebalancing
in loadings on the Fama–French five factors (Fama and frequency took into account practical limitations (e.g.,
French 1993, 2015) except the market factor (Mkt-Rf ). restrictions on redemptions and subscriptions) and the
In fact, average loadings on the size (SMB), value (HML), possibility of significant changes in a fund, such as in its
profitability (RMW), and investment (CMA) factors strategy or key personnel.
across Q1–Q4 portfolios are all very close to zero. Funds How sensitive are our findings to the rebalancing
in the winner portfolio have relatively smaller loading on frequency? Exhibit 14 presents the performance statistics
the market factor compared to funds assigned to other shown in Exhibit 6 when portfolios instead are rebal-
quartile portfolios, with an average loading almost half anced every 3 or 18 months, along with the original
that in the loser portfolio. Because the market has been results. The exhibit suggests that a higher rebalancing
mostly f lat or going up between 1997 and 2019, the frequency magnifies the dispersion among the portfolios
smaller market factor loading of winner funds cannot in terms of returns and volatilities and thus Sharpe ratios.
explain their superior performance. For example, in the case of quarterly rebalancing, the
Systematic risk. If the performance difference Sharpe ratio of the winner portfolio increases from 1.23
between the loser and winner portfolios is driven at to 1.47, whereas that of the loser portfolio decreases from
least in part by different exposures to systematic risk, the 0.42 to 0.26. This supports the view that the MMS scores
correlations of the two portfolios with major asset classes are informative and thus that using them with more
should be different. However, as Exhibit 13 indicates, timely information would lead to more effective ranking
the correlations are remarkably similar, suggesting that of funds. It is important to note, however, that the rela-
our findings are not driven by systematic exposures. In tion between past rankings and future performance was
fact, the largest difference is in comparison to the S&P preserved even when portfolios were rebalanced over a
500 Index, with which the loser portfolio actually has a relatively long horizon of 18 months.
higher correlation than the winner portfolio (0.82 versus Data. Because reporting by hedge funds to a
0.72, respectively). commercial database is voluntary, they often choose to
Manager Fund Selection 2020 The Journal of Portfolio Management   13
 Exhibit 14 Exhibit 15
Effects of Rebalancing Frequency, 1997–2019 Effects of Inclusion Bias on Performance of Quartile
Portfolios, May 1997–December 2019
4 4
/RVHU 4 4 :LQQHU 4 4
3DQHO$0RQWK5HEDODQFLQJ /RVHU 4 4 :LQQHU
0HDQ5HWXUQ PR     3DQHO$8VLQJDOO$YDLODEOH'DWD
9RODWLOLW\ PR     $YHUDJH5HWXUQ PR    
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

WK3HUFHQWLOH  ± ± ± ± 9RODWLOLW\ PR    
:RUVW0RQWK  ± ± ± ± 3HUFHQWLOH  ± ± ± ±
0D[LPXP'UDZGRZQ  ± ± ± ± :RUVW0RQWK  ± ± ± ±
6KDUSH5DWLR $QQ     0D[LPXP'UDZGRZQ  ± ± ± ±
3DQHO%7KUHH0RQWK5HEDODQFLQJ $QQ6KDUSH5DWLR    
0HDQ5HWXUQ PR     3DQHO%8VLQJ3RVWLQFOXVLRQ'DWD
9RODWLOLW\ PR     $YHUDJH5HWXUQ PR    
WK3HUFHQWLOH  ± ± ± ± 9RODWLOLW\ PR    
:RUVW0RQWK  ± ± ± ± 3HUFHQWLOH ± ± ± ±
0D[LPXP'UDZGRZQ  ± ± ± ± :RUVW0RQWK ± ± ± ±
6KDUSH5DWLR $QQ     0D[LPXP'UDZGRZQ ± ± ± ±
3DQHO&0RQWK5HEDODQFLQJ $QQ6KDUSH5DWLR    
0HDQ5HWXUQ PR    
9RODWLOLW\ PR     Notes: Funds are assigned into quartile portfolios Q1–Q4 every year
WK3HUFHQWLOH  ± ±
based on their MMS scores computed over the previous 12 months.
± ±
Winner and loser funds were ranked in the top and bottom 25% of funds,
:RUVW0RQWK  ± ± ± ± respectively. Performance statistics are calculated based on postformation
0D[LPXP'UDZGRZQ  ± ± ± ± returns between May 1997 and December 2019 because the HFR data-
6KDUSH5DWLR $QQ     base includes information on each fund inclusion date as far back as May
1996. To make results comparable, we use the same rebalancing dates and
Notes: Funds are assigned into quartile portfolios Q1–Q4 every 3, 12, sample period for the results in Panel A.
or 18 months based on their MMS scores computed over the previous Source: HFR, Barclays Research.
12 months. Winner and loser funds were ranked in the top and bottom
25% of funds, respectively. Performance statistics are calculated based on
postformation returns. The results indicate that the effect of the inclusion
Source: HFR, Barclays Research. bias in absolute terms is material (the Sharpe ratio of the
winner group drops from 1.26 to 0.88 after correcting
initiate it after a streak of good performances. As a result, the bias), but the decline is quite uniform. In particular,
using return data prior to the date of inclusion in the the drop in Sharpe ratio for the loser portfolio is 0.26,
database has been shown to generate upward-bias esti- which is similar to the 0.38 drop in the winner portfolio.
mates of hedge funds returns. Although our objective Consequently, the performance difference between the
is not to evaluate the performance of hedge funds as winner and loser portfolio is almost unchanged.
an asset class, the inclusion bias may still inf luence our
results if its effect on performance is not uniformly dis- CONCLUSION
tributed across portfolios.
To eliminate the possible effect of inclusion bias, Hedge funds are an increasingly important asset
we eliminate return data prior to the funds’ inclusion class but present unique challenges for investors. Owing
date.9 Exhibit 15 compares the performance statistics to their variety and complexity, simple measures using
of the Q1–Q4 portfolios based on the full dataset with past returns offer limited insight into their future perfor-
those using only the postinclusion returns. mance and are unable to identify managers who exhibit
persistent skill over time.
9
The earliest available fund inclusion date in the HFR database We discuss some of the challenges in identifying
is May 1996. As a result, the first rebalance date (with annual rebal- performance persistence in hedge funds and introduce
ancing) was April 30, 1997, instead of December 31, 1996. All subse- MMS—a new measure of a manager’s skill that addresses
quent rebalancing was at the end of each April instead of December.

14   Identifying Performance Persistence in Hedge Funds Using a Measure of Manager Skill Manager Fund Selection 2020
 these difficulties. We form portfolios of hedge funds Brown, S. J., W. N. Goetzmann, and R. G. Ibbotson.
based on MMS scores and find that funds in the top 1999. “Offshore Hedge Funds: Survival and Performance
quartile of MMS scores have a 62% probability of 1989–1995.” Journal of Business 72 (1): 91–117.
being ranked above the median in the following year.
Furthermore, an equally weighted portfolio of these Cook, R. D. 1977. “Detection of Inf luential Observations in
Linear Regression.” Technometrics 19 (1): 15–18.
funds delivered a Sharpe ratio of 1.23 over a 23-year
period since 1997, compared with 0.42 for the portfolio Edwards, F. R., and M. O. Caglayan. 2001. “Hedge Fund
of funds with the lowest 25% of MMS scores over the
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

Performance and Manager Skill.” Journal of Futures Markets

same period. 21 (11): 1003–1028.
MMS scores are computed for individual funds and
can be easily applied to portfolios composed of multiple Fama, E. F., and K. R. French. 1993. “Common Risk Fac-
funds by aggregating the scores. Because the measure tors in the Returns on Stocks and Bonds.” Journal of Financial
is probabilistic in nature and the empirical success rates Economics 33 (1): 3–56.
can be estimated on average (i.e., the likelihood a fund
with a certain ranking is above a certain percentile of the ——. 2015. “A Five-Factor Asset Pricing Model.” Journal of
population in the following period), an investor can use Financial Economics 116 (1): 1–22.
MMS scores to compute the probability that a portfolio
Fung, W., and D. A. Hsieh. 2004. “Hedge Fund Benchmarks:
of hedge funds will outperform its peers, given its cur-
A Risk-Based Approach.” Financial Analysts Journal 60 (5):
rent composition. 65–80.

REFERENCES Harvey, C. R., Y. Liu, and H. Zhu. 2016. “… and the Cross-
Section of Expected Returns.” Review of Financial Studies
Agarwal, V., and N. Y. Naik. 2000. “Multi-Period Perfor- 29 (1): 5–68.
mance Persistence Analysis of Hedge Funds.” Journal of Finan-
cial and Quantitative Analysis 35 (3): 327–342. Jagannathan, R., A. Malakhov, and D. Novikov. 2010. “Do
Hot Hands Exist among Hedge Fund Managers? An Empir-
Ambastha, M., and A. Ben Dor. “Barclays Capital Hedge ical Evaluation.” Journal of Finance 65 (1): 217–255.
Fund Replicators in 2009: Performance Review and Anal-
ysis.” Barclays Capital, 2010. Joenvaara, J., M. Kauppila, R. Kosowski, and P. Tolonen.
2019. “Hedge Fund Performance: Are Stylized Facts Sensi-
Ammann, M., O. Huber, and M. Schmid. 2013. “Hedge tive to Which Database One Uses?” Critical Finance Review
Fund Characteristics and Performance Persistence.” European (forthcoming).
Financial Management 19 (2): 209–250.
Kosowski, R., N. Y. Naik, and M. Teo. 2007. “Do Hedge
Bares, P. A., R. Gibson, and S. Gyger. 2003. “Performance Funds Deliver Alpha? A Bayesian and Bootstrap Analysis.”
in the Hedge Fund Industry: An Analysis of Short- and Journal of Financial Economics 84 (1): 229–264.
Long-Term Persistence.” The Journal of Alternative Investments
6 (3): 25–41. Liang, B. 1999. “On the Performance of Hedge Funds.”
Financial Analysts Journal 55 (4): 72–85.
Ben Dor, A., L. Dynkin, and A. Gould, “The Nature of
Hedge Fund Alpha.” Global Relative Value, Lehman Mitchell, M., and T. Pulvino. 2001. “Characteristics of Risk
Brothers, March 20, 2006. and Return in Risk Arbitrage.” Journal of Finance 56 (6):
2135–2175.
Ben Dor, A., R. Jagannathan, and I. Meier. 2003. “Under-
standing Mutual Funds and Hedge Funds Styles Using Teo, M. “Does Size Matter in the Hedge Fund Industry?”
Return-Based Style Analysis.” Journal of Investment Manage- Working paper, Singapore Management University, 2009.
ment 1 (1): 94–134.

Ben Dor, A., R. Jagannathan, I. Meier, and Z. Xu. 2012. To order reprints of this article, please contact David Rowe at
“What Drives the Tracking Error of Hedge Fund Clones?” d.rowe@pageantmedia.com or 646-891-2157.
The Journal of Alternative Investments 15 (2): 54–74.

Manager Fund Selection 2020 The Journal of Portfolio Management   15

 What Drives the Tracking Error of Hedge Fund
ADDITIONAL READING
Clones?
A rik B en D or , R avi Jagannathan, Iwan M eier ,
Performance in the Hedge Funds Industry and Z he Xu
The Journal of Alternative Investments
An Analysis of Short- and Long-Term Persistence
https://jai.pm-research.com/content/15/2/54
P.-A. Barès, R ajna Gibson, and S. Gyger
The Journal of Alternative Investments ABSTRACT: Hedge fund clones provide a liquid, efficient, and
https://jai.pm-research.com/content/6/3/25 transparent alternative to investing in hedge funds. As a group, how-
Downloaded from https://jpm.pm-research.com by Flavio Ferreira on March 2, 2020. Copyright 2020 Pageant Media Ltd.

ever, their recent performance has been disappointing, despite the

ABSTRACT: In this study, we analyze the performance persis-
large variation in the replication methodologies used. The author
tence of hedge funds over short- and long-term horizons. Using a
investigates hedge fund clones’ tracking errors and finds that contrary
non-parametric test, we first observe that the Relative Value and the
to common belief, the reliance on historical data to “reverse engineer”
Specialist Credit strategies contain the highest proportion of outper-
hedge fund allocation is not the primary cause. Instead, the author
forming managers. We next analyze the performance persistence of
identifies two important drivers of tracking errors of hedge fund clones.
portfolios ranked according to their average past returns. Persistence is
One is changes in marketwide liquidity levels as measured by the
mainly observed over one- to three-month holding periods but rapidly
basis between derivatives and cash securities. The second is biases in
vanishes as the formation or the holding period is lengthened. We
measuring the returns that arise due to attrition among hedge funds
finally examine long-term risk-adjusted returns persistence of hedge
that affect the performance of commonly used hedge fund indices.
fund portfolio within an APT framework. This leads us to detect a
Together, they account for about half of the variation in hedge fund
slight overreaction pattern that is more pronounced among the direc-
clones’ tracking errors over time.
tional hedge fund strategies.

16   Identifying Performance Persistence in Hedge Funds Using a Measure of Manager Skill Manager Fund Selection 2020