Upload
others
View
8
Download
0
Embed Size (px)
Citation preview
1
Performance Persistence and Flow Restrictions in Hedge Funds
Kee-Hong Bae*
York University
and
Junesuh Yi**
Dongguk University
This draft: January 2012
* Department of Finance, Schulich School of Business, York University, Toronto, ON, Canada, e-mail: kbae@schulich.
yorku.ca ** Department of Finance, Dongguk Business School, Dongguk University, Seoul, Korea, e-mail: [email protected]
2
Performance Persistence and Flow Restrictions in Hedge Funds
Abstract
This paper examines the impact of flow restrictions on superior persistence of performance in hedge funds.
We find that not only money outflow restrictions such as redemption notice period, lock period, and payout
period, but also money inflow restrictions such as minimum investment amount, closed-end, and closing to
individual investors are positively associated with winners‟ persistence. Out of two resources with respect to
persistence, outflow restrictions are observed to be a more important factor for persistence than inflow
restrictions.
We also document that managerial incentives have influence on winners‟ persistence. Thus, we explore which
feature is a more salient determinant for continuous excellent performance persistence. Interestingly, while
each of funds with higher restrictions and funds with higher incentives display higher ratio of persistence,
funds with lower incentives given higher restrictions exhibit higher ratio of persistence than funds with higher
incentives. This suggests that persistence is subject to flow restrictions rather than to incentives. The empirical
results in this paper are resonant with the theoretical prediction of performance persistence by Glode and
Green (2011) that reveal a determinant to affect persistence not from managerial skills but from unique
structure of hedge funds composed of limited and general partners.
Key words: Hedge fund, performance persistence, flow restriction, managerial incentive
JEL classification: G11, G23
3
I. Introduction
What is an essential source to hold superior performance in hedge fund? While the hedge fund
has been considered to be an alternative investment vehicle compared with traditional investment
instruments and many academic papers have examined the related issues, there is little known about
the concrete features of hedge funds generating continuously excellent performance. Under the
circumstances with little information or severe information asymmetry due to the unregulated
natures of hedge funds, the investors would of course wish to invest in trustworthy funds
consistently performing well.
In fact, several previous researches have documented the performance in hedge funds. On the
contrary, only a few studies have attempted to examine the persistence of performance. Even most
of them investigate whether performance persistence exists in hedge fund industry, but hardly pay
attention to the features of hedge funds showing persistence by individual hedge fund. Performance
is not necessarily identical to performance persistence. It is possible that the performance of a
particular fund is inferior in a subsequent period even if the performance of a fund delivers a
superior performance in a previous period. Hence, funds may present differential extent of
persistence at different estimation periods of returns. Furthermore, most of literature investigates
persistence mainly focused on investment managers as mutual fund studies do. It is well known that
performance of mutual funds depends on stock picking skill and timing ability of fund managers.
However, hedge funds have natures distinctive from mutual funds in many aspects. Aside from
basic differences such as unregulated investment companies, usage of various investment
techniques including short selling, derivatives, or leverage, and pursuing absolute returns, hedge
funds have a unique, organized structure as a limited partnership. Investors generally contribute
capital as limited partners and investment managers operate funds serving as general partners.
Accordingly, funds are mostly raised from large and accredited institutional investors and fund
managers receive compensation based on their performance like in a limited partnership corporation.
In this way, money flows and compensation of hedge funds particularly present pretty differentiated
structures compared with other investment vehicles.
Money flows are typically restricted not only at set up but at withdrawal stage as funds are
solicited from limited investors. Regarding money inflows to hedge funds, most funds require the
minimum investment amount of $1 million beyond the qualification on net income for individual
investors and on net worth for institutions. Hedge funds even may not be open to individual
4
investors. The funds also operate in a closed-end form that does not accept further investment after
inception, maintaining the limited number of shares. Besides, there are more restrictive terms
associated with money outflows from funds. Investors cannot withdraw their money during lockup
period when investors are initially required to hold their money in the fund. Although the lockup
period ends, investors are not free to exit their money due to redemption period and notice period.
Hedge funds do not redeem investors‟ shares on a daily basis like mutual funds but on a quarterly or
yearly base. Therefore, investors who desire to take the money out should notify hedge fund in
advance of redemption period. If investors want to receive any return from their investment in the
hedge funds‟ capital, they are also required to wait during payout period. Such the various
restrictive provisions related to money inflows and outflows of hedge funds lead to share illiquidity
for investors, compared to mutual funds where investors can buy and sell their shares whenever
they want.
From the viewpoint of general partners in limited partnership, hedge fund managers are
rewarded by performance-based fee contracts. The standard fee arrangement of hedge fund industry
is “2 and 20”. Namely, hedge fund managers receive a fixed management fee corresponding to 2
percent of assets under management and an incentive fee of 20 percent of the profits above a
predetermined rate. This rate is named as hurdle rate generally using Treasury bill rate plus a spread.
Because fund managers can have incentive fee only if they perform higher than the hurdle rate, this
asymmetric compensation contract is called as option-like structure. Another important provision of
compensation for general partners is high water mark. If fund managers make a loss in some period,
they can receive the incentive fee only after they have recovered the highest net asset value
previously seen at the end of the fiscal year.
Under the circumstances of restrictive money flows and incentive-centered fee structures on
distinguished organizational form, the performance persistence of hedge funds can be justified
through other resources rather than the ability of fund managers. In particular, share illiquidity
caused by money flow restrictions may be more likely to affect performance persistence. Investors
can neither easily withdraw their money after exhibition of worse performance nor make new
investment in the fund after confirmation of better performance. These rigid fund flows allow fund
managers to establish a more creative and innovative trading strategy and more flexible portfolio
construction, possibly resulting in superior performance persistence.
Hence, our paper conjectures that the magnitude of flow restrictions may be one of the
determinants for fund presenting good performance persistence and investigates the relationship
5
between flow restrictions and performance persistence. We separate flow restriction provisions into
inflow and outflow restrictions, and explore the impact of each of the flow restrictions on
performance persistence. We also understand the likelihood that the incentive structure among
several characteristics of funds can influence on performance persistence or that the incentive
structure and flow restriction can simultaneously affect persistence. Thus, we analyze the
relationships among these three variables.
As the results of empirical tests, flow restrictions are positively associated with performance
persistence. Outflow restrictions in particular are observed to have more impact on performance
persistence. As performance measures for performance estimation, we use parametric method by
regression of excess returns on risk factors including liquidity premium and non-parametric method
by contingency table. We also find that funds with aggressive incentive structure exhibit superior
performance persistence and that managerial incentives are highly correlated with flow restrictions.
Consequently, we discover that flow restrictions are a more important determinant than managerial
incentives to retain better performance.
This study is in line with Glode and Green (2011) in that the source of performance persistence
in hedge fund is not drawn from manager‟s ability but from innovative trading strategies or
techniques. They provide a theoretical model that rationalizes the performance persistence of hedge
fund based on hedge funds‟ unique organizational form differentiated from other investment
vehicles. Performance persistence of hedge funds can be explained through information secrecy
between limited partners and general partners. According to them, fund managers need to share
informational rents with the incumbent investors to prevent them from revealing information about
trading strategies to other managers. Consequently, general partners and limited partners need to
restrict fund flows even when expected profitability is high to maintain performance persistence.
We consider the information secrecy between limited partner and general partner in their paper as
flow restrictions of fund managers to investors. This is because the effect to performance
persistence with minimization of money movement by keeping up confidentiality about valuable
information is almost identical the flow restrictions on agreement between funds and investors‟
contract. Therefore, the empirical test results of this paper are resonant with the theoretical
prediction of performance persistence in hedge funds by Glode and Green (2011).
To our knowledge, this is the first research about the relationship between performance
persistence and flow restrictions in hedge funds. Previous literatures have covered in a fragmentary
manner compared to this topic such as share restriction and performance, restrictions and incentives,
6
and persistence and estimation intervals. The second contribution of this paper is to figure out the
source of performance persistence reflecting specific organizational form composed of limited and
general partners on hedge funds. Thirdly, this paper takes into account restrictions and incentives
simultaneously to estimate their impact on performance persistence and demonstrates which factor
is salient. Finally, we scrutinize money inflow restrictions such as close to public or closed-end
provisions as well as money outflow restrictions.
The paper proceeds as follows. Section 2 discusses the background of objectives and the
purpose of the paper as well as existing related literature. Section 3 describes the data and
methodologies including parametric and non-parametric method. Section 4 presents the empirical
results and Section 5 concludes the paper.
II. Background
Although a number of academic papers on hedge funds have been recently published, most
studies are incorporated to find out risk factors to correctly estimate performance. Considering the
interest of investors, practitioners, and academicians, and the ripple effect to hedge fund industry, it
is a wonder that there is no comprehensive study that investigates determinants for performance
persistence. Of course, there are several studies related to persistence, share restrictions and
performance, and managerial incentives and performance respectively.
In terms of performance persistence of hedge funds, most papers have examined it focusing on
fund managers‟ ability. These studies provide the mixed results depending on types of estimated
returns, evaluation period, and superior or inferior funds. Especially, it is suggested that there are
several layers with respect to the existence periods of persistence. While the early researches insist
that performance persistence happens at quarterly level. (Brown, Goetzmann, and Ibbotson (1999),
Agarwal and Naik (2000), Bares, Gibson, and Gyger (2003), and Baquero, Horst, and Verbeek
(2005)), the recent studies assert that the period being performance persistence is longer than
quarterly horizon. Kosowski, Naik, and Teo (2007) provide evidence it over a 1 year horizon and
Jaganathan, Malakhov, and Novikov (2010) report it on even 3 years. Kosowski et al. (2007)
explain the reason that prior researches do not find long-term persistence is because they
imprecisely measure performance too much relying on frequency probability of short-term period.
Taking a step forward, Jaganathan et al. (2010) contend that estimation period of persistence should
7
be as at least 3 years due to the specific circumstances related to share illiquidity such as lockup,
redemption, and notice period.
Regarding flow restrictions, the literature mainly concentrates on share restrictions related to
money outflow. Ding et al. (2009) and Baquero and Verbeek (2009) examine the impact of share
restrictions on the flow-performance relationship and find conflicting results. Ding et al. (2009) find
that flow-performance relation exhibit concave or convex form depending on share restrictions. In
contrast, Baquero and Verbeek (2009) demonstrate a linear relationship between flow and
performance irrelevant to share restrictions. Aragon (2007) examines the relation between share
restriction and fund return. He presents that funds with restrictions deliver higher excess return than
those without restrictions and these abnormal returns are disappeared after controlling the
restrictions. He interpret that the differences of excess return between two fund groups are because
the liquidity reservoir caused by share restrictions allow fund managers to operate their assets more
freely. On the contrary, Sadka (2010) demonstrate that liquidity risk is an important determinant in
hedge fund returns but the impact of liquidity risk on returns is independent of the share restrictions,
inconsistent with Aragon (2007)
The empirical analyses of relationship between managerial incentives and performance present
conflict results. While Ackermann, McEnally, and Ravenscraft (1999) and Edwards and Caglayan
(2001) provide evidence that funds with higher incentive fees structure exhibit better performance,
Brown et al. (1999) demonstrate that performance is not affected by incentive fees. Also Panageas
and Westerfield (2009) discover that fund managers would not place large weight on risky assets, in
spite of option-like compensation arrangement by high water mark contract.
The influence of both managerial incentives and restrictions on performance is documented by
Agarwal, Daniel, and Naik (2009) that find that a higher degree of managerial discretion as well as
greater managerial incentives deliver superior performance. The level of incentives are calculated
based on manager‟s option delta, managerial ownership proxied by cumulative value of the
incentive fee reinvested on fund, and high water mark provision, and the degree of managerial
discretions are estimated by lockup, notice, and redemption period. It is more likely that they
conduct comprehensive analyses about managerial discretion and incentives connected with
performance. This paper provides additional contribution to their study in that we estimate
persistence not performance, consider inflow restrictions as well as outflow restrictions, and reveal
more salient one factor of the two, restrictions and incentive, to keep performance persistence.
Meanwhile, Brown et al. (2008) analyze other characteristics of hedge funds such as capital
8
structure (leverage, margin) and governance structure (75% ownership, domestic direct corporation,
indirect owner) for examination of the value of disclosure by SEC requirements. They find that
operational risk factors are correlated with lower leverage and higher ownership. In other words,
funds with regulatory problem have less leverage and more concentrated management structure.
Aragon, Liang, and Park (2011) exploit the differences of capital flow and performance between on-
and off-shore funds and find that risk adjusted abnormal returns are observed in stand-alone onshore
funds more imposed share restrictions than offshore funds.
III. Data and Methodology
3-1. Data
We use the TASS hedge fund data that is known as the most comprehensive database comparing
to other hedge fund data venders1. The TASS database provides monthly returns, net asset value
(NAV), asset under management (AUM), fund strategies, fund characteristics related restrictions
(open or close to public, open-end or closed-end, minimum investment amount, redemption notice
period, lockup period, payout period), and other fund characteristics (management fee, incentive fee,
high water mark, leverage, margin, domicile country, management company location, and so on).
Our Sample period extends from January 1994 to March 2008 including dead funds as well as
lives. To alleviate survivorship bias, the period begins with 1994 because the TASS database
supplies the information of dead funds after January 1994 named as defunct fund. As of March
2008, the number of sample funds is 10,838 consisting of 6,327 live and 4,511 dead funds.
Out of total sample funds, we basically exclude funds that report returns as foreign currency and
gross of all fees, and whose primary category is „fund of hedge funds‟ and „undefined‟. Composed
of portfolios with other hedge funds, „fund of hedge funds‟ exhibit different features from individual
hedge funds in fee structure, trading strategy, risk management, and due diligence cost. Afterwards,
we employ two ways, non-parametric and parametric method as performance measure for
persistence. In non-parametric method, performance persistence of funds is determined by
comparing style-adjusted return of each year for two years. On the other hand, in parametric method
„winners‟ or „losers‟ are resolved using abnormal returns computed by regression analysis with 30
monthly returns.
1 Liang (2000)
9
Therefore, for the non-parametric method based on yearly interval, we eliminate funds with less
than 12 months of return history and with inception date after 2007. For parametric test based on 3
years interval, we discard funds with inception date after 2003 and with less than 60 months of
return data. Moreover, we delete the observations of funds with less than 6 monthly returns a year
and with missing value or zeros of asset under managements at the end of year in both non-
parametric and parametric tests.
In addition, we remove the observations of inception and termination year of funds in order to
mitigate backfill bias that funds report returns prior to their added date to database and liquidation
bias that returns are not reported prior actually closing the funds. Consequently, through the
combination of selection procedure for sample and observation controlling several biases frequently
happened in hedge funds, we finally extract 3,756 sample (2,276 live and 1,486 dead funds) and
17,365 yearly observations for non-parametric test, and 2,001 (1,035 live and 966 dead funds)
sample and 211,258 monthly observations for parametric test.
Table 1 reports descriptive statistics2. The average (median) yearly return of entire sample is
12.23% (9.92%) and flows increase 45.74% a year. They show quite different percent by strategy.
While emerging market achieves 17.49% yearly returns on average as the highest among strategies,
dedicated short bias delivers only 1.85% of returns as the lowest. On the other hand, dedicated short
sale records the smallest fund size but largest fund flows. Long/short equity hedge accounts for the
most number of samples.
Insert Table 1 about here
3.2 Methodology
To investigate the impact of flow restrictions on persistence, once we identify funds with
superior persistence using non-parametric and parametric methods. Next, we examine the
relationship between restrictions and persistence using restriction difference analyses among funds
classified by the extent of persistence drawn by non-parametric method, and using regression
analyses of persistence drawn by parametric method on restrictions. After exploring other
characteristics beyond restrictions and extracting uncovered character to affect persistence, we
examine the relationship between restrictions and uncovered character. Finally, we demonstrate
2 This is about samples drawn from non-parametric method. The patterns of descriptive statistics for samples drawn from
parametric method are very close to those for non-parametric method.
10
which feature, restrictions or uncovered character, is salient factor to influence on persistence in
hedge funds through comparing winners‟ ratio among the groups classified by restrictions and
incentives.
3.2.1 Performance persistence
Performance persistence is basically estimated by comparing the performance measures in
previous period with the performance measures in subsequent period. We use two different
performance measures, non-parametric method based on contingency table and parametric on
regression.
For non-parametric test, we calculate style-adjusted returns of funds. Unlike other investment
vehicles, hedge funds employ the various trading strategies, showing that return distributions of
funds within a particular fund strategy are remarkably different from those within other fund
strategy. Hence, we need to adjust returns of individual fund with returns of funds within same
strategy. Style-adjusted return of a fund is calculated by deduction of average return of all funds
employing identical strategy with a fund from a fund‟s raw return. Based on style-adjusted returns,
we construct contingent table of winners and losers. If a style-adjusted return on yearly base of a
fund is higher than median return of all funds employing same strategy with a fund, it becomes a
winner (W), otherwise, it is a loser (L). Therefore, funds denoted as winners for two consecutive
periods (WW) are classified by funds with superior performance persistence3. We also extend
evaluation period to multi-period, three consecutive years. In multi-period analysis, funds being
winners for three periods (WWW) are identified as funds existing performance persistence.
Panel A of Table 2 describes summary statistics of yearly returns on both previous and
subsequent periods by three classes (WW, WL or LW, and LL) in two periods persistence. The ratio
of funds displaying superior performance persistence is 28.59% out of total observed funds. Funds
displaying superior persistence achieve yearly returns of 14.06% and 12.94% on each period.
Overall returns of the second period are lower than returns of the first period in all classes, implying
that returns in hedge fund industry have been decreasing in recent years. On the other hand,
skewness and kurtosis of funds with superior persistence are higher in the second period, suggesting
that funds with extreme positive returns have increased.
3 If a fund is losers in two consecutive periods, it is denoted as LL. WL denotes winners in previous period but losers in
subsequent period, and LW denotes the reverse.
11
Insert Table 2 about here
The relative performance persistence by non-parametric test cannot necessarily provide genuine
superior persistence of funds because the entire peer group may have inferior performance.
Therefore, we estimate absolute performance persistence using performance measure by the
multifactor model. For this parametric test, we employ modified Fung and Hsieh model (hereafter
MFH model) that liquidity factor is added to Fung and Hsieh (2004) seven factor model as in
equation (1).
(1)
where is the monthly return of a fund i in excess of one month T-bill rate. EMKT (equity
market) is S&P 500 minus risk free rate. SIZE (size spread) is Wilshire small cap 1750 minus
Wilshire large cap 750 till the end of 2006 and Russell 2000 minus S&P 500 thereafter. BMKT
(bond market) is the monthly change in the 10 year treasury constant maturity yield. CRED (credit
spread) is the monthly change in the Moody's Baa yield less 10-year treasury constant maturity
yield. TFB (bond trend-following factor), TFCR (currency trend-following factor), and TFCM
(commodity trend-following factor) are the return of PTFS (primitive trend-following strategy) of a
bond lookback straddle, of a currency lookback straddle, and of a commodity lookback straddle,
respectively4.
Meanwhile, is Sadka (2006) liquidity factor on permanent-variable price impact
component extracted from the Trades and Quotes (TAQ) tick-by-tick data. Liquidity is important
factor to estimate performance of hedge funds especially associated with share restrictions. Aragon
(2007) indicates the reason that fund returns with share restrictions are higher than those without
restrictions is due to liquidity premium by share illiquidity. With respect to his finding, however,
Sadka (2010) argues that the liquidity premium could be controlled by using existing liquidity risk
factor.
4 The three trend-following factors of Fung and Hsieh (2001) are provided by David Hsieh web site at
http://faculty.fuqua.duke.edu/~dah7/DataLibrary/TF-FAC.xls
12
We calculate alphas over consecutive non-overlapping 3 year intervals. That is, to be estimated
performance persistence, funds have at least 60 monthly returns5. This is a quite long time and may
lead to likelihood of survivorship bias. However, hedge funds commonly restrict money outflows
by provisions such as lockup and redemption notice period for 2 or 3 years. Moreover, it is observed
that the duration of persistence has been lengthened by developing new estimation methods with
respect to short-term period returns in recent studies. In addition, because parametric test evaluates
not relative persistence but absolute persistence, it is hard to say that survivorship bias is
problematic. In other words, it does not matter whether funds survive in the long time on the
parametric test since this paper examines the existence of persistence on individual funds not on
entire hedge fund industry.
As the result of regression analysis in equation (1), if an alpha of a fund is positive and
statistically significant in one period, it becomes a winner (WW)6. To be classified as the funds
existing performance persistence, funds should display positive and statistically significant alphas in
two consecutive periods (WWWW)7.
Panel B of Table 2 displays summary statistics of monthly returns on both previous and
subsequent periods by five classes in two periods persistence8. The number of funds displaying
superior performance persistence is 1,313 out of 8,005 fund observations, which is only 16.4% of
the total observed funds. Regardless of existence of persistence, average returns of funds are
commonly lower in second period as in panel A. The percentage of funds with persistence is lower
than 22% for alpha persistence by April 2005 of Jananathan (2010). This is more likely to be caused
by recently decreasing rate of returns over the entire hedge fund industry as shown in panel A and B.
Moreover, liquidity risk factor used in this study is attributable to depreciate the abnormal return as
in Sadka (2010). Average yearly return of funds with superior performance is 12.48% in the first
and 11.04% in the second period, which is a bit lower than those in panel A.
5 Funds have minimum 30 monthly observations in each period to calculate alpha. 6 If an alpha of a fund is positive but insignificant, the fund is denoted as W, negative but insignificant, it is L. and
negative and significant, then it is LL. 7 As the results of regression analyses for two estimation periods, we have nine different outcomes depending on signal
and significance of alphas, WWWW, WWW, WW, WWL(=LWW), WWLL(=WLLW= LLWW), WLL(=LWW), LL, LLL,
and LLLL. 8 We classify nine cases in drawn from note 7 by five depending on the extent of performance persistence. WWWW:
alphas are positive and statistically significant in two consecutive periods, WWW and WW: alphas are positive and at
most one alpha is significant, LLL and LL: alphas are negative and at most one alpha is significant, LLLL: alphas are
negative and statistically significant in two consecutive periods, and the others (WWL, LWW, WWLL, WLLW, LLWW,
WLL, LWW): one of alphas is positive and the other of alpha is negative.
13
3.2.2. Persistence and restrictions
We estimate flow restrictions through six related features of funds. Open- or closed-end,
opening or closing to public, and minimum investment amount are used as proxies for inflow
restrictions and redemption notice period, lockup period, and payout period are used as proxies for
outflow restrictions. Summary statistics related to flow restrictions exhibit in Table 3.
Insert Table 3 about here
Median of redemption notice period and payout period are 30 days and 10 days, respectively.
On the contrary, the median lockup period is 0 month and 68.8% of funds do not impose lockup
period. Average (median) minimum investment amount is $8.64 ($0.5) millions. 58% of funds are
open-end and only 17% of funds allow individual investors to invest in the funds.
The relationship between performance persistence and flow restrictions is examined using
probit and ordered-probit regressions as in equation (2). Dependent variable is persistence and
explanatory variables are proxies representing flow restrictions. Size, flow, and volatility of funds
are included in explanatory variables for control purposes. Strategy dummies are also added to
control the variability of performance relying on fund category.
(2)
where is 1 if a fund i is a winner in previous and subsequent periods and is 0
otherwise in probit model. In ordered probit model, it ranges from 1 to 5 depending on the extent of
performance persistence9. equals 1 if a fund i is open-end and 0 if a fund i is closed-end.
equals 1 if a fund i is open to public and 0 otherwise. is the log of the
minimum investment amount of fund i. and are redemption
notice period (days), lockup period (months), and payout period (days) of fund i, respectively.
9 It equals 5 if a fund i is denoted as WWWW in note 8, 4 denoted as WWW and WW, 3 denoted as WWL, LWW,
WWLL, WLLW, LLWW, WLL, and LWW, 2 denoted as LL and LLL, and 1 denoted as LLLL.
14
is the log of estimated assets of fund i , is the money flows of fund i10
, and
is the standard deviation of monthly returns of fund i at the end of year t-211
. Meanwhile,
we witness that there are considerable missing value or zeros for estimated assets, implying that size
and flow variables might not be very accurate. Therefore, funds with the top and bottom 1% of them
are winsorized to control for outliers. is a strategy dummy that equals 1 if fund i
belongs to strategy s and equals 0 otherwise12
.
One of drawbacks in analyses by equation (2) is likely that the parameters of restriction
variables are extracted from the last available date of fund data, even if these variables may change
over time. However, hedge funds are observed to rarely alter their contract features throughout the
life. Ding et al. (2009) compare these features using two different version of the TASS database,
January 2001 and September 2005 and find that only 10% (9%) of individual hedge funds change
their subscription (redemption) period. Liang and Schwatz (2008) also find no more than 2% of
funds reform the incentive and management fee structure from 1998 to 2006.
Another possible shortcoming in equation (2) is that a particular fund may be able to have
multiple observations in dependent variable despite having an identical explanatory variable.
Therefore, we run probit and ordered probit regressions by clustering the observations by funds,
because clustered standard errors are able to correct for within fund correlation13
.
3.2.3 Persistence and other characteristics
The impact of other characteristics beyond flow restrictions on persistence is analyzed using
equation (3). Explanatory variables are classified by four categories, incentives, ownership, capital
structures, and locations. Incentives include incentive fee, management fee, and high water mark
provision. Even if management fee is devised to cover the operating cost of the managers, it can be
considered as one of the incentive features because management fee of larger funds can deliver a
substantial part of a manager‟s profit. Ownership category includes a variable whether principals
have money invested in funds. Using leverage and margin in leverage are included in capital
structure category. Location is composed of variables whether funds are on- or off-shore and
10 Flows are calculated by the following equation.
where is the assets under
management of fund i at the end of year t-1 and is the return for fund i during the year t. 11 t-2 is not a calendar year but a year prior to beginning of the persistence estimation because estimation period is three
years interval. For instance, if we estimate persistence of fund i from 1996 to 2001, t-2 becomes 1995. 12 To be full rank for the equation, the number of dummy variables for ten strategies is nine. 13 We also make clustering by two dimensions, fund and year at the same time using Cameron, Gelbach, and Miller
(2006).
15
whether management companies are located in big cities. Control variables used in equation (2) are
also employed in equation (3).
where is as defined in equation (2). and are
management fee and incentive fee charged by fund i , respectively. equals 1 if
fund i has a high water mark provision, and equals 0 otherwise. equals 1 if
principals of management company of fund i have invested in a fund i and equals 0 otherwise.
is 1 if a fund i use leverage, and is 0 if a fund i does not use leverage.
equals 1 if a fund i leverages using margin borrowing and equals 0 other wise. equals 1 if
a fund i is onshore and equals 0 if a fund i is offshore. equals 1 if headquarter of
management company of fund i is located in big city, and equals 0 otherwise14
. ,
, , and are as defined as in equation (2).
3.2.4 Persistence, restrictions, and incentives
To analyze the relationship between persistence and both of restrictions and incentives at the
same time, we use two kinds of ordered probit regression models whose dependent variable is the
extent of persistence. As explanatory variables, one includes scaled inflow restrictions, outflow
restrictions, and incentives, and the other contains a classification variable representing the extent of
restrictions and incentives.
We scale the parameters of restriction and incentive in a fund from 0 to 1 by percentile of
probability distribution on each variable. Because inflow restrictions, outflow restrictions, and
incentives include three variables, scaled them provide the range from 0 to 3, respectively, implying
that the higher scales score is the stronger restrictions or the more incentives. For binary distribution
of open end and open to public variables, scale score is endowed with 1 if a fund is closed-end or
does not raise money from individual investors.
14 Big cities are defined as cities where more than 30 funds‟ management companies are located and are following 8 cities;
New York, London (UK), San Francisco, Hamilton (Canada), Greenwich, Boston, Chicago, and Stamford.
16
To investigate which factor between restrictions or incentives is more important to persistence,
we also conduct probit regression of persistence on the magnitude of restrictions and incentives. For
this, we construct portfolios based on the degree of restrictions and incentives. Funds are
categorized to be high (low) if scaled incentives or restrictions of funds are above (below) median
of total funds, resulting in four kinds of value in classification variable (HH, HL, LH, and LL). For
instance, if a fund whose incentives and restrictions are above median, then the fund is classified
HH. We create matrix columns using parameterization method for the classification variable and
specify less than full rank. Therefore, parameter estimates of each class by GLM (generalized linear
method) present the difference in the effect of each class compared to the last class.
IV. Empirical results
4-1. Performance persistence and flow restrictions
4.1.1. Non-parametric method
The results of difference analyses on flow restrictions among funds classified based on the
extent of persistence drawn by non-parametric method are shown in Table 4. We find obviously
clear results that the funds with superior performance persistence exhibit higher degree of flow
restrictions. In panel A analyzing two period persistence, the funds belong to WW display higher
minimum investment amount, more closed-end, less opening to public, and longer redemption,
payout, and lockup periods. Differences among classes of persistence are statistically significant in
all restriction variables15
. Restrictions of WW are substantially more rigid than those of LL. Funds
with persistence show 5.6 days longer redemption period, 4.3 days longer payout period and 1.4
months longer lockup periods than funds belongs to LL. The fund ratios of open-ended and opening
to public for WW are less than those for LL. In panel B that we extend the persistence estimation
period to three consecutive years, all restriction variables except „open to public‟ also present
differences with statistically significant among classes of persistence16
.
15 Basically we conduct significant test using GLM (general linear model) for difference analyses among funds classified
by persistence. To control heteroscedasticity that presents difference variances of restriction variables among classes for
persistence, we also use Welch‟s and Kruskal-Walli test with respect to variables showing heteroscedasticitic variance as a
result of Levin‟s test. The results using these tests are very close to those reported in the paper. We can provide results on
request. 16 In three periods persistence, we have four classes depending on the number of winners, WWW: winners in three
periods, WW (=WWL, WLW, LWW): winners in two periods, LL (=WLL, LLW, LWL): winners in one period, and LLL:
losers in three periods.
17
Insert Table 4 about here
Consistent with the above results, funds with persistence have more rigid flow restrictions
provisions when we use the scaled restrictions in lieu of individual restriction proxies in panel C
and D of Table 4. In two and three periods analyses, scaled inflow and outflow restrictions as well
as scaled total restrictions exhibit statistically significant differences between WW and LL or
WWW and LLL. Especially, the difference of outflow restrictions between two classes is bigger
than that of inflow restrictions. While the average scaled inflow restrictions of WWW and LLL in
three estimation periods are 1.79 and 1.66, the average scaled outflow restrictions of WWW and
LLL are 1.60 and 1.36, respectively, meaning that the difference of outflow restrictions is as almost
twice as that of inflow restrictions.
Table 5 shows flow restrictions and performance persistence of funds by trading strategies. As
we expect, in most strategies, the funds with persistence present the higher total flow restrictions as
well as inflow and outflow restrictions17
. In equity market neutral, event driven, fixed income,
global, and long/short equity strategies, funds of winners for two consecutive periods show
statistically significant differences from funds of losers in all three scaled restrictions. Especially,
the average of scaled total restrictions of WW is as much as 0.5 higher than that of LL in fixed
income and global strategies. The scaled outflow restrictions difference between WW and LL is
maximized in fixed income strategy as 0.35. Meanwhile, funds belong to emerging markets,
managed futures, and multi-strategy do not display the difference of scaled inflow restrictions.
Insert Table 5 about here
4.1.2. Parametric method
Subsequent to restrictions difference analyses among funds classified by relative performance
persistence drawn by non-parametric method, we investigate the relation impact of restrictions on
persistence estimated by absolute performance computed through parametric method. Table 6
shows the results of relation between flow restrictions and persistence drawn by MFH model in
equation (1).
17 We remove „dedicated short bias‟ among ten trading strategies due to the insufficient number of samples.
18
Insert Table 6 about here
In both simple/multiple probit and ordered probit regression analyses using equation (2)18
, we
find that restriction provisions have influence on performance persistence. In detail, minimum
investment amount, redemption notice period, and lockup period are positively associated with
superior performance persistence in all models. Payout period is also positively related to
persistence and open to public is negatively related to persistence as expectation in simple model.
These relationships exhibit clearer in ordered probit model. Except open-ended in simple probit,
and open-ended and open to public in multiple probit model, all restriction variables are observed to
affect performance persistence. Open ended and open to public variables provide negative sign as
expectation but insignificant parameter estimates in some models. It implies that outflow restriction
provisions are more influence on persistence rather than inflow restriction provisions.
By strategy, the impact of scaled inflow or outflow restrictions on persistence is estimated using
probit and ordered probit model that dependent variable is classes sorted on persistence as in Table
6 and explanatory variables are scaled inflow and outflow restrictions including control variables.
Inconsistent with results of non-parametric test, the scaled restrictions are positively associated with
persistence in not as many as strategies in Table 7. Only funds belong to event strategy present that
both inflow and outflow restrictions have significantly positive relations with persistence in probit
model. In ordered probit model, while fixed income, global, long/short equity hedge, managed-
futures, and multi-strategy exhibit significant estimates for scaled outflow restrictions, emerging
market and event driven show significant estimates for scaled inflow restrictions. In analyses with
respect to entire samples, we find that not only scaled outflow restriction but inflow restrictions
have significant influence on persistence.
As the results of test through both non-parametric and parametric tests, we reveal that
restrictions affect performance persistence. Moreover, outflow restrictions are more positively
associated with persistence of hedge fund, which is line with Aragon (2007) and Agarwal, Daniel,
and Naik (2009) that provide the evidence the impact of individual flow outflow restrictions such as
lockup period and redemption period on performance.
Insert Table 7 about here
18 Dependent variables are 0 or 1 for probit, and 1 to 5 as defined in note 9 for ordered probit model.
19
4.2 Performance persistence and fund characteristics
In this section, we explore the relationship between other characteristics beyond flow
restrictions and persistence. For this, we segregate other characteristics into four categories,
incentives, ownership, capital structure, and location. Table 8 displays the results of these analyses.
Insert Table 8 about here
We conduct probit /ordered probit regression of persistence on each category in model 1 to 4
and on all variables including four categories in model 5. In panel A for probit model of Table 8,
only high water mark among variables within incentive category in model 1 and 5 display positively
significant estimate, consistent with Agarwal et al. (2009). In addition, onoff variable belongs to
location category in model 4 and 5 exhibits significant estimate, consistent with Aragon et al. (2011)
that onshore funds deliver higher performance than offshore funds. No variables connected with
ownership and capital structure display significant estimate.
On the other hand, in ordered probit model of panel B, we find that all variables included in
incentive provisions in model 1 and 5 show positively significant estimates. Namely, funds with
persistence have more likely to impose high water mark, higher incentive fee, and higher
management fee. In model 5 including variables in all category, we cannot find any other variables
with statistically significant estimates except variables included in incentives. In turn, we suggest
that incentives including incentive fee, high water mark, and management fee may be determinants
to affect persistence of performance in hedge funds.
4.3. Persistence and Incentives
In the above sections, we observe that flow restrictions and incentives among characteristics
affect performance persistence of hedge funds. We examine the relationship between restrictions
and incentives in this section.
Panel A of Table 9 display the results of regression of restrictions on individual proxies
representing incentives. Dependent variables are scaled total, inflow, and outflow restrictions, and
explanatory variables are management fee, incentive fee, high water mark, and control variables.
While incentive fee and high water mark are observed to have positive relation with flow
restrictions, management fee have negative connection with them. It implies that funds with strict
20
flow restrictions are likely to enjoy more their incentives after satisfying with investors‟ expected
rate of returns. It also indicates that funds provide the opportunity for investors to choose
appropriate funds on their investment purpose, either higher returns attributable to an illiquidity of
their money by strict flow restrictions or lower returns with low management fee. In the analyses
that explanatory variables are scaled from 0 to 1 by percentile of probability distribution on each
incentive variables, the results are exactly same as those drawn from analyses using original figures.
Insert Table 9 about here
Panel B of Table 9 display correlation coefficient of scaled restrictions and scaled incentives.
Consistent with the above outcomes, scaled total, inflow, and outflow restrictions are positively
correlated with scaled incentive fee and scaled high water mark, and are negatively correlated with
scaled management fee.
4.4 Performance persistence, restrictions, and incentives
We recall that flow restrictions and managerial incentives are determinants for superior
performance persistence and restrictions are positively correlated with incentives. In this section, we
investigate which feature is more salient factor to affect persistence by comparing winner ratios on
groups constructed by the extent of restrictions and incentives, and by probit regression of
persistence on class variable whose value is determined by the degree of restriction and incentives.
Panel A of Table 10 displays the results of regression of persistence on scaled inflow restrictions,
outflow restrictions, and incentives. While only scaled outflow has significantly positive estimate in
probit model, the above three variables are positively associated with superior persistence in
ordered probit model. Namely, outflow restriction is discovered as more important determinant than
any other factors.
Insert Table 10 about here
For the search for more salient factor between restrictions and incentive with respect to superior
persistence, we construct four groups classified by the extent of incentives and restriction. That is, if
a fund has scaled score of restriction above the median of entire sample, a fund is denoted as high
(H) and is denoted as low (L) otherwise. If a fund‟s scaled incentive score is higher than median
21
incentive score of all samples, a fund belongs to high (H) and belongs to low (L) otherwise. Thereby,
all funds are classified as four groups (HH, HL, LH, and LL)19
. Median of scaled total, inflow,
outflow restrictions, and incentives are 3.17, 1.75, 1.43, and 1.83 respectively.
Panel B of Table 10 displays frequencies and proportion of winners on each group. As we
expect, funds classified by high restriction or high incentive show more frequency and higher
proportion of winners. For example, while 22.6% (18.1%) of funds with high total restrictions
(incentives) exhibit superior performance persistence, 10.5% (14.7%) of funds with low total
restrictions present it. However, if we analyze frequency and proportion of winners on each group,
the results are changed. Interestingly, funds with high restrictions and high incentives (HH) have
lower winner ratio than funds with high restrictions and low incentives (LH). While winner ratio for
HH is 21.2%, winner ratio of LH is 25.4% in groups classified by scaled total restrictions and
incentives. The differences between two groups are widened in groups classified by scaled outflow
restrictions and incentives. On the other hand, funds with high restrictions and high incentives (HH)
have always higher winner ratio than funds with low restrictions and high incentives (HL)
regardless of total, inflow, and outflow restrictions. Namely, although funds with higher restrictions
and higher incentives respectively display higher ratio of persistence, funds with lower incentives
exhibit higher ratio of persistence than funds with higher incentives if we consider interact effect
between restrictions and incentives and fix condition as high restrictions. These results suggest that
total restrictions and outflow restrictions are more important factor than incentives to affect
persistence of performance in hedge funds.
Finally, we conduct probit regression of persistence on class variables categorized four groups
including control variables. For dependent variable of class, we create matrix columns using
parameterization method. The results are shown in Panel C of Table 10. The parameter estimates of
each group present the difference in the effect of each class compared to funds with high incentives
and restrictions (HH). Funds included in LL and HL classified by total restrictions or outflow
restrictions obviously display lower superior persistent ratio than HH. However funds belong to LH
show higher winner ratios comparing to ratios of HH in spite of insignificant estimates20
. Therefore,
we can conclude that restrictions matter more to superior persistence of performance in hedge funds
than incentives. In the mean time, in groups classified by inflow restrictions and incentives, funds
19 The first letter represents incentives and the second represents total, inflow, and outflow restrictions. 20 In analyses with explanatory variables excluding control variables, funds belong to LH exhibit statistically significant
higher winner ratio than funds in HH.
22
with high inflow restrictions and high incentives do not display higher winner ratio than those of
funds with low inflow restrictions and high incentives.
4.5 Robustness
This section addresses the robustness of the main results. We conduct regression analyses using
interaction variables with proxies for restrictions and performance in previous period. That is, we
can investigate whether performance persistence is attributable to restrictions by analyzing the
relationship between performance of subsequent period and combining variables with proxies for
restrictions and performance of previous period as following equation (4).
where and are abnormal returns of subsequent and previous period for three year
intervals of fund i, drawn from MFH model. , , ,
and of fund i are defined as in equation (2).
and of fund i are same as defined in equation (2) at time t21
.
Panel A of Table 11 displays the relation between performance at subsequent period and
combined effect of restrictions and performance at previous period for funds with superior
persistence. We find that combined variables of minimum investment amount, redemption period,
payout period, and lockup period with performance at previous period exhibit significantly positive
estimates. It is interpreted that performance at the first period interacted with restrictions are
associated with performance at the second period. The combined variables with open-ended and
open to public report negative sign but insignificant.
Insert Table 11 about here
On the other hand, we do not find any positive relations in Panel B of Table 11 analyzing the
relations for funds without superior persistence. No combined variables with individual proxy for
restrictions exhibit positive relation with performance at the second period. Rather, the combined
21 t is last year of subsequent period for three year intervals.
23
variables of previous performance with minimum investment amount and open to public show
significant reversed sign to Panel A. Hence, we suggest that the results provide the clear evidence
that funds with superior persistence restrict fund flows.
V. Conclusions
This paper examines the impact of flow restrictions on superior persistence of performance in
hedge funds. We find that money outflow restrictions such as redemption notice period, lock period,
and payout period are positively associated with winners‟ persistence. In addition, we provide the
evidence that funds with barrier to entry for investment by high minimum investment amount,
closed-end, and opening to only institutions are more likely to deliver the consistent superior
performance. It implies that fund managers who are confident in achievement of good performance
seem to impose rigid contractual provisions associated with flow restrictions. Out of two resources
with respect to persistence, outflow restrictions are observed to more important factors for
persistence than inflow restrictions.
We also document that managerial incentives such as incentive fees, high water mark, and
management fees have influence on winners‟ persistence. Thereby, we explore which feature
between flow restrictions or managerial incentives is more salient determinant for continuous
excellent performance persistence. Interestingly, while funds with higher restrictions and higher
incentives respectively display higher ratio of persistence, funds with lower incentives exhibit
higher ratio of persistence under higher restrictions condition. This result suggests that persistence
is subject to flow restrictions rather than incentives.
This paper makes several contributions in the following ways. First, our study is the first one
about the relation between performance persistence and flow restrictions or persistence and both
restrictions and incentives simultaneously. Second, we provide the empirical evidence with the
theoretical prediction by Glode and Green (2011) that reveal determinant to affect persistence not
from managerial skills but from unique structure form of hedge funds composing of limited and
general partners. Third, we consider entry barrier to invest in hedge funds such as closed-end form
and closing to individual investors beyond restrictions related to withdrawn.
We believe that these results have important implications for investors to choose an appropriate
fund and for practitioners to make contractual features in hedge fund industry. We expect that future
research issues associated with this paper seem to be characters of funds with rigid flow restrictions.
24
In other words, if we understand that funds whose restrictions are strict deliver superior
performance persistence, then which funds are willing to impose these restrictions?
25
<References>
Ackermann, G., R. McEnally, and D. Ravenscraft, 1999, The performance of hedge funds: Risk,
return, and inventives, , Journal of Finance 54, 833-874.
Agarwal, V., N. D. Daniel, and N. Y. Naik, 2009, Role of managerial incentives and discretion in
hedge fund performance, Journal of Finance 64, 2221-2256.
Agarwal, V. and N. Y. Naik, 2000, Multi-period performance persistence analysis of hedge funds,
Journal of Financial Quantitative Analysis 35, 327-342.
Aragon, G. O., 2007, Share restrictions and asset pricing: Evidence from the hedge fund industry,
Journal of Financial Economics 83, 33-58.
Aragon, G. O., B. Liang, and H. Park, 2011, Onshore and offshore hedge funds: Are they twins?,
working paper.
Avramov, D., R. Kosowski, N. Y. Naik, and M. Teo, 2011, Hedge funds, managerial skill, and
macroeconomic variables, Journal of Financial Economics 99, 672-692.
Bali, T. G., S, J. Brown, and M. O. Caglayan, 2011, Do hedge funds‟ exposures to risk factors
predict their future returns?, Journal of Financial Economics 101, 36-58.
Baquero, G., J. R. Horst, and M. Verbeek, 2005, Survival, look-ahead bias and the persistence in
hedge fund performance, Journal of Financial Quantitative Analysis 40, 493-517.
Baquero, G. and M. Verbeek, 2009, A portrait of hedge fund investors: Flows, performance, and
smart money, working paper.
Bares, P. A., R. Gibson, and S. Gyger, 2003, Performance in the hedge fund industry: An analysis of
short and long-term persistence, Journal of Alternative Investment 6, 25-41.
Brown, S. J., W. N. Goetzmann, and R. G. Ibbotson, 1999, Offshore hedge funds: Survival and
performance 1898-1995, Journal of Business 72, 91-118.
Brown. S., W. Goetzmann, B. Liang, and C. Schwarz, 2008, Mandatory disclosure and operational
risk: Evidence from hedge fund registration, Journal of Finance 63, 2785-2815.
Ding, B., M. Getmansky, B. Liang, and R. Wermers, 2009, Share restrictions and investor flows in
the hedge fund industry, working paper.
Edwards, F. R. and M. O. Caglayan, 2001, Hedge fund performance and manager skill, Journal of
Futures Markets 21, 1003-1028.
Fung, W. and D. A. Hsieh, 2004, Hedge fund benchmarks: a Risk-based approach, Financial
Analysts Journal 60, 65-80.
Glode, V. and R. C. Green, 2011, Information spillovers and performance persistence for hedge
funds, Journal of Financial Economics 101, 1-17.
26
Hong, H., J. D. Jubic, and J. C. Stein, 2005, Thy neighbor‟s portfolio: Word-of-mouth effects in the
holdings and trades of money managers, Journal of Finance 60, 2801-2824.
Jaganathan, R., A. Malakhov, and D. Novikov, 2010, Do hot hands exist among hedge fund
managers? An empirical evaluation, Journal of Finance 65, 217-255.
Kosowski, R., N. Y. Naik, and M. Teo, 2007, Do hedge funds deliver alpha? a Bayesian and
bootstrap analysis, Journal of Financial Economics 84, 229-264.
Liang, B., 2000, Hedge funds: The living and the dead, Journal of Financial Quantitative Analysis
35, 309-326.
Panageas, S. and M. Westerfield, 2009, High water marks: high risk appetites? Convex
compensation, long horizons, and portfolio choice, Journal of Finance 64, 1-36.
Sadka., R., 2006, Momentum and post-earnings-announcement drift anomalies: The role of liquidity
risk, Journal of Financial Economics 80, 309-349.
Sadka, R., 2010, Liquidity risk and the cross-section of hedge fund returns, Journal of Financial
Economics 98, 54-71.
27
Table 1. Descriptive statistics
This table reports descriptive statistics of 3,756 (2,276 live and 1,486 dead funds) sample funds and 17,365
yearly observations by trading strategy. The TASS database classifies hedge funds into ten primary categories.
Sample period is from January 1994 to March 2008. Mean and median returns are the average and mean of
yearly returns for funds belong to identical strategy. Size is average assets under management (AUM) at the
end of year of funds belong to identical strategy. Flow is calculated by the following equation.
where is the assets under management of fund i at the end of year t-1 and
is the return for fund i during the year t. Volatility is annualized monthly standard deviation of all funds within
identical strategy.
Trading Strategy N Observations Return Return
Size($M) Flow
(%)
Volatility
(%) Mean(%) Median(%)
Convertible Arbitrage 168 804 9.01 8.39 168.05 47.94 20.84
Dedicated Short Bias 36 177 1.85 2.85 35.07 67.36 78.45
Emerging Markets 294 1,523 17.49 13.82 157.99 36.21 67.82
Equity Market Neutral 273 1,157 8.55 7.12 109.87 54.11 24.07
Event Driven 433 2,139 12.15 10.86 231.33 52.41 22.52
Fixed Income Arbitrage 218 955 9.41 8.53 220.94 58.10 17.56
Global Macro 232 899 10.06 7.29 235.78 49.00 46.11
Long/short Euiqty Hedge 1,452 6,711 13.68 11.03 119.53 45.09 51.99
Managed Futures 439 2,058 10.57 8.70 70.01 30.83 72.89
Muti-Strategy 211 942 12.14 10.23 294.43 51.25 27.62
Total 3,756 17,365 12.23 9.92 152.84 45.74 45.96
28
Table 2. Summary statistics for performance persistence
This table reports summary statistics of performance persistence for two estimation periods in which
performance is measured in non-parametric and parametric methods. Panel A shows yearly returns statistics of
three groups (WW, LL, WL or LW) classified by the extent of persistence drawn by non-parametric method.
If a style-adjusted return (raw return-average return of all funds within same strategy) on yearly base of a fund
is higher than median return of all funds employing same strategy with a fund, it becomes a winner (W),
otherwise, it is a loser (L). Therefore, if funds are winners (losers) in two consecutive periods, funds are
denoted as WW (LL). WL denotes winners in previous period but losers in subsequent period, and LW
denotes the reverse. Panel B shows yearly returns statistics of five groups (WWWW, WWW or WW, LL or
LLL, LLLL, the others) classified by the extent of persistence drawn by parametric method. Performance of
funds are estimated by modified Fung and Hsieh model (MFH model) that liquidity factor is added to Fung
and Hsieh (2004) seven factor model as following equation.
where is the monthly return of a fund i in excess of one month T-bill rate. EMKT (equity market) is
S&P 500 minus risk free rate. SIZE (size spread) is Wilshire small cap 1750 minus Wilshire large cap 750 till
the end of 2006 and Russell 2000 minus S&P 500 thereafter. BMKT (bond market) is the monthly change in
the 10 year treasury constant maturity yield. CRED (credit spread) is the monthly change in the Moody's Baa
yield less 10-year treasury constant maturity yield. TFB (bond trend-following factor), TFCR (currency trend-
following factor), and TFCM (commodity trend-following factor) are the return of PTFS (primitive trend-
following strategy) of a bond lookback straddle, of a currency lookback straddle, and of a commodity
lookback straddle, respectively. is Sadka (2006) liquidity factor on permanent-variable price impact
component extracted from the Trades and Quotes (TAQ) tick-by-tick data. Therefore, the definition of each
group is as follows. WWWW: alphas are positive and statistically significant in two consecutive periods,
WWW and WW: alphas are positive and at most one alpha is significant, LLL and LL: alphas are negative
and at most one alpha is significant, LLLL: alphas are negative and statistically significant in two consecutive
periods, and the others (WWL, LWW, WWLL, WLLW, LLWW, WLL, LWW): one of alphas is positive and
the other of alpha is negative.
Panel A. Group classified using Non-parametric performance measure
Previous period
Subsequent period
n mean median std skew kurtosis
mean median std skew kurtosis
WW 3,836 14.06 8.90 15.33 2.00 4.91
12.94 7.72 14.70 2.29 6.38
WL+LW 5,891 2.32 0.69 16.91 1.17 4.65
0.69 -0.62 16.63 1.20 4.57
LL 3,692 -8.75 -5.89 8.97 -1.47 4.93
-9.51 -6.60 9.37 -1.70 2.91
Panel B. Group classified using parameter performance measure
Previous period
Subsequent period
n mean median std skew kurtosis
mean median std skew kurtosis
WWWW 1,313 12.45 10.03 8.02 1.49 2.68
11.04 8.60 8.18 2.35 7.45
WW+WWW 3,514 10.58 7.78 9.55 1.86 4.26
8.78 6.15 8.71 2.08 5.57
The others 2,660 5.03 3.91 13.32 0.31 2.40
0.13 -1.34 11.55 0.68 2.88
LL+LLL 505 -7.94 -5.43 8.09 -2.28 7.77
-8.18 -5.86 8.68 -2.27 6.45
LLLL 13 -18.40 -14.83 15.43 -0.81 -0.54
-24.38 -28.47 19.53 -0.24 -1.30
29
Table 3. Summary statistics for flow restrictions provisions
This table reports summary statistics of flow restrictions provisions. Redemption notice period is time period
that has to be given to the fund before shares can be redeemed. Lockup period is minimum period an investor
is initially required to hold its money in the fund. Payout period is time period before an investor will receive
money as returns back. Open ended equals 1 if a fund is open-end and equals 0 if a fund is closed-end. Open
to public equals 1 if a fund opens to public and equals 0 if a fund closes to public. The parameters of
restrictions are scaled from 0 to 1 by percentile of probability distribution on each variable. For binary
distribution of open-end and open to public variables, scale score is endowed with 1 if a fund is closed-end or
closing to public and is endowed with 0 otherwise. Inflow restrictions and outflow restrictions include three
variables, so that scaled them provide the range from 0 to 3, respectively.
Mean SD Min Q1 Median Q3 Max Mode
Minimum Investment
Amount($M) 8.64 2.30 0 0.25 0.50 1 100 1
Redemption Notice
Period(days) 33.75 26.93 0 14 30 45 180 30
Payout Period(days) 14.83 21.53 0 0 10 30 640 0
Lockup Period(Months) 3.82 6.50 0 0 0 12 90 0
Open Ended 0.58 0.49 0 0 1 1 1 1
Open To Public 0.17 0.37 0 0 0 0 1 0
Scaled Total Restrictions 3.24 1.11 0.61 2.36 3.19 4.12 5.9 2.79
Scaled Inflow Restrictions 1.74 0.74 0 1.16 1.75 2.49 2.99 1.75
Scaled Outflow Restrictions 1.49 0.61 0.61 1.04 1.44 1.98 2.97 0.61
30
Table 4. Flow restrictions and performance persistence for non-parametric method
This table reports flow restrictions among funds classified by the extent of persistence drawn by non-
parametric method. Panel A shows the magnitude of each restriction variable on three groups (WW, LL, WL
or LW) classified by two periods persistence drawn by non-parametric method. If funds are winners (losers)
in two consecutive periods, funds are denoted as WW (LL). WL denotes winners in previous period but losers
in subsequent period, and LW denotes the reverse. Panel B shows the magnitude of each restriction variable
on four groups (WWW, WW. LL, LLL) classified by three periods persistence drawn by non-parametric
method. If funds are winners (losers) in three consecutive periods, funds are denoted as WWW (LLL). If
funds are winners (losers) in two periods, funds are denoted as WW (LL). Panel C displays the scaled inflow
and outflow restriction score on three groups classified by two periods persistence drawn by non-parametric
method. Panel D displays the scaled inflow and outflow restriction score on four groups classified by three
periods persistence drawn by non-parametric method. Inflow restrictions and outflow restrictions include
three variables, so that scaled them provide the range from 0 to 3, respectively. Significant test use GLM
(general linear model) for difference analyses among funds classified by persistence. To control
heteroscedasticity that presents difference variances of restriction variables among classes for persistence, we
also use Welch‟s and Kruskal-Walli test with respect to variables showing heteroscedasticitic variance as a
result of Levin‟s test. ***
,**
, and * indicate significance at the 1%, 5%, 10% level.
Panel A. Two period (raw)
WW LW+WL LL Fvalue WW-LL t
Open Ended 0.5938 0.6089 0.6498 13.54*** -0.0559 -5.01***
Open To Public 0.1491 0.1536 0.1666 2.39*** -0.0175 -2.08**
Minimum Investment($M) 1.0330 0.8439 0.8768 14.59*** 0.1562 3.61***
Redemption Notice Period 38.22 34.81 32.57 42.55*** 5.64 8.93***
PayOut Period 15.55 13.81 11.23 40.87*** 4.32 9.56***
LockUp Period 4.67 3.79 3.28 42.38*** 1.39 8.93***
Panel B. Three period (raw)
WWW WW LL LLL Fvalue WWW-LLL t
Open Ended 0.6028 0.6069 0.6306 0.6695 7.64*** -0.0667 -4.05***
Open To Public 0.1494 0.1456 0.1517 0.1649 1.11 -0.0155 -1.24
Minimum Investment($M) 1.1054 0.8999 0.8225 0.9528 10.61*** 0.1526 2.28***
Redemption Notice Period 39.58 36.16 33.62 32.70 25.65*** 6.88 7.12***
PayOut Period 15.84 14.17 12.36 9.97 27.33*** 5.87 9.3***
LockUp Period 5.13 4.00 3.52 3.07 31.8*** 2.06 8.7***
Panel C. Two period (scaled)
WW WL+LW LL Fvalue WWW-LLL t
Total Restriction 3.36 3.22 3.08 62.5*** 0.28 11.16***
Inflow Restriction 1.79 1.74 1.68 21.35*** 0.11 6.47***
Outflow Restriction 1.56 1.47 1.39 81.69*** 0.17 12.77***
Panel D. Three period (scaled)
WWW WW LL LLL Fvalue WWW-LLL t
31
Total Restriction 3.39 3.26 3.16 3.02 37.94*** 0.37 9.98***
Inflow Restriction 1.79 1.76 1.72 1.66 10.96*** 0.13 5.2***
Outflow Restriction 1.60 1.49 1.43 1.36 56.33*** 0.24 12.21***
32
Table 5. Flow restrictions and performance persistence for non-parametric method by trading strategy
This table reports scaled total restriction, inflow restriction, and outflow restriction among three groups (WW,
LL. WL or LW) classified by the extent of two periods persistence drawn by non-parametric method. If funds
are winners (losers) in two consecutive periods, funds are denoted as WW (LL). WL denotes winners in
previous period but losers in subsequent period, and LW denotes the reverse. The parameters of restrictions
are scaled from 0 to 1 by percentile of probability distribution on each variable. For binary distribution of
open-end and open to public variables, scale score is endowed with 1 if a fund is closed-end or closing to
public and is endowed with 0 otherwise. Inflow restrictions and outflow restrictions include three variables, so
that scaled them provide the range from 0 to 3, respectively. Panel A shows the magnitude of scaled total
restriction composed of six variables, Panel B is associated with scaled inflow restriction, and Panel C is
related to scaled outflow restrictions. Significant test use GLM (general linear model) for difference analyses
among funds classified by persistence. ***
,**
, and * indicate significance at the 1%, 5%, 10% level.
Panel A. Scaled total restrictions
WW WL+LW LL Fvalue WW-LL t
Convertible Arbitrage 3.54 3.50 3.28 4.04**
0.26 2.62***
Emerging Markets 2.64 2.52 2.59 1.45 0.06 0.75
Equity Market Neutral 3.49 3.58 3.23 11.92***
0.27 3.37***
Event Driven 3.94 3.69 3.48 25.34***
0.46 6.94***
Fixed Income Arbitrage 3.72 3.59 3.17 16.69***
0.55 5.36***
Global Macro 3.04 2.94 2.51 14.24***
0.53 4.98***
Long/short Euiqty Hedge 3.56 3.42 3.31 21.58***
0.25 6.53***
Managed Futures 2.36 2.29 2.19 4.18**
0.17 2.85***
Muti-Strategy 3.40 3.35 3.16 3.28**
0.24 2.5**
Panel B. Scaled inflow restrictions
Convertible Arbitrage 1.99 1.92 1.68 10.65***
0.31 4.32***
Emerging Markets 1.25 1.24 1.29 0.5 -0.04 -0.81
Equity Market Neutral 1.95 2.05 1.81 9.86***
0.13 2.21**
Event Driven 2.04 1.94 1.90 5.53***
0.14 3.18***
Fixed Income Arbitrage 1.97 1.99 1.78 6***
0.19 2.75***
Global Macro 1.69 1.67 1.41 8.2***
0.28 3.56***
Long/short Euiqty Hedge 1.89 1.82 1.79 6.89***
0.10 3.57***
Managed Futures 1.38 1.36 1.32 0.94 0.06 1.36
Muti-Strategy 1.75 1.76 1.65 1.76 0.10 1.49
Panel C. Scaled outflow restrictions
Convertible Arbitrage 1.55 1.57 1.58 0.19 -0.03 -0.59
Emerging Markets 1.38 1.26 1.27 5.69***
0.11 2.67***
Equity Market Neutral 1.55 1.53 1.41 4.92***
0.14 2.93***
Event Driven 1.89 1.74 1.57 40.98***
0.32 8.98***
Fixed Income Arbitrage 1.74 1.61 1.39 22.81***
0.35 6.64***
Global Macro 1.34 1.26 1.08 14.11***
0.26 5.2***
Long/short Euiqty Hedge 1.67 1.60 1.52 27.69***
0.15 7.46***
Managed Futures 0.97 0.93 0.87 8.02***
0.10 3.95***
Muti-Strategy 1.65 1.59 1.51 3.33**
0.14 2.57**
33
Table 6. Flow restrictions and performance persistence for parametric method
This table reports the results of probit or ordered probit regression of persistence on flow restriction variables
as following equation.
where is 1 if a fund i is a winner in previous and subsequent periods and is 0 otherwise
in probit model. In ordered probit model, it ranges from 1 to 5 depending on the extent of performance
persistence. It equals 5 if a fund i is denoted as WWWW in note 8, 4 denoted as WWW and WW, 3 denoted as
WWL, LWW, WWLL, WLLW, LLWW, WLL, and LWW, 2 denoted as LL and LLL, and 1 denoted as LLLL.
equals 1 if a fund i is open-end and 0 if a fund i is closed-end. equals 1 if a fund i
is open to public and 0 otherwise. is the log of the minimum investment amount of fund i.
and are redemption notice period (days), lockup period (months), and
payout period (days) of fund i, respectively. is the log of estimated assets of fund i , is the
money flows of fund i , and is the standard deviation of monthly returns of fund i at the end of year
t-2. is a strategy dummy that equals 1 if fund i belongs to strategy s and equals 0 otherwise.
Panel A show the results of probit regression, and Panel B exhibit the results of ordered probit regression. ***
,**
, and * indicate significance at the 1%, 5%, 10%
Panel A. Probit regression
simple multiple
coefficient Chi-sq coefficient Chi-sq
OpenEnded -0.0521 0.81 -0.1177 1.63
OpenToPublic -0.1591 3.35* -0.0086 0.00
Minimum investment 0.1926 80.16***
0.1029 6.78***
RemptionNoticePeriod 0.0118 118.09***
0.0068 17.33***
PayOutPeriod 0.0047 12.87***
0.0005 0.07
LockUpPeriod 0.0189 18.24***
0.0153 5.48**
size
-0.0291 1.07
flow
-0.0049 0.05
std
-10.3386 41.61***
Panel B. Ordered Probit regression
simple Multiple
coefficient Chi-sq coefficient Chi-sq
OpenEnded -0.0597 1.78 -0.0413 0.44
OpenToPublic -0.1735 6.87***
-0.0564 0.39
Minimum investment 0.1705 143.51***
0.1202 24.49***
RemptionNoticePeriod 0.0104 104.23***
0.0050 9.95***
PayOutPeriod 0.0078 40.98***
0.0039 7.83***
LockUpPeriod 0.0199 33.17***
0.0107 4.72**
size
-0.0597 9.76***
flow
-0.0009 0.00
std
-4.5425 39.11***
34
Table 7. Flow restrictions and performance persistence for parametric method by trading strategy
This table reports the results of probit or ordered probit regression of persistence on scaled inflow and outflow
restrictions by trading strategy as following equation.
where is 1 if a fund i is a winner in previous and subsequent periods and is 0 otherwise in
probit model. In ordered probit model, it ranges from 1 to 5 depending on the extent of performance
persistence. It equals 5 if a fund i is denoted as WWWW in note 8, 4 denoted as WWW and WW, 3 denoted as
WWL, LWW, WWLL, WLLW, LLWW, WLL, and LWW, 2 denoted as LL and LLL, and 1 denoted as LLLL.
and are scaled inflow and outflow restriction of fund i.The parameters of
restrictions are scaled from 0 to 1 by percentile of probability distribution on each variable. For binary
distribution of open-end and open to public variables, scale score is endowed with 1 if a fund is closed-end or
closing to public and is endowed with 0 otherwise. Inflow restrictions and outflow restrictions include three
variables, so that scaled them provide the range from 0 to 3, respectively. Panel A show the results of probit
regression, and Panel B exhibit the results of ordered probit regression. For total fund analyses, strategy
dummy variable is added in the equation. ***
,**
, and * indicate significance at the 1%, 5%, 10%.
Panel A. Probit regression
Restriction Inflow Restriction outflow
coefficient chi-sq coefficient chi-sq
Convertible Arbitrage 0.1886 0.78 -0.1076 0.11
Emerging Markets 0.1569 1.19 -0.0163 0.01
Equity Market Neutral -0.0405 0.02 0.4466 0.85
Event Driven 0.2726 3.47* 0.4058 4.66**
Fixed Income Arbitrage -0.2031 0.88 0.9158 10.26***
Global Macro -0.2810 0.66 1.2087 6.09**
Long/short Euiqty Hedge 0.1526 1.65 0.1663 0.93
Managed Futures 0.2307 1.64 1.1795 9.95***
Muti-Strategy 0.0490 0.07 0.6546 8.93***
Total 0.1096 2.79* 0.3848 16.55***
Panel B. Ordered Probit regression
Restriction Inflow Restriction outflow
coefficient chi-sq coefficient chi-sq
Convertible Arbitrage 0.2415 1.73 -0.0820 0.08
Emerging Markets 0.3052 4.89** 0.1716 0.87
Equity Market Neutral -0.0880 0.15 0.4167 1.70
Event Driven 0.2644 4.20** 0.2503 2.20
Fixed Income Arbitrage -0.2208 1.32 0.7188 7.21***
Global Macro 0.1814 0.70 0.8350 7.72***
Long/short Euiqty Hedge 0.0050 0.00 0.2586 7.06******
Managed Futures 0.1447 2.14 0.8224 12.23***
Muti-Strategy 0.2001 0.96 0.6909 6.59**
Total 0.0988 4.88** 0.3803 35.18***
35
Table 8. Performance persistence and fund characteristics
This table reports the results of probit or ordered probit regression of persistence on fund characteristics such
as incentives, ownership, capital structures, and location as following equation.
where is 1 if a fund i is a winner in previous and subsequent periods and is 0 otherwise in
probit model. In ordered probit model, it ranges from 1 to 5 depending on the extent of performance
persistence. It equals 5 if a fund i is denoted as WWWW in note 8, 4 denoted as WWW and WW, 3 denoted as
WWL, LWW, WWLL, WLLW, LLWW, WLL, and LWW, 2 denoted as LL and LLL, and 1 denoted as LLLL.
and are management fee and incentive fee charged by fund i .
equals 1 if fund i has a high water mark provision, and equals 0 otherwise. equals 1 if
principals of management company of fund i have invested in a fund i and equals 0 otherwise.
is 1 if a fund i use leverage, and is 0 if a fund i does not use leverage. equals 1 if a fund i
leverages using margin borrowing and equals 0 other wise. equals 1 if a fund i is onshore and
equals 0 if a fund i is offshore. equals 1 if headquarter of management company of fund i is located in
big city, and equals 0 otherwise. Explanatory variables are fund‟s contractual provisions associated with
incentive in model 1, ownership in model 2, capital structure in model 3, and location in model 4. All
characteristic variables are included in model 5. Panel A show the results of probit regression, and Panel B
exhibit the results of ordered probit regression. ***
,**
, and * indicate significance at the 1%, 5%, 10%.
Panel A. Probit
model 1 model 2 model 3 model 4 model 5
coeff chi-sq coeff chi-sq coeff chi-sq coeff chi-sq coefft chi-sq
ManagementFee 0.0594 0.60
0.1011 1.59
IncentiveFee 0.0005 0.00
0.0027 0.09
HighWaterMark 0.1748 4.17**
0.1600 2.79*
PersonalCapital
-0.0974 1.44
-0.0703 0.67
Leveraged
0.0561 0.16
0.0214 0.02
Margin
-0.0491 0.13
0.0070 0.00
onoff
0.2057 6.39** 0.1814 4.17**
city
0.1133 2.10 0.1421 3.00
Panel B. Ordered probit
ManagementFee 0.0764 2.56
0.0938 3.43*
IncentiveFee 0.0118 3.31*
0.013 3.66*
HighWaterMark 0.2092 11.36***
0.2133 10.26**
PersonalCapital
-0.0436 0.55
-0.0159 0.07
Leveraged
0.105 1.41
0.0243 0.07
Margin
-0.0401 0.23
0.0156 0.04
onoff
0.1097 3.16* 0.0901 1.97
city
0.051 0.77 0.0825 1.84
36
Table 9. Restrictions and Incentives
This table reports the relationship between restrictions and incentives. Panel A display the results of
regression of restrictions on individual proxies representing incentives. Dependent variables are scaled total,
inflow, and outflow restrictions, and explanatory variables are management fee, incentive fee, high water
mark, and control variables. mfee and ifee are scaled management fee and incentive fee. Panel B exhibits
correlation coefficient of scaled restrictions and scaled incentives. ***
,**
, and * indicate significance at the 1%,
5%, 10%.
Panel A. Regression Analyses
(1) raw explanatory variables
Total Restriction Inflow Restriction Outflow Restriction
coefficient t coefficient t coefficient t
ManagementFee -0.2450 -8.02*** -0.1297 -5.46*** -0.1100 -8.25***
IncentiveFee 0.0149 3.5*** 0.0089 2.87*** 0.0076 3.37***
HighWaterMark 0.8482 18.1*** 0.3450 10.09*** 0.5069 20.57***
(2) scaled explanatory variables
mfee -0.4914 -6.33*** -0.3078 -5.36*** -0.1688 -4.15***
ifee 0.4751 4.61*** 0.3299 4.33*** 0.1783 3.32***
HighWaterMark 0.8927 20.19*** 0.3530 10.79*** 0.5468 23.48***
Panel B. Correlation coefficients between scaled restrictions and scaled incentives
Total Restriction
Inflow
Restriction
Outflow
Restriction mfee ifee
Inflow Restriction 0.8607***
Outflow Restriction 0.7603*** 0.3237***
mfee -0.1351*** -0.1169*** -0.0892***
ifee 0.1773*** 0.1397*** 0.1702*** 0.1004***
HighWaterMark 0.4404*** 0.2647*** 0.4888*** -0.0316 0.2374***
37
Table 10. Performance persistence, Restrictions, and Incentives
This table reports the relationship between performance persistence and both of restrictions and incentives. Panel A displays the results of probit or ordered probit regression of persistence on scaled inflow restrictions,
outflow restrictions, and incentives. Panel B shows frequencies and proportion of winners on groups
classified by the extent of restrictions and incentives. If a fund has scaled score of restriction above the
median of entire sample, a fund is denoted as high (H) and is denoted as low (L) otherwise. If a fund‟s scaled
incentive score is higher than median incentive score of all samples, a fund belongs to high (H) and belongs to
low (L) otherwise. Thereby, all funds are classified as four groups (HH, HL, LH, and LL: fist letter is
incentives). We conduct probit regression of persistence on class variables categorized four groups. (if control
variables are included (excluded) in model, it is denoted as full (multiple)). For dependent variable of class,
we create matrix columns using parameterization method. The results are shown in Panel C. The parameter
estimates of each group present the difference in the effect of each class compared to funds with high
incentives and restrictions (HH). ***
,**
, and * indicate significance at the 1%, 5%, 10%.
Panel A. Regression of persistence on scaled inflow, outflow, and incentives
probit
ordered probit
coefficient chi-sq coefficient chi-sq
Inflow Restriction 0.1075 2.70 0.0801 3.37*
Outflow Restriction 0.3782 14.20*** 0.295 18.14***
Incentives 0.0186 0.06 0.1323 6.12**
Panel B. Frequencies and Proportion of Winners
Total Restrictions
Low High Total
Incentives Low 9.57%(256/2675) 25.39%(322/1268) 14.66%(578/3943)
High 12.22%(172/1407) 21.21%(563/2655) 18.09%(735/4062)
Total 10.48%(428/4082) 22.56%(885/3923)
Inflow Restrictions
Low High Total
Incentives Low 10.69%(245/2291) 20.16%(333/1652) 14.66%(578/3943)
High 13.85%(201/1451) 20.45%(534/2611) 18.09%(735/4062)
Total 11.92%(446/3742) 20.34%(867/4263)
Outflow Restrictions
Low High Total
Incentives Low 10.32%(290/2809) 25.40%(288/1134) 14.66%(578/3943)
High 12.44%(157/1262) 20.64%(578/2800) 18.09%(735/4062)
Total 10.98%(447/4071) 22.01%(866/3934)
Panel C. Regression
Multiple
Full
coefficient chi-sq
coefficient chi-sq
(1) Total Restrictions
LL -0.9333 46.33***
-0.6229 15.70***
HL -0.6587 18.46***
-0.384 5.84**
LH 0.2349 2.88*
0.1775 1.57
(2) Inflow Restrictions
LL -0.7641 26.68***
-0.367 5.03**
HL -0.4693 10.15***
-0.0142 0.01
LH -0.0182 0.02
0.1002 0.50
(3) Outflow Restrictions
LL -0.8151 38.80***
-0.489 11.15***
HL -0.6048 13.48***
-0.4222 5.55**
LH 0.269 3.52*
0.1483 0.99
38
Table 11. Robustness test
This table reports the results of regression of the abnormal return at subsequent period on restriction variables
interacted with the abnormal returns at the previous period as following equation.
where and are abnormal returns of subsequent and previous period for three year intervals of
fund i, drawn from MFH model. , , , and of fund i are defined as in equation (2). and of fund i are same as
defined in equation (2) at time t. Panel A displays the relation between performance at subsequent period and
combined effect of restrictions and performance at previous period for funds with superior persistence. Panel
B shows this relationship for funds without persistence. ***
,**
, and * indicate significance at the 1%, 5%, 10%.
Panel A. Funds with performance persistence
coefficient t
lag alpha*Open Ended -0.0002 -0.01
lag alpha*Open To Public -0.0376 -1.43
lag alpha*Minimum Investment 0.0627 8.12***
lag alpha*Redemption Notice Period 0.0018 5.99***
lag alpha*PayOut Period 0.0005 2.58***
lag alpha*LockUp Period 0.0052 4.37***
Panel B. Funds without performance persistence
lag alpha*Open Ended -0.0034 -0.14
lag alpha*Open To Public 0.0606 2.01**
lag alpha*Minimum Investment -0.0461 -5.31***
lag alpha*Redemption Notice Period -0.0007 -1.35
lag alpha*PayOut Period 0.0005 0.94
lag alpha*LockUp Period 0.0016 0.93