PRELIMINARY AND INCOMPLETE
If a quali ed investor has a choice between investing in a secretive fund and a transparent fund with the same investment objective, which should she choose? Prior work suggests that the secretive fund is better. Hedge fund managers generally use their discretion for the bene t of their investors (Agarwal, Daniel and Naik, 2009, Agarwal, Jiang, Tang and Yang, 2013). In this study we identify a subset of hedge funds managers, which appear to use their discretion to feign skill. Using a proprietary dataset obtained from a fund of funds, we document that hedge funds that are more secretive earn somewhat higher returns than their investment-objective-matched peers during up markets, consistent with earlier papers documenting skill-based performance, but signi cantly worse returns during down markets. This evidence suggests that at least part of the superior performance that secretive funds appear to generate is in fact compensation for loading on additional risk factor(s) as compared to their objective-matched peers.
Top Rated Pune Call Girls Sinhagad Road ⟟ 6297143586 ⟟ Call Me For Genuine S...
Fooling the Savvy Investor: Secrecy and Hedge Fund Performance
1. Fooling the Savvy Investor:
Secrecy and Hedge Fund Performance
Sergiy Gorovyy
Morgan Stanley
Patrick J. Kelly
New Economic School
Olga Kuzmina
New Economic School
This draft: February 2016
PRELIMINARY AND INCOMPLETE
Abstract
If a quali…ed investor has a choice between investing in a secretive fund and a transparent
fund with the same investment objective, which should she choose? Prior work suggests
that the secretive fund is better. Hedge fund managers generally use their discretion for the
bene…t of their investors (Agarwal, Daniel and Naik, 2009, Agarwal, Jiang, Tang and Yang,
2013). In this study we identify a subset of hedge funds managers, which appear to use
their discretion to feign skill. Using a proprietary dataset obtained from a fund of funds,
we document that hedge funds that are more secretive earn somewhat higher returns than
their investment-objective-matched peers during up markets, consistent with earlier papers
documenting skill-based performance, but signi…cantly worse returns during down markets.
This evidence suggests that at least part of the superior performance that secretive funds
appear to generate is in fact compensation for loading on additional risk factor(s) as compared
to their objective-matched peers.
Keywords: G01, G11, G23, G32
JEL codes: Hedge funds, Risk premia, Disclosure, Transparency
We thank the fund of funds for making this paper possible by providing access to the proprietary data on hedge
fund returns and characteristics. We are obliged for many conversations and suggestions to Andrew Ang, Emanuel
Derman, Roger Edelen, Andrew Ellul, Maria Guadalupe, Robert Hodrick, Greg van Inwegen, Norman Schurho¤,
James Scott, and Suresh Sundaresan, and to seminar participants at the New Economic School, Gaidar Institute for
Economic Policy, IE Business School, SKEMA Business School and the University of Rhode Island. We are thankful
to conference participants at the 3rd International Moscow Finance Conference.
1
2. 1 Introduction
Hedge funds in the U.S. are exempt from many disclosure requirements funds under the rational
that the savvy and sophisticated clientele permitted to invest in hedge funds is well quali…ed to
evaluate funds’governance and investment strategies without the interference of government regula-
tion.1
While the greater secrecy a¤orded hedge funds allows them to pursue proprietary investment
strategies with less risk that other investors might mimic and free ride on their strategies, there is a
natural tension between secrecy and the ability of a hedge fund’s investors to monitor the managers,
who in the absence of monitoring may deviate from strategies which are optimal for the investors.
Prior research provides evidence that mangers often use their discretion for the bene…t of their
investors. Agarwal, Daniel and Naik (2009) …nd that hedge fund returns are higher when managers
have more discretion as proxied by the length of lockup, notice and redemption periods. Aragon,
Hertzel and Shi (2013) and Agarwal, Jiang, Tang and Yang (2013) provide evidence that mangers
use their discretion to delay the reporting of fund holdings to the U.S. Securities and Exchange
Commission (SEC) for the bene…t of their investors, generating higher abnormal returns.2
Each of these two aspects of managerial discretion have built-in disciplining mechanisms, which
may reduce the ability of managers to abuse their discretion for their own bene…t. With regard
to regulated disclosure, managers have much less scope to conceal information as it will eventually
be revealed, albeit, with some delay. In the case of contractually stipulated lock-up, notice and
redemption periods, the money investors are withdrawing will eventually be returned. The fact
that there are these built-in disciplining mechanisms may be important. Other work suggests that
when managers have greater discretion they eschew their …duciary responsibility in order to secure
hire fees. Agarwal, Daniel and Naik (2009) …nd that managers with more incentive and opportunity
to do so, in‡ate their returns in December.
In this paper we consider a situation over which managers have full discretion: how secretive
they are vis a vis their own investors. Whether more disclosure is good or bad largely depends on
the source of a fund’s performance. If secretive funds attract more skillful managers that employ
1
Investment Company Act of 1940 carves out an exception from some disclosure requirements for investment
companies which only accept funds from “accredited investors”. Accredited investors are those income greater
than $200,000 (or $300,000 with a spouse a net worth greater than $1 million (https://www.investor.gov/news-
alerts/investor-bulletins/investor-bulletin-accredited-investors). Senate Report No. 293, 104th Cong., 2d. Sess. 10
(1996) and Sta¤ Report to the United States Securities and Exchange Commission, September 2003, “Implications
of the Growth of Hedge Funds”comment on the reasoning for this exception.
2
Section 13(f) of the Securities Exchange Act of 1934 require investment companies with more that $100 million
in assets to report holdings on a quarterly basis. Managers may request to delay discloser of the holdings for up to
a year
2
3. proprietary know-how strategies and invest into acquiring more information about the instruments
they trade (i.e. generate an "alpha"), more disclosure would not be necessarily good, since it
might allow other funds or investors to free-ride on these more skillful managers, reducing their
competitive advantage and incentives for providing superior performance. If on the other hand,
secrecy allows hedge funds to misbehave and take more systematic risk (i.e. high "beta"), than
they claim the take, or perhaps more unpriced risk, such as uncovered option writing, then there
may be a rationale for increasing disclosure requirements, so that investors understand what they
are being compensated for when they receive their seemingly superior returns.
We argue that during relatively good times, the high-alpha and the high-beta/high-risk explana-
tions for secretive funds may be observationally equivalent as long as we do not know the full model
of hedge fund returns or do not observe all possible risk factors that explain variation in returns. On
the other hand, during bad times, the high-alpha and the high-beta/high-risk explanations yield
very di¤erent predictions under the assumption that the risks, on which the high-beta/high-risk
funds load, realize during these bad times.
Using a proprietary data base …rst used by Ang, Gorovyy and van Inwegen (2011), we compare
the performance of secretive and transparent hedge funds during good and bad times and …nd
that during an up market secretive funds signi…cantly outperform transparent funds, controlling
for the investment style; however, during a down market the secretive funds perform dramatically
worse consistent with secretive funds, relative to their investment-style matched peers, loading on
additional risks, which realize during the down market. The bene…t of our empirical setup lies in
the opportunity for making such an assessment irrespective of knowing the true model that drives
hedge fund returns, but instead by relying on the assumption that some of the risk factors that
secretive funds may have loaded more aggressively on, also su¤ered low returns during the period
of the global …nancial crisis.
We also examine the relation between ‡ows and performance across secretive and transparent
funds, because fund ‡ows, especially out‡ows, can act as a disciplining mechanism for hedge fund
managers. We hypothesize that secretive funds will be less sensitive to past performance than
transparent funds as it may be harder for investors to infer deviations from the secretive fund’s
declared strategy. Consistent with this hypothesis, we …nd that ‡ow to performance sensitivity
is greater for transparent funds than secretive. However, secretive funds become more responsive
following their severe poor down-market performance.
This paper contributes to three areas of the literature. First, we contribute to the literature
3
4. on disclosure and managerial incentive alignment by examining whether hedge fund managers use
their discretion for the bene…t of their clients. Because of our proprietary data set obtained from
a fund of funds spanning April 2006 to March 2009, we are able to directly measure the secrecy
level of a fund – a qualitative characteristic that is missing in public hedge fund databases and
describes the willingness of the hedge fund manager to disclose information about its positions,
trades and immediate returns to investors. We document that secretive funds signi…cantly out-
perform transparent funds during good times consistent with the …ndings of Aragon, Hertzel and Shi
(2013) and Agarwal, Jiang, Tang and Yang (2013) which examine a di¤erent aspect of transparency,
but provide evidence of managerial skill (alpha) driving the superior performance. Our …ndings
di¤er in part because we are able to use the crisis period to investigate the source of this out
performance. The fact that secretive funds signi…cantly under-perform transparent funds during
the bad times suggests that at least a part of the performance di¤erential between secretive and
transparent funds during good times can be attributed to a higher risk taking by secretive funds,
which earned a premium during good times but faced these realized risks during bad times. In
this way our work makes its second contribution by contributing to the literature on hedge fund
performance measurement emphasizing the value of measuring performance across up and down
markets.
Finally, our paper contributes to the market e¢ ciency literature. In this literature institutional
and other accredited investors are treated as more savvy than typical retail investors. This paper
provides evidence that among the distribution of savvy investors are those who appear to be unable
to identify the fact that secretive funds are loading up on risks that give the investor greater exposure
to market down turns than their style-matched peers. While wealthy and institutional investors
may be smarter on average, this paper makes plain that such investors are not uniformly superior,
as they appear to be leaving their money with funds that claim one style of investing, but in fact
are following a di¤erent style.
While few papers in the asset pricing literature have raised the issue of transparency as related
to hedge funds, presumably due to the absence of adequate data to explore this question, some
prior research has examined aspects of this important issue. Anson (2002) outlines di¤erent types
of transparency and discusses why investors may want a higher degree of transparency; Hedges
(2007) overviews the key issues of hedge fund investment from a practitioners perspective; Goltz
and Schroder (2010) survey hedge fund managers and investors on their reporting practices and
…nd that the quality of hedge fund reporting is considered to be an important investment criterion.
4
5. Aggarwal and Jorion (2012) study the e¤ects of hedge funds’decisions whether to provide or not to
provide managed accounts for their investors. They interpret the incidence of accepting managed
accounts as an indicator of the willingness of the fund to o¤er transparency. In contrast, we are able
to directly measure the level of transparency of a fund by using proprietary fund of funds scores that
are based on formal and informal interactions with hedge funds, such as internal reports, meetings
with managers and phone calls.
Our paper is closely related to Agarwal et al. (2013) and Aragon, et al (2013), which explore the
con…dential …lings of equity hedge funds. Using the data up to 2007, they …nd that the con…dential
("secretive") holdings of hedge funds out-perform regular ("transparent") …lings on a risk-adjusted
basis (e.g. using Carhart’s, 1997, four-factor alpha). They interpret it as a higher stock-picking skill
in hedge funds con…dential holdings. We consider a broader span of funds across di¤erent strategies
(for which four factors may not explain large portions of cross-sectional variation in returns), as well
as transparency with respect to fund investors, rather than more public …lings with SEC. Further,
we propose to evaluate performance di¤erentials during good and bad times separately. This enables
us to infer the presence of risk premia with respect to potentially unobserved factors, which would
not be distinguishable from skill during good times.
The paper is also close in spirit to Brown, Goetzmann, Liang, and Schwarz (2008) who use SEC
…ling data to construct a so called !-score, which is a combined measure of con‡ict of interests,
concentrated ownership, and leverage, and show that it is a signi…cant predictor of the projected
fund life. In a subsequent paper, Brown, Goetzmann, Liang, and Schwarz (2012) use proprietary
due diligence data to construct an operational risk variable as a linear combination of variables that
correspond to mistakes in statements, internalized pricing, and presence of an auditor in the Big 4
group. We consider operational risk in a broader sense, where the willingness of hedge fund managers
to provide details of their strategies, as well as hedge fund liquidity, investment concentration, and
the ability of the investors to understand fund’s operations are important.
Our paper is organized as follows: Section 2 describes the data and explains the details of the
identi…cation strategy; Section 3 discusses the main results regarding the return premia associated
with highly secretive funds; Section 4 examines ‡ow-to-performance sensitivity of hedge funds based
on the secrecy of funds and Section 5 concludes.
5
6. 2 Data Description and Identi…cation Strategy
2.1 Data Description
We use a unique data set obtained from a fund of funds that contains detailed monthly fund
information over the period from 2006 to 2009. This fund of funds is one of the largest in the U.S.
The data provide information on hedge fund returns net of fees, their assets under management,
their long and short exposure, and the principal strategy of the fund. Most importantly, these
data include scores for hedge fund non-transparency, illiquidity, concentration, and complexity, as
rated by the fund of funds on a scale from 1 to 4. Once a year at the end of March the fund
of funds grades all the hedge funds it invests in based on its interactions with them during the
previous twelve months. These interactions consist of weekly or monthly reports to the fund of
funds, meetings with managers, phone calls, etc. Due to the nature of the scoring process and a
signi…cant level of e¤ort put into the construction of the scores, we feel con…dent that they represent
unique information about funds’operation that cannot be captured by the quantitative data alone.
Such qualitative measures are not present in public hedge fund databases, such as CISDM, HFR,
or TASS. Therefore, we think our data are especially well-suited for studying the return premia
associated with di¤erent qualitative characteristics of hedge funds.
The de…nitions of non-transparency (secrecy), illiquidity, concentration, and complexity as used
by the fund of funds are natural and intuitive. In particular, hedge fund non-transparency represents
a relatively low willingness of the hedge fund manager to share information about the fund’s current
activities and investments with its investors and, for example, provide the return instantaneously
(e.g. upon a call) when a certain market event happens. Hedge fund illiquidity measures the
illiquidity of investments with the hedge fund from the point of view of investors. It comprises of
both the illiquidity of fund’s assets and restrictions on investment withdrawal, such as the presence
and the length of lockup periods. Hedge fund concentration represents the level of concentration
of hedge fund investments. Finally, hedge fund complexity corresponds to the complexity of hedge
fund strategy and its operations. For example, a hedge fund that uses derivative instruments and
swap agreements is considered to be complex, since it is harder for investors to understand exactly
the kinds of exposures they face by investing with such a fund.
In order to avoid arti…cial signi…cance in our results, we conduct our analysis at the level of
fund families, where each "family" corresponds to a number of funds (usually 2 or 3) that are
characterized by the same returns in all periods, same strategy, same long and short exposures and
6
7. same qualitative scores in every period. Essentially, these are di¤erent copies of the same fund
having the same portfolio, but targeted at di¤erent investors: e.g. for quali…ed vs. regular partners,
onshore vs. o¤shore funds, funds denominated in di¤erent currencies, and additional fund copies
potentially created after the maximum of the number of partners has been achieved.3
This way we are left with 4,847 monthly observations of 200 di¤erent hedge fund families ("hedge
funds" in what follows) that are evenly spread across the three years, with 1,663 observations
between April 2006 and March 2007, 1,610 observations between April 2007 and March 2008, and
1,573 observations between April 2008 and March 2009. Since our qualitative grades are assigned
at the end of March, we use yearly periods starting every April. For example, the monthly returns
of a fund from April 2006 to March 2007 are matched to transparency, liquidity, concentration, and
complexity grades that the fund of funds issued at the end of March 2007. This approach ensures
that all interactions with the hedge fund that constitute the basis for the grades are conducted
in the same period when the fund return is delivered. Although primarily emerging as a result of
the grading month, this March to April time frame also corresponds nicely to three very distinct
periods, that would allow us to shed light on whether there is any risk component associated with
certain characteristics of funds. In particular, as we show in the next subsection, the risk premium
story should reveal di¤erent subsets of funds to perform better in good versus bad states of the
economy.
2.2 Illustration of the Identi…cation Strategy
To illustrate the use of di¤erent periods in identifying the risk premia associated with di¤erent
hedge fund characteristics, and hedge fund non-transparency in particular, suppose that the true
model for hedge fund returns consists of n factor returns:
Rit = i + i
1F1t + i
2F2t + ::: + i
nFnt + it; (1)
where Rit is the excess return of fund i in month t, i ("fund alpha") is the fund-speci…c
performance excess of what can be explained by factor loadings i
j on (excess) factor returns Fjt
(j = 1; 2; :::n). 4
3
This approach is very similar to conducting the analysis at the fund level, but properly accounting for perfect
correlation within each fund family in a given month (e.g. by clustering errors) to avoid arti…cial signi…cance of the
results. We decided to look at the family level instead, because we are also interested at looking at assets under
management that, given the same fund manager and strategy, are ultimately a fund family characteristic.
4
These factors may or may not be priced in the cross-section, they may also be non-linear functions of returns.
For example, if hedge funds agressively short index put options, one of Fjt could be the return on shorting index
7
8. If the econometrician knows the true model and observes all n factor returns, then she can
obtain unbiased and consistent estimates of i and j
i from historical data. However, not knowing
the true model (or observing fewer than n factors) can make inference about i incorrect, even if
the omitted factors are orthogonal to the observable ones.
To be speci…c, suppose the econometrician does not observe F1t and estimates a misspeci…ed
model containing the other n 1 factors only, so that F1t implicitly ends up in the error term (for
simplicity assume it is orthogonal to the included factors). In this case, the estimate of fund alpha,
bi = i + i
1F1t +
nX
j=2
( i
j
bi
j)Fjt + it, will be over-estimating the true i when omitted factor
is performing relatively well (F1t > 0), while under-estimating the true i when omitted factor
performs relatively poorly.(F1t < 0).
Therefore, if we do not know the true model then –by estimating an abridged (misspeci…ed)
model during times when realized returns on the omitted factors are generally positive –we would
erroneously attach the risk premium with respect to these omitted factors to fund alpha (e.g.
managerial skill). We can further think about one of these omitted factors being related to tail risk.
In this case, realized returns during "good" times (when tail risk earns a premium) could not be
empirically distinguished from fund alpha. This is especially important because market crashes –
when tail risk realizes or when the strategy of shorting put options goes bust –do not happen often,
and hence with respect to these potential omitted factors most of the times are actually "good"
times. In the example above, even adjusting for risk premia associated with all observable factors
(F2t to Fnt) does not entail an unbiased estimation of skill, as long as the times are on average
"good" with respect to the omitted factors.
Similarly, if we compare performance of any two groups of funds (e.g. secretive vs transparent
funds) and …nd that one group of funds over-performs the other in a particular period (even on a
risk-adjusted basis), we cannot disentangle the two explanations: either the …rst group has better
managers and earns an alpha and/or it simply loaded more on unobserved risk factors that earned
a premium and did not crash during this particular period.5
To illustrate the point, rewrite (1) for the average realized returns of secretive and transparent
funds and consider their di¤erence:6
RSEC
t RTRAN
t = ( SEC TRAN ) + ( SEC
1
TRAN
1 )F1t
puts, with j
then related to the relative weight of this strategy in fund’s total portfolio.
5
There is also a trivial alternative explanation for any observed empirical performance relation between two groups
of funds: time-varying luck of both groups (average epsilons). In a su¢ ciently long estimation period they should
average out, so for the sake of exposition, we dropped these from the expressions that follow.
6
RSEC
t RT RAN
t in this expression can represent the di¤erence of returns that are risk-adjusted for all the
8
9. If we …nd that in a particular period secretive funds over-perform transparent ones (RSEC
t
RTRAN
t > 0), then without observing F1t we cannot know, whether secretive funds had a higher
alpha ( SEC TRAN > 0) and/or they loaded more on an unobserved factor that did relatively well
during the period of estimation (( SEC
1
TRAN
1 )F1t) –the two explanations would be observationally
equivalent. Because of this conceptual impossibility of quantifying or even establishing the overall
existence of the skill component –in the absence of the true risk model of funds –we take a di¤erent
approach to deducing whether there were any signi…cant risk components associated with particular
groups of funds (e.g. with secretive funds).
In particular, we attempt to identify 2 periods in the data when we would be comfortable
assuming that an omitted factor F1t has di¤erential performance (i.e. there is a "good" period
when F1t > 0 and a "bad" period when F1t < 0). Then if it turns out that RSEC
t RTRAN
t > 0
in the "good" period while RSEC
t RTRAN
t < 0 in the "bad" period, it follows that secretive funds
load more on F1t than do transparent ones, with the proof amounting to noticing that the two
inequalities on returns can be satis…ed simultaneously only when SEC
1 > TRAN
1 .7
Given relatively low levels of disclosure among hedge funds and the virtual absence of information
on what exactly hedge funds may be doing at any particular moment of time, this approach has
the advantage of not requiring the complete knowledge or observation of all factors in the model,
but instead of assuming omitted factors in particular periods do relatively well or relatively poorly.
It thereby poses an empirical challenge of identifying such periods in the data.
The March to April time frame, introduced by the fund of funds grading scheme corresponds
nicely to three very distinct periods, so that it is relatively easy to select a "good" and a "bad"
period. Because the "good" and "bad" is always relative to the omitted factors, it is especially
compelling that our data covers the period of the global …nancial crisis, where we feel comfortable
to assume that risk factors on which hedge funds may have loaded did indeed realize – simply
because so many things crashed during this period. Although we may have in mind some of the
omitted factors being potentially related to rare events and tail risk (as also supported by loadings
on strategies associated with option-based returns as in Agarwal and Naik, 2004), they may well
represent other risks that were likely to realized during the crisis period. We therefore label April
observable factors: RSEC
t RT RAN
t
nX
j=2
( SEC
j
T RAN
j )Fjt. Alternatively, if we think that all factors in the true
model are not observable or they are measured with error, or the length of the time-series does not allow for a credible
estimation of loadings on all observable factors, RSEC
t RT RAN
t can represent the di¤erence in raw performance
without changing the conclusion conceptually.
7
With more than one omitted factor, it becomes a conditional statement on at least one factor performing
su¢ ciently bad in the "bad" period to overturn the di¤erence in average returns.
9
10. 2008 to March 2009 as the "bad" period –a recession period according to NBER, highlighted by
the bankruptcy …ling by Lehman Brothers in September 2008 and some of the largest drops of stock
market indices in history.
The period between April 2006 and March 2007, on the other hand, can be considered a "good"
period: according to the Financial Crisis Inquiry Report (2011) it was a normal growth period, a
growth period according to NBER, and a period of rapid rise of the U.S. stock market indices. This
period also followed a period of steady growth, so it is relatively safe to assume that at least some
of the omitted risks were not realizing during this period, but were instead earning a compensation.
Finally, the period between April 2007 and March 2008 is a somewhat intermediary period, as it
ends with the collapse of Bear Stearns that declared the beginning of the …nancial crisis, but was an
NBER growth period for much of the period. Since we cannot safely assume whether the possible
omitted risks realized during this period or were earning a compensation, this period would not be
of a particular help in trying to disentangle skill from the risk loadings.
We argue that comparing and contrasting the returns of di¤erent types of hedge funds (e.g.
secretive vs transparent) in di¤erent states of nature ("good" vs "bad" periods) is essential to
understanding whether there are risks associated with these types of hedge funds –in the situation
when the true model is unknown. The exogenous nature of the global …nancial crisis presents us
with a unique opportunity to observe hedge funds returns during a truly bad event realization when
the two explanations (skill vs risk loadings) would not be observationally equivalent.8
2.3 Summary Statistics and Variable De…nition
Hedge funds in our data set represent a broad set of strategies, with each fund being identi…ed
by a single strategy. This characteristic is time-invariant for a given hedge fund (at least during
the periods considered), which is not surprising given that funds are created in order to pursue a
particular strategy and investors expect the fund to follow it continuously over time. Panel A of
Table 1 tabulates the number of hedge funds by various strategies for each of the three periods
considered. In particular, there are credit (CR), event-driven (ED), equity (EQ), relative-value
(RV), and tactical trading (TT) hedge funds. Credit hedge funds trade mostly corporate bonds
and CDS on those bonds; event-driven hedge funds seek to predict market moves based on speci…c
news announcements; equity hedge funds trade equities (e.g. high/low net exposure to sectors
8
The idea of using "good" and "bad" periods having di¤erent informational content is not completely new. For
example, Schmalz-Zhuk (2013) argue that stocks should be more sensitive to news during bad periods: good or bad
performance during bad times is a clearer signal for investors than good or bad performance during good times.
10
11. and regions); relative value hedge funds seek pair trades where one asset is believed to outperform
another asset independent of macro events (e.g. capital structure or convertible bond arbitrage);
and tactical trading funds seek to establish favorable tactical positions using various combinations
of the above strategies. As we see, approximately half of the hedge funds in the database are equity
funds, with relative-value and event-driven as the next popular strategies. This distribution of
strategies across funds is comparable to other databases, as reported, for example, by Bali, Brown,
and Caglayan (2011) for TASS.
Panel B of Table 1 reports the mean, median, standard deviation, and the number of observations
for hedge fund monthly returns and volatility (in annualized percentage points), and assets under
management (AUM) separately for each of the periods considered. Hedge funds performed well as
a group during the good period from April 2006 to March 2007 delivering on average an excess
return of 7.66% per year with a within-fund 6.80% standard deviation. During the intermediate
period they delivered on average a –1.27% return with a higher 10.86% volatility, while during the
crisis period they delivered on average a negative –21.90% return with a 15.66% volatility.
The funds in our data set appear to be somewhat larger, than funds in CISDM, HFR, or TASS
databases, because we aggregate total assets under management across funds in the same family
(corresponding to the same managed portfolio). Ang, Gorovyy, and van Inwegen (2011) use the
same data to explore hedge fund leverage. They note that the composition of funds by strategy
is similar to the overall weighting (as reported by TASS and Barclays Hedge), and the aggregate
performance of the fund of funds is similar to that of the main hedge fund indexes. Importantly, all
funds in our database report their returns and those that terminate due to poor performance are
also covered in the data. Ang et al. (2011) describe the hedge fund selection criteria and note that
the criteria are not likely to introduce selection bias. In addition, they note that these hedge fund
data include both funds that are listed in the common hedge funds data sets, such as TASS, CISDM,
and Barclay Hedge, and funds that are not. This mitigates concerns about selection bias associated
with voluntary performance disclosure (Agarwal et al, 2010, among others). Finally, survivorship
bias is mitigated by the fact that hedge funds enter the data base several months prior to the fund
of fund’s investment and the hedge funds exit the data base several months after disinvestment.
Therefore, we are con…dent that our data set is broadly representative of the hedge fund industry
and su¤ers from less bias than is typical.
Finally, for each of the fund qualitative characteristics (secrecy, illiquidity, concentration, and
complexity) we de…ne a set of dummy variables that represent their High, Medium, and Low levels,
11
12. based on the original grades assigned by the fund of funds.9
,10
Panels C and D of Table 1 report
the distribution of fund secrecy levels (which are of our primary interest) across periods and across
strategies. As we see most funds are rated as being moderately secretive, and this pattern generally
appears across di¤erent fund strategies. Importantly, we have both high- and low-secretive funds in
each strategy (this is also true for "good" and "bad" period separately), which means that we can
identify di¤erences in performance across these two groups of funds within each strategy, and do
not simply rely on some fund strategies performing di¤erently in various periods and accidentally
being also intrinsically di¤erent in terms of their secrecy.
Lastly, Panel E reports pair-wise Spearman rank correlations between all of our qualitative
scores (using one observation per fund-year). We observe that more secretive funds are also more
illiquid, with the correlation statistically signi…cant at 1% level. More secretive and more illiquid
funds are also more complex on average. Finally, more illiquid funds are also more concentrated.
These relations between our qualitative scores are quite expected and give even more credibility to
our measures of secrecy, illiquidity, concentration and complexity. In our empirical estimation we
will account for these within-fund correlations accordingly.
2.4 Implementation of the Identi…cation Strategy
We study the hedge fund return premia associated with secrecy, illiquidity, concentration, and
complexity estimating the following empirical speci…cation for di¤erent periods of our data ("good"
and "bad"):
Rit = aM
SecSecM
i + aH
SecSecH
i + (2)
aM
IlliqIlliqM
i + aH
IlliqIlliqH
i + (3)
aM
ConConM
i + aH
ConConH
i + (4)
aM
ComComM
i + aH
ComComH
i + X0
it + dt + it; (5)
9
Although rated on a scale from 1 to 4, there is a very small percentage of funds that are ever rated with any grade
of 4. Henceforth we combine the grades of 3 and 4 into one category in order to ensure that we have a reasonable
number of observations in each category.
10
Interestingly, the original scale for grades represents levels of "problem" for the fund of funds associated with each
characteristic. As we check with the fund of funds, transparency score of 1 means lowest problem with transparency
(i.e. these are least secretive funds), and concenration score of 1 means lowest problem with concentration (i.e.
least concentrated funds). Based on how the original scale is constructred, fund of funds views secrecy, illiquidity,
concentration, and complexity as problematic characteristics of the fund.
12
13. where Rit is the excess return of fund i in month t, SecM
i (SecH
i ) are dummy variables that equal
to 1 if the fund is rates as moderately (highly) secretive in the period of estimation, and dummy
variables for illiquidity, concentration, and complexity are de…ned accordingly.
Coe¢ cient aM
Sec (aH
Sec) identi…es the mean di¤erence in performance between moderately (highly)
secretive funds and the least secretive funds, which are used as a base category in this estimation,
controlling for the di¤erences in other characteristics and control variables.11
Very similarly, the
estimates of aM
Illiq (aH
Illiq), aM
Con (aH
Con), and aM
Com (aH
Com) will correspond to cross-sectional di¤erences
in performance between moderately (highly) illiquid funds and the least illiquid funds, moderately
(highly) concentrated funds and the least concentrated funds, moderately (highly) complex funds
and the least complex funds, respectively.
As discussed previously, such a cross-sectional di¤erence in performance between any two groups
of funds can result from both managerial skill (alpha) and/or higher risk-taking (beta). By esti-
mating this speci…cation on both "good" and "bad" periods, we will see whether there is a risk
component associated with more secretive, illiquid, concentrated and complex funds, without actu-
ally having to observe all possible risk factors. In particular, under the assumption of certain risks
realize during the "bad" period, but not realizing during the "good" period (for which we selected a
stable growth period), more secretive (illiquid/concentrated/complex) funds are expected to over-
perform during good times, but under-perform during bad times if they take more of these risks.
That is, if we …nd that aM
Sec (aH
Sec), and/or aM
Illiq (aH
Illiq), and/or aM
Con (aH
Con), and/or aM
Com (aH
Com)
are positive during the "good" period and negative during the "bad" period, we will interpret it as
evidence of more secretive (illiquid/concentrated/complex) taking more risk that their less secretive
(illiquid/concentrated/complex) counterparts. If, on the other hand, we do not …nd a di¤erential
pattern in over- and under-performance in two periods, then we would not be able to assign perfor-
mance di¤erences to skill or risk-taking without knowing the true model of returns and observing
all factors.
In most of the speci…cations we include monthly …xed e¤ects, dt, to account for macroeconomic
conditions that are common to all hedge funds. In some of speci…cations we further include a vector
of controls, X0
it , which includes the natural logarithm of fund’s assets under management, net ‡ows
11
In fact, if we do not include any other qualitative characteristics and controls, and have only the two dummies
for Medium- and High-Secrecy funds, it would be exactly analogous to doing a 2-group test on the means twice (for
M vs L, and for H vs L). We model it in a regression framework, because it can accomodate additional characteristics
and controls that we would like to include into some speci…cations.
13
14. to the fund over the last three months –to account for a potential di¤erence in performance of funds
that have di¤erent size or have recently experienced abnormal ‡ows, as well as strategy-month …xed
e¤ects –to account for di¤erent loadings on risk factors across strategies. Finally, it denotes the
error term in the above-speci…ed regression model.
Finally, in all our speci…cations we report standard errors that are robust to heteroskedasticity,
as well as within-fund correlation over time (i.e., clustered at the fund level).12
3 Do More Secretive Hedge Funds Take More Risk?
Before turning to formal analysis, we …rst use the data to explore the time-series performance of
secretive and transparent funds graphically. Figure 1 plots monthly returns of the equally-weighted
portfolio of the most secretive funds (blue line) and the equally-weighted portfolio of the most
transparent funds (red line), averaging across the funds that are present in the data for the entire
period between April 2006 and March 2009 to minimize concerns about survival. Figure 2 plots
performance of similar equally-weighted portfolios averaging across funds that, on top of being
active over the entire period, are also given the same transparency score all three times they are
rated by the fund of funds. This makes a somewhat cleaner comparison, since we only look at
funds that did not adapt their reporting levels to di¤erent market conditions, potentially re‡ecting
a change in their underlying risk strategy.
As can be seen from these …gures, when both portfolios of funds are doing relatively well (having
a positive return), e.g. during 2006 and most of 2007, the portfolio of secretive funds over-performs
the portfolio of transparent funds almost in every month. On the other hand, when all funds
do poorly, e.g. during most of the year 2008, the portfolio of secretive funds under-performs the
portfolio of transparent funds. This pattern is especially pronounced during some of the largest
crashes of secretive funds in August 2007 and in the end of 2008: transparent funds did not crash
together with secretive funds, but on the contrary delivered a slightly positive, or a moderately
negative return (as compared to secretive funds).
These …gures convey the main message of the paper: at least part of the over-performance of
secretive funds that could be observed in good times, was in fact due to higher loading on certain
risks that realized during the crisis and made secretive funds fall more as compared to transparent
12
As a robustness check we also tried clustering at both fund and month level at the same time (as in Petersen,
2009) –to account simultaneously for correlation over time within the same fund, as well as cross-sectional correlation
across funds in any given month that may arise due to common shocks to all funds. The results were the same.
14
15. ones. Secretive funds could of course have had a positive alpha with respect to their true model of
returns but one would need to know the full model to accurately measure alpha. Nonetheless, they
would still have had to take more risk for their performance to be consistent with the observable
pattern.
We now turn to a more systematic analysis in order to see if there are risk components associated
with di¤erent characteristics of hedge funds and take into account observable di¤erences between
secretive and transparent funds. Following this analysis we turn to examining di¤erences in hedge
fund performance controlling both for secrecy and for exposure to risk factors. In both cases, we
separately examine the good period (2006-2007) and the bad period (2008-2009) in order to allow for
time variation in factor loadings, which makes our tests much more conservative, as they implicitly
assume that investors realize before hand, how hedge funds will alter their risk exposure during up
and down markets.
3.1 Di¤erences in raw performance of secretive and transparent funds
3.1.1 Performance of hedge funds in the "good" period
We start by considering the "good" period –the normal growth period of April 2006 to March
2007 –to see if there are any performance di¤erences across di¤erent types of funds during good
times, which could be later attributed to skill or risk-taking, once we have also explored the "bad"
period.
Table 2 column 1 reports the results of the simplest speci…cation that regresses hedge fund
performance on the indicator variables corresponding to medium and high levels of secrecy –our
primary variable of interest. We see that during good times highly secretive hedge funds out-
performed the most transparent hedge funds by 4.83% on an annual basis. Although there were no
statistically signi…cant di¤erences between moderately secretive and the most transparent funds (as
indicated by the di¤erence of 1.85% per year with a standard error of 1.58), highly secretive funds
also signi…cantly outperformed the moderately secretive funds by 2.98% (the unreported di¤erence
between the two coe¢ cients is signi…cant at 10% level).
We proceed by adding monthly …xed e¤ects in column 2 to account for average loadings on
all time-varying factors, both observable and unobservable. Accounting for time-series di¤erences
across months immediately helps explain 17% of the total variation in returns, but our cross-sectional
comparison between secretive and transparent funds remains similar in magnitude and statistical
signi…cance: highly secretive funds signi…cantly over-perform both the least secretive funds and the
15
16. moderately secretive funds by 4.88% and 2.97% per year, respectively.
Since our sample consists of funds in which the fund of funds invested in a particular year, a
potential concern for our results is that the fund of funds selected a di¤erent subsample of funds
every year and for this reason the sample of funds in the "good" period may be di¤erent from the
sample of funds in the "bad" period. In this case, we may be picking skill and/or risk-taking by the
fund of funds itself, rather than by hedge funds it invests it. To rule out this concern, in column 3
we consider only funds that are present in the data for the entire period from April 2006 to March
2009. We see that even within this balanced panel of funds, highly secretive funds still signi…cantly
over-perform the most transparent ones by 4.09% per year. The number of observations drops by
almost a half, so the estimate becomes less precise, but is still signi…cant at 10% level.
We now turn to considering other qualitative characteristics of funds, since as we learned from
Table 1 Panel E, many of them are highly correlated. In particular our cross-sectional division into
secretive and transparent funds may simply be picking up illiquid or more complex funds. In column
4 we estimate a speci…cation with all of our qualitative characteristics at the same time on the full
sample of funds. We observe highly secretive funds delivering a 4.64% higher return than the most
transparent funds in this speci…cation, so our results cannot be explained by a correlation of secrecy
levels with other qualitative characteristics. Highly secretive funds also signi…cantly over-perform
moderately secretive funds –by 3.17%, which is signi…cant at 5% level (unreported).
In column 5 we add hedge fund control variables to the speci…cation on column 4, such as
the logarithm of total assets under management and percentage net ‡ows during past 3 months
to control for potential di¤erences in the size that may exist across di¤erent types of funds, and
that could also be responsible for the performance di¤erence. The results are very similar in both
magnitude and are signi…cant at 5% level. Even when we estimate this very tight speci…cation on
a subsample of funds that are present in the data for all three years of the data –in column 6 –the
coe¢ cient estimate remains similar in magnitude.
Finally, in column 7 we estimate our fullest speci…cation on the subsample of funds that on top
of being in the data for all three years, are also rated with the same transparency score every year.13
We do so to make a cleaner comparison across funds that are likely to be characterized by the
same investment strategy in both "good"and "bad" periods, since a within-fund change in secrecy
13
This reduces the number of observations slightly, since qualitative characteristics of funds are generally persistent.
In particular, among all funds that have a secrecy levels present for two years or more, 89% actually have the same
level of secrecy in all years. Similarly, 91%, 83%, and 94% of funds have the same level of illiquidity, concentration,
and complexity, respectively, in all years. The observation that the fund disclosure level and its structure in general
are rarely modi…ed after the fund’s initiation is not surprising, because fund investors expect the fund to maintain
the same con…guration over time.
16
17. score may signal an adaptation of its reporting levels to a new underlying risk strategy or to the new
manager with a di¤erent skill. We …nd that funds that remained highly secretive over the entire
period over-performed those that remained the most transparent over the entire period by 8.32%
annually during the "good" period. This estimate is signi…cant at 1% level.
Overall, we …nd very robust results with respect to the over-performance of most secretive funds
during "good" times. So far this evidence is consistent with secretive funds either having better skill
and/or taking more risk that earns a premium during good times. In the next subsection, we will
consider whether this pattern in performance survives during the period of crisis, but …rst, let us
brie‡y discuss the results on other qualitative characteristics of the funds in the last four columns.
We observe that during the "good" period the most illiquid funds out-perform the most liquid
funds by 5%-8% per year, depending on the speci…cation, and the moderately illiquid funds –by
4%-6% per year. This suggests that funds that are rated as highly illiquid may indeed load more
on the illiquidity factor, but in order to investigate this further, we would need to look at whether
performance di¤erences revert in the "bad" period (under the assumption that illiquidity factor also
collapsed during this period).
There is some evidence of highly complex funds under-performing less complex funds, which may
be related to higher transactions costs when executing more complicated trading strategies. Finally,
in the fullest speci…cations we also …nd that highly concentrated funds on average out-perform less
concentrated funds by 7%-8% per year. Although in standard …nance theory, concentrated portfolios
should not bear a premium, this result is in line with a recent empirical study by Ivkovi´c, Sialm and
Weisbenner (2008) who …nd that individuals with more concentrated portfolios out perform those
with more diversi…ed portfolios.14
3.1.2 Performance of hedge funds in the "bad" period
In order to see whether there is any risk component associated with investing in more secretive
funds, we re-estimate the same set of speci…cations as before –now during the "bad" period from
April 2008 to March 2009. The results are reported in Table 3. We see a striking di¤erence to
performance premia observed during the "good" period.
Speci…cally, across all speci…cations the most secretive funds now under-perform the most trans-
parent ones. The most secretive funds on average earned 7%-17% lower return than the least
secretive funds, with most speci…cations being statistically signi…cant at conventional levels.15
14
In unreported results we …nd that highly concentrated funds are on average more volatile than less concentrated
funds.
15
In speci…cations 4 and 7 the coe¢ cient at highly-secretive funds is marginally signi…cant.
17
18. At the same time, even the moderately secretive funds signi…cantly under-performed the least
secretive funds –by 11%-21% per year. Since the di¤erence between the moderately secretive and
the highly secretive funds is not signi…cant at any reasonable level in any of the speci…cations, this
suggests that it is really the least secretive funds that were able to beat all other funds during bad
times.
Importantly, the last three speci…cations among other variables also control for net fund ‡ows,
so that the under-performance of highly- and moderately-secretive funds cannot be explained by
investors pulling money more out of these funds during the "bad" period. As we discuss in the
previous subsection, such di¤erential performance also cannot be attributed to di¤erent strategies
potentially loading di¤erently on various time-varying factors, or a time-varying pattern of hedge
funds selection by the fund of funds.
We interpret these results as strong evidence of secretive funds loading more on certain risk
factors that earn a risk premium during good times and that manifest in risk realizations during
bad times. Higher managerial skill or superior proprietary strategies of secretive funds on their own
are not consistent with the observed pattern of performance. Secretive funds may of course have a
higher skill on top of taking more risks. However, in order to disentangle the two quantitatively, one
would need to know the full model of fund returns. The bene…t of our empirical setup lies in the
opportunity to assess whether secretive funds load more on risk factors or their over-performance
during good times can be fully attributed to higher managerial skill and superior strategies, without
having to know the full model of fund returns. In case of hedge funds such a model may be especially
di¢ cult to construct, since much of the variation in hedge fund returns remains unexplained even
after accounting for many risk factors. Instead, we only rely on the assumption that some of the
risk factors that secretive funds may have loaded more aggressively on, crashed during the period
of the global …nancial crisis.
For the purpose of illustrating the identi…cation strategy in the Section 2 we assumed that
factor loadings were constant over time. However, a possible explanation for secretive funds over-
performing transparent funds during good times, could be a better market-timing ability of the
managers of secretive funds. In particular, they could be optimally adjusting their loadings upwards
on factors that perform well in good periods as would be optimal to do in order to earn a higher
return. However, if secretive funds were indeed better market timers, than transparent ones, they
should have adjusted loadings on factors that perform poorly during the bad times downwards, in
order to not lose as much. The observed performance pattern during bad times is not consistent
18
19. with secretive funds being purely better market timers. If anything, the ability of secretive funds
to time some of the factors well, would mean they loaded even more on factors that they couldn’t
time (and that crashed during the bad period).
Going back to other qualitative characteristics, we note that in the "bad" period the most illiquid
funds dramatically under-performed the most liquid funds, with magnitudes between 34%-47% on
an annual basis depending on the speci…cation. At the same time, moderately illiquid funds under-
performed the most liquid funds on average by 12%-23% per year, suggesting that the most illiquid
funds could beat the other two types of funds during this period. Since the performance di¤erence
reverts in the "bad" period, as compared to the "good" period, we conclude that this qualitative
characteristic is a good proxy for fund loading on the illiquidity factor, which was likely to crash
during the "bad" period as well. This also suggests that the performance di¤erences between
secretive and transparent funds cannot be attributed purely to a di¤erent loading on illiquidity
factor, but rather on some other factor that also collapsed during the "bad" period.
Finally, we …nd some performance reversal with respect to concentration as well: highly concen-
trated funds on average under-performed less concentrated funds during the "bad" period. These
di¤erences are not statistically signi…cant at conventional levels, but they are clearly not positive
and signi…cant as we observed during the "good" period. Such pattern may also signal of concentra-
tion commanding a risk premium due to some limited ability of investors to diversify concentrated
risks.
In summary, in Tables 2 and 3 we document that secretive …rms earn more positive returns than
transparent funds during an up market, but signi…cantly worse returns during a down market. This
is consistent with the funds loading on risk(s) which carry a premium during the up market, but
the funds su¤er severe losses when that risk is realized.
3.2 Di¤erences in risk-adjusted performance of secretive and transpar-
ent funds
In this section we use several methods to control of risk of the funds strategies. We begin with
a model-free approach, which uses strategy- and sub-strategy-month …xed e¤ects. The results of
the model-free approach can be interpreted in one of two ways. Either as a risk adjustment for
any possible, even unknown, factors or as the equivalent of strategy and sub-strategy matching.
The former interpretation allows us to identify more clearly the skill (or lack of skill) of the hedge
funds. The later allows us to examine whether secretive funds out or under-perform their strategy
19
20. or sub-strategy matched peers; that is, are their returns di¤erent from other funds which follow the
same investment strategy. Once again we separately examine the good and bad periods to allow
funds to adjust their risk exposure across periods.
Our second approach is model based. For the second approach we will try di¤erent priced and
unpriced factors to see if we can explain the return the prior analyses identi…ed as excess. If we can
then we may be able to explain the di¤erences in risk that secretive and non-secretive funds take.
3.2.1 Abnormal Performance of hedge funds in the "good" period
In table 4 we examine the di¤erence in abnormal returns to secretive funds controlling for the
risk of the fund. In Column 1 we regress hedge fund returns on a dummy for secretive funds and
include strategy-month …xed e¤ects. This is equivalent to matching secretive and transparent funds,
which claim to follow the same investment strategy. Speci…cally, we model excess returns as the
following:
Rit = aM
SecSecM
i + aH
SecSecH
i + dst + "it
where dst is a group-month (strategy-month) …xed e¤ect. The advantage of this method is that
it automatically subsumes any group-speci…c loadings on all factors, including unknown factors,
and, in this sense it is “model-free”because we do not need to know all the factors that are relevant
for a particular hedge fund style. The downside of this approach is that it leaves open the possibility
that di¤erences are nonetheless due to di¤erences in risk. That said, this method does e¤ectively
match funds on strategy and substrategy, allowing us to compare the performance of funds that
purport to follow the same strategy and which, therefore ought to have similar risk exposure and
similar performance.
The evidence in Column 1 is similar in magnitude to the raw results of Table 2. Secretive
funds earn 4.61% more on an annual basis than do transparent funds, controlling for the investment
strategy of the fund. As in Table 2, one might be concerned that we confound the skill of hedge
fund managers in picking assets and the skill of the fund-of-fund manager in picking hedge funds,
so in Column 2 we restrict the same to only those funds which are present in the entire sample. In
addition we require the same transparency measure across all years. Notably the results are even
more pronounced as returns are nearly 10.5% higher for secretive funds than non-secretive on a
strategy-matched basis.
Concerned that the strategy classi…cations might be too broad, we also examine sub-strategy-
month …xed e¤ects. Substrategy-month …xed e¤ects are superior to the strategy-month …xed e¤ects
20
21. because we are much more likely to have as a comparison group exactly comparable funds; but, this
comes at a cost: there are many fewer funds in each substrategy, which almost certainly reduces the
power of our tests. In Column 3 we see that even with a narrower de…nition of strategy secretive
funds out perform non-secretive in the “good”period. Even the magnitude is similar at 3.66% per
year, though it is only statistically signi…cant at the 10% level. As before in Column 4 we require
that the fund exist through the entire sample and have the same transparency score. Once again
the results are more pronounced, yielding a 7.74% di¤erence between secretive and transparent
funds per year. If we were to focus only on the up-market, it would appear that secretive managers
are using their discretion for the bene…t of their investors and they appear to have superior stock
picking skill.
One might want us to control for risk using standard asset or hedge fund pricing models. We
believe this is tantamount to assuming that the model we measure ex-post is identical to the model
investors assumed ex-ante. We …nd it much more credible that investor believe that funds with
similar strategies are similarly risky. Nonetheless, in Columns 5 –8 we control for risk using two
standard models. Columns 5 and 6 we use a market model to control for risk. Similar to the
presentation above, Column 5 present the market model for the full sample and Column 6 requires
that the fund be present in the full sample with an unchanging secrecy measure. If anything, the
results are more pronounced, which suggests that this simple market model is missing important risk
exposure. We repeat the exercise for the Fama-French (1993) three-factor model and …nd similar
results. Unfortunately, models with more factors, such as the Fund and Hsieh (2001) seven-factor
model, quickly use up degrees of freedom leading to over …t.
3.2.2 Abnormal Performance of hedge funds in the "bad" period
As in Table 4, in Table 5 we examine the di¤erence in abnormal returns to secretive funds
controlling for the risk of the fund, but this time in the bad period from April 2008 to March
2009. In Columns 1 and 3 we regress hedge fund returns on a dummy for secretive funds and
include strategy-month …xed e¤ects in Column 1 and sub-strategy …xed in Column 3. While the
magnitude of the abnormal performance of secretive funds is somewhat larger than for raw returns,
the statistical signi…cance is weaker. At the 10% level in Column 1 and at the 11% level in Column
3. The weaker results are in part due to the lower power of the tests. In Columns 2 and 4, as above
we restrict the sample to only those funds which are include through the entire sample and do not
change the level of secrecy. Once again the results are more pronounced. Finally, In Columns 5
through 8, we again control for a simple market model and the Fama-French three-factor model.
21
22. Results are similar in strength to the raw results.
The important take away from these …ndings is that while secretive funds appear to earn higher
returns during the up market, they appear to do so at the expense of high losses during the down
market. This is consistent with the funds advertising one strategy or sub-strategy, but actually
doing another higher risk strategy.
4 Are investors fooled? Flow-to-performance sensitivity
Flows may act as a disciplining mechanism, keeping hedge fund managers incentives aligned with
their investors, under the threat of losing the asset based from which they derive their fees. We
hypothesize that secretive funds will be less sensitive to past performance than transparent funds
as it may be harder for investors to infer deviations from the secretive fund’s declared strategy (see
Huang, Wei and Yan, 2012). In Table 6 we examine ‡ow to performance, regressing ‡ow on the
return over the past quarter plus controls. The speci…cation is similar to other ‡ow-to-performance
regressions in the literature (Bollen and Pool, 2012, Sialm, Starks and Zhang, 2015, and Sirri and
Tufano,1998) except that instead of rank-based performance measures, we use the past quarter’s
return.
NetFlowi;t+3 = c + aM
SecSecM
i + aH
SecSecH
i + L
SecLi;t (Ri;t rf;t) + M
SecMi;t (Ri;t rf;t) +
H
SecHi;t (Ri;t rf;t) + X0
i;t + ds + "it
We opt for the raw return instead of a relative performance rank as in prior research because,
while we believe our database is representative of the universe of hedge funds, it is not the entire
universe. Columns 1 and 3 of Table 6 show that ‡ows generally chase past quarterly, but that
during the up market the sensitively is entirely driven by transparent funds, controlling for illiquidity,
strategy …xed e¤ects and various other controls. This is consistent with the hypothesis that investors
are better able to make inferences about managerial quality for transparent funds. It is notable that
during the down market secretive funds become sensitive to past ‡ows, even when controlling for
fund illiquidity, consistent with the notion that investors are only able to learn about the relative
skill of management during the down-market, much in the same way earlier in the paper we use the
di¤erent market performance to identify di¤erences in performance due to exposure to unobserved
factors.
22
23. 5 Concluding Remarks
In this paper we use proprietary data obtained from a fund of funds to document that funds
that are less willing to share the information about their holdings and trades with their investors
signi…cantly out-perform the less secretive funds during an up market. This …ndings holds both
controlling for the strategy and substrategy of the funds. We further investigate the source of this
out-performance and conclude that it cannot be explained purely by superior instrument-picking
skill, trading strategy or market-timing ability of more secretive funds. By looking separately at
good and bad periods we infer that at least part of this out-performance is explained by secretive
funds loading more than transparent ones on risk factors that earn a risk premium during good
times, but crash during bad times.
The bene…t of our empirical setup lies in the opportunity of making such an assessment irre-
spective of knowing the true model that drives hedge fund returns, but instead by relying on the
assumption that some of the risk factors that secretive funds may have loaded more aggressively
on, crashed during the period of the global …nancial crisis.
Our evidence on the ‡ow to performance sensitivity of the funds shows that transparent funds
are more sensitive to past performance than secretive funds, consistent with investors having a more
di¢ cult time making inferences when signals are obscured. Perhaps, this suggests that these savvy
investors were fooled. However, the evidence is consistent with the investors in secretive hedge
funds using the crash to infer skill in much the same way we do in this paper.
23
24. References
[1] Acharya, V. V., Pedersen, L. H., 2005, “Asset Pricing with Liquidity Risk,”Journal of Financial
Economics, 77, 2, 375-410.
[2] Aggarwal, R.K., and Jorion P., 2012, “Is There a Cost to Transparency?," Financial Analysts
Journal, 68(2), 108-123.
[3] Agarwal, V., and Naik N.Y., 2004, “Risks and portfolio decisions involving hedge funds,"
Review of Financial Studies, 17(1), 63-98.
[4] Agarwal, V., W.Jiang, Y.Tang, and B.Yang, 2013, "Uncovering Hedge Fund Skill from the
Portfolio Holdings They Hide, Journal of Finance, 68(2), 739-783.
[5] Amihud, Y., Mendelson, H., 1986, “Asset Pricing and the Bid-ask Spread,”Journal of Financial
Economics, 17, 223-249.
[6] Ang, A., Gorovyy, S., and van Inwegen, G. B., 2011, “Hedge Fund Leverage,” Journal of
Financial Economics, 102, 1, 102-126.
[7] Ang, A., Hodrick, B., Xing, Y., and Zhang, X., 2009, “High Idiosyncratic Volatility and Low
Returns: International and Further U.S. Evidence” Journal of Financial Economics, 91, 1,
1-23.
[8] Anson, M.J.P., 2002, “Hedge Fund Transparency,”The Journal of Wealth Management, 5, 2,
79-83.
[9] Aragon, G.O., 2007, “Share Restrictions and Asset Pricing: Evidence from the Hedge Fund
Industry,”Journal of Financial Economics, 83, 1, 33-58.
[10] Bali, T. G., Brown, S. J., and Caglayan, M. O., 2011, "Do hedge funds’exposures to risk
factors predict their future returns?," Journal of Financial Economics, 101, 1, 36-68.
[11] Brown, S., Goetzmann, W., Liang, B., Schwarz, C., 2008, “Estimating Operational Risk for
Hedge Funds: The !-Score,”working paper, Yale.
[12] Brown, S., Goetzmann, W., Liang, B., Schwarz, C., 2012, “Trust and Delegation,”Journal of
Financial Economics, 103, 221-234.
24
25. [13] Carhart, Mark M., 1997, On persistence in mutual fund performance, Journal of Finance 52,
57–82.
[14] Financial Crisis Inquiry Commission, 2011, “The Financial Crisis Inquiry Report: …nal report
of the national commission on the causes of the …nancial and economic crisis in the United
States,”PublicA¤airs, Perseus Books Group.
[15] Glode, V., Green, R.C., 2011, “Information Spillovers and Performance Persistence for Hedge
Funds,”Journal of Financial Economics, 101, 1-17.
[16] Goltz, F., Schroder, D., 2010, “Hedge Fund Transparency: Where Do We Stand?,”The Journal
of Alternative Investments, 12, 4, 20-35.
[17] Hedges, J.R.IV, 2007, “Hedge Fund Transparency,”The European Journal of Finance, 11, 5,
411-417.
[18] Huberman, G., and Halka, D., 2001, “Systematic Liquidity,” The Journal of Financial Re-
search, 24, 2, 161-178.
[19] Markowitz, H., 1952, “Portfolio Selection,”Journal of Finance, 7, 1, 77-91.
[20] Liang B., 1999, “On the Performance of Hedge Funds,” Financial Analysts Journal, 55, 4,
72-85.
[21] Markowitz, H., 1952, “Portfolio Selection,”Journal of Finance, 7, 1, 77-91.
[22] Merton, R.C., 1987, “Presidential address: a simmple model of capital market equilibrium with
incomplete information,”. Journal of Finance, 42, 483-510.
[23] Pastor, L., Stambaugh, R.F., 2003, “Liquidity risk and expected stock returns,” Journal of
Political Economy, 111, 642-685.
[24] Petersen, Mitchell A., 2009, "Estimating Standard Errors in Finance Panel Data Sets: Com-
paring Approaches", Review of Financial Studies, 22, 435-480.
[25] Schmalz, Martin C., and Zhuk, Sergey, 2013, "Revealing Downturns", working paper available
at SSRN: http://ssrn.com/abstract=2176600.
25
26. Figures and Tables
Figure 1
Note: This figure plots average performance of the most secretive and the most transparent
hedge funds (in annualized percentage points) over the period from April 2006 to March 2009,
for funds that have non-missing returns over the entire period.
Figure 2
Note: This figure plots average performance of the most secretive and the most transparent
hedge funds (in annualized percentage points) over the period from April 2006 to March 2009,
for funds that have non-missing returns and the same transparency score over the entire period.
25
27. Panel A: Number of observations by fund strategy and year
Strategy April 06-March 07 April 07-March 08 April 08-March 09 April 06-March 09
CR 152 156 131 439
ED 236 237 266 739
EQ 924 899 745 2568
RV 260 246 322 828
TT 91 72 110 273
Total 1663 1610 1574 4847
Panel B: Hedge fund summary statistics
Variable Period Mean Median Std. Deviation N
Excess Return April 06-March 07 7.66 8.25 28.47 1663
April 07-March 08 -1.27 1.44 44.92 1610
April 08-March 09 -21.90 -5.29 76.85 1574
Volatility April 06-March 07 6.80 6.03 4.55 1661
April 07-March 08 10.86 9.05 6.54 1610
April 08-March 09 15.66 12.37 10.24 1573
AUM March 07 1.99b 0.81b 3.19b 149
March 08 2.38b 1.04b 3.76b 138
March 09 1.85b 0.66b 2.98b 136
Notes: The sample includes all fundswith non-missing returns and a non-missing transparency score in a given
year. Excess return is the fund return net of a risk-free rate, and volatility is the within-fund return standard
deviation during the period (both measured in annualized percentage points). AUM is total assets under
management summed across all funds in the same family at the end of the corresponding period.
Table 1. Distribution of Hedge Funds and Summary Statistics
26
28. Panel C: Number of observations by fund secrecy score and year
Secrecy April 06-March 07 April 07-March 08 April 08-March 09 April 06-March 09
Low 326 308 210 844
Medium 1118 1062 1083 3263
High 219 240 281 740
Total 1663 1610 1574 4847
Panel D: Number of observations by fund secrecy score and strategy
Strategy Low Medium High Total
CR 144 190 105 439
ED 101 482 156 739
EQ 415 2043 110 2568
RV 72 450 306 828
TT 112 98 63 273
Total 844 3263 740 4847
Panel E: Pairwise rank correlations of fund scores
Secrecy Illiquidity Concentration Complexity
Secrecy 1.000
Illiquidity 0.1807*** 1.000
Concentration -0.0355 0.2049*** 1.000
Complexity 0.2580*** 0.1230** -0.1009** 1.000
Table 1 (cont'd). Distribution of Hedge Funds and Summary Statistics
Secrecy score
Notes: This panel reports pairwise Spearman rank correlations between secrecy, illiquidity,
concentration, and complexity. *, **, and *** indicate significance at the 10%, 5%, and 1% levels
correspondingly.
27
29. (1) (2) (3) (4) (5) (6) (7)
Secrecy Medium 1.85 1.91 1.75 1.47 1.51 1.09 4.40***
(1.58) (1.61) (1.96) (1.50) (1.55) (2.03) (1.36)
High 4.83*** 4.88*** 4.09* 4.64*** 3.96** 4.16* 8.64***
(1.76) (1.83) (2.39) (1.75) (1.96) (2.28) (1.95)
Illiquidity Medium 1.01 0.51 2.89 1.47
(1.24) (1.48) (1.79) (1.91)
High 6.51*** 6.51*** 7.69*** 4.97**
(1.55) (1.87) (1.91) (2.38)
Concentration Medium 2.88** 3.40** 3.11** 3.23**
(1.29) (1.35) (1.31) (1.36)
High 5.24 6.28* 7.85*** 7.05***
(3.59) (3.72) (2.10) (1.88)
Complexity Medium -0.09 -0.30 -1.75 -2.71**
(1.39) (1.47) (1.47) (1.28)
High -3.34** -2.52* -3.86** -3.16**
(1.28) (1.36) (1.49) (1.47)
Month FE Yes Yes Yes Yes Yes Yes
Hedge Fund Controls Yes Yes Yes
Present in 2007-2009 Yes Yes Yes
Same Secrecy in 2007-2009 Yes
Observations 1,663 1,663 996 1,663 1,642 995 779
Number of funds 150 150 83 150 149 83 65
Adjusted R2
0.001 0.170 0.185 0.181 0.226 0.253 0.265
where Rit-rft is the excess return of fund i in month t ; SecMi (SecHi) is an indicator variables that equals 1 if a
fund is rated as moderately (highly) secretive, and 0 otherwise; Xit are fund-level controls (log of assets under
management and percentage net flows during past month), in specifications 5 to 7; and dt are month fixed
effects, in specifications 2 to 7. Specifications 4 to 7 additionally include indicator variables for
moderately/highly illiquid funds, moderately/highly concentrated funds, and moderately/highly complex funds.
Specifications 3, 4, 6, and 7 estimate the results using only the funds that are present in the sample in all months
from April 2006 to March 2009. Specifications 6 and 7 estimate the results using only the funds that have the
same level of secrecy in all months from April 2006 to March 2009. Standard errors are clustered at the fund
level and are reported below the coefficients. * indicates 10% significance; ** 5% significance; *** 1%
significance.
Rit - rft = c + αM
SecMi + αH
SecHi + Xit′γ + dt + εit ,
This table reports the results of estimating the following specification during the period between April 2006 and
March 2007:
Table 2. Hedge fund performance: April 2006 to March 2007 ("good" period)
28
30. 1 2 3 4 5 6 7
Secrecy Medium -15.35*** -14.86*** -17.75*** -11.20*** -15.54*** -21.67*** -22.69***
(4.93) (5.08) (6.69) (3.94) (4.49) (6.30) (6.95)
High -9.98* -10.10* -16.73** -7.02 -14.12** -15.40* -14.25
(5.41) (5.53) (6.41) (4.28) (5.80) (8.08) (10.01)
Illiquidity Medium -13.36*** -22.49*** -24.67** -22.63**
(4.14) (6.95) (9.96) (10.30)
High -33.70*** -44.02*** -41.25*** -42.99***
(4.94) (6.19) (7.21) (8.26)
Concentration Medium -4.69 -2.40 -3.28 -3.57
(4.10) (4.51) (5.35) (6.30)
High -8.65 -2.89 -6.88 -9.10
(7.94) (7.41) (9.15) (11.79)
Complexity Medium 3.51 5.35 7.21 14.72**
(5.99) (6.22) (7.10) (7.26)
High 10.69** 11.89** 3.84 8.35
(4.90) (5.88) (6.30) (7.20)
Month FE Yes Yes Yes Yes Yes Yes
Hedge Fund Controls Yes Yes Yes
Present in 2007-2009 Yes Yes Yes
Same Secrecy in 2007-2009 Yes
Observations 1,574 1,574 1,020 1,574 1,560 1,012 780
Number of funds 140 140 85 140 138 85 65
Adjusted R2
0.003 0.177 0.217 0.207 0.304 0.332 0.307
Table 3. Hedge fund performance: April 2008 to March 2009 ("bad" period)
This table reports the results of estimating the following specification during the period between April 2008 and
March 2009:
Rit - rft = c + αM
SecMi + αH
SecHi + Xit′γ + dt + εit ,
where Rit-rft is the excess return of fund i in month t ; SecMi (SecHi) is an indicator variables that equals 1 if a
fund is rated as moderately (highly) secretive, and 0 otherwise; Xit are fund-level controls (log of assets under
management and percentage net flows during past month), in specifications 5 to 7; and dt are month fixed
effects, in specifications 2 to 7. Specifications 4 to 7 additionally include indicator variables for
moderately/highly illiquid funds, moderately/highly concentrated funds, and moderately/highly complex funds.
Specifications 3, 4, 6, and 7 estimate the results using only the funds that are present in the sample in all months
from April 2006 to March 2009. Specifications 6 and 7 estimate the results using only the funds that have the
same level of secrecy in all months from April 2006 to March 2009. Standard errors are clustered at the fund
level and are reported below the coefficients. * indicates 10% significance; ** 5% significance; *** 1%
significance.
29
31. (1) (2) (3) (4) (5) (6) (7) (8)
Secrecy Medium 1.65 5.78*** 1.59 3.36 6.98*** 8.39*** 5.66*** 5.90***
(1.60) (1.62) (1.79) (2.51) (0.00) (0.00) (0.00) (0.00)
High 4.61** 10.49*** 3.66* 7.74** 9.57*** 7.16*** 13.04*** 7.62***
(2.10) (2.71) (2.18) (3.28) (0.00) (0.00) (0.00) (0.00)
Present in 2007-2009 Yes Yes Yes Yes
Same Secrecy in 2007-2009 Yes Yes Yes Yes
Observations 1,663 780 1,663 780 1,663 780 1,663 780
Number of funds 150 65 150 65 150 65 150 65
Adjusted R2
0.255 0.322 0.480 0.612 0.390 0.415 0.546 0.565
where Rit-rft is the excess return of fund i in month t ; SecMi (SecHi) is an indicator variables that equals 1 if a fund is
rated as moderately (highly) secretive, and 0 otherwise. The Risk-Adjustment term depends on the specification: 1 to 4
are "model-free", while 5 to 8 assume a specific factor model. Columns 1 and 2 inlude strategy-month fixed effects to
control for strategy-specific loadings on any factors. Columns 3 and 4 inlude substrategy-month fixed effects to control
for substrategy-specific loadings on any factors. Columns 5 and 6 control for fund-specific loadings on MktRf factor.
Columns 7 and 8 control for fund-specific loadings on MktRf, SmB, and HmL. Specifications 2, 4, 6, and 8 estimate the
results using only the funds that are present in the sample and have the same level of secrecy in all months from April
2006 to March 2009. Standard errors are clustered at the fund level and are reported below the coefficients. * indicates
10% significance; ** 5% significance; *** 1% significance.
Rit - rft = c + αM
SecMi + αH
SecHi + Risk-Adjustment + εit,
This table reports the results of estimating the following specifications during the period between April 2006 and March
2007:
Table 4. Hedge fund risk-adjusted performance: April 2006 to March 2007 ("good" period)
Strategy-specific
loadings
Substrategy-specific
loadings
Model-free Risk Adjustment Factor model Risk Adjustment
Market model
3-factor Fama-French
model
30
32. (1) (2) (3) (4) (5) (6) (7) (8)
Secrecy Medium -12.00** -27.26*** -7.05 -12.22 -2.60*** -9.61*** -3.84*** -15.90***
(5.26) (9.34) (4.42) (7.46) (0.00) (0.00) (0.00) (0.00)
High -10.68* -31.51*** -5.79 -19.52** -14.18*** -9.56*** -13.55*** -13.78***
(5.65) (10.76) (6.47) (9.35) (0.00) (0.00) (0.00) (0.00)
Present in 2007-2009 Yes Yes Yes Yes
Same Secrecy in 2007-2009 Yes Yes Yes Yes
Observations 1,574 780 1,574 780 1,574 780 1,574 780
Number of funds 140 65 140 65 140 65 140 65
Adjusted R2
0.247 0.253 0.617 0.813 0.442 0.382 0.642 0.575
Strategy-specific
loadings
Substrategy-specific
loadings
Market model
3-factor Fama-French
model
Table 5. Hedge fund risk-adjusted performance: April 2008 to March 2009 ("bad" period)
This table reports the results of estimating the following specifications during the period between April 2008 and March
2009:
Rit - rft = c + αM
SecMi + αH
SecHi + Risk-Adjustment + εit,
where Rit-rft is the excess return of fund i in month t ; SecMi (SecHi) is an indicator variables that equals 1 if a fund is
rated as moderately (highly) secretive, and 0 otherwise. The Risk-Adjustment term depends on the specification: 1 to 4
are "model-free", while 5 to 8 assume a specific factor model. Columns 1 and 2 inlude strategy-month fixed effects to
control for strategy-specific loadings on any factors. Columns 3 and 4 inlude substrategy-month fixed effects to control
for substrategy-specific loadings on any factors. Columns 5 and 6 control for fund-specific loadings on MktRf factor.
Columns 7 and 8 control for fund-specific loadings on MktRf, SmB, and HmL. Specifications 2, 4, 6, and 8 estimate the
results using only the funds that are present in the sample and have the same level of secrecy in all months from April
2006 to March 2009. Standard errors are clustered at the fund level and are reported below the coefficients. * indicates
10% significance; ** 5% significance; *** 1% significance.
Model-free Risk Adjustment Factor model Risk Adjustment
31
33. (1) (2) (3) (4) (3) (4)
Net Performance 0.0698*** 0.0396***
(0.0109) (0.00658)
SecL*Net Performance 0.105*** 0.0920*** 0.0518*** 0.0751***
(0.0143) (0.0350) (0.0165) (0.0194)
SecM*Net Performance 0.0499*** 0.0256 0.0425*** 0.0606***
(0.0116) (0.0325) (0.00686) (0.0159)
SecH*Net Performance 0.0474 0.0174 0.0263** 0.0492***
(0.0410) (0.0544) (0.0102) (0.0186)
SecM 3.462 -4.777** 3.671 -4.534**
(2.366) (2.010) (2.429) (1.894)
SecH -0.968 -1.902 -0.713 -1.457
(3.727) (2.458) (3.986) (2.468)
Median Strategy Flow 0.490** 0.479** 0.453** 0.826*** 0.841*** 0.839***
(0.210) (0.209) (0.206) (0.0559) (0.0579) (0.0558)
lnAUM -1.416** -1.423** -1.371* -0.480 -0.606 -0.446
(0.668) (0.707) (0.697) (0.549) (0.510) (0.495)
Annual volatility -0.0935 -0.0989 -0.106 0.0563 0.0765 0.0764
(0.0607) (0.0621) (0.0705) (0.0481) (0.0474) (0.0526)
Strategy FE Yes Yes Yes Yes Yes Yes
Hedge Fund Controls Yes Yes Yes Yes Yes Yes
Illiquidity and Illiquidity*Net Performance controls Yes Yes
Observations 1,113 1,113 1,113 1,273 1,273 1,273
Number of funds 113 113 113 114 114 114
Adjusted R2
0.146 0.163 0.169 0.473 0.483 0.491
Table 6. Hedge fund flow-to-performance sensitivity.
April 2006 to March 2007 ("good" period) April 2008 to March 2009 ("bad" period)
where NetFlowit+3 is the net quarterly flow to the fund from month t to t+3 , Rit-rft is the quarterly excess return of fund i
from month t-3 to t ; SecLit (SecMit, SecHit) is an indicator variables that equals 1 if a fund is rated as low- (moderately,
highly) secretive, and 0 otherwise; Xit are fund-level controls (log of assets under management, measured at t-3 , annual
volatility, measured from t-15 to t-3 , and median percentage net flows for funds in the same strategy, from t to t+3 ), and
ds are strategy fixed effects. Specifications 3 and 6 additionally include indicator variables for illiquidity, as well as their
interactions with performance. Specifications 1 to 3 estimate the model for months from April 2006 to March 2007 ("good"
period); specifications 4 to 6 estimate the model for months from April 2008 to March 2009 ("bad" period). Standard errors
are clustered at the fund level and are reported below the coefficients. * indicates 10% significance; ** 5% significance; ***
1% significance.
NetFlowit+3 = c + αM
SecMit + αH
SecHit + γL
SecLit *(Rit - rft)+ γM
SecMit*(Rit - rft) + γH
SecHit*(Rit - rft) + Xit′δ + ds + εit ,
This table reports the results of estimating the following specification for various periods:
32