Você está na página 1de 16

Analysis of hedge fund strategies using slack-based DEA models

Author(s): UD Kumar, AB Roy, H Saranga and K Singal


Source: The Journal of the Operational Research Society, Vol. 61, No. 12 (December 2010),
pp. 1746-1760
Published by: Palgrave Macmillan Journals on behalf of the Operational Research Society
Stable URL: http://www.jstor.org/stable/40926105
Accessed: 06-07-2016 21:10 UTC

REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/40926105?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.

Operational Research Society, Palgrave Macmillan Journals are collaborating with JSTOR to
digitize, preserve and extend access to The Journal of the Operational Research Society

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
Journal of the Operational Research Society (2010) 61, 1746-1760 2010 Operational Research Society Ltd. All rights reserved. 0160-5682/10 ^

www.palgrave-jou rnals.com/jors/

Analysis of hedge fund strategies using slack-based


DEA models
UD Kumar, AB Roy, H Saranga* and K Singal
Indian Institute of Management, Bangalore, India

Hedge funds have made a significant impact on the performance of world financial markets in recent times. Our
objective in this paper is to develop a robust framework for the evaluation of hedge funds by incorporating a
maximum number of performance measures through public data sources. We analyse the hedge fund strategies
(styles) using a variety of classical risk-return measures with the help of slack-based Data Envelopment
Analysis (DEA) models to determine a unique performance indicator. The main thrust is to investigate the risk
return profile of 4730 hedge funds classified under 18 different strategies using multiple inputs and outputs. The
originality of the work lies in applying Slack-Based DEA to decipher the risk-return profile of these strategies
using advanced risk-return measures such as Value at Risk, drawdown, lower and higher partial moments and
skewness. We find that the correlation between the ranking of hedge fund strategies based on Sharpe ratio and
the DEA models is very low; at the same time, there is a significant correlation between rankings obtained
by the application of DEA using different sets of input/output measures. We have also compared the DEA
rankings with other traditional financial ratios such as modified Sharpe ratio, Sortino ratio and Calmar ratio.
The paper also studies the impact of events such as the Asian financial crisis on the performance of hedge
funds. The study around the event shows that only a relatively small number of strategies performed better
during times of turmoil.
Journal of the Operational Research Society (2010) 61, 1746-1760. doi:10.1057/jors.2009.143
Published online 11 November 2009

Keywords: data envelopment analysis; hedge fund; Sharpe ratio; Calmar ratio; Sortino ratio

Introduction (Eling and Schuhmacher, 2007). Scholz (2007) has discussed


several modifications to the Sharpe ratio to overcome some
Hedge funds have always intrigued regular investors. With
of the limitations of Sharpe ratio, but all these modifications
the outlandish fees that hedge funds charge, it is natural
still consider single risk and single return measure to capture
that investors expect high returns compared to other invest-
ment choices such as mutual funds. The sector's allure is in the performance of an asset.
The lack of a comprehensive methodology to evaluate
its supposed ability to make positive returns irrespective of
hedge fund performance is attributed to the complex nature
market conditions. The potential for absolute returns, plus the
of trading strategies adopted by the hedge fund managers,
lack of correlation between hedge funds and almost all other
which are often not revealed in the public domain. Although
asset classes, is an irresistible mix for most investors (Prosser,
most hedge funds use many of the same liquid asset classes
2007). As a result, a significant portion of academic research
as mutual funds, they use dynamic trading strategies that
on hedge funds focuses on the performance evaluation of
frequently involve short sales, leverage and derivatives with a
hedge funds, which has resulted in a multitude of perfor-
mandate to make absolute returns, irrespective of the market
mance measures that have cropped up over the last few years.
environment (Fung and Hsieh, 1997). Therefore, any perfor-
However, a classical measure like the Sharpe ratio (Sharpe,
mance evaluation methodology for hedge funds needs to
1966, 1994), which was initially developed for mutual fund
take into account both the underlying strategy as well as the
evaluation, is one of the most widely used metrics in the hedge
risk-return structure of fund performance in a comprehensive
fund performance measurement till date, despite the obvious
manner. The single input, single output performance indices
unsuitability of the Sharpe ratio for hedge fund evaluation due
that are in popular use today, such as Omega, Sortino Ratio,
to the non-normal nature of hedge fund returns (Kao, 2002;
Calmar ratio, Sterling ratio, Jensen measure, Treynor ratio
Gregoriou and Gueyie, 2003). This raises questions about the
and Modified Sharpe ratio etc. (Sortino and van der Meer,
suitability of the multitudes of these other metrics, which most
1991; Gregoriou and Gueyie, 2003; Agarwal and Naik, 2004;
often result in an identical ranking to that of the Sharpe ratio
Alexander and Baptista, 2004) consider only one aspect of
return and one aspect of risk, forcing one to consider all these
* Correspondence: H Saranga, D-108, Faculty Block D, UM Bangalore,
Bannerghatta Road, Bangalore, Karnataka 560076, India. performance indices simultaneously to incorporate other
E-mail: haritha.saranga@gmail.com aspects of risk and return, creating further confusion in the

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
UD Kumar et a- Analysis of hedge fund strategies 1 747

minds of investors. To address this drawback in hedge fund study we have found a relatively high correlation between
performance measurement, some researchers have proposed performances of hedge funds within a particular strategy as
the application of Data Envelopment Analysis (DEA)-based compared to that of funds across different strategies. Here, we
models, which have the ability to incorporate multiple inputs would like to draw attention to the fact that we have also used

and multiple outputs simultaneously and result in a single a larger data set (around 5000 hedge funds) spread across
performance measure. Eling (2006) has listed all the existing many strategies (18 strategies), as compared to previous
DEA-based studies on the hedge funds and proposes an DEA studies on hedge funds (Gregoriou, 2003; Gregoriou
appropriate framework for the selection of inputs and outputs, et al 2005; Wilkens and Zhu, 2005; Eling, 2006;
which plays a very critical role in DEA efficiency evaluation. Nguyen-Thi-Thanh, 2006). We have also compared hedge
Selection of inputs and outputs in DEA is a highly researched fund strategy ranking of DEA models with traditional finan-
topic in DEA as the selected inputs and outputs can change cial ratios such as modified Sharpe ratio, Sortino ratio and
the efficiency scores of Decision Making Units (DMUs) Calmar ratio.

(Norman and Stocker, 1991; Casu et al, 2005). In the case


of hedge funds, the DEA efficiency scores should be treated Growth of hedge funds and performance measurement
as the performance indicators of hedge funds, however,
Hedge fund evolution
throughout this paper, we use the terminology 'efficiency'
to be consistent with the DEA literature. Eling (2006) also The first known hedge fund was started by Alfred Jones in
discusses the most suitable DEA models for hedge fund 1949. Jones used a combination of both leverage and short-
performance evaluation based on a variety of factors, such selling strategies to enable investing in both rising and falling
as scale advantages/disadvantages, ranking requirements markets. During the 1950s and 1960s Jones's fund consis-
etc. However, firstly, none of the existing DEA applications tently outperformed the best equity mutual funds and thus
have considered an integrated approach to evaluate both the attracted more players in the hedge fund arena. By 1968 there
hedge fund strategies and the fund performance in tandem. were 200 hedge funds in the US. The 1990s saw a stellar
Secondly, most of the proposed DEA models so far are appli- growth in hedge funds with exceptional performance by few
cable only for non-negative input/output data and no DEA star managers. The number of hedge funds grew from 610
application to hedge funds has made use of slack-based DEA in 1990 to more than 9400 in 2006 (Naik and Tapley, 2007).
models, thus the multi-dimensional performance evaluation Figure 1 shows the growth of hedge funds between 1990 and
capability of DEA is severely underutilized. One exception is 2005. Throughout this period, hedge funds continued to attract
the study by Nguyen-Thi-Thanh (2006), who suggests use of investment from high net worth individuals, private banks,
a selection of classical DEA models depending upon whether insurance companies, pension plans and hedge fund of funds.
the negative data is present in the input or output variables. However, in the process, hedge funds have also attracted a
In the current study, we try to bridge this gap by firstly lot of bad reputation to their name. A speculation regarding
proposing an integrated framework that evaluates strategy their large carry trade positions in yen led the meltdown of
wise performance in Stage 1 and identifies the best funds the financial markets on 27th February 2007. The case of the
from the most efficient strategies in Stage 2, by incorpo- Bayou Group hedge fund has been making rounds in courts
rating multiple aspects of risk-return measures simultane- after the group declined to return the principal amounts to
ously. Secondly, we propose the application of Slack-based the investors who had exited from the fund before its crash

DEA models that are capable of analysing negative data for (Justin, 2005). The more publicized Amaranth hedge fund has
both efficiency estimations and super efficiency estimations been found guilty of taking huge bets on natural gas futures.
in order to rank the respective strategies and corresponding When faced with a liquidity crunch in the gas futures market,
hedge funds. the fund had to file for bankruptcy, bringing bad repute to
Hedge fund managers use more than 30 investment strate- the whole hedge fund industry (White, 2006). Hedge funds
gies. The criteria used to classify hedge funds into various over the years have seen many scandals; the recent one is
categories are discussed in Getmansky et al (2004). Fung the Bernard Madoff scandal that resulted in a loss to the

and Hsieh (1997) identified dominant investment strategies tune of $50 billion to the investors. However, Gregoriou and
in hedge funds by analysing 409 hedge funds. From an Lhabitant (2009) claim that Madoff never claimed to be a
investor's perspective it is worth analysing the performance hedge fund manager. Bernard Madoff Investment securities
of hedge fund strategies, as most often strategy is the only (BMIS) claimed that they are able to execute a conserva-
information revealed to the investor by the fund managers. In tive strategy that would deliver annual return of 10%- 12%
the second stage, we club all the hedge funds under the most by actively trading a very specific portfolio of stocks and
efficient strategies into one sample and evaluate their relative options. Gregoriou and Lhabitant (2009) also point out that
performance through application of slack-based super effi- several banks refused to do business with Madoff and one of

ciency DEA models. This two-stage analysis not only allows them even blacklisted BMIS. In spite of all these problems,
us to identify the best strategies but also enables us to zero in the hedge funds industry has seen a phenomenal growth till
on the most efficient funds among the best strategies. In our 2008 and industry experts predict hedge funds to grow further

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
1748 Journal of the Operational Research Society Vol. 61, No. 12

Figure 1 Growth of hedge funds between 1990 and 2005.


Source: Figure 1 is plotted using the hedge fund data from Bloomberg Database.

Returns - Stock market vs Hedge funds

1600000 -i ^^
1400000 - w00000000"^ ^^

| 800000 - h**<***+^^^000^'
600000 - ^"T^ ^"T^
400000 -
200000 -
0 -I

2000 2001 2002 2003 2004 2005 2006 2007


Year

A Stock returns B Hedge funds returns

Figure 2 Returns of Hedge funds vis--vis stock market index.

(Nayar, 2009). We would like to point out that scandals are Hedge funds are speculative investments and the strategies
not unique to hedge funds; many frauds are reported in the used by the fund managers play a crucial role. Connor and
highly regulated mutual fund market as well. Lasarte (2005) provide a detailed analysis of various strate-
Hedge fund investors are concerned with two simple gies used by hedge fund managers and how they are used in
questions - the risk of investing in a hedge fund and the practice. For example, 'equity market neutral strategy' is the
expected returns. A look at the data regarding hedge fund original hedge fund strategy that uses a combination of buys
performance reveals that apprehensions about the underper- and short-sales (sometimes augmented by options and future
formance of hedge funds relative to the stock market are position) to offset correlation between portfolio return and
anything but valid. From the year 2001 to 2006, hedge funds the overall market return (Connor and Lasarte, 2005). Event-
have yielded excess returns to the tune of 30% on an average driven strategies look for events that are expected to make an
over the traditional investor's return from stock markets, as impact over a relatively short period of time.
can be noted from Figure 2, which compares the cumulative
returns that would be earned by an investor when investing The risk return measures of hedge funds
in a stock market portfolio vis--vis hedge funds with an The most commonly used measure of risk is volatility, that is,
initial investment of $1000000. We have used the net asset the annualized standard deviation of returns. Most academic
values (NAV) data downloaded from Bloomberg Database studies (Harper, 2003) demonstrate that hedge funds, on
to arrive at the Figure 2. The stock market performance is average, are less volatile than the market. For example, based
estimated based on the performance of S&P 500 index. It on our calculations using data from the Bloomberg database,
is seen that hedge funds have ubiquitously outperformed for the bull market period of 1994-2000, volatility of the
the stock market during the period. As the average investor S&P 500 was about 14%; whereas volatility of the aggre-
might suspect, the high returns come at considerably larger gated hedge funds was only about 10%. That is, for about
risks. It is a popular belief that most hedge funds use global two-thirds of the time, the expected returns were within 10%
macro strategies and simultaneously place large directional of the average return. Thus, contrary to popular belief, hedge
bets on stocks, currencies, bonds, commodities and gold, fund returns are relatively less volatile and more concentrated
thereby using a lot of leverage. than the market returns.

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
UD Kumar et a/- Analysis of hedge fund strategies 1749

However, the negative skewness in hedge funds has classical Sharpe ratio, which is widely considered as an
attracted the attention of several researchers and many theories unsuitable method of measuring the hedge fund performance
have been put forth to explain it (Mitchell and Pulvino, 2001; (Eling and Schuhmacher, 2007). Consequently, a single
Goetzmann et al, 2003; Taleb, 2004). The non-normality of measure that comprehensively reports the performance of a
hedge fund returns makes it imperative to devise new tech- hedge fund is still wanting. It is here that DEA can be used
niques of hedge fund risk and return measurement. As the to incorporate the numerous 'risk and return' measures for
typical distribution of a hedge fund is negatively skewed, a hedge fund and thereby arrive at a single efficiency score
measures of downside risk are more potent in describing describing the performance.
the actual risk associated with hedge funds (Jorion, 2000;
Estrada, 2001, Agarwal and Naik, 2004; Alexander and
Data envelopment analysis models for hedge funds
Baptista, 2004). Specifically, measures like Semi Deviation
(Estrada, 2001), Extreme Value Theory (EVT), Value at Risk Charnes et al (1978) introduced DEA to measure the scale
(VaR), Conditional Value at Risk (CVaR) etc, are more appro- efficiencies of various public sector firms. Central to a DEA
priate measures of risk. For return measures, Arithmetic model is the notion of a DMU, which is basically an abstract
Excess Returns (AER), Geometric Excess Returns (GER) and representation (of a firm in a given industry), that produces
Kurtosis have been found to be more suitable among others. a collection of outputs by consuming a set of inputs. In the
Liang and Park (2007) analysed the risk-return trade off current paper, different DMUs constitute the different strate-
in hedge funds using semi-deviation, VaR, expected shortfall gies used by different fund managers. The DEA models are
and tail risk using the deciles portfolio approach, mentioned based on a linear programming formulation that determines
in Fama and French (1992). They found that during the period the relative efficiency of a specific DMU with respect to the
1995-2004, hedge funds with high Expected shortfall (ES) remaining DMUs.
outperformed those with low ES by an annual return differ- The first DEA model, introduced by Charnes et al (1978),
ence margin of 7% after adjusting for auto-correlation and assumes that the size of a DMU does not affect the efficiency
heteroskedasticity. Agarwal and Naik (2004) used empirical of a firm. It is aptly called the Constant Returns to Scale
distributions of returns of funds to determine expected short- (CRS) model, also known as the CCR model, and provides
fall or CVaR. Their observations clearly reveal that the mean- the global technical efficiency measure. As, this assumption
variance framework is inadequate to measure the downside may not always hold good in practice, Banker et al (1984)
risk and hence they suggest the ES -optimization technique developed a DEA model called the variable returns to scale
instead. Bali and Gokcan (2004) estimated VaR for hedge fund (VRS) model, also known as the BCC model, which calcu-
portfolios using a normal distribution, a fat-tailed generalized lates local pure technical efficiency, under the assumption of
error distribution (GED), the Cornish-Fisher (CF) expansion VRS. The VRS model includes an additional constraint on
and the EVT. They used the hedge fund research indexes and 'jS (J^j^j - 1)' which while restricting the feasible region
found that the EVT approach and the CF expansion capture to a convex hull, ensures that the variable returns to scale
tail risk better than the other approaches. However, VaR is also assumption is satisfied.
subject to severe criticism. Lo (2001) points out that merely The use of risk as input and return as output to analyse
several years of historical data may not show the distribution the performance of financial assets was practised by many
of returns and therefore questions the usefulness of VaR-based researchers in the past. The reason for this is that the fund
risk management. Traditionally, VaR has always suffered from managers have the option of choosing a strategy that has
these theoretical shortcomings. It does not provide for the higher or lower risk (input) to achieve certain rate of return
magnitude of the possible losses below the threshold it iden- (output). Several authors have used DEA to analyse the
tifies. VaR also lacks convexity and monotonicity along with performance of hedge funds (Gregoriou, 2003; Gregoriou
continuity (Artzner et al, 1999). et al, 2005; Eling, 2006; Nguyen-Thi-Thanh, 2006).
Most models of performance analysis of financial assets Gregoriou et al (2005) have used BCC and super efficiency
use variance as the measure of risk and mean return as the BCC models to identify the number of efficient and ineffi-
measure of return from a portfolio. Variance does not separate cient funds with eight different strategies using a dataset of
the upside risk (that investors seek) from the downside risk 641 funds. Eling (2006) makes use of three different models,
(the detrimental one). Hence, a classical measure of perfor- namely, Constant Returns to scale (CCR), BCC and super-
mance like the Sharpe Ratio has been modified to incorpo- efficiency DEA to analyse the performance of hedge funds
rate alternative risk measures like drawdown, lower partial using inputs and outputs chosen with the help of rank corre-
moments and Modified Value at Risk, (MVaR) leading to lation and principal component analysis. In the case of hedge
the formulation of new measures like the Sortino Ratio and funds, the input and output measures, such as skewness of
Omega or Calmar Ratio. returns and the principal components of inputs and outputs
All financial ratios, whether classical or modern, allow can take negative values. As the traditional DEA models
for the integration of only one dimension of risk and like the CCR and the BCC cannot accommodate negative
return, which often results in similar rankings to that of the data, some data transformation methodology is usually used

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
1750 Journal of the Operational Research Society Vol. 61, No. 12

before the application of DEA models when negative data x >xo (7)
exists in both inputs and outputs. Nguyen-Thi-Thanh (2006),
for example, suggests adding a constant to outputs of all y^yo (8)
DMUs when negative values are present to transform them
into positive outputs. Seiford and Zhu (2002), however, state 3U>o
that this kind of transformation of data may affect DEA
We have used DEA Solver Pro 5.0 to solve the SBM and
results and rankings. Portela et al (2004) point out that the
Super-SBM models. The scheme used for transforming nega-
most frequently used DEA model in the presence of nega-
tive outputs in SBM models is discussed in the manual of DEA
tive data is the VRS additive model developed by Chames
et al (1985). However, the VRS additive model is not units Solver Pro 5.0 and by Duzakin and Duzakin (2007). In the
following section, we have briefly discussed the scheme used
invariant (Portela et al, 2004).
by DEA Solver Pro 5.0 to deal with negative output values.
Various DEA models have been proposed to deal with nega-
Assume that output r of DMU o is negative, that is, yro ^ 0.
tive data (Lovell and Pastor, 1995; Portela et al, 2004). Tone
Then we define two variables y+ and y~ as shown below:
(2001) and Duzakin and Duzakin (2007) have demonstrated
the use of slack-based DEA models to solve problems with
vr+ = max7=i,2,...,w{^; 'yrj > 0} (9)
negative outputs. The DEA models used in this paper have
been inspired by the works of Cooper et al (2000) and Tone y~ = min;=i)2,...,{yr7 |vr; > 0} (10)
(2001). The Slack-Based Measure (SBM) of Efficiency, which
If the output r has no positive elements, then we define
is described below in Model 1, gives a non-radial measure that
y+ = y~ = 1. Once the values of y+ and y~ are calculated
allows for non-proportional reductions in inputs or augmen-
using Equations (9) and (10), we replace the term s+/yroin
tations in outputs (Cooper et al, 2000; Tone, 2001).
the objective function (Equation 1) as stated below.
Model 1: Slack-based DEA model If v+ > y~ then we replace s+ /yro by the term

The slack-based DEA model with n DMUs, m inputs X=jc17 e


Rmxn and s outputs y = yu e Rsxn is given by: y7{y-y7)/{yt-yro) z z

If y+ = y- then we replace s+/yro by the term


1 - SLi^r/^
minp= s+

s {y7)2/B{y+ - yro) (U)


Subject to constraints: where B (in Equation 12) is a large value (in the DEA Solver
Pro 5.0 the value of B = 100). Note that the values of yro
X + s-=xo (2)
in the constraints are not changed. This scheme takes into
YX - s+ = y0 (3) account the magnitude of the non-positive output positively
and the score obtained is units invariant. Recently, Sharpe
et al (2007) presented a modified slack-based DEA model to
deal with negative inputs and outputs, generalizing the model
A DMU is SBM efficient if p* = 1, which is equivalent
to s~* = 0 and s+* = 0. The efficient DMUs can be further by Portela et al (2004).

discriminated using Super-SBM models (Tone, 2002).


Hedge fund data and DEA inputs and outputs
Model 2: Super-SBM model In this section, we describe the data collection methodology
Super slack-based DEA model is given by: for individual hedge funds and the corresponding hedge fund
strategies. We also describe two different methodologies used
to choose input/output sets for DEA application in the current
study.
min 6 = ^
-jyr=iyr/yro
s
Hedge fund data description
Subject to constraints:
For this study, data was collected from the Bloomberg
n
database on net asset values (NAVs) for 4730 hedge funds,
J2 jXj^x (5) spanning a time period of 14 years, from 1995 to 2007.
We initially began with a sample of 5520 hedge funds
n and withdrew 790 funds from the sample due to non-
E xjyj>y (6) availability of continuous data for the study period. The
data collected was on a monthly basis, which commenced

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
UD Kumar et a/ - Analysis of hedge fund strategies 1751

Table 1 DEA inputs-outputs


Authors Inputs Outputs

Gregoriou (2003) Lower partial moments of order 1 Higher partial moments of order 1
Gregoriou and Gueyie (2003) Lower partial moments of order 2 Higher partial moments of order 2
Lower partial moments of order 3 Higher partial moments of order 3
Wilkens and Zhu (2005) Standard deviation Average return
Lower partial moments of order 0 Minimum return
Skewness
Nguyen-Thi-Thanh (2006) Standard deviation Average return
Excess kurtosis Skewness
Eling (2006) Conditional value at risk Higher partial moment of order 0
Lower partial moments of order 0 Higher partial moment of order 3

Table 2 DEA inputs and outputs for hedge fund strategies

Risk Measures - Input Return Measures - Output

Standard Deviation Arithmetic Excess Return (AER)


Lower Partial Moment of order 0 Geometric Excess Return (GER)
Lower Partial Moment of order 1 Higher Partial Moment of order 0
Lower Partial Moment of order 2 Higher Partial Moment of order 1
Lower Partial Moment of order 3 Higher Partial Moment of order 2
Maximum Drawdown (MD) Higher Partial Moment of order 3
Average Drawdown (AD) Skewness
Standard Deviation of Drawdown (SDD)
Value at Risk (VaR)
Conditional Value at Risk (CVaR)
Modified Value at Risk (MVaR)

on January 1995 and continued till April 2007. These Pastor et al (2002) used the concept of efficiency contri-
hedge funds were grouped according to fund strategies. The bution measure that compares the efficiency scores of two
following 18 different strategies were found: Convertible, DEA models differing in either one input or output. To
Corporate/ Preferred, Currency, Derivative, Emerging market, reduce the curse of dimensionality Adler and Golany (2002)
Equity Directional, Equity Market Neutral, Event Driven, developed a model combining DEA and principle compo-
Fixed income directional, Fixed income relative value, nent analysis (PCA) to strengthen the discriminating power
Flexible portfolio, Geographically focused, Global Macro, of DEA. Serrano Cinca and Mar Molinero (2004) proposed
Government/ Corporate, Managed Futures, Multi-Strategy, a methodology that evaluates efficiencies for all input and
Sector Funds and Various Assets. After the grouping was output combinations that are further analysed using PCA.
done, the NAV for each strategy was arrived at using the Appropriate input and output selection has been one of the
average NAV of all funds within the strategy bucket. The most critical issues in the application of DEA to financial
NAV data was then normalized with the April 2007 NAV set assets (Eling, 2006; Nguyen-Thi-Thanh, 2006). Table 1 shows
to 1 . Subsequently, the monthly NAV data was used to calcu- the inputs/outputs that have been hitherto used for the analysis
late the returns for each of the 18 strategies for the period of hedge funds. Eling (2006) suggests the use of rank corre-
between January 1995 and April 2007. lation analysis and PCA to decide inputs and outputs, as they
One major argument against hedge funds is that only funds help not only in reducing the number of inputs and outputs,
that are performing well provide the data. Also, there could but also in identifying the most appropriate input/output set.
be a significant difference between the performance of live In the current study, while computing the performance scores
and dead funds. However, Ding and Shawky (2007) show in both stages of DEA implementation (for each strategy and
that at the strategy level there is no difference between the each fund within the most efficient strategy), risk measures
performance of dead funds and live funds. were taken as inputs and return measures as outputs. An
exhaustive list of the input and output parameters are given
Selection of inputs and outputs for DEA analysis in Table 2. Traditionally, researchers in the field of financial
Selection of appropriate DEA models, especially the inputs asset management have used risk measures such as Sharpe
and outputs has been a focus of DEA research for many ratio, modified Sharpe ratio, Sortino ratio, Calmar ratio etc to
years (Banker and Morey, 1986; Norman and Stocker, 1991; measure the performance of financial assets. However, many
Pastor et al, 2002; Serrano Cinca and Mar Molinero, 2004). of these ratios have limitations as almost all of them assume

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
1752 Journal of the Operational Research Society Vol. 61, No. 12

that the return follows a normal distribution and use one 2. Skewness. For fund /, skewness is given by:
risk and one return measure to calculate the performance of
an asset. For example, Sharpe ratio assumes that the return 1T
follows a normal distribution and does not differentiate the Si = jY,^-rlf/SD] (17)
t='
good volatility from bad volatility. In our model, we have
used secondary variables listed in Table 2, as we wanted
to include all types of risks and returns that may or may Next, we calculated all the input/output measures listed in
not follow a normal distribution to provide flexibility to our Table 2 for all the 18 strategies and carried out a PCA to arrive
model. For example, lower order partial moments capture the at two input components and two output components that
downside risk of an asset. Bawa and Lindenberg (1977) argue explained more than 70% of the variation in the input/output
that models based on mean-lower partial moment have better data, respectively.
ability to explain the market data.
For each hedge fund strategy, several input and output Analysis of SBM DEA results
measures were calculated using the NAV data and finally
Summary statistics for all the 18 strategies, including the
a performance score has been arrived at for each strategy.
number of funds in each strategy are reported in Table 3,
Although it is possible that funds within the same strategy
wherein the DEA inputs and outputs selected through SRC
may have low correlations, evaluation of our data shows
analysis, are also reported. The average return for each
that correlations between funds within the same strategy are
strategy was calculated by taking the average of returns
markedly higher than between strategies. Thus, it is reason-
corresponding to all the funds within that strategy.
able to assume that funds within a strategy have relatively
more similarities and can therefore be grouped together to
Discussion of DEA output and comparison with Sharpe ratio
provide a rough measure of the underlying strategy. We
used two different methodologies to arrive at the appropriate The results of application of DEA SBM with both sets of
input/output sets for the DEA application: (i) Spearman's input and output measures (as given by SRC and PCA) are
Rank Correlation (SRC) and (ii) Principal Component Anal- enumerated in Table 4, which also contains, for the purpose
ysis (PCA). Based on SRC, the following inputs and outputs of comparison, the rankings based on Sharpe ratio, Sortino
were short listed for use in the DEA models: ratio, Calmer ratio and modified Sharpe ratio. It is interesting
to note that the DEA ranking has low correlation with tradi-
Risk measures - inputs tional performance measures such as Sharpe ratio (12.79%).
1. Standard Deviation of Drawdown (SDD). SDD for fund i Many researchers in the past have argued that Sharpe ratio is
not relevant for hedge funds as the hedge fund returns do not
is given by:
follow normal distribution (Kao, 2002; Gregoriou and Gueyie,
K
2003) and is not suitable for alternative investment strate-
SDDik= YsDfk (13) gies. Bertrand (2005) has listed several reasons why Sharpe
'j*=i ratio is not a suitable measure for hedge funds and claims
that Sortino ratio is a better measure compared to Sharpe
Where Dik is the drawdown of fund /, and K is the number
of drawdowns. ratio. The Wall Street Journal on 1 February 2001 reported
that the Art Institute of Chicago lost $43 million by investing
2. Value at Risk (VAR). VAR for fund i is given by:
in a hedge fund called 'Integral Investment Management'
VARi = -(rfta + Z0LSDi) (14) in which the fund manager claimed that their fund had the
highest Sharpe ratio in the industry (Bertrand, 2005). In our
Where, SDt is the standard deviation, and is given by:
opinion, the main reason for low correlation between Sharpe
ratio and DEA model is due to the fact that Sharpe ratio does
1 T
not differentiate good volatility from bad, whereas in the DEA
V " l t=i model we have differentiated good and bad volatility by calcu-
lating higher and lower order partial moments. Another reason
rit is the return of fund / for month t (= 1 , 2, . . . , T) and rfa
being the consideration of multiple risk return measures used
is the average return of fund i over a long duration of time in our DEA model vis--vis the single input - single output
V and Za is the critical value of standard normal variate.
measures used in Sharpe ratio calculations. The correlation
between rankings based on DEA rank correlation model with
Return measures - outputs
Sortino ratio is 32.02%, a significant improvement over corre-
1. Higher Partial Moment of order 0 (HPMO). Where: lation between DEA and Sharpe ratio (12.79%). The better
correlation in this case is probably due to the fact that Sortino
HPM0 = Pr{rit^r} (16)
ratio differentiates the good and bad volatility and thus we
Where, t is the minimum acceptable return. observe better correlation between DEA and Sortino ratio.

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
UD Kumar et al- Analysis of hedge fund strategies 1753

Table 3 Summary statistics of strategies and the Spearman rank correlation input/output values

Fund strategy Number of funds Average return (%) SDD VAR HPMO Skewness
Convertible 1 0.85 0.026 -0.032 0.619 -0.464
Corporate/Preferred 1 0.76 0.034 -0.036 0.551 -0.065
Currency 1 0.41 0.010 -0.011 0.457 -0.218
Derivative 1 -0.02 0.039 -0.075 0.462 -0.245
Emerging Market 132 0.24 0.042 -0.068 0.565 -2.129
Equity Directional 921 0.74 0.025 -0.027 0.585 2.113
Equity Market Neutral 290 1.38 0.122 -0.022 0.524 10.311
Event Driven 141 1.77 0.204 -0.037 0.673 10.446
Fixed Income Directional 89 0.30 0.021 -0.036 0.503 -1.436
Fixed Income Relative Value 195 0.79 0.096 -0.028 0.524 6.024
Flexible Portfolio 15 0.46 0.017 -0.017 0.500 3.887
Geographically Focused 3 0.40 0.014 -0.010 0.446 0.451
Global Macro 51 0.53 0.045 -0.029 0.463 5.740
Government/Corporate 7 1.48 0.047 -0.002 0.762 7.075
Managed Futures 351 0.87 0.081 -0.047 0.449 8.869
Multi-Strategy 2516 1.13 0.096 -0.050 0.571 8.944
Sector Funds 1 0.08 0.009 -0.013 0.253 -4.376
Various Assets 14 0.64 0.034 -0.017 0.563 2.659

Table 4 Ranking of hedge fund strategies using DEA-SBM and Sharpe ratio

Strategy Sharpe DEA- Rank DEA-PCA Sonino Calmar Modified sharpe


ratio correlation ratio ratio ratio

Value Rank Value Rank Value Rank Value Rank Value Rank Value Rank

Convertible 0.176 2 1 1 0.01 18 86.20 2 0.34 2 0.33 2


Corporate/Preferred 0.102 4 0.21 18 0.06 16 44.97 6 0.14 4 0.23 7
Currency -0.007 13 1 11 1 -2.81 14 0.04 10 0.41 1
Derivative -0.114 17 1 11 1 0.29 12 -0.001 15 -0.005 18
Emerging Market -0.043 15 1 11 1 -16.94 16 0.051 7 0.06 17
Equity Directional 0.129 3 1 11 1 76.59 3 0.05 8 0.30 4
Equity Market Neutral 0.079 5 1 11 1 72.44 4 -0.016 16 0.11 12
Event Driven 0.066 8 1 11 1 15.56 9 0.012 14 0.09 15
Fixed Income Directional -0.055 16 1 1 0.02 17 -19.84 17 0.032 11 0.14 9
Fixed Income Relative Value 0.039 10 0.68 15 1 1 -0.007 13 0.019 13 0.08 16
Flexible Portfolio 0.026 11 1 11 1 17.80 8 0.098 5 0.27 6
Geographically Focused -0.015 14 0.33 17 0.32 15 -4.93 15 0.162 3 0.28 5
Global Macro 0.024 12 0.94 14 1 1 2.99 11 0.043 9 0.12 10
Government/Corporate 0.223 11 11 1 287.04 1 0.35 1 0.31 3
Managed Futures 0.055 9 1 11 1 34.90 7 -0.019 17 0.11 13
Multi-Strategy 0.0734 6 1 11 1 52.64 5 -0.025 18 0.12 11
Sector Funds -0.376 18 0.99 13 1 1 -37.36 18 0.085 6 0.09 14
Various Assets 0.066 7 0.66 16 0.53 14 11.186 10 0.030 12 0.19 8

We have also compared the results with Calmar and modi- corresponding efficiency frontier. When inputs and outputs
fied Sharpe ratio. In our opinion, the results based on DEA were selected by the PC A, 13 strategies emerged as efficient
should be treated as different from the results based on other strategies. The results obtained from the application of DEA
risk ratios such as Sharpe ratio, Sortino ratio, Calmar ratio models are quite contradictory to the results given by the
etc. For example, the Calmar ratio is usually calculated using Sharpe Ratio analysis. As mentioned earlier, it is interesting
the latest 36 months return whereas DEA results are based on to note that the ranking of hedge funds using Sharpe ratio and
the returns during the entire period. 12 other performance measures resulted in nearly identical
The results of DEA SBM (using inputs and outputs as deter- ranking (Eling and Schuhmacher, 2007). From an investor's
mined by the SRC) showed that, out of the total of 18 strate- viewpoint, the rankings arrived at using DEA are more mean-
gies, 12 were found to be efficient strategies and defined the ingful as more than one measure of risk and return are taken

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
1754 Journal of the Operational Research Society Vol. 61, No. 12

Table 5 Ranking of hedge fund strategies using Super SBM* and Sharpe ratio

Strategy DEA-Principal DEA-Spearman Rank Sharpe ratio


Component (PCA) Correlation (SRQ

Value Ranking Value Ranking Value Ranking


Convertible 1.02 6 0.17 2
Currency 1 5 1.34 2 -0.01 11
Derivative 1 5 1 8 -0.11 14
Emerging Market 15 1 8 -0.04 12
Equity Directional 1 5 1 8 0.13 3
Equity Market Neutral 1.06 2 1.05 5 0.08 4
Event Driven 1.45 1 1.12 4 0.07 6
Fixed Income Directional 1 8 -0.05 13
Fixed Income Relative Value 1 5 0.04 8
Flexible Portfolio 1 5 1.41 1 0.03 9
Global Macro 1 5 0.02 10
Government/Corporate 1.01 4 1.32 3 0.22 1
Managed Futures 1.02 3 1.02 7 0.06 7
Multi-Strategy 15 1 8 0.07 5
Sector Funds 1 5 -0.38 15

* Please note that some of the strategies listed in the table are inefficient either w.r.t PCA model or w.r.t SRC model and hence the cells corresponding
to these strategies under that DEA model are empty. For example, Convertible Strategy is found to be inefficient under PCA model and hence is not
included in the super SBM evaluation with PCA input/output set.

into account simultaneously, especially when the Sharpe ratio consisting of 141 funds; and (iv) Government/Corporate,
has failed in the past in predicting the performance (Bertrand, consisting of seven hedge funds. Note that these four strate-
2005). gies were shortlisted based on their super efficiency ranking
As may be noted from Table 4, it was difficult to with PCA input/output set, for the sake of demonstration.
differentiate between the 12 efficient strategies under the We then carried out SBM DEA and Super-SBM DEA on
DEA-Rank Correlation or the 13 efficient strategies under this set of hedge funds using both sets of inputs/outputs. The
the DEA - Principal Component Analysis, as all of them top 20 efficient firms with the Spearman Rank Correlation
have an efficiency score of T. In order to arrive at a input/output set are shown in Table 6, and the results from
ranking that has greater discriminatory power, we applied the PCA results are shown in Table 7. As can be noted from
the Super-SBM model on these two sets of efficient strate- Table 6, there are 15 efficient funds from Managed Futures,
gies using input/output sets from DEA-Rank Correlation and four funds from Equity Market Neutral and one efficient fund
DEA-PCA respectively. The results of Super-SBM applica- from Event Driven. It is notable that there is no efficient fund
tion are listed in Table 5. from the Government/Corporate strategy. Similarly, from
The results from Stage 1 analysis, shown in Tables 4 Table 7, one can note that there are nine efficient funds from
and 5 differentiate efficient strategies from inefficient strate- Managed Futures, eight from Equity Market Neutral, two
gies, but as the investors are interested in knowing a fund's from Event Driven and one from the Government/Corporate
performance, it is necessary to analyse the individual perfor- strategy. Note that the number of efficient funds from each
mance of the hedge funds. This can be done in two ways: strategy in this case is proportional to the number of funds
(1) On the basis of Stage 1 analysis, if an investor has already in each strategy.
decided on which strategy to invest in, but is not sure as to The most important factor to note from Table 6 however
which fund to invest in under this strategy, she can apply the is that, IMJICMD GU Equity fund, which tops the list of
DEA Super SBM to all the funds under this strategy and find efficient funds is from the Managed Futures strategy that
the most efficient fund; (2) On the other hand, if the investor ranks seven in the strategy level Super efficiency analysis.
is not even sure about which strategy to invest in, she could This is to be expected, as we have taken averages of fund
shortlist the top four to five strategies and apply the DEA level performance measures to compute the risk-return pattern
Super SBM to all the funds under these strategies and find of strategies during the strategy level analysis. Thus, it is
the most efficient fund. quite possible that, a fund whose performance is significantly
To demonstrate the latter concept, we carried out the above average in a given strategy turns out to be more effi-
Stage 2 analysis by collecting all the hedge funds (a total cient than the efficient funds from the higher ranking strate-
of 775) from four efficient strategies, namely, (i) Managed gies. As can be noted from Table 7, the results corresponding
futures, consisting of 342 hedge funds; (ii) Equity Market to the PCA input/output also show that the STILNWF KY
Neutral, consisting of 285 hedge funds; (iii) Event Driven, Equity Fund, the KINGLBB VI Equity and the ERGBLRG

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
UD Kumar et a/ - Analysis of hedge fund strategies 1755

Table 6 Super-SBM efficiency ranks and corresponding scores of individual funds with Spearman Rank Correlation (SRC)
inputs/outputs

Hedge fund SRC score SRC rank

IMJICMD GU Equity- Managed Futures 1.358 1


MONEQPO KY Equity- Equity Market Neutral 1.005 2
COVETS A VI Equity- Managed Futures 1.000 3
BEACH3X KY Equity- Managed Futures 0.961 4
ACERCBE FP Equity- Managed Futures 0.736 5
IQSPERF KY Equity- Managed Futures 0.701 6
RABSPSF KY Equity- Event Driven 0.565 7
OKUOPPA VI Equity- Equity Market Neutral 0.514 8
ESROGLB BM Equity- Managed Futures 0.473 9
SAILFXL BM Equity- Managed Futures 0.403 10
MATADOR VI Equity- Managed Futures 0.378 1 1
ORSYFFF FP Equity- Managed Futures 0.339 12
SLFPUIK FP Equity- Managed Futures 0.328 13
SUPERFU KY Equity- Managed Futures 0.301 14
FRIEDDIV CN Equity- Managed Futures 0.260 15
CONBFII ID Equity- Managed Futures 0.181 16
RMNRTAA LX Equity- Equity Market Neutral 0.164 17
SCOGNIU KY Equity- Equity Market Neutral 0.158 18
GI ANEXE VI Equity- Managed Futures 0.154 19
ERGMXXL BM Equity- Managed Futures 0.148 20

Table 7 Super-SBM efficiency ranks and corresponding scores of the individual funds with Principal Component Analysis (PCA)
inputs/outputs

Hedge fund PCA score PCA rank Frequency of reference

STILNWF KY Equity- Equity Market Neutral 3.397 1 25


KINGLBB VI Equity- Equity Market Neutral 2.341 2 733
ERGBLRG BM Equity- Equity Market Neutral 1.076 3 87
ORFFUT2 FP Equity- Managed Futures 1.006 4 458
SMONREG VI Equity- Equity Market Neutral 1.000 5 12
MOALWDA BH Equity- Event Driven 1.000 5 4
GIAALGD VI Equity- Managed Futures 1.000 5 38
STRAFOR VI Equity- Managed Futures 1.000 5 56
ITCORPL BZ Equity- Government/Corporate 0.995 9 0
DAIBCCF KY Equity- Managed Futures 0.690 10 0
BEAUMOF VI Equity- Managed Futures 0.568 11 0
RMNRTAA LX Equity- Equity Market Neutral 0.522 12 0
PRODOPP KY Equity- Event Driven 0.475 13 0
IMJICMD GU Equity - Managed Futures 0.395 14 0
PAPHDAU BM Equity- Managed Futures 0.390 15 0
SGLSQAE ID Equity- Equity Market Neutral 0.378 16 0
BEACMLP KY Equity- Equity Market Neutral 0.336 17 0
FAISIGM VI Equity- Equity Market Neutral 0.181 18 0
HENGCUU KY Equity- Managed Futures 0.156 19 0
ORSYFFF FP Equity- Managed Futures 0.100 20 0

BM Equity, which rank one, two and three, respectively in the performances. Investors, who are convinced about investing
PCA Super-efficient list, in fact belong to the Equity Market in a given strategy, can find the most efficient fund within that
Neutral strategy, which ranks second in the strategy level strategy; and investors who are looking for the best fund to
super efficient analysis using PCA input/output set. Two funds invest in, irrespective of a strategy, can identify the most effi-
from the Event Driven strategy, which tops the list of effi- cient fund among the top ranking strategies, so as to accrue
cient strategies in the PCA analysis, in fact rank five and 13, both strategy level advantages and the fund level performance
respectively in the fund level PCA analysis. Thus, this two- benefits.

stage DEA framework provides investors with more specific These results also suggest that the performance of hedge
insights into both the strategy level as well as the fund level funds is not dictated solely by the underlying strategy. Funds

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
1756 Journal of the Operational Research Society Vol. 61, No. 12

belonging to relatively inefficient strategies may achieve this paper, and the interested reader can refer to related DEA
stellar performances due to various other factors. Thus, on literature for more details (Cooper et al, 2000; Saranga, 2009).
the surface at least, it does seem that the fund managers play
a pivotal role in generating above average returns and hence
Comparison with the z-test and a look at the 1997 Asian
there is some justification to the high performance fees that
market crisis
hedge funds charge.
DEA evaluation also provides further details on the perfor- We carried out the standard z-test to compare the results of
mance of strategies and funds by providing reference sets the DEA application. A test of the hypothesis using the z-test
and target input/output values for the inefficient units. The for difference of mean returns showed that at 95% confidence
reference sets are basically a set of frontier units, which are interval,
closest in nature to the inefficient unit under consideration
and provide benchmarking information. For example, in the 1. Convertible Strategy is better than Currency, Emerging
current context, the fund in the reference set of an ineffi- Market, Fixed Income Directional, Geographically
cient fund, say 'O', may either have similar risk structure Focused and Sector Funds.

(inputs) to that of 'O' and performs better than 'O' in its 2. Corporate/Preferred Strategy is better than Sector Funds.
returns (outputs), or it may have similar returns as fund 'O' 3. Currency is better than Fixed Income Directional and
but with a lower risk level than fund 'O'. Therefore, fund 'O' Sector Funds.

can benchmark itself against its reference set while targeting 4. Emerging market is better than Currency, Derivative,
its performance improvement and all DEA models, including Flexible, Geographically Focused and Sector Funds.
the SBM, can provide a target input/output set to fund 'O' 5. Equity Directional is better than Fixed Income Direc-
to catch up with its reference set of funds. More importantly, tional.

for the investor, who is looking for an efficient fund to invest 6. Fixed Income Directional is better than Sector Funds.
in, this information sheds more light by providing informa- 7. Fixed Income Relative Value is better than Sector Funds.
tion about the efficient funds that become part of a reference 8. Flexible Portfolio is better than Global Macro and Sector
set for the inefficient funds. The more times an efficient fund Funds.

figures in a reference set, the better it is considered to be, as it 9. Geographically Focused is better than Sector Funds.
means that there are so many inefficient funds in its horizon 10. Government/Corporate is better than Currency, Fixed
that are performing with a similar risk-return structure but are Income Directional, Fixed Income Relative Value, Flex-
lagging behind. ible Portfolio, Geographically Focused, Global Macro,
To demonstrate how this additional information can be used Multi Strategy and Sector Funds.
in the latter case for identifying the efficient funds, we have
listed the number of times an efficient fund has figured in As one may note from the comparison of Tables 4 and 5 with
a reference set in column 4 of Table 7. As one can note, the above results, except for statements 2 and 9 and a few
KINGLBB VI Equity Market Neutral, which is ranked two subparts of statements 1 and 4, all the remaining statements
in the super efficiency analysis has appeared in significantly hold good in the DEA results in terms of efficiency. Thus, the
more number of reference sets (733) than the STILNWF KY strategies that form the DEA efficiency frontiers do seem to
Equity fund (only 25 reference sets), which has topped the dominate the inefficient strategies in terms of greater returns.
list. Similarly, 0RFFUT2 FP Equity, which is ranked four, Hence, the z-test results, to a large extent, substantiate the
has figured in more reference sets (458) than the ERGBLRG DEA framework for investors looking for better investment
BM Equity (87), which is ranked three. In some contexts, this opportunities.
kind of information can significantly aid the decision-making Another main area that interests investors is regarding the
process. Similarly, the inefficient funds can make use of the performance of the funds during times of financial turmoil.
reference sets coupled with input/output targets to improve The results of the 1997 crisis hypothesis testing showed
their performance in the long run so as to catch up with that:
the more efficient funds and also to become more attractive

to the investors. DEA also provides Returns to Scale infor- 1. Convertible was better than Emerging Market, Flexible
mation, which specifies whether a particular fund is in the Portfolio and Geographically Focused.
Increasing Returns to Scale, the Constant Returns to Scale or 2. Corporate/Preferred is better than Emerging Market and
the Decreasing Returns to Scale region. This kind of informa- Flexible portfolio.
tion is very relevant because it can aid the funds in determining 3. Currency is better than Geographically Focused and
whether to aim for a higher scale or not, and the investors Fixed Income Directional.

can rely on this information to make decisions regarding their 4. Emerging Market is better than Geographically Focused.
investments, which is of great significance especially because 5. Equity Directional is better than Currency, Geograph-
hedge funds by nature deal with huge scales of investments ically Focused, Emerging Market and Fixed Income
(Eling, 2006). However, these topics are beyond the scope of Directional.

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
UD Kumar et al- Analysis of hedge fund strategies 1757

Table 8 Hurst ratio of hedge fund strategies

Fund strategy Hurst ratio Percentage funds with Hurst ratio > 0.5
Convertible 0.614928836
Corporate/Preferred 0.6759 1 9455
Currency 0.63104202
Derivative 0.422106608
Emerging Market 0.548469488 0.643939394
Equity Directional 0.522237473 0.541893362
Equity Market Neutral 0.522139864 0.555555556
Event Driven 0.545091 1 17 0.70212766
Fixed Income Directional 0.516935304 0.595505618
Fixed Income Relative Value 0.537536424 0.649484536
Flexible Portfolio 0.511 225024 0.4
Geographically Focused 0.574467373
Global Macro 0.531611215 0.652
Government/Corporate 0.533457288
Managed Futures 0.500021886 0.477011494
Multi-Strategy 0.520759735 0.551875499
Sector Funds 0.673799402
Various Assets 0.48203474 0.357142857

*Note that Convertible, Corporate/Preferred, Currency, Derivative and Sector Funds constitute a single fund (refer to Table 3) and hence the percentages
are not given, while Geographically Focused and Government/Corporate also have only seven and three funds, respectively.

6. Equity Market Neutral is better than Geographically the z-test substantiate this finding with a 95% confidence
Focused, Emerging Market and Fixed Income Direc- level, as shown in Table 9.
tional.

7. Event Driven is better than Emerging Market. Stability of hedge fund returns
8. Fixed Income Directional is better than Emerging Market.
Stability of returns is a major concern for any investor, espe-
9. Flexible Portfolio is better than Fixed Income Directional.
cially in the case of hedge funds. One of the metrics used
10. Global Macro is better than Emerging Market.
for checking stability is the Hurst ratio, which is usually
11. Managed Futures is better than Emerging Market and used to check whether the returns of the funds are stable
Flexible Portfolio.
or just random. It differentiates the funds that are consistent
12. Multi-Strategy is better than Emerging Market and Flex-
performers from the ones that experienced strokes of luck.
ible Portfolio.
Hurst ratio as defined by Koh Francis et al (2004) is calcu-
lated as
A close look at the results reveals that nine strategies
performed distinctly better than the Emerging market strategy Hurst Ratio =

during the Asian finacial crisis. The Asian financial crisis LogAf - Log a
started in July 1997 in Thailand and South Korea with the where M(t) = (max(0 - min(t))/S(t); N = length of shorter
financial collapse of Thai Bant and Kia, and affected curren- sub-periods into which a manager's return record has
cies, stock markets and other asset prices in Asian countries, been sub-divided; t = number of sub-periods into which a
including the Four Asian Tigers (Taiwan, South Korea, Hong manager's return record has been sub-divided; S(t) = standard
Kong and Singapore). Indonesia, South Korea and Thailand deviation of data over sub-period t; a = constant term that is
were the countries most affected by the crisis; Hong Kong, negligible if track record is 5 years or less (Ignored for our
Malaysia, Laos and the Philippines were also hit by the case).
slump; while China, India, Taiwan, Singapore and Vietnam The results obtained for different hedge fund strategies for
were relatively unaffected. Japan was not affected much by Hurst ratio are shown in Table 8. It can be seen that most of
this crisis but was going through its own long-term economic the strategies have Hurst ratio greater than 0.5. This implies
difficulties. However, all the nations mentioned above saw that the returns obtained by the fund managers in these strate-
their currencies dip significantly relative to the US dollar, gies are persistent and the managers can be said to have 'hot
though the harder hit nations saw extended currency losses. hands'. A value less than 0.5 means that the returns tend to
Out of all the countries affected, South Korea was hit the fluctuate randomly but converge to a stable value over time.
hardest. Thus, it is not difficult to see that the hedge funds The Hurst ratio for S&P 500 comes out to be 0.57. Out of all
investing in Emerging Markets of Asia were liable to be more the strategies - Emerging Markets, Event Driven and Fixed
adversely affected than most other strategies. The results of Income Relative Value are the strategies that can be said to

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
1758 Journal of the Operational Research Society Vol. 61, No. 12

' t? ^ co^HOOt-iomrocoo^toov.o have higher percentage of funds with persistent returns. The
correlation between beta and Hurst ratio comes out to be
0.004; hence it can be said that the beta values are not corre-
& > '- ir-mio^oocsoNcnocn^-H -^j- lated to the Hurst ratio values. It is worthwhile noting that
except for the Derivative Strategy, all the other strategies that
were found to be efficient through DEA models have Hurst
"So ^oocnoooooor-'inoovo on <N ratio greater than 0.5 (Table 9).

Conclusions

1"8
The significant growth in hedge funds fuelled by greater
interest from investors has increased the need for appropriate
II W
methodologies to measure performance of hedge funds and
"as
the corresponding trading strategies. This paper shows with a
O
^ 5 ^ ^-Hoor-Hoor-^t-- r- <n o o brief review of existing literature on hedge funds, that tradi-
* ^5> ^ fi h es O' in cN^o
I tional measures of performance like mean-deviation ratio,
which use single risk-return aspects, are not sufficient to
'S S 's s ^ <NO*-iooor-(N^H co vo io f- r- capture the complete risk return profile of hedge funds. A
I * s w -^ s *S ^ovo^HooiOTfvor- Tt ^ m ^t m
recently proposed approach that incorporates multiple risk-
k^
return measures through the PCA and the Spearman Rank
" ^ c Correlation technique in performance measurement is adopted
i to compare the hedge funds, both at the strategy level and at
s the fund level. Application of slack-based DEA models that
I are capable of utilizing negative input/output data to compute
S '? ^ 'mvoO'-H > m o' a m ri es the efficiency scores has been clearly demonstrated, with the
I help of more than 4730 hedge fund data. The framework
"s-^-b^^ n io m O ^ rtoomoooo>o proposed in the current paper serves as a comprehensive guide
abb^ onooono^o o'o'mooOMnoo
S3
to an investor looking for alternative investments in hedge
^*
funds. As the proposed framework, through DEA-PCA and
c
DEA-Rank Correlation, incorporates multi-dimensional risk

i
> OnOnOnO ^OO^OOOOCNTi-m
-2-is^ oor^<^o moN^inONO'^h
return analysis, it has the potential to account for the complex-
'S S ities involved in hedge fund transactions and also the dynamic
nature of trading strategies. One can also use this framework to
1
^0f3v "Ht- ooomcNOsooomo gather additional information needed to support the decision-
ta making process, which many a time may still use traditional
measures like the Sharpe's ratio (with two-dimensional risk
I return analysis) due to various reasons. The additional infor-
"SO
^^ in in cno'ooo'OHMrf) mation like reference sets and input/output targets will supple-
N
ment the primary knowledge and aid the decision process,
ON especially in situations where multiple funds/strategies are
S found to be equally profitable and compete for similar invest-
ment opportunities.
In today's global economic environment where hedge
s cnmO'-HOOvooNoooo^oaNr^ funds have a significant influence on macro-economic events
C^ -i o' o h lor-moocnracs
such as the 1997 Asian financial crisis, the 2000 asset bubble
burst and the Yen carry trade and the more recent sub-prime

S
mortgage crisis, it is imperative to develop comprehen-
sive performance measurement methodologies that aid the
decision-making process of large investors. Through use of
standard z-tests and Hurst ratio analysis, we find that the DEA
1 il! i-jjj, application results are not just robust, but more importantly,

i m |-| g ||| f can determine the most consistent and profitable performers.
One of the limitations of DEA is that as the number of
I || tf f ItJtJ! 11 1
<3 uuuwwmwEEouSS
inputs and outputs increase, the number of efficient firms also
increases. To some extent, this problem can be addressed by

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
UD Kumar et al- Analysis of hedge fund strategies 1759

using super efficiency DEA models, but further research is Fama E and French K (1992). The cross-section of expected stock
returns. J Financ 47: 427-465.
required to tackle this problem, especially if sample size of
DMUs is small. Fung W and Hsieh DA (1997). Empirical characteristics of dynamic
trading strategies: The case of hedge funds. Rev Financ Studies
10: 275-302.

Acknowledgements - We thank the anonymous referees and the editor for Getmansky M, Lo AW and Mei SX (2004). Sifting through
their constructive comments that have helped us to improve the quality the wreckage: Lessons from recent hedge-fund liquidations.
of this research work. / Investment Mngt 2(4): 6-38.
Goetzmann W, Ingersoll J and Ross S (2003). High-water marks and
hedge fund management contracts. J Financ 58: 1685-1717.
References Gregoriou GN (2003). Performance appraisal of funds of hedge funds
using data envelopment analysis. J Wealth Mngt 5(4): 88-95.
Gregoriou GN and Gueyie JP (2003). Risk-adjusted performance of
Adler N and Golany B (2002). Including principal component weights
funds of hedge funds using a modified Sharpe ratio. J Wealth Mngt
to improve discrimination in data envelopment analysis. J Opl Res
Soc 53: 985-991. 6(Winter): 77-83.
Gregoriou GN and Lhabitant FS (2009). Madoff: A riot of red flags.
Agarwal V and Naik N (2004). Risks and portfolio decisions involving
hedge funds. Rev Financ Stud 17: 63-98. Research report, EDHEC Risk and Asset Management Research
Center, February 2009.
Alexander G and Baptista A (2004). A comparison of VaR and CVaR
constraints on portfolio selection with the mean-variance model. Gregoriou GN, Sedzro K and Zhu J (2005). Hedge fund performance
Mngt Sci 50: 1261-1273. appraisal using data envelopment analysis. Eur J Opl Res 164:
555-571.
Artzner P, Delbean F, Eber JM and Heath D (1999). Coherent measures
of risk. Math Financ 9: 203-228. Harper D (2003). Introduction to Hedge Funds- Part Two,
Bali TG and Gokcan S (2004). Alternative approaches to estimating Investopedia, httpV/www.investopedia.com/articles/OS/l 2 1 003 .asp,
VaR for hedge fund portfolios. In: Schachter B (ed). Intelligent accessed on 31 August 2007.
Hedge Fund Investing. Risk Books: London, pp 253-277. Jorion P (2000). Value at Risk: The New Benchmark for Managing
Financial Risk. McGraw Hill: New York.
Banker RD and Morey RC (1986). Efficiency analysis for exogenously
fixed inputs and outputs. Mngt Sci 43: 513-521. Justin P (2005). Bayou Hedge Fund: The story so far, Hedge Fund
Banker RD, Chames A and Cooper WW (1984). Some Models for Street, http://www.fundstreet.org/2005/09/bayou JiedgeJ'm.html,
accessed 1 October 2007.
estimating technical and scale efficiencies in data envelopment
analysis. Mngt Sci 30: 1078-1092. Kao DL (2002). Battle for alphas: Hedge funds versus long-only
Bawa VS and Lindenberg EB (1977). Capital market equilibrium portfolios. Financial Analysts Journal 58(2): 16-36.
in a mean-lower partial moment framework. J Financ Econ 5: Koh Francis, Koh Winston TH, Lee David KC and Phoon K
189-200. (2004). Investing in hedge funds: Risks, returns and performance
Bertrand J (2005). In the land of hedge fund, the Sharpe ratio management. In: Gregoriou G, Papageorgiou N, Hubner G
is no longer the king, HSBC Global Asset Management article and Rouah F (eds). Hedge Funds: Insights in Performance
last cited on 4th March 2009 at http://www.sinopia-group.com/ Measurement. John Wiley & Sons: USA, pp 341-364.
internet/Sinop2007/docs/NewRiskIndicator.pdf. Liang B and Park H (2007). Risk measures for hedge funds: A cross-
Casu B, Shaw D and Thanassoulis E (2005). Using a group support sectional approach. Eur Financ Mngt J 13: 317-354.
system to aid input-output identification in DEA. / Opl Res Soc 56: Lo A (2001). Risk management for hedge funds: Introduction and
1363-1372. overview. Financ Anal J 57: 16-33.

Charnes A, Cooper WW and Rhodes E (1978). Measuring the Lovell CAK and Pastor JT (1995). Units invariant and translation
efficiency of decision making units. Eur J Opns Res 2: 429-444. invariant DEA models. Opns Res Lett 18: 147-151.
Charnes A, Cooper WW, Golany LM, Seiford S and Stutz J (1985). Mitchell M and Pulvino T (2001). Characteristics of risk and return
Foundations of data envelopment analysis for Pareto-Koopmans in risk arbitrage. J Financ 56: 2135-2176.
efficient empirical production functions. J Econ 30: 91-107. Naik N and Tapley M (2007). Demystifying hedge funds. Bus Strategy
Connor G and Lasarte T (2005). An introduction to hedge fund Rev 18(2): 68-72.
strategies. Research report by Institute of Asset Management, Nayar S (2009). US hedge fund industry: Experts predict growth by
London School of Economics. fourth quarter. Hedge week special report, February 2009, pp 3-4.
Cooper WW, Seiford LM and Tone K (2000). Data Envelopment Nguyen-Thi-Thanh H (2006). On the use of data envelopment analysis
Analysis. Kluwer Academic Publishers: USA, pp A2-A6. in hedge fund selection. Working paper. Universit d'Orlans:
Ding B and Shawky HA (2007). The performance of hedge fund Orlans, France.
strategies and the asymmetry of return distributions. Eur Financ Norman M and Stocker B (1991). Data Envelopment Analysis: The
Mngt 13: 309-331. Assessment of Performance. John Wiley and Sons: Chichester, UK.
Duzakin E and Duzakin H (2007). Measuring the performance Pastor JT, Ruiz JL and Sirvent I (2002). A statistical test for nested
of manufacturing firms with super slacks based model of data radial DEA models. Opns Res 50: 728-735.
envelopment analysis: An application of 500 major industrial Prosser D (2007). David Prosser's outlook: Hedge funds have plenty
enterprises in Turkey. Eur J Opl Res 182: 1412-1432. to answer for. The Independent on Sunday, http://news.independent
Eling M (2006). Performance measurement of hedge funds using .co.uk/business/comment/article2851457.ece, accessed on 17 April
data envelopment analysis. Financ Markets Portfolio Mngt 20: 2008.
442-471. Portela MS, Thanassoulis E and Simpson G (2004). Negative data in
Eling M and Schuhmacher F (2007). Does the choice of performance DEA: A directional distance approach applied to bank branches.
measure influence the evaluation of hedge funds? J Bank Financ J Opl Res Soc 55: 1111-1121.
31: 2632-2647. Saranga H (2009). The Indian auto component industry - estimation
Estrada J (2001). The cost of equity of internet stocks: A downside of operational efficiency and its determinants using DEA. Eur
risk approach. Working paper, IESE Business School, Spain. J Opl Res 196: 707-718.

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms
1760 Journal of the Operational Research Society Vol. 61, No. 12

Scholz H (2007). Refinements to the Sharpe ratio: Comparing Tone K (2001). Slack based measure of efficiency in data envelopment
alternatives for bear markets. J Asset Mngt 7: 347-357. analysis. Eur J Opl Res 130: 498-509.
Serrano Cinca C and Mar Molinero C (2004). Selecting DEA Tone K (2002). A slack-based measure of super-efficiency in data
specifications and ranking units via. PC A J Opl Res Soc 55: envelopment analysis. Eur J Opl Res 143: 32-41.
521-528. White B (2006). Amaranth outlines its liquidation plans. Financial
Sharpe WF (1966). Mutual fund performance. J Bus 39(1): 119-138. Times, http://www.ft.eom/cms/s/0/540c3fd8-4dbc-l ldb-8704-0000
Sharpe WF (1994). The Sharpe ratio. J Portfolio Mngt 21(1): 49-58. 779e2340.html, accessed 1 October 2007.
Sharpe JA, Meng W and Liu W (2007). A modified slacks-based Wilkens K and Zhu J (2005). Classifying hedge funds using data
measure model for data envelopment analysis with 'natural' envelopment analysis. In: Gregoriou GN, Rouan F and Karavas
negative outputs and inputs. J Opl Res Soc 58: 1672-1677. VN (eds). Hedge funds: Strategies, Risk Assessment and Returns.
Seiford ML and Zhu J (2002). Modelling undesirable factors in Beard Books: Washington, pp 161-175.
efficiency calculations. Eur J Opl Res 142: 16-20.
Sortino FA and van der Meer R (1991). Downside risk. J Portfolio
Mngt 17(Spring): 27-31.
Taleb NN (2004). Blowup versus Bleed: What does empirical
psychology say about the preference for negative skewness? J Received July 2008;
Behavioral Financ 5(1): 2-7. accepted September 2009 after one revision

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 21:10:48 UTC
All use subject to http://about.jstor.org/terms

Você também pode gostar