Mahdi-Thesis

OULU BUSINESS SCHOOL
Mahdi Yadipur
HEDGE FUND PERFORMANCE DUE TO SKILL OR LUCK?
Master’s Thesis
Department of Finance
March 2014

UNIVERSITY OF OULUABSTRACT OF THE MASTER’S THESIS
Oulu Business School
Unit: Department of Finance
Author: Mahdi Yadipur Supervisor: Juha Joenvaara
Title: HEDGE FUND PERFORMANCE DUE TO SKILL OR LUCK?
Subject:
Finance
Type of the degree:
Master of Science (Economics and
Business Administration)
Time of publication:
Spring 2014
Number of pages:
55
Abstract:
In this study we examine the proportion of false discovery rate exists amongst the individual
funds in Hedge Fund Research (HFR) database. Applying the Fung and Hsieh (2004) seven-
factor model in a time series regression along with a statistical false discovery rate
methodology construct the main framework of this study. False discovery rate helps to
measure the proportion of lucky funds among hedge funds that have statistically significant
alphas and explains how many percentages of funds with significant alphas would be
achieved due to luck compared to skill.
Even after adjusting for the backfill bias, the proportion of false discovery rate states that the
hedge funds outperform due to skill compared to luck and underperforms due to be unlucky
compared to be unskilled. Results for strategies demonstrate that the proportion of false
discovery rate in Event Driven, Relative Value, and Multi Strategy is very low in the right
tail respectively and the manager has skill compared to luck. In contrast, strategies such as
CTA, Relative Value, and Short Bias have the lowest proportion of false discovery rate in
the left tail respectively which implies the manager is more unskilled compared to be
unlucky in his performance. The proportion of false discovery rate for small funds is greater
than large funds in the right tail of the distribution and it implies that for small funds the
manager outperforms mostly by luck compared to skill. Contrarily, the proportion of false
discovery rate for large funds is greater than small funds in the left tail of the distribution and
it implies that for large funds the manager is more unlucky compared to be unskilled to
outperform in the market.
Keywords: Hedge Fund Performance, Skill and Luck, Performance Measurement, P-Value and
Q-Value, False Discovery Rate.
Additional information:

CONTENTS
1. INTRODUCTION ................................................................................................... 3
2. LITERATURE REVIEW AND CONTRIBUTION................................................ 7
2.1. Hedge Funds Review........................................................................................ 8
2.2. Mutual Funds Review .................................................................................... 14
2.3. Contribution of the Study............................................................................... 16
3. DATA AND METHODOLOGY........................................................................... 18
3.1. Data Description and its Availability............................................................. 18
3.2. Performance Measurement............................................................................. 21
3.3. Statistical Measurements................................................................................ 22
3.3.1. False Discovery Rate............................................................................ 24
3.3.2. Estimation Procedure ........................................................................... 27
4. EMPIRICAL RESULTS........................................................................................ 30
4.1. The Impact of Luck on Performance before and after Adjusting for Backfill
bias ………………………………………………...…………………………….30
4.2. The Impact of Luck on Performance for Different Strategies........................ 35
4.3. The Impact of Luck on Performance Based on Time Period......................... 40
4.4. The Impact of Luck on Performance Based on Size...................................... 43
5. CONCLUSION...................................................................................................... 49
REFERENCES: ......................................................................................................... 51

FIGURES
FIGURE I: COMPARING SKILL VS. LUCK IN SIGNIFICANT FUNDS............ 34
FIGURE II: THE PROPORTION OF DIFFERENT STRATEGIES........................ 35
TABLES
TABLE I: ALL INDIVIDUAL FUNDS BEFORE AND AFTER ADJUSTING FOR
BACKFILL BIAS...................................................................................................... 31
TABLE II: THE FUNDS OF HEDGE FUNDS IN DIFFERENT STRATEGIES.... 36
TABLE III: THE FUNDS OF HEDGE FUNDS IN DIFFERENT TIME PERIODS40
TABLE IV: THE FUNDS OF HEDGE FUNDS IN DIFFERENT SIZE GROUPS. 44

3
1. Introduction
A hedge fund can be defined as an investment fund which can embrace a diverse
range of investment and financial activities. The type of investors is mostly specified
by regulators and they are available for approved and ascertained investors rather
than to be sold or offered in general public. Hedge funds employ a wide variety of
techniques and strategies to achieve a premium for their investors. They commonly
use a management and incentive fees to avoid the agency costs and meet the
preferences of both investors and managers. Focusing on absolute return provides
this potential for hedge funds to have abnormal return in any condition of the market
(either up or down) compared to similar investments like mutual funds. Unlike other
investments, hedge funds have advantages like low correlation with other traditional
asset classes, Low volatility and high loss protection in equity bear markets,
flexibility in investment options, access to greater liquidity and leverage, and using
mostly often derivatives that makes them superior and attractive compared to other
options for investment. Hedge funds have a stronger performance incentive
compared to other investments like mutual funds and it is believed that they have
most talented and active managers. The development and high increasing growth of
hedge funds have attracted the attention of many investors and scholars in financial
markets over the past two decades. This industry has experienced a dramatic growth
and development since 1990s alongside other alternative investments. The estimated
number of hedge funds has increased from 3500 with around $1 trillion in assets in
1998 to 8500 active hedge funds with over $2.8 trillion in assets recently. One of the
reasons may be the strong faith of the investors to hedge funds to beat the market and
get abnormal return through the skill of the managers. Although there are many
arguments whether this abnormal returns are due to the skill of the manager or his
luck to outperform the corresponding benchmark.
One of the criteria for investors to invest in hedge funds is the past performance of
manager. Because hedge funds target the absolute return, hence, the handy way to
select a fund for investment is through accessing the past performance. Many studies
have stated that sentiments in financial market are strong driven forces for investors.
Therefore they would be attracted to invest in funds with highest past performance as

4
it is the most publicly accessible information about hedge funds. This conversation
leaves enough room for this question whether the past performance indicates and
guarantees the future performance in the case of hedge funds or not. Some theories
like CAPM explains outperforming in the market is by chance and market is
efficient. Contrarily, there is the possibility of active management and it means
beating the market and outperforming by skill compared to luck. In hedge funds and
other sorts of investments the scholars define the excess return based on risk factors
and if these factors couldn’t define the whole abnormal return it would be called
alpha which is the initial and handy measure of skill for the manager. Some scholars
like Cochrane believe that we have a correct beta and incorrect beta. Therefore, in
the case of having alpha it would be the mispriced beta.
The main idea of this study is to answer this question whether the performance of
the manager is due to skill or luck. The statistical methodology used here is false
discovery rate (FDR) which is defined and proposed by John Storey in 2002. The
false discovery rate can be defined as the likelihood of the proportion of significant
results to be truly null zero. False discovery rate proposes whenever we have
multiple hypotheses testing to correct for multiple comparisons. In other words, false
discovery rate has a procedure to control the error type I which is the expected
proportion of incorrectly rejected null hypotheses. In a hypothesis test there is the
possibility of error type I which is rejecting the true null or error type II which is
accepting the false null. In practice, sometimes it happens to have multiple tests with
thousands or even millions of simultaneous hypothesis tests and many possible ways
to combine them. In result, the probability of error type I and II increase sharply. In
respond to this problem some methods have proposed to control these errors. Some
of the correction approaches for error type I that proposed by scholars are; Per
Comparison Error Rate (PCER), Per Family Error Rate (PFER), Family Wise Error
Rate (FEWR), False Discovery Rate (FDR), and Positive False Discovery Rate
(PFDR). The first problem is to choose one of these approaches which have the best
power and control for testing. The applying method should not be too conservative to
exclude funds and not too liberate to provide many funds so that makes the process
of finding right performance very difficult and confusing. The direct approach of
false discovery rate proposed by Storey (2002) is a balanced methodology that takes
into consideration all aspects of a standard technique. This method attempts to find

5
the proportion of significant funds according to a significant level  and then seeking
the proportion of false discovery funds among them. That is what is called truly null
(Benjamini and Hochberg 1995, Storey 2002 and Storey, Taylor and Siegmund
2004). Putting it in a nut shell, false discovery rate measures the proportion of lucky
funds among the proportion of funds with significant alphas and explains how many
percentages of funds with significant alphas would be achieved due to luck compared
to skill.
To distinguish luck from skill, we apply the procedure followed by Storey (2002)
and Storey and Tibshirani (2003), for finding false discovery rate. Specifically, we
conduct a multiple hypothesis test to examine simultaneously the test statistic of each
hedge fund in the population instead of just one fund. In this study we provide a new
approach to control the false discovery rate in a multiple fund setting. Our approach
much more accurately identifies the proportion of positive and negative significant
funds, lucky and unlucky funds, skilled and unskilled funds, and false discovery rate
with respect to five different significance levels of 5, 10, 15, and 20 percents in the
right and the left tail of the distribution. This procedure will be applied for all
individual funds before and after adjusting for backfill bias, different types of
strategies, different time periods, and different fund size. The results of the study
demonstrate that the proportion of positive significant funds is greater than the
negative significant funds and these results increase along the increase in
significance level. Also, the proportion of skilled funds is greater than lucky funds in
the right tail but the proportion of unlucky funds is greater than unskilled funds in the
left tail of the distribution. The proportion of false discovery rate which is the
indicator for separating skill and luck for the performance of the manager is greater
in the left tail. It implies that for positive significant funds the manager has more skill
rather than luck in his performance in the right tail, contrarily, he is more unlucky to
be unskilled in his performance in the left tail. These results are not the same in some
cases but mostly they follow this procedure. Results for different types of strategies
demonstrate that the proportion of false discovery rate increases in Event Driven,
Relative Value, Multi Strategy, Global Macro, Sector, Long/Short, CTA, Emerging
Markets, Market Neutral, and Short Bias respectively. It implies that by increasing
the false discovery rate the proportion of skilled funds decreases and the proportion
of lucky funds increases. In other words, the skill of the manager decreases in the

6
strategies respectively. Comparing the results for different time periods reveals that
the manager has the highest skill from 1999 to 2003 and 1994 to1998 respectively
and he has been luckier to outperform in time periods from 2004 to 2008 and from
2009 to 2013 respectively. Analyzing the size effect demonstrates that the proportion
of false discovery rate for small funds is greater than large funds in the right tail of
the distribution and it implies that for small funds the manager outperforms mostly
by luck compared to skill. Contrarily, the proportion of false discovery rate for large
funds is greater than small funds in the left tail of the distribution and it implies that
for large funds the manager is more unlucky compared to be unskilled to outperform
in the market.
In the following study, chapter two provides the literature review and contribution
of this study. Because hedge funds and mutual funds have many similarities we will
separate the literature review into two parts and we will have a short review on both
fields to provide a broad perspective for the readers. Finally, at the end of this
chapter we present the contribution of this study which distinguishes it from the past
studies. Chapter three explains the data and the applied methodology of the study.
This chapter after a short review about the availability of the data and explaining the
possibility of data biases will focus on the methodology. The methodology of the
study includes two parts; the first part focuses on time series regression based on
Fung and Hsieh (2004) seven-factor model and deriving corresponding results. The
second part focuses on the statistical part which is the innovation of this study. It
explains the procedures of measuring false discovery rate in order to separate the
skill versus the luck of the performance of the manager. Chapter four demonstrates
the empirical results and presents them in four parts. The procedure of measuring
false discovery rate will be applied for all individual funds before and after adjusting
for backfill bias, based on different strategies which have been applied by the
manager, different time periods which includes different circumstances of the
market. We examine the results based on fund size as well, therefore, we separate
data into three subgroups small, medium and large funds and then we evaluate the
impact of luck on them. In chapter five we will give the including remarks and will
make a conclusion of this study.

7
2. Literature Review and Contribution
Hedge fund is an investment vehicle which is slightly regulated and can invest in a
variety range of investments using multiple complex strategies and techniques. Being
less regulated and usually targeting the absolute return provide the possibility of
playing an essential role by the manager in the market to outperform in the market
and increase the return of hedge fund. Although the manager can achieve a high
absolute return applying different types of strategies and techniques, there is a
possibility of remarkable losses at the same time as well. The range of hedge funds
investment is so broad and extended and it includes investing within the financial
markets through stocks, bonds, commodities, currencies, derivatives and etc. Also,
hedge funds can use different types of strategies for managing the underlying
investment such as short selling, arbitrage, leveraging, swaps and so many other
strategies. The history of hedge funds show that they were designed in the United
Stated in 1948 and they became one of the essential vehicles for investment in the
financial market. They are commonly used in a portfolio beside other alternative
investments rather than being used individually investment. Hedge funds usually take
a long term investment in the portfolio of stocks and bonds. The portfolio that
consists of hedge funds needs skilled managers to apply the best combination of
strategies and approaches to beat the market and get abnormal returns. These
approaches should be in such a way that provides appropriate diversification and
fitting risk- return ratio (Man 2013).
The flexibility of using a wide variety of strategies in hedge funds in one hand
gives the managers this opportunity to achieve remarkable return but on the other
hand it makes their assessment so complex and challenging. Employing options like
futures makes the return of hedge funds distinguishable compared to standard asset
classes. Therefore, evaluating the return of hedge funds needs more attention to carry
out a comprehensive and solid assessment. In finance, to measure the abnormal
return of a security or a portfolio of securities using the Jensen’s alpha is the most
commonly criterion. Jensen (1968) for the first time used this measure to examine
the returns of mutual funds. Alpha is one of the risk-adjusted criteria for measuring
the abnormal return in an investment. In other words, the return in CAPM model is

8
risk adjusted which means an achieved return is connected to its corresponding
relative risk and if there is any return excess to the adjusted risky return it will be
called positive alpha or abnormal return. Hence, alpha explains the return for an
active manager who has been able to carry out excess to the benchmark index or risk
free rate in an investment. The origin of this idea comes from the studies which are
done in the middle of the twentieth century. The idea of efficient market generated
the formation of market capitalization based on the weighted index and replicating
the performance of investing the same proportion in the entire market. Therefore,
investing in the mostly high held equity securities like replicating the same weight of
S&P 500 and Wilshire 5000 which approximately represent the 500 largest equities
and the largest 5000 securities respectively would be an index for passive
management. These two benchmarks disclose 80 and 99 percent of the total market
capitalization of the US market as a whole. Based on this idea, the active
management should not only be able to avoid losing money and prove the return
according to the standard investing but also to make more money compared to the
passive management. The passive management is established on the basis of efficient
market hypothesis that asserts financial markets are informationally efficient and
there is no more return excess to the risk adjusted basis. The efficiency of prices is
not only because of the correct behavior of the agents in the financial market but also
because of luck. As always in finance, there are two extreme viewpoints about the
same idea. In one side of the extreme there are those who believe any remarkable
performance is an evidence of exceptional performance as almost super natural. On
the other side of the extreme there are those who explain the performance according
to the luck and believe that the manager is only lucky to outperform in the market
(Fama and French 2009).
2.1. Hedge Funds Review
Most of the studies, in general, which have been done on the manager’s performance
seeking the answer for three types of questions; are the average managers outperform
in the market, whether this performance comes out from skill or luck, and if this
performance is persistence or not. In the following study we will present some

9
studies which take into consideration one of the above mentioned targets and by
applying different strategies try to answer the questions. As the studies in hedge
funds progress new things will be discovered and sometimes they challenge the
previous results. For example, in many studies the assumption is that the hedge funds
are market neutral and they have low correlation with the usual market indexes.
Patton (2008) examines this idea and by finding evidences against the neutrality of
hedge funds which are exposed to the market risk criticizes the concluded remarks.
Patton defines the market neutrality and dissects the process of measuring market
neutrality into five different concepts to be measured. He proposes mean neutrality,
variance neutrality, value-at-risk neutrality, tail neutrality, and complete neutrality.
He applies the statistical tests for monthly return of 1,423 hedge funds according to
the defined neutrality groups. Results demonstrate that one quarter of funds in the
market neutral actually are significantly non neutral at the 0.05 percent level.
Another study that tries to challenge the previous studies refers to the work is done
by Kosowski et al (2007). This study is related to the current study and tries to
examine whether the performance of the top hedge funds can be explained by luck or
not. They apply a powerful bootstrap and Bayesian method to evaluate the fund
returns according to the corresponding risk factors. Findings reveal that the
performance of top hedge funds cannot be explained only by luck and this
performance is persistence at annual horizon. Another study which examines the
alpha of hedge funds is done by Ibbotson et al (2011). After excluding all the biases
in data, they evaluate the corresponding return by decomposing it into three
components. Alpha which is the value added by hedge funds, betas which are the
systematic risks, and the cost which is the hedge fund fees. Applying the model for
3,500 hedge funds reveals that there would be significant positive alpha even after
excluding the biases. This excess return will be approximately equal to the fees
which states that this abnormal return has been shared roughly between the manager
and the investors.
Joenvaara (2010) examines hedge fund performance applying a unique
methodology to study the evolution of different approaches for choosing optimal
portfolio. The author proposes that the future performance of a portfolio is
predictable through using the conditional t-statistic of the corresponding funds
weight. He also connects the funds alpha and portfolio weight, since their t-statistics

10
are equal with almost the same significant degree. Results demonstrate that the
predictability of the portfolio weights for individual funds and along different types
of strategies changes in a statistically significant movement. Evidences show that
funds with the highest conditional t-statistic of the portfolio weight reveal more
significant outperformance. In another study Joenvaara et al. (2012), provide new
stylized facts about the hedge fund industry by using a large developed hedge fund
database. They survey the sensitivity of hedge fund performance and data bias in
different databases. They demonstrate the importance of using a large developed
database by examining different aspects of hedge funds such as average
performance, performance persistency, and the cross sectional relation between fund
characteristics and hedge fund performance. Findings of the study show that the
manager outperforms specifically adding value after the fees. Results for average
excess returns and achieved alphas are economically significant in both scenarios of
equal and value weight basis. The main core result is that the hedge fund
performance results are subject to database selection. Moreover, in the case of
making equal weighted portfolio, hedge funds performance is persistence at annual
horizon. But in the case of value weighted portfolios there is persistency at monthly
horizon. Therefore, based on the results the performance will be long stable for equal
weighted portfolios and short stable for value weighted portfolios which is driven
with small funds. Generally, this study demonstrates that several stylized facts were
affected by the type of database chose for analyzing. Hence, it is very important to
take into consideration that for controlling the corresponding biases there should be a
developed database with high quality data combination. In another study Joenvaara
et al. (2013) revised the stylized facts about the performance of hedge funds and their
sensitivity to the database is selected for assessment. They try to find the source of
distinct quantitative and qualitative results for the performance of hedge funds in
different studies. The results reveal that the differences in underlying database such
as backfill bias, the quality of information about asset under management and dead
funds affect on the achieved results. Testing the performance of hedge funds under
circumstances of controlling underlying database demonstrate that shares constraints
or liquidity proxies do not have any relationship with higher outperformance while
greater managerial incentives have a significant affects on the performance
specifically to achieve higher performance in the market. Also in the case of using

11
equal weighting funds there is no evidence of performance persistence for investors
who are seeking excess return and beating the market.
Avramov et al. (2010) study and apply a framework to evaluate the forecasting
power of economic variables in an extensive database including individual funds. In
this study they concentrate on the origin and the economic factors of predicting
model and try to develop an admirable out of sample performance for their study. To
be more exact about the performance of hedge funds in different types of strategies
they examine the economic and statistical sources of effect and their distinctive
aspect on selected funds by the strategy. Results demonstrate some evidences of
return predictability for hedge funds. Also, the time varying alphas is one of the
sources for return predictability which provide access to the advantage of running
numerous regressions instead of one single or multiple regressions for predicting the
return. Avramov et al (2013) in another study evaluate hedge fund performance
through portfolio strategies that form predictability using macroeconomic variables.
Forming the model in such a way that includes predicts ability would improve the
out of sample performance for all hedge funds and different investment styles.
Applying the strategies that have more flexibility and give predict ability to the
managerial alpha, risk factors, and benchmark returns would outperform the
strategies that ignore all the factors with together and focus on them separately or
individually. The results demonstrate that the employed strategy outperforms other
portfolio strategies by choosing funds that spawn statistically significant alpha.
Moreover, the selected styles of funds cannot describe the relative performance of
hedge funds over time. Incorporating applied strategies and their comparisons
present that the long only strategy has the ability to predict the managerial skills and
outperform the Fung and Hsieh (2004) benchmarks by over 17 percent per year.
Jame (2012) in his paper examines the trading skill of hedge funds using the data
according to transaction level. He believes that assessing the skill based on
transaction data would exclude the biases in data bases such as unreliable returns,
backfill bias, survivorship bias and etc. transaction data includes intra quarter trading
and short selling and provides a more robust estimation of skill compared to
quarterly holdings data. Results reveal no evidence of outperforming the average
hedge funds during the holding periods from one month to one year. However, using
the bootstrap simulation implies that the top trading skill of hedge funds cannot be

12
interpreted by only luck. Similarly, the results demonstrate that the persistence exists
in the top percentage of hedge funds and the intra quarter data which includes trading
skills is the source of this phenomenon. Examining the skilled hedge funds reveal
that the minority of them shows evidence of skill for the manager and mostly often
they are more illiquid stocks. In contrast to these findings that suggest the average
hedge funds are unskilled, there are some evidences of persistent skilled funds
providing liquidity in the portfolio and in the market.
Another subject that has attracted scholars and researchers in hedge funds is to
examine the relationship between diversification and performance. Many studies
have been done on how to make benefit and achieve performance through
diversification but there has been less attention on the factors that would cause using
diversification. Figueiredo et al. (2008) attempt to provide a good understanding and
interpretation what causes using diversification. They believe that both skill and luck
affect making decision about diversifying in the hedge funds population. The paper
depicts that when the manager confront an abnormal positive return which is not
persistent tries to use the benefit of diversification in portfolio. Comparing firms in
terms of the way they analyze and make decision about diversifying portfolio reveals
that firms with ex ante view outperform the firms with ex post view on short term
performance. They do this study by focusing on the profit maximization goal applied
by firms and using a huge Panel data set of hedge funds.
There are so many other scholars that examined hedge funds performance and
their corresponding persistence. For example, in short, Brown et al. (1999) measure
Sharpe ratio and single factor Jensen’s alpha. Results show Offshore hedge funds
have positive outperformance as a group in terms of Sharpe ratio or Jensen’s alpha,
but there is no evidence of persistence. Agarwal et al. (2000) measure the alpha over
the hedge funds with the same strategy and evaluate the ratio in terms of different
aspects. Findings show hedge fund performance persists at quarterly horizon, before
and after fees as well as there is very limited persistence after one year or more. In
another study they measure alpha over an eight factor model and find that hedge
funds outperform their benchmark consistently and there is persistence in hedge fund
performance at the quarterly level. Edwards et al. (2001), measure simple return,
excess return, Sharpe ratio, and alpha over a six factor model and find both positive
and negative hedge fund performance persists up to two years when alpha is

13
measured against a six factor model. Bares et al. (2003) measure alpha over
investment strategy peers and find there is evidence of short-term performance
persistence of hedge fund portfolios and long-term performance reversal and also
persistence does not stem from volatility. Brown et al. (2003) measure simple return
and find evidence of return persistence is very limited at the yearly level and there is
persistence in risk levels in the different investment styles. The same study has done
by Kat et al. (2003) which they find hedge fund returns do not persist much, but risk
and correlations with the market do. Koh et al. (2003) also study the simple return
and alpha over a seven-factor model and find Asian hedge funds’ performance
persists at the monthly and quarterly horizon but not longer. Capocci et al. (2004)
measure Alpha over market index, over Carhart’s factors, and over an 11 factor
model and find no performance persistence in extreme docile hedge funds but some
persistence in average funds. Harri et al. (2004) study Simple return, Sharpe ratio,
and an eight factor model and find there is evidence of hedge fund performance
persistence at the quarterly horizon, but most persistence is observed at a one-month
horizon. Baquero et al. (2005) measure Simple return and alpha over strategy index
and find hedge fund performance persists at horizons from one to four quarters and
there is performance reversal at the two-year horizon. Malkiel et al. (2005) study
simple returns hedge fund return persistence at the yearly horizon is limited and
greatly varies from year to year. Boyson et al. (2008) study alpha over a 19 factor
model and find hedge fund performance persists at a quarterly horizon if managers
are selected on past performance and manager tenure. Zhong et al. (2008) study
alpha over the Fung and Hsieh (2004) seven- factor model and find Hedge fund
aggregate alpha has been decreasing over time. This is due to a small right tail of the
returns distribution rather than to a bigger left tail. Aggarwal et al. (2010) study alpha
over style index and find after inception, individual hedge fund performance persists
up to five years and so on.
As it is clear many works have been done but there is not enough room here to
explain all of them. Moreover, the current study tries to add a new step forward to
the process of evaluating hedge funds performance. The past studies tried to evaluate
the hedge fund performance by running a regression and examining different ideas.
We would do the same procedure but combining it with a statistical process of
finding false discovery rate. However, reviewing the past studies would help to

14
understand and distinguish the current and the past procedure of assessing hedge
funds performance. In the next section we explain a close field of study which
focuses on mutual funds.
2.2. Mutual Funds Review
Although hedge funds are different form mutual funds as they target absolute return,
follow complex strategies, different limitation and so on, there are some studies that
focus on performance measurement in mutual funds and the manager’s skill. As the
focus of these studies is on the performance of manager we can apply the same
methodology with some modification for hedge funds. For example, Barras et al.
(2010), measure luck in estimated alphas for mutual fund performance. They apply
the false discovery rate technique to show that the proportion of skilled fund
managers has diminished rapidly over the past 20 years, while the proportion of
unskilled fund managers has increased considerably. This paper also shows that the
underperformance of managed mutual funds is due to survival of truly
underperformance funds for a longer time period. Most actively managed future
funds present positive or zero net of expense alphas and provide a balance with
passive funds. To distinguish between luck and skill for individual funds, Fama et al.
(2010), use three-factor model and four-factor (Carhart) model as well. They
compare the distribution of t-alpha achieved from actual fund returns and the
bootstrap simulation method that includes funds with zero true alphas. Results
demonstrate that net returns show evidence of skill that would cover the cost of
funds. Examining the gross fund returns in the bootstrap simulation reveals that the
manager has skill in his performance. Therefore, there is strong evidence of manager
skill before the costs in expense ratio and assessing it after the costs is not dramatic.
Similarly Kosowski et al. (2006) use bootstrap simulations to examine the
performance of the manager for net fund returns. They run a cross section regression
with a four-factor model to capture the t-distribution of alpha. They find that in the
higher percentiles there is strong evidence of skill among the best funds. Many of
them have more sufficient skill to cover costs. Compared to the results of previous
study, Fama-French’s study reveals much less evidence of skill for the manager.

15
Chen et al. (2010) revise this study and they examine the style timing abilities for
mutual funds in details. They focus on US growth oriented equity mutual funds and
assess the growth timing as a new contributor to previous studies. They provide
evidences of skill for growth timing like the way previous studies have done for
market timing.
Cuthbertson et al. (2010), investigate the performance of the German equity
mutual fund industry. They use false discovery rate (FDR) to examine both model
selection and performance measurement. This method tries to measure the proportion
of lucky funds among a group of funds with statistically significant performance.
Results demonstrate that around 80 percent of funds neither are superior nor inferior
to their benchmark. In other words, they follow the style of their indexes in the
market and manager is not able to beat the market and get abnormal return. Also, the
proportion of false discovery rate amongst the best funds is greater than the worst
performing funds. It implies that the conventional method which is counting the
number of funds with statistically significant test is more misleading for winners
compared to losers. Cuthberson et al. (2011) in another paper study the false
discovery rate in UK. They use a multiple testing framework to estimate the
proportion of truly zero alpha funds in the total population of UK funds. Also, they
estimate the false discovery rate amongst the best and worst UK equity mutual funds.
Findings reveal that around 75 percent of the mutual funds in UK neither
underperform nor outperform their benchmarks. These results are like the study they
did for Germany and it implies they follow their indexes benchmark.
Berk et al (2013) measure the skill and luck for mutual funds. They find that the
average mutual funds add value to the portfolio and it cannot be easily attributed to
the luck as it is persistent for 10 years. Results also demonstrate that the best funds
would collect higher manager fee and it would be a criteria for the future
performance. Investors are able to provide positive alpha return by investing in the
best funds or winners but in average abnormal return in the market are equal to zero.
In another study Guimond (2006) study the skill of the manager for mutual funds.
They use Black-Litterman model to examine whether the weight of underlying
portfolio deviate from the passive benchmark dramatically or not. Deviation from the
passive management and related benchmark imply an evidence of skill for the
manager. Results demonstrate that there is manager skill for the best funds and this

16
percentage is higher for value funds compared to growth funds. The results for all
funds are statistically significant and robust.
There are some more studies that evaluate the performance of mutual funds and
specially the skill of manager for outperforming or underperforming the benchmark.
However, this study concentrate on hedge funds and there is not enough space to
elaborate all the aspects and studies of mutual funds. In the next section we express
the contribution of this study which distinguishes the current study from the previous
studies which reviewed on hedge funds and mutual funds.
2.3. Contribution of the Study
This study in line with the previous studies to evaluate the performance of the
manager focuses on the abnormal return which is captured by alpha. We apply the
Fung and Hsieh (2004) seven-factor model to capture the alphas and their
corresponding t-statistics for the individual funds in our sample. The improvement
of this study is to step forward to a new stage which is to apply a statistical approach
proposed by Storey (2002) and Storey and Tibshirani (2003) to find the proportion of
false discovery rate. Another superiority of this study is to use q-values instead of p-
values in our analysis. Therefore, we use a statistical procedure applying bootstrap
method and robustness with a predetermined significant level to transform the p-
values. The reason is that the distribution of p-values may not follow a normal
pattern and by using this method we are able to adjust the p-values for measuring the
proportion of truly null. Moreover, this study takes into consideration all the tests for
soundness and robustness like applying the backfill bias test. This bias occurs when
the data collectors start to report the performance of a hedge fund which performs
very well but its past performances is not reported. Usually hedge fund managers
choose to report data to an index to attract the attention to the recent strong
performance. Since the favorable and successful early returns have been reported are
not unfavorable ones they provide an upward bias to the overall return data. Many
studies like Fung and Hsieh (2000) examined this bias and its effect on the results
extracted from biased databases. We adjust this bias for our sample data by deleting
the first 12 return observations, the first 24 return observations, and the time data

17
added to the database from the monthly return series. Also, this study revises the
procedure for different types of strategies which has been applied by the manager
and different time periods as well as different size groups. All the above mentioned
steps give this superiority to the current study to provide thorough and
comprehensive study as well as perfect results.

18
3. Data and Methodology
In this section, there is a short review about the available data and their stipulated
characteristics. The data under study include some biases and there should be some
modification process to appropriate them for the current study. In this section all the
necessary steps and the calculation formulas are presented for measuring the
proportion of false discovery rate, skilled and unskilled funds, as well as lucky and
unlucky funds. The methodology is presented in two parts. The process of measuring
alpha and corresponding p-values and t-values is stated in the first part and the
process of converting p-values to q-values and measuring the proportion of false
discovery rate and other rates is demonstrated in the second part.
3.1. Data Description and its Availability
We evaluate the hedge funds using monthly net of fee returns of live and dead
reported in HFR database over January 1994 to March 2013, a time period that
includes both market upturns and downturns as well as relatively calm and turbulent
periods. Our initial fund universe contains more than 12,000 live and dead hedge
funds. To have better and reasonable results we exclude funds that have less than 36
monthly return observations. Thereafter, the database contains 8,046 individual funds
of hedge funds that we apply the Fung and Hsieh (2004) seven-factor model and the
False Discovery Rate technique to separate luck and skill in the performance of the
manager. To minimize the impact of some biases such as survivorship, self-selection
and backfill biases which hedge fund database suffers we take into consider some
analysis. Ackerman et al. (1999) note that hedge fund performance measures are
affected by a survivorship bias when the database contains only live funds. Since in
our sample period the database contains also the defunct hedge funds, there shouldn’t
be any effect of survivorship bias. A self-selection bias may arise when only the
hedge funds with superior performance want to market their achieved performance
and participate in the survey. This may lead to an upward bias in the reported
historical performance. Jagannathan et al. (2010), show that the self-selection bias

19
has a significant effect on the performance measures. Another bias is backfill bias
which refers to the reporting only new and successful funds performance and not
reporting their poor performance. Historically, one of the aspects of hedge funds is
that they do not have very tight regulation and they are not required to report their
return. Therefore, they provide this incentive to report only the good periods of
getting high returns which transmit an upward bias in the estimated measure of
performance specifically backfill bias. Backfill bias is also known as instant history
bias. It is closely related to self-selection bias and survivorship bias that distorts the
indices. Three different scenarios are applied to minimize the impact of this bias in
the database; excluding the first 12 return observations, the first 24 return
observations, and finally deleting the monthly returns based on the time data added
to the database from the monthly return series. The procedure of finding the
proportion of false discovery rate is done for different types of strategies in a
separate scenario. All the individual funds are grouped into: Commodity Trading
Advisor (CTA), Emerging Markets, Long/Short Equity, Equity Market Neutral,
Event Driven, Fixed Income Relative Value, Global Macro, Multi-Strategy, Sector
and Short Bias category.
Commodity Trading Advisor is known also as Managed Futures Funds. They
have some similarity with macro funds in terms of using relatively liquid instruments
such as futures and following directional strategies. But they are not exactly the same
since they employ more high frequency and quantitative signals. Emerging Markets
funds invest in all types of securities in emerging countries with the prospect of high
return, including equities, bonds and sovereign debt. This strategy takes much greater
risk due to political instability, economic infrastructure, exchange rate volatility, and
so on. Long-Short Equity funds take long and short positions in undervalued and
overvalued stocks, respectively, and reduce systematic risks in the process. Equity
Market Neutral funds are similar to Long/Short Equity funds since they take long and
short positions but differ in that they typically follow more high frequency signals
and systematic trading strategies. Event Driven funds which also known as corporate
life cycle is investing. It includes Merger Arbitrage funds which monitor corporate
events and reestablishment to apply multiple strategies. This strategy usually
contains investing in opportunities created by significant transferable events, such as
spin-offs, mergers and acquisitions, bankruptcy reorganizations, recapitalizations,

20
and share buybacks. Fixed Income Relative Value funds seek to profit by exploiting
pricing inefficiencies between related fixed income securities while at the same time
neutralizing exposure to interest rate risk. Global macro strategies generally focus on
changes in global economics. They mostly concentrate on changes in government
policies which affect financial instrument and subsequently currency, interest rates,
and stock index. Hence, global macro strategies make abnormal returns by
forecasting the changes in financial markets due to the systemic factors and taking
large leveraged positions in financial derivatives. Global Macro funds try to apply
top-down global approach and their target usually is to achieve high returns with a
more flexible attitude towards risk than other hedge fund classes. Multi Strategy
hedge funds try to deliver normally positive returns regardless of the directional
movement in equity, interest rate or currency markets. In general, the risk profile of
multi-strategy is significantly lower than equity market risk. Multi-strategy funds
employ a wide variety of investment strategies and the diversification benefits help
to reduce the volatility and obtain smooth returns as well as avoiding the risk of
single strategy. Multi-strategy funds apply different techniques such as convertible
bond arbitrage, equity long/short, statistical arbitrage and merger arbitrage. Sector
Strategy select sectors by giving more weight on the basis of their attractiveness such
as their size and rate of growth, ease of entry, competitive dynamics and availability
of targets. Short Bias strategy provides a net short exposure to the market by
combining the short and long positions. A devoted short bias investment strategy
tries to capture profits when the market declines, by holding investments that are
overall biased to the short side.
There are around 8,046 individual funds in the total population after excluding
biases through stipulated constraints. This number of funds is distributed into
different applied strategies as follows. Strategy Long/Short includes 2,370 funds
(29.46%), Multi Strategy 1,402 funds (17.43%), Relative Values 937 funds
(11.65%), Emerging Markets 936 funds (11.63%), Event Driven 611 funds (7.59%),
Market Neutral 564 funds (7.01%), Commodity Trading Advisor 426 funds (5.29%),
Sector 409 funds (5.08%), Global Macro 342 funds (4.25%), and Short Bias 49 funds
(0.61%) respectively. Chapter four presents the detail information about each
strategy.

21
3.2. Performance Measurement
To compute the fund performance, we apply the seven-factor model proposed by
Fung and Hsieh (2004) which is the core asset pricing model for hedge funds. This is
the standard equation which has been developed during the time. We employ a time
series regression on the net of fee monthly excess returns of hedge funds. We use all
the factors on the basis of excess to the risk free rate and the equation for capturing
the factors is as follow:
tiff
fffti
RPTFSCOMbRPTFSFXb
RPTFSBDbTYBAAbRTYbSPRLbRSPbaR
,76
54321,
)()(
)()()()()(


These seven factors are defined as the excess return of the S&P 500 index (SP −
RF); the return spread between the Russell 2000 index and the S&P 500 index (RL −
SP); the excess return of 10-year U.S. Treasuries (TY − RF); the return of Moody’s
BAA corporate bonds minus 10-year Treasuries (BAA − TY); and the excess returns
of look-back straddles on bonds (PTFSBD − RF), currencies (PTFSFX − RF), and
commodities (PTFSCOM − RF). The defined alpha ( a ) in the model is an intercept
which measures the average abnormal return of the underlying portfolio. The
stochastic term captures the error of the model. Based on the studies it seems that
applying a multifactor model for assessing the performance of hedge funds,
specifically their excess returns should be compulsory.
Hedge funds can apply different types of investment strategies and taking short
and long position at the same time. Hence, there would the possibility that a variety
of economic risk factors affect the returns of individual hedge funds. Moreover,
individual hedge funds have distinct sensitivities to the same risk factors even in a
peculiar style. It varies according to their trading strategy and the leverage they apply
to achieve abnormal returns in the market. The number of factors which have been
used in the multiple- model depends on the researcher. There is no exact rule to
define this number but adding more factors to the model makes it more complicated
for analyzing and interpreting (Fung and Hsieh 1997). Using single factor model and
other alternative measures to assess hedge funds do not give significant results and
provide a good reason to apply multiple factor model to estimate the excess return

22
for hedge funds. The multi factor model may have better estimation and
interpretation of risk factors which hedge funds are exposed and in result provides a
more reasonable comparison with the real achieved excess returns by hedge funds.
Using the above mentioned seven-factor model for individual hedge funds along
adding the necessary constraint provides the variables for assessing the performance
of the manager. Running the time series regression based on the defined model
provide some factors such as alphas, p-values and t-values. There is a possibility to
estimate the proportion of positive and negative significant funds (S-
, S+
) on the basis
of achieved t-values. This estimation is done for different significant level of 5, 10,
15, and 20 percent. This procedure is repeated for all individual funds before and
after adjusting for backfill bias. To mitigate the backfill bias some new constraint
such as deleting the first monthly 12 and 24 return observations as well as deleting
the data added to the database are applied to modify the underlying data. Another
evaluation is to revise the procedure on the basis of different types of strategies. In
this stage, after running the time series regression the individual data are sorted based
on their main strategy. Also there are constraints to separate individual funds to
different time periods and different size groups such small, medium and large based
on the asset under management. Finding p-values is another purpose of this
regression which helps us to find the proportion of truly null zero funds in the
population of M-funds. It is the corner stone for estimating the proportion of false
discovery rate, lucky, unlucky, skilled and unskilled funds in the population.
3.3. Statistical Measurements
The normal way of determining if there is any evidence of skill or luck in a
population is to make a hypothetical test through a null and alternative hypothesis. In
this statistical method a threshold is defined and according to the predetermined
significant level the null will be rejected if it falls in the rejection region and will be
accepted other way around. Therefore, we test a null hypothesis H0 versus an
alternative H1 for a statistic x with different threshold. We reject H0 when xγ in a
rejection region of γ and we accept H0 when x γ and it is in an accepted region. A
type I error (false positive, rejection of a true hypothesis) occurs when x is in the

23
rejection region but it is actually true, and a type II error (false negative, non-
rejection of a false hypothesis) occurs when x does not belong to rejection region but
actually it does. For finding the rejection region γ a preferable level of type I error
will be chosen and on the basis of different threshold all the rejection regions which
are less or equal to preferred level will be tested. The one with lowest type II error
satisfies the criteria to be the selected rejection region. Hence, the procedure is
established in a way to control type I and II error in this case. This procedure cannot
be applied in a multiple hypotheses when many tests apply at the same time. It would
be so complicated and needs an easier and more significant procedure. There are type
I and II error and at the same time many tests which make it unclear how to measure
them (Story, 2001).
Multiple hypothesis testing includes controlling the errors which are more
complicated than single hypothesis testing. The type I error in a single test
hypothesis would be compounded in a multiple testing. Dudiot et al. (2002), asserts
that the p-value for a single hypothesis does not have the same signal and significant
meaning since with increasing the number of hypothesis tests the probability of
getting small p-values will increase sharply. Therefore, the main focus of multiple
hypothesis testing is to deal with type I error (false positive) and develop new
methods for adjusting simultaneous tests. The probability of rejecting the null
hypothesis when it is actually false shows the power of test. In other words, the
probability of avoiding type II error which is Power = 1- P (type II error) defines the
power of test. The outcome of m multiple hypothesis testing is being summarized as
follows:
Decision Accept Reject
True Null Correct Decision (1- α)
Incorrect Decision
Type I error (α)
False Null Incorrect Decision
Type II error (β)
Correct Decision (1- β)
Which α = P (Type I Error) and β = P (Type II Error)

24
As mentioned above many tests run simultaneously and correspondingly the p-
values need to be adjusted. There are three ways to apply this adjustment: the single
step, the step down, and the step up procedure. In the single step procedure each test
has its own critical value and the results of tests are not affected by one another. In a
step down procedure the unadjusted p-values will be ordered by the most significant
one which is not very conservative compared to the single step and develops the
power. In a step up procedure the opposite way will be followed and unadjusted p-
values will be ordered by the least significant one. This procedure is less
conservative compared to the step down procedure in contingent of independence
tests. Many methods and ways are proposed to control the multiple hypothesis
testing. But to answer which one of these procedures is the best depends how much
the test is supposed to be conservative (Demissie 2008).
An approach known as the false discovery rate (FDR) tries to provide a balance
between controlling type I error and the power of test by classifying funds as
significant at a chosen significance level of γ in test. The driving force question is
what proportion of significant funds are false discoveries. The approach we use in
this study follows Storey (2002) and Storey, Taylor, and Siegmund (2004)
procedure. We propose the recently introduced q-value by Storey (2002) to analyze
statistical tests on thousands hypothesis test on funds. One of the most commonly
used concepts in this field is the q-value which represents the quantity of false
rejected null. This term also is known as false discovery rate (FDR) that stems from
the developments of the other fields of studies such as genomics. The technology of
microarray is an example of applied techniques to solve the problem of multiple
hypotheses testing which enables testing thousands of gens simultaneously.
3.3.1. False Discovery Rate
The concept of false discovery rate is proposed in cases like microarray that one is
interested to control the number of false positives among the total number of
positives. In applying false discovery rate we deal with the q-value which is similar
to the well known p-value. It attributes a significance level to each hypothesis under
a certain error rate. As p-value estimates the minimum false positive rate in a

25
significant test, q-value measures the minimum false discovery rate in a significant
test. Comparing these two criteria reveals that p-value is a common tool for
measurement in single hypothesis test but q-value provides a good measurement for
a multiple hypothesis testing. Regardless of this different usage both of them has a
null and an alternative hypothesis. The concept of significance here is the amount of
deviation from the null. Before any more explanation about false discovery rate we
should distinguish two confusing concepts. The false positive and false discovery
rate seems to be the same but they are two different concepts. The false positive
shows the rejection of null when it is actually true. Therefore a false positive rate
demonstrates the proportion of true null hypotheses which have been estimated to be
significant incorrectly. A false discovery conveys the same false positive but
concentrating on the false positive among the discoveries of significant tests. We can
see this difference in the following definitions:
False Positive Rate ≈ (# of false positives) / (# of true null tests)
False Discovery Rate ≈ (# of false positives) / (# of significant tests)
Therefore while the false positive measures how likely it is for a truly null to be
significant like what it is observed, false discovery deal with the information about
the false null hypotheses. In a testing many hypotheses simultaneously the false
discovery rate gives more useful information. However, the false positive rate is an
easy way to calculate and measure p-values. To apply false discovery rate and have
an exact measurement there have to be some calculation for finding which tests are
truly significant and what kind of distribution is attributed to them (Storey, 2004).
There are different approaches in defining false discovery rate (FDR) which
among them Benjamini and Hochberg (1995), and Storey (2002) provide
conservative procedures which BH approach provides a less conservative estimation
by including the proportion of true null hypotheses. In BH approach they select a
predetermined false discovery rate and then they follow the procedure to find a data
dependent threshold rule so that the FDR would be less or equal to the pre-chosen
level. They define the false discovery rate as follows:
)0Pr(0
1









 RR
R
V
E
R
V
EFDR , Where )1,max(1 RR  ,

26
With an assumption that p-values follow the null distribution with an independent
and uniformly distribution there would be three steps:
Step 1: let )(...)1( mpp  observed p-values based on their order.
Step 2: calculate }./)(:1max{ˆ mkkpmkk 
Step 3: if kˆ exists, the null will be rejected corresponding to )ˆ(...)1( kpp  .
Otherwise, nothing will be rejected.
The procedure of BH was originally introduced by Simes (1986) to control
weakly the Family Wise Error Rate (FEWR) under the assumption that all p-values
are independent. The problem for this approach is that it provides a strong control on
false discovery rate. The second approach was proposed by Storey (2002) that a
threshold is fixed beforehand and based on that a false discovery rate in the
significant region which is less or equal to the true false discovery rate will be
estimated. In the Storey approach there is another option to measure the positive
false discovery pFDR and it is defined as: 



 0R
R
V
EpFDR . Bothe FDR and
pFDR are equivalent in an asymptotic setting.
Storey (2002) defines a conservative estimation of false discovery rate for a non-
random significance threshold t as:
mtR
t
tRDF
/}1)({
)(ˆ
)(ˆ 0




The term )(ˆ0  is the proportion of true null hypotheses. This estimation of false
discovery rate is dependent to the parameter λ and is defined as:
m
W
)1(
)(
)(ˆ0





Storey in his first theorem asserts that under an independent and identically
distribution of p-values corresponding to the true null hypotheses, for fixed )1,0[ ,
)()](ˆ[ tFDRtRDFE 
Another important concept about false discovery rate is the q-value. The q-value is
somehow similar to the well known p-value. It provides each hypothesis test an
estimation of significance corresponding to a certain error rate. While the p-value
measures the minimum false positive rate that is occurred in a significant test, q-

27
value measures the minimum false discovery rate which is occurred in a significant
test. The quantity of p-value in a test with a rejection region  is defined to be
}0min{Pr()(  HTtvaluep . This quantity demonstrates the strength of the
observed statistics which can occur with the minimum type I error. In a multiple
testing these p-values can be adjusted to reveal again the strength of an observed
statistic with corresponding type I error. Therefore, for an observed statistic T=t the
q-value of t is defined to be )}(inf{)(  pFDRtq . The q-value can be defined as the
minimum pFDR which is achieved when rejecting a statistic with value t in a
rejection region. According to the above mentioned definitions both of p-value and
q-value have the same concept but the q-value is being adjusted for the multiple
testing. The q-value in a multiple test hypothesis with identically independent p-
values will be defined as:
.
)Pr(
inf)}(inf{)( 0











p
pFDRpq
In other words, the q-value can be defined as the pFDR analogue of the p-value.
Storey (2001) asserts that the q-value is the same whether it is obtained from the
original statistics or from their corresponding p-values. By the procedure proposed
by Storey it is an easy way to adjust the obtained p-values and estimate the false
discovery rate as well as it makes this approach a widely available method.
3.3.2. Estimation Procedure
Barras et al. (2010), attempt to measure the mutual fund performance and its
persistence through separating the skilled and unskilled fund managers by controlling
the false discovery. They try to implement this identification by correcting the usual
hypotheses testing. A classical method is to classify the whole distribution into three
groups. The alpha which is the measurement of the skill of the managers is located in
the right tail, zero, or left tail. These regions will demonstrate the skill, no skill, and
unskilled of the manager respectively. In this methodology the funds reveal the skill
of the manager by locating in the right tail and rejecting the null hypothesis
accordingly. Also the funds demonstrate that the manager is unskilled by locating in

28
the left tail and rejecting the null hypothesis. By accepting this assumption which
each category (unskilled, zero alpha, and skilled funds) has its own distribution,
makes it possible to assess their population. Therefore, the only problem is to find
the proportion of them in the total population of funds.
At a given significance level of , the probability that a zero alpha fund
demonstrates luck equals 2/ . If the proportion of the zero alpha funds in the
population is 0 , the expected proportion of lucky funds is equal to: 2/.)( 0  
FE
If )( 
SE is the proportion of significant positive alpha funds, then we simply adjust
)( 
SE to find the expected proportion of truly skilled funds (at a significance
level ):
)2/()()()()( 0   
SEFESETE
The expected proportion of unlucky funds in the left tail, denoted by )( 
SE , is
achieved applying a similar procedure. Choosing different levels for  makes it
possible to see whether the number of truly skillful funds rises noticeably with  or
not, which tells us whether skilled funds are concentrated or dispersed in the right tail
of the distribution. It provides helpful information in choosing an ex-ante portfolio of
skilled funds. Since the probability of being unlucky is also equal to 2/ therefore,
)( 
FE is equal to )( 
FE . As a result, the expected proportion of unskilled funds
)( 
TE is similarly given by:
).2/.()()()()( 0   
SEFESETE
The expected false discover rate amongst the statistically significant positive alpha
funds is:
)(
)2/(
)(
)( 0








SESE
FE
FDR
To calculate all the above statistics, the key point is to estimate the proportion of
truly zero alphas 0 in the population. To find 0 we use p-values achieved by the

29
seven factor regression model. To apply these p-values for a multiple testing
hypotheses they should be adjusted to q-values. In this stage we apply the method of
Storey (2004) for adjusting p-values. Each distribution of funds in the right and left
tail will be assessed separately with four different confidence level. To have a normal
distribution of q-values we apply the bootstrap method along with a robust option to
estimate the truly zero alphas 0 in the population of each group. This procedure will
be repeated for all individual alphas before and after adjusting for backfill bias,
different types of strategies, different time periods, and finally different size groups.

30
4. Empirical Results
This chapter reports the main results of this study. In this study, the risk-adjusted
performance is measured using the Fung and Hsieh (2004) seven-factor model. This
step is able to derive the alphas and their corresponding t-values which determine
whether they are significant or not. For this step of time series regression, there are
some tricky steps which are important, like taking into consideration enough data for
each fund and backfill bias which may have an effect on the results. Therefore, we
separated this chapter into four subtitles in order to have a better interpretation and
analysis of our findings. Firstly, we evaluate the impact of luck on hedge funds
performance before and after adjusting for backfill bias. This gives us the
opportunity to study the effect of backfill bias which is one of the most important
analyses in hedge funds. Secondly, we study the impact of luck on hedge funds
performance based on different strategies which has been applied in practice by the
manager. Thirdly, the impact of luck on hedge funds performance based on time in
which the whole period will be separated into four groups: 1994-1998, 1999-2003,
2004-2008, and finally 2009-2013. The final aspect of our study is to examine the
impact of luck on hedge funds performance based on size, which for every one of
time periods we divide it into three subgroups: small, medium and large funds. The
criterion for separating the funds for size effect is asset under management.
4.1. The Impact of Luck on Performance before and after Adjusting for
Backfill bias
We first require that a hedge fund has at least 36 return observations. There are 8,046
hedge funds that fulfill this condition. We find a mean alpha of 4.097 and a median
alpha of 3.796. We next test the significant positive and negative funds in two sides
(left tail and right tail) with four different significance levels of 5, 10, 15 and 20
percent. The factors for evaluation are false discovery rate, significant funds, lucky
or unlucky funds, and finally skilled or unskilled funds.

31
Table I: All Individual Funds before and after Adjusting for Backfill bias
In this table all individual funds are separated into four Panels. In each Panel we examine the
proportion of positive and negative significant funds, lucky and unlucky funds, skilled and unskilled
funds, and finally, the false discovery rate. We show the results in four different significance levels of
5, 10, 15, and 20 percent in the right and left tails. In Panel A the results are derived before backfill
bias, in Panel B after deleting the first 12 monthly return observations, in Panel C after deleting the
first 24 monthly return observations, and in Panel D after deleting the data added into the database.
Panel A: All Individual Funds
N=8,046, Mean Alpha=4.097, Median Alpha=3.796
Left tail Right tail
Sig. Level (  ) 0.05 0.10 0.15 0.20 0.20 0.15 0.10 0.05 Sig. Level (  )
FDR-
(%) 55.77 68.51 78.02 58.15 14.58 12.19 9.22 5.867 FDR+
(%)
Signif. S-
(%) 2.548 4.151 5.543 6.738 39.62 35.53 30.85 24.22 Signif. S+
(%)
Unlucky F-
(%) 1.421 2.844 4.332 5.776 5.776 4.332 2.844 1.421 Lucky F+
(%)
Unskilled T-
(%) 1.127 1.307 1.211 0.962 33.84 31.19 28.01 22.79 Skilled T+
(%)
Panel B: Backfill Bias ≤ 12
FDR-
(%) 59.03 76.73 84.55 92.01 17.89 14.99 11.55 6.979 FDR+
(%)
Signif. S-
(%) 2.509 4.190 5.688 6.970 35.83 32.07 27.83 21.22 Signif. S+
(%)
Unlucky F-
(%) 1.481 3.215 4.809 6.413 6.413 4.809 3.215 1.481 Lucky F+
(%)
Unskilled T-
(%) 1.028 0.975 0.879 0.557 29.42 27.26 24.62 19.74 Skilled T+
(%)
Panel C: Backfill Bias ≤ 24
FDR-
(%) 55.06 64.35 70.32 78.00 19.31 16.06 12.76 8.417 FDR+
(%)
Signif. S-
(%) 2.917 4.993 6.735 8.192 33.09 29.49 25.19 19.08 Signif. S+
(%)
Unlucky F-
(%) 1.606 3.213 4.736 6.390 6.390 4.736 3.213 1.606 Lucky F+
(%)
Unskilled T-
(%) 1.311 1.780 1.999 1.802 26.70 24.75 21.98 17.47 Skilled T+
(%)

32
Panel D: Date Added
Sig. Level (  ) 0.05 0.10 0.15 0.20 0.20 0.15 0.10 0.05 Sig. Level ( )
FDR-
(%) 49.66 61.38 68.73 75.63 25.33 21.68 17.13 11.10 FDR+
(%)
Signif. S-
(%) 3.542 5.733 7.541 9.208 27.49 23.91 20.54 15.84 Signif. S+
(%)
Unlucky F-
(%) 1.759 3.519 5.183 6.964 6.964 5.183 3.519 1.759 Lucky F+
(%)
Unskilled T-
(%) 1.783 2.214 2.358 2.244 20.53 18.73 17.02 14.08 Skilled T+
(%)
The results demonstrate that significant funds concentrated on the right tail and
the numbers of positive significant funds are more than the negative significant
funds. The numbers of significant funds increase alongside with the increase in
significance level in both tails. The percentage of lucky and unlucky funds will
increase correspondingly with the increase in significance level, and it is the same for
both tails because our assumption is to have a normal distribution. The percentage
and the number of funds to have truly significant funds are more concentrated on the
right tail. Also, truly significant funds increase when the significance level increases
in the right tail but decreases in the left tail. The results for false discovery rate show
that it is greater for the left tail and in the right tail we have a very small percentage
of FDR. This percentage increases correspondingly with increasing the significance
level. Panel A shows that for individual funds the impact of luck is low and the
performance can be explained by the skill of the manager.
We follow the same process for studying the individual funds performance but we
adjust for backfill bias by deleting the first 12 return observations. Therefore the
number of funds in Panel B will be 6,417 with a mean alpha of 3.329 and a median
alpha of 3.329. Deleting the first 12 return observations causes reduction of some
significant funds and its effects on FDR that measures the proportion of lucky funds
among significant funds to increase. The percentage and the number of positive
significant funds are greater than negative significant funds, the percentage of lucky
funds are greater than unlucky funds, and the proportion of skilled funds are greater
than unskilled funds. But for FDR, the concentration is on the left tail. The
significant funds and their luck factors will increase correspondingly with an increase

33
in the significance level. Skilled funds follow the same pattern but unskilled funds
follow the opposite pattern. The results are similar to the before backfill bias but the
difference is that the number of significant funds would decrease and the proportion
of truly significant funds would decrease in result. The proportion of unlucky funds
among negative significant funds is considerable while the proportion of lucky funds
among positive significant funds is lower in comparison to the left tail. The
percentage of FDR is higher after adjusting for backfill bias which deletes the first 12
return observations.
In Panel C, we review all individual funds with regard to backfill bias after
deleting the first 24 return observations. After controlling backfill bias, we have
5,188 funds with a mean alpha of 2.524 and a median alpha of 2.737. The proportion
of significant funds is greater in the right tail and it increases correspondingly with
the increase in significance level in both tails. In comparison with the first backfill
bias (deleting 12 first return observations), the percentage and the number of
significant funds decreased in the right tail but it increased smoothly in the left tail.
The percentages of lucky and unlucky funds are the same as those in Panel B. In
contrast to Panel B, the percentage of skilled funds decreased in the right tail but
increased in the left tail smoothly. The evidence of the false discovery rate in contrast
with Panel B shows that FDR increases in the right tail and it decreases in the left
tail, however in both tails with increasing the significance level, the FDR increases.
Another aspect of examining funds is to check the backfill bias based on the data
added into the database. In Panel D we added a new constraint before running time
series regression and we deleted the data which added during that time to the
database. Therefore, we would have around 4,746 funds with a mean alpha of 1.609
and a median alpha of 1.945. The new constraint causes more funds to be killed off
in both sides of the distribution. Although the proportion of significant funds
decreases in both tails, the proportion of significant funds in the right tail is still
greater than the left tail. This proportion increases correspondingly with the increase
in significance level. The proportion of skilled funds in the right tail decreased in
contrast with previous Panels and the unskilled funds increased in the left tail. The
proportion of skilled and unskilled funds in both tails increased with the increase in
significance level. The proportion of false discovery increased considerably in the
right tail and it decreased in the left tail.

34
In conclusion, we may say that all individual funds, before applying backfill bias
(Panel A), after applying the first adjustment for backfill bias (deleting the first 12
return observations - Panel B), and second backfill bias (deleting the first 24 return
observations - Panel C), and finally third backfill bias (data added - Panel D) has the
higher number of funds respectively and their mean and median decreases with
respect to decreasing the number of funds. The proportion of positive significant
funds in the right tail decreases and the proportion of negative significant funds
increase in each Panel relatively. In the right tail, along with killing more funds with
different backfill bias, the proportion of lucky funds increases and the proportion of
skilled funds decreases. Contrarily, in the left tail the proportion of unlucky and
unskilled funds increase but does not follow a smooth transition and it is not very
clear. The proportion of false discovery rates increases dramatically in all
significance levels in the right tail. In the left tail, except Panel B, we have a decrease
in the proportion of unlucky funds among negative significant funds. In general,
whether it is before or after adjusting for backfill bias, the proportion of lucky funds
among positive significant funds is greater than unlucky funds among negative
significant funds and for positive significant funds we have the skill of the manager
but not, however, in the left tail and for negative significant funds.
Figure I: Comparing Skill vs. Luck in Significant Funds
In this figure we can see that the manager has more skill compared to luck in his performance in the
market. This comparison is done in four different significance levels of 5, 10, 15, and 20 percent.
Significant Skilled Lucky

35
4.2. The Impact of Luck on Performance for Different Strategies
In this section we follow the same procedure for 10 different strategies. The
distribution of all funds in each strategy is as follows: the CTA includes 426 funds
(5.29%), the Emerging Markets includes 936 funds (11.63%), the Event Driven
includes 611 funds (7.59%), the Global Macro includes 342 funds (4.25%), the
Long/ Short include 2370 funds (29.46%), the Market Neutral includes funds (7%),
the Multi Strategy includes 1,402 funds (17.42%), the Relative Value includes 937
funds (11.65%), the Sector strategy includes 409 funds (5.08%), the Short Bias
includes 49 funds (0.61%).
Figure II: The Proportion of Different Strategies
In this figure the proportion of all 10 strategies applied by the manager is shown in percentage.
Long/Short and, Multi Strategy, Emerging Markets, Relative Value, Market Neutral, CTA, Sector,
Global Macro, and Short Bias have the highest amount relatively.

36
As it is demonstrated in figure II, the strategies of Long/Short, Multi Strategy,
Relative Value, Emerging Markets, Event Driven, Market Neutral, CTA, Sector,
Global Macro, and Short Bias have the highest proportion of funds respectively. For
every one of the above strategies we examine the proportion of positive and negative
significant funds, lucky and unlucky funds, skilled and unskilled funds, and the false
discovery rate of individual funds. Similarly, we evaluate the results in different
significance levels of 5, 10, 15, and 20 percent. The results are presented in Table II
for different strategies. The results of this study are derived from individual funds
and we have evaluated all the strategies which are applied by the manager.
Therefore, comparing the results with studies which put them in a portfolio is not at
that much easy. But in terms of their proportion and the way they have been used
they follow the same pattern of previous studies such as Kosowski et al. (2007),
Joenvaara (2010), and similar studies.
Table II: The Funds of Hedge Funds in Different Strategies
In this table all individual funds is separated into different Panels based on the strategy applied by the
manager. We have 10 different strategies, which, in each strategy we examine the proportion of
positive and negative significant funds, lucky and unlucky funds, skilled and unskilled funds, and the
false discovery rate with different significance levels of 5, 10, 15, and 20 percent in the right and left
tails.
Strategy: CTA
N=426, Mean Alpha=4.871, Median Alpha=3.259
FDR-
(%) 32.82 43.75 57.91 67.31 17.98 15.75 12.38 8.416 FDR+
(%)
Signif. S-
(%) 4.695 7.042 7.981 9.155 34.27 29.34 24.88 18.31 Signif. S+
(%)
Unlucky F-
(%) 1.541 3.081 4.622 6.162 6.162 4.622 3.081 1.541 Lucky F+
(%)
Unskilled T-
(%) 3.154 3.961 3.359 2.993 28.11 34.72 21.79 16.77 Skilled T+
(%)
Strategy: Emerging Markets
FDR-
(%) 45.07 59.63 72.77 80.63 18.35 15.67 12.29 8.687 FDR+
(%)
Signif. S-
(%) 3.419 5.128 6.303 7.585 33.33 29.27 24.89 17.74 Signif. S+
(%)
Unlucky F-
(%) 1.541 3.058 4.587 6.116 6.116 4.587 3.058 1.541 Lucky F+
(%)
Unskilled T-
(%) 1.878 2.070 1.716 1.469 27.21 24.68 21.83 16.19 Skilled T+
(%)

37
Strategy: Event Driven
FDR-
(%) 59.67 59.71 66.22 71.68 5.907 5.603 3.693 2.247 FDR+
(%)
Signif. S-
(%) 1.473 2.946 4.583 4.909 59.57 54.17 47.63 39.12 Signif. S+
(%)
Unlucky F-
(%) 0.879 1.759 3.035 3.519 3.519 3.035 1.759 0.879 Lucky F+
(%)
Unskilled T-
(%) 0.594 1.187 1.548 1.390 56.05 51.14 45.87 38.24 Skilled T+
(%)
Strategy: Global Macro
FDR-
(%) 49.01 75.40 88.29 98.75 17.03 14.57 11.01 7.429 FDR+
(%)
Signif. S-
(%) 2.924 3.801 4.971 5.848 33.92 30.12 26.02 19.29 Signif. S+
(%)
Unlucky F-
(%) 1.433 2.866 4.389 5.775 5.775 4.389 2.866 1.433 Lucky F+
(%)
Unskilled T-
(%) 1.491 0.935 0.582 0.073 28.15 25.73 23.15 17.86 Skilled T+
(%)
Strategy: Long / Short
FDR-
(%) 73.54 79.05 94.23 101.9 17.47 14.86 11.35 7.625 FDR+
(%)
Signif. S-
(%) 2.109 3.924 5.063 6.245 36.41 32.11 27.34 20.34 Signif. S+
(%)
Unlucky F-
(%) 1.551 3.102 4.771 6.361 6.361 4.771 3.102 1.551 Lucky F+
(%)
Unskilled T-
(%) 0.558 0.822 0.292 -0.12 30.05 27.34 24.24 18.79 Skilled T+
(%)
Strategy: Market Neutral
FDR-
(%) 53.38 78.98 77.65 85.55 22.26 17.84 13.56 8.563 FDR+
(%)
Signif. S-
(%) 3.014 4.078 6.028 7.979 30.67 26.24 23.76 18.79 Signif. S+
(%)
Unlucky F-
(%) 1.609 3.221 4.681 6.826 6.826 4.681 3.221 1.609 Lucky F+
(%)
Unskilled T-
(%) 1.405 0.857 1.347 1.153 23.84 21.56 20.54 17.18 Skilled T+
(%)

38
Strategy: Multi Strategy
FDR-
(%) 41.43 63.09 67.51 72.00 12.14 10.00 7.597 3.845 FDR+
(%)
Signif. S-
(%) 2.496 3.994 5.706 7.133 42.29 38.52 33.17 26.89 Signif. S+
(%)
Unlucky F-
(%) 1.034 2.520 3.852 5.136 5.136 3.852 2.520 1.034 Lucky F+
(%)
Unskilled T-
(%) 1.462 1.474 1.854 1.997 37.15 34.67 30.65 25.86 Skilled T+
(%)
Strategy: Relative Value
FDR-
(%) 46.61 56.27 59.33 62.51 9.979 8.110 5.832 3.354 FDR+
(%)
Signif. S-
(%) 2.581 4.269 6.083 7.684 48.13 44.50 41.19 35.86 Signif. S+
(%)
Unlucky F-
(%) 1.203 2.402 3.609 4.803 4.803 3.609 2.402 1.203 Lucky F+
(%)
Unskilled T-
(%) 1.378 1.867 2.474 2.881 43.33 40.89 38.79 34.66 Skilled T+
(%)
Strategy: Sector
FDR-
(%) 95.03 147.8 171.6 237.5 18.39 14.20 11.18 7.229 FDR+
(%)
Signif. S-
(%) 1.711 2.200 2.934 2.934 37.89 35.45 29.09 22.49 Signif. S+
(%)
Unlucky F-
(%) 1.626 3.252 5.035 6.968 6.968 5.035 3.252 1.626 Lucky F+
(%)
Unskilled T-
(%) 0.085 -1.05 -2.10 -4.03 30.92 30.42 25.84 20.86 Skilled T+
(%)
Strategy: Short Bias
FDR-
(%) 85.30 45.31 51.19 72.53 24.17 17.06 16.48 12.18 FDR+
(%)
Signif. S-
(%) 2.041 8.163 10.20 10.20 30.61 30.61 22.45 14.29 Signif. S+
(%)
Unlucky F-
(%) 1.741 3.699 5.222 7.398 7.398 5.222 3.699 1.741 Lucky F+
(%)
Unskilled T-
(%) 0.300 4.464 4.978 2.802 23.21 25.39 18.75 12.55 Skilled T+
(%)

39
The results demonstrate that the proportion of positive significant funds in the
right tail is greater than the left tail and it increases correspondingly with the increase
in significance level in both sides. The proportion of skilled funds is greater than
lucky funds in the right tail and it increases along with the increase in the
significance level. In contrast, the proportion of unlucky funds is greater than the
proportion of unskilled funds in the left tail and it increases along with the increase in
the significance level. In terms of positive significant funds, Event Driven, Relative
Value, Multi Strategy, Sector, Long/Short, Global Macro, CTA, Market Neutral,
Emerging Market, and Short Bias have the highest proportion respectively in the
right tail. Strategies Short Bias, CTA, Emerging Markets, Market Neutral, Relative
Value, Multi Strategy, Long/Short, Global Macro, Event Driven, and Sector have the
highest negative significant funds respectively in the left tail. In terms of lucky funds,
Short Bias, Sector, Market Neutral, Long/Short, CTA, Emerging Markets, Global
Macro, Multi Strategy, Relative Value, and Event Driven has the highest proportion
respectively in the right tail. Strategies Short Bias, Sector, Market Neutral,
Long/Short, CTA, Emerging Markets, Global Macro, Multi Strategy, Relative Value,
and Event Driven have the highest proportion of unlucky funds respectively in the
left tail. Strategies Event Driven, Relative Value, Multi Strategy, Sector, Long/Short,
Global Macro, Market Neutral, CTA, Emerging Markets, and Short Bias has the
highest proportion of skilled funds respectively in the right tail but these results are
not so clear for making a decision about the proportion of unskilled funds in the left
tail. Comparing the strategies in terms of false discovery rate shows that strategies
Short Bias, Market Neutral, Emerging Markets, CTA, Long/Short, Sector, Global
Macro, Multi Strategy, Relative Value, and Event Driven has the highest proportion
respectively in the right tail and strategies Sector, Long/Short, Global Macro, Market
Neutral, Emerging Markets, Multi Strategy, Event Driven, Short Bias, Relative
Value, and CTA has the highest proportion of false discovery rate respectively in the
left tail. In conclusion the manager has been more skillful compared to be lucky in
the strategies of Event Driven, Relative Value, Multi Strategy, Global Macro, Sector,
Long/ Short, CTA, Emerging Markets, Market Neutral, and Short Bias respectively.
In contrast, the manager has been more unlucky compared to be unskilled in the
strategies of Sector, Long/Short, Global Macro, Market Neutral, Emerging Markets,
Multi Strategy, Event Driven, Short Bias, Relative Value, and CTA respectively.

40
4.3. The Impact of Luck on Performance Based on Time Period
In this section we examine the individual funds of hedge funds to find the impact
of luck on performance in different time periods. Hence, we separate the sample into
four time periods and we study the proportion of positive and negative significant
funds, lucky and unlucky funds, skilled and unskilled funds, and the false discovery
rate in different significance levels of 5, 10, 15, and 20 percent.
The results of this section are in consistent with the study of Fung et al (2008).
Fung et al (2008) find that during the year 1998 and 2000 there are a significant
proportion of alpha. For the time period 1999-2003, we also find the highest
proportion of positive significant alpha compared to other time periods.
Table III: The Funds of Hedge Funds in Different Time Periods
In this table we separate individual funds into four different time periods; 1994-1998, 1999-2003,
2004-2008, and 2009-2013. In each time period we examine the proportion of positive and negative
significant funds, lucky and unlucky funds, skilled and unskilled funds, and the false discovery rate
with different significance level of 5, 10, 15, and 20 percent in right and left tails.
Panel A: Time Period 1994-1998
FDR-
(%) 36.32 52.92 58.24 60.07 12.94 11.15 7.992 4.898 FDR+
(%)
Signif. S-
(%) 3.442 4.724 6.800 8.323 38.64 35.52 31.28 25.52 Signif. S+
(%)
Unlucky F-
(%) 1.250 2.500 3.960 5.000 5.000 3.960 2.500 1.250 Lucky F+
(%)
Unskilled T-
(%) 2.192 2.224 2.840 3.323 33.64 31.56 28.78 24.27 Skilled T+
(%)
Panel B: Time Period 1999-2003
FDR-
(%) 114.0 175.0 206.9 198.9 7.568 6.685 4.674 2.794 FDR+
(%)
Signif. S-
(%) 0.885 1.165 1.584 2.050 53.87 49.02 43.62 36.11 Signif. S+
(%)
Unlucky F-
(%) 1.009 2.039 3.277 4.077 4.077 3.277 2.039 1.009 Lucky F+
(%)
Unskilled T-
(%) -0.12 -0.87 -1.69 -2.03 49.79 45.74 41.58 35.10 Skilled T+
(%)

41
Panel C: Time Period 2004-2008
FDR-
(%) 103.1 121.4 123.4 133.6 22.66 19.37 15.75 11.25 FDR+
(%)
Signif. S-
(%) 1.693 2.944 4.342 5.348 31.53 27.67 22.69 15.51 Signif. S+
(%)
Unlucky F-
(%) 1.745 3.573 5.359 7.145 7.145 5.359 3.573 1.745 Lucky F+
(%)
Unskilled T-
(%) -0.05 -0.63 -1.02 -1.79 24.39 22.31 19.12 13.77 Skilled T+
(%)
Panel D: Time Period 2009-2013
FDR-
(%) 33.74 36.26 40.34 43.35 24.83 20.88 15.39 10.09 FDR+
(%)
Signif. S-
(%) 4.700 8.751 12.18 15.11 26.38 23.53 20.62 15.71 Signif. S+
(%)
Unlucky F-
(%) 1.586 3.173 4.913 6.550 6.550 4.913 3.173 1.586 Lucky F+
(%)
Unskilled T-
(%) 3.114 5.578 7.267 8.560 19.83 18.62 17.45 14.12 Skilled T+
(%)
In Panel A which is the time period, 1994-1998, we have 1,250 funds with a mean
of 4.489 and a median of 4.247. The proportion of positive significant funds in the
right tail is greater than the proportion of negative significant funds in the left tail and
it increases alongside with the increase in significance level. The proportion of
skilled funds is greater than lucky funds in the right tail but the proportion of unlucky
funds is greater than unskilled funds in the left tail and all results increases alongside
with the increase in significance level. The proportion of false discovery rate in the
left tail is greater than the right tail and it shows that we have more unlucky funds
than unskilled funds. In the right tail and for positive significant funds, however, we
have more skill for the manager than luck.
In Panel B which is time period 1999-2003 we have 2,146 funds with a mean of
9.322 and a median of 7.711. The results demonstrate that the proportion of positive
significant funds in the right tail is greater than the negative significant funds in the
left tail which increases correspondingly with the increase in significance levels. The
proportion of skilled funds is greater than lucky funds in the right tail but the
proportion of unlucky funds is greater than unskilled funds in the left tail which

42
results in an increase alongside with the increase in significance level. The
proportion of false discovery rate is greater in the left tail in comparison to the right
tail and we have a large number which shows more funds among negative significant
funds are unlucky rather than to be unskilled. Results are not so considerable in the
right tail and we have a smaller percentage for false discovery rate which implies
there is a skill for the manager.
In Panel C which is time period 2004-2008, we have 4,976 funds with a mean of
4.681 and a median of 3.834. The results demonstrate that the proportion of positive
significant funds in the right tail is greater than negative significant funds in the left
tail and the results increases alongside with the increase in significance level. The
proportion of skilled funds is greater than lucky funds in the right tail but the
proportion of unlucky funds is greater than unskilled funds in the left tail and results
increases alongside with the increase in significance level. The proportion of false
discovery rate is greater for the left tail and the number are larger which it implies
more unlucky funds rather than unskilled funds in the left tail and more skilled funds
rather than lucky funds in the right tail. Results show evidences of skill for the
manager in the right tail and less unskilled in the left tail.
In Panel D which is time period 2009-2013 we have 3,851 funds with mean 0.632
and median 0.952. The results demonstrate that the proportion of positive significant
funds in the right tail is greater than the proportion of negative significant funds in
the left tail and the results increase correspondingly with the increase in significance
level. The proportion of skilled funds is greater than lucky funds in the right tail and
the proportion of unskilled funds is greater than unlucky funds in the left tail. This
Panel in contrast with previous Panels, shows that the right and left tail follow the
same procedure. The proportion of false discovery rate is greater for the left tail but
the percentage is lower in comparison with previous Panels which imply we have
more unskilled funds rather than unlucky funds. The evidence for the skill of the
manager exists like the previous Panels and it follows the same procedure.
In conclusion, Panels B, A, C and D has the highest proportion of positive
significant funds in the right tail and Panels D, A, C, and B has the highest
proportion of negative significant funds in the left tail respectively. The proportion of
lucky funds in Panels C, D, A and B in the right tail and the proportion of unlucky
funds in Panels C, D, A, and B is the highest in the left tail respectively. The

43
proportion of skilled funds in Panels B, A, C, and D in the right tail and the
proportion of unskilled funds in Panels D, A, C, and B in the left tail has the highest
amount respectively. The proportion of false discovery rate in Panels C, D, A, and B
in the right tail and the proportion of false discovery rate in Panels B, C, A, and D in
the left tail is the highest amount respectively. Hence, the performance of the
manager reveals more skill compared to luck in Panels B, A, D, and C respectively.
In contrast, the performance of the manager has been more unlucky compared to be
unskilled in Panels B, C, A, and D respectively.
4.4. The Impact of Luck on Performance Based on Size
Another aspect of examining luck on the hedge funds’ performance is based on size.
In this part in every one of four time groups we separate funds in three subgroups,
Small, Medium, and Large with respect to asset under management. The study
follows the same procedure which takes into consideration different significance
level of 5, 10, 15, and 20 percent in the right and left tail. Also, we study the size
effect for factors like false discovery rate, significant funds, lucky and unlucky funds,
as well as skilled and unskilled funds. All the results are shown in Table IV. Our
findings demonstrate that large funds include more positive significant alphas. Also
on the basis of false discovery rate they reveal more skill compared to luck. Although
the results of this section is in contrast with studies like Fung and Hsieh (2008) and
some other studies, there are studies such as Ibbotson et al. (2011), that are in
consistent with this study. There can be some reason behind this evidence. Ibbotson
et al. argue that if one eliminates the look back bias for the asset under management
and we calculate the return for equally weighted portfolio the largest funds on
average outperform all the categories. This bias arises when a fund includes data
after demonstrating its performance. It may happen in several ways, like reporting
the return very late when a fund does not have a good performance and vice versa or
dropping out a fund from the database in case it continues to have a low return. It is
widely assumed that the hedge funds with larger asset under management are more
likely to underperform. Because finding investment for bigger size is difficult for the
manager or maybe the transaction cost would increase with size. Although it may be
true there are some other explanations to confront this hypothesis. First, the largest

44
funds may have more access to leverage or seeking to take extra risk. Second,
managers of larger funds may have greater skill which in this study we found that it
is actually true and they have more skill than small funds.
Table IV: The Funds of Hedge Funds in Different Size Groups
In this table we separate each Panel of table III into three subgroups. Based on asset under
management, which gives the size of the funds in each time period, we separate individual funds into
small, medium, and large funds. Similarly, we examine the proportion of positive and negative
significant funds, lucky and unlucky funds, skilled and unskilled funds, and false discovery rate with
different significance level of 5, 10, 15, and 20 percent in right and left tails.
Panel A: Size Effect for Time Period 1994-1998
Year Size Sig. Level (  ) 0.05 0.10 0.15 0.20 0.20 0.15 0.10 0.05 Sig. Level (  )
1994-1998
Small
FDR-
(%) 29.84 46.21 47.14 52.38 18.18 14.60 10.59 6.858 FDR+
(%)
Signif. S-
(%) 4.533 5.866 9.333 11.20 32.27 30.13 25.60 19.73 Signif. S+
(%)
Unlucky F-
(%) 1.353 2.711 4.400 5.867 5.867 4.400 2.711 1.353 Lucky F+
(%)
Unskilled T-
(%) 3.180 3.155 4.933 5.333 26.40 25.73 22.89 18.38 Skilled T+
(%)
Medium
FDR-
(%) 30.06 34.00 58.00 63.23 14.52 12.84 7.144 5.215 FDR+
(%)
Signif. S-
(%) 4.521 6.649 7.713 9.043 39.36 34.84 31.65 26.06 Signif. S+
(%)
Unlucky F-
(%) 1.359 2.261 4.474 5.718 5.718 4.474 2.261 1.359 Lucky F+
(%)
Unskilled T-
(%) 3.162 4.388 3.239 3.325 33.64 30.37 29.39 24.70 Skilled T+
(%)
Large
FDR-
(%) 48.87 66.68 69.44 66.67 10.39 9.100 6.453 4.443 FDR+
(%)
Signif. S-
(%) 2.400 3.200 5.067 6.400 41.07 38.67 33.07 26.40 Signif. S+
(%)
Unlucky F-
(%) 1.173 2.134 3.519 4.267 4.267 3.519 2.134 1.173 Lucky F+
(%)
Unskilled T-
(%) 1.227 1.066 1.548 2.133 36.80 35.15 30.94 25.23 Skilled T+
(%)
Panel B: Size Effect for Time Period 1999-2003
1999-2003
Small
FDR-
(%) 128.66 174.08 201.6 189.8 11.77 10.29 7.796 4.884 FDR+
(%)
Signif. S-
(%) 0.949 1.424 1.898 2.689 43.35 37.18 31.80 25.00 Signif. S+
(%)
Unlucky F-
(%) 1.221 2.479 3.827 5.103 5.103 3.827 2.479 1.221 Lucky F+
(%)
Unskilled T-
(%) -0.27 -1.06 -1.93 -2.41 38.25 33.35 39.32 23.78 Skilled T+
(%)
Medium
FDR-
(%) 140.7 281.4 290.2 225.1 6.321 6.195 3.764 2.297 FDR+
(%)
Signif. S-
(%) 0.632 0.6319 1.106 1.579 56.24 51.82 47.24 38.70 Signif. S+
(%)
Unlucky F-
(%) 0.889 1.778 3.210 3.555 3.555 3.210 1.778 0.889 Lucky F+
(%)
Unskilled T-
(%) -0.26 -1.15 -2.10 -1.98 53.69 48.61 45.46 37.81 Skilled T+
(%)
Large
FDR-
(%) 75.06 123.1 153.5 173.2 6.365 5.202 3.867 2.158 FDR+
(%)
Signif. S-
(%) 1.263 1.579 1.895 2.212 60.19 55.92 50.24 43.92 Signif. S+
(%)
Unlucky F-
(%) 0.948 1.943 2.909 3.831 3.831 2.909 1.943 0.948 Lucky F+
(%)
Unskilled T-
(%) 0.315 -0.36 -1.01 -1.62 56.36 53.01 48.29 42.97 Skilled T+
(%)

45
Panel C: Size Effect for Time Period 2004-2008
2004-2008
Small
FDR-
(%) 77.94 107.5 110.8 126.5 27.57 23.65 19.82 14.32 FDR+
(%)
Signif. S-
(%) 2.439 3.613 5.149 6.143 28.18 24.12 19.60 13.28 Signif. S+
(%)
Unlucky F-
(%) 1.901 3.885 5.704 7.769 7.769 5.704 3.885 1.901 Lucky F+
(%)
Unskilled T-
(%) 0.538 -0.27 -0.56 -1.63 20.41 18.42 15.72 11.38 Skilled T+
(%)
Medium
FDR-
(%) 99.67 113.9 130.0 142.4 24.39 20.63 16.08 12.23 FDR+
(%)
Signif. S-
(%) 1.805 3.159 4.152 5.054 29.51 26.17 22.38 14.71 Signif. S+
(%)
Unlucky F-
(%) 1.799 3.599 5.399 7.198 7.198 5.399 3.599 1.799 Lucky F+
(%)
Unskilled T-
(%) 0.006 -0.44 -1.25 -2.14 22.31 20.77 18.78 12.91 Skilled T+
(%)
Large
FDR-
(%) 151.5 129.6 136.7 112.9 20.23 19.84 13.78 9.111 FDR+
(%)
Signif. S-
(%) 0.994 2.439 4.065 5.601 31.26 28.00 22.94 16.53 Signif. S+
(%)
Unlucky F-
(%) 1.506 3.162 5.556 6.323 6.323 5.556 3.162 1.506 Lucky F+
(%)
Unskilled T-
(%) -0.51 -0.72 -1.49 -0.72 24.94 22.44 19.78 15.02 Skilled T+
(%)
Panel D: Size Effect for Time Period 2009-2013
2009-2013
Small
FDR-
(%) 27.28 33.84 42.62 48.53 32.20 27.42 19.78 13.32 FDR+
(%)
Signif. S-
(%) 6.282 10.13 12.97 15.19 22.89 20.16 17.33 12.87 Signif. S+
(%)
Unlucky F-
(%) 1.714 3.428 5.528 7.371 7.371 5.528 3.428 1.714 Lucky F+
(%)
Unskilled T-
(%) 4.568 6.702 7.442 7.819 15.519 14.632 13.902 11.156 Skilled T+
(%)
Medium
FDR-
(%) 49.83 46.13 43.45 53.52 22.94 16.59 14.15 10.16 FDR+
(%)
Signif. S-
(%) 3.239 6.984 9.818 12.45 29.05 25.71 22.77 15.89 Signif. S+
(%)
Unlucky F-
(%) 1.614 3.222 4.266 6.663 6.663 4.266 3.222 1.614 Lucky F+
(%)
Unskilled T-
(%) 1.625 3.762 5.552 5.787 22.387 21.444 19.548 14.276 Skilled T+
(%)
Large
FDR-
(%) 68.25 65.17 66.72 67.22 16.58 13.79 9.686 5.974 FDR+
(%)
Signif. S-
(%) 2.126 4.453 6.883 9.109 36.94 33.29 29.96 24.29 Signif. S+
(%)
Unlucky F-
(%) 1.451 2.902 4.592 6.123 6.123 4.592 2.902 1.451 Lucky F+
(%)
Unskilled T-
(%) 0.675 1.551 2.291 2.986 30.817 28.698 27.058 22.839 Skilled T+
(%)
In Panel A we study the size effect for the time period 1994-1998. In this Panel
there are 375 small funds with a mean alpha of 4.533 and a median alpha of 4.408,
376 medium funds with a mean alpha of 4.656 and a median alpha of 4.429, and
finally 375 large funds with a mean alpha of 3.689 and a median alpha of 3.716. The
results demonstrate that there are more significant funds in the right tail and for

46
larger funds which increases alongside with the increase in the significance level. But
these results are opposite for the left tail and we have more significant funds for
smaller funds which increases alongside with the increase in the significance level.
The proportion of lucky and unlucky funds has results close to each other but it is a
little bit greater for small funds as compared to large funds. The proportion of skilled
funds in the right tail is greater for large funds and it increases correspondingly with
the increase in the significance level. These results are in reverse in the left tail and
we have greater proportion of unskilled funds for small funds. False discovery rate,
which shows the proportion of lucky and unlucky funds among the significant funds,
is greater for small funds which increase alongside the increase in significance level.
Contrarily, in the left tail the false discovery rate is greater for large funds rather than
small funds which increase correspondingly with the increase in significance level.
In Panel B we have size effect for time period 1999-2003. Small funds include
632 funds with a mean alpha of 9.133 and a median alpha of 7.347, medium funds
include 633 funds with a mean alpha of 10.12 and a median alpha of 8.369, and large
funds include 633 funds with a mean alpha of 8.491 and a median alpha of 7.013.
The proportion of significant funds in the right tail is greater than those of the left
tail. Also, large funds have greater significant funds as compared to small funds
which increase alongside the increase in significance level. We have the same
procedure in the left tail but this difference is not dramatic and it’s negligible. The
proportion of lucky funds is greater for small funds in the right tail but we cannot
conclude clearly how the procedure is in the left tail. We can conclude, however, that
unlucky funds are greater for small funds in the left tail as well, but the results are
not so clear and the differences are not so huge. The proportion of skilled funds is
greater for large funds in the right tail and it increases alongside the increase in
significance level. Contrarily, the proportion of unskilled funds is greater for large
funds but does not follow a smooth way and the difference is not so huge and
dramatic so as to render a clear decision. Therefore, the final calculation for false
discovery shows that this proportion is greater for small funds rather than large funds
which increase alongside the increase in significance level. The results again do not
follow a smooth path for making a clear conclusion about this proportion as it
increases from small funds to medium funds and then decreases from medium funds
to large funds.

47
In Panel C we examine the size effect for the time period 2004-2008. Small funds
in this time period have 1,107 funds with a mean alpha of 4.852 and a median alpha
of 3.294, medium funds have 1,108 funds with a mean alpha of 4.069 and a median
alpha of 3.273, and large funs have 1,107 funds with a mean alpha of 4.204 and a
median alpha of 3.431. The proportion of significant funds does not change
significantly for the different sized groups, however, large funds have a little bit
greater significant funds as compared to small funds in the right tail and it increases
alongside the increase in significance level. These results are in an opposite way in
the left tail and small funds have a little bit greater significant funds as compared to
large funds. The proportion of lucky funds is greater for small funds as compared to
large funds and it increases correspondingly with the increase in significance level.
Similarly the proportion of unlucky funds for small funds is greater than large funds
in the left tail but this difference is not so huge and dramatic. The proportion of
skilled funds for large funds is greater than small funds and they increase alongside
the increase in significance level in the right tail. Contrarily, the proportion of
unskilled funds for small funds is greater than large funds but they do not follow a
clear move alongside the increase in significance level. Finally, the proportion of
lucky funds among the significant funds for small funds is greater than large funds
and they increase correspondingly with the increase in the significance level in the
right tail. False discovery rate follows the opposite way in the left tail and this
proportion for large funds is greater than small funds and they increase alongside the
increase in significance level for small and medium funds but not for large funds.
In Panel D we examine the size effect for time the period 2009-2013. Small funds
in this time period include 987 funds with a mean alpha of 0.805 and a median alpha
of 0.952, medium funds include 988 funds with a mean alpha of 1.439 and a median
alpha of 1.809, and large funs include 988 funds with a mean alpha of 2.523 and a
median alpha of 2.999. The proportion of significant funds is greater for large funds
and they increase alongside the increase in significance level in the right tail.
Contrarily, the proportion of significant funds is greater for small funds as compared
to large funds and they increase alongside the increase in significance level. The
proportion of lucky funds is greater for small funds and they increase alongside the
increase in significance level in the right tail but the difference is not so large or
dramatic. This proportion for unlucky funds is greater for small funds and they

48
increase alongside the increase in the significance level in left tail. The proportion of
skilled funds is greater for large funds and they increase alongside the increase in
significance level in the right tail. Contrarily, the proportion of unskilled funds for
small funds is greater than large funds and they increase alongside the increase in
significance level in left tail. Finally, the false discovery rate for small funds is
greater for small funds in the right tail which increases alongside the increase in
significance level. These results follow the opposite way in the left tail and FDR for
large funds is greater than small funds.
In comparing all Panels we see that the proportion of positive significant funds in
the right tail is greater than the proportion of negative significant funds in the left tail
and they increase alongside the increase in significance level. The proportion of
positive significant funds for large funds is greater than small funds in the right tail
and the proportion of negative significant funds for small funds is greater than large
funds in the left tail. The proportion of lucky funds for small funds is greater than
large funds in the right tail and the proportion of unlucky funds for small funds is
greater than large funds in the left tail. The proportion of skilled funds for large funds
is greater than small funds in the right tail and the proportion of unskilled funds for
small funds is greater than large funds in the left tail. The proportion of false
discovery rate for small funds is greater than large funds in the right tail and this
proportion for large funds is greater than small funds in the left tail. Another
conclusion is that all results increase correspondingly with the increase in
significance level. Therefore, we have higher FDR for small funds as compared to
large funds in the right tail and this implies that there is more luck than skill for the
performance of the manager in this size group. Contrarily, we have higher FDR for
large funds as compared to small funds in the left tail and this implies that the
manager has been more unlucky on his performance rather than to be unskilled in
this size group.

49
5. Conclusion
Hedge funds as a specific type of alternative investment have attracted the
attention of many investors and scholars. The normal signal for investors to find out
an investment vehicle interesting is to outperform its benchmark. The
outperformance of a fund with respect to its benchmark may be due to chance or it
may be due to the skill of active management. In either case, alpha is the way to
measure the outperformance of a fund. In the hedge funds industry usually the past
performance of the managers is almost the only handy and accessible public
information on the fund. The question arises here is whether the past performance
guarantee the future performance? To address this question we considered Fung and
Hsieh (2004) seven-factor model for a time series regression. Alongside this
regression we conduct a multiple hypothesis testing to examine simultaneously the
test statistic of each hedge fund in the population. The second method applied is
False Discovery Rate which was proposed by Storey (2003). Results demonstrate
that the performance of the manager refers to his skill instead of luck and the
manager is more unlucky rather than unskilled for negative performance.
The proportion of skilled funds is greater than lucky funds for the significant
positive funds before and after adjusting for backfill bias. The proportion of false
discovery rate is low which implies past performance of the manager is because of
skill rather than luck and it may guarantee the future performance. The proportion of
unlucky funds is greater than unskilled funds for the negative significant alphas
before and after adjusting for backfill bias. The proportion of false discovery is high
which implies for negative performance the manager is more unlucky rather than to
be unskilled. These results for time periods would follow the same pattern. Applying
the same procedure for different types of strategies follow the same pattern as well.
Sorting results reveal that the strategy of Event Driven, Relative Value, Multi
Strategy, Global Macro, Sector, Long/Short, CTA, Emerging Markets, Market
Neutral, and Short Bias have the highest skill compared to luck and the manager has
been able to show more skill in each of the strategies respectively. The proportion of
false discovery rate for negative significant funds demonstrate that the strategy of
Sector, Long/Short, Global Macro, Market Neutral, Emerging Markets, Multi

50
Strategy, Event Driven, Short Bias, Relative Value, and CTA have the highest FDR
respectively. It implies that the manager in his performance has been more unlucky
rather than unskilled in these strategies respectively. Results for different time
periods follow the same pattern but we cannot have clear comparison in different
groups.
The proportion of false discovery rate for the positive significant funds for small
funds is greater than large funds. It implies that the manager has been more lucky
compared to be skillful in his performance. In other words, small funds compared to
large funds outperform the benchmark because of the luck of the manager and not
the skill. The proportion of false discovery rate for the negative significant funds for
large funds is greater than small funds. It implies that the manager has been more
unlucky compared to be unskilled in his performance. In other words, large funds
compared to small funds underperform because of the unlucky compared to unskilled
of the manager.

51
References:
Agarwal, V., Daniel, N.D., Naik, N.Y., (2011). Do hedge funds misreport their reported returns?
Review of Financial Studies. 24 (10).
Agarwal, V., Daniel N.D., Naik N.Y., (2009). Role of managerial incentives and discretion in hedge
fund performance. The Journal of Finance 64(5), 2221-2256.
Agarwal, V., Fos, V., Jiang, W., (2010). Inferring Reporting-Related Biases in Hedge Fund databases
from Hedge Fund Equity Holdings. Forthcoming in Management Science.
Agarwal, V., Naik, N.Y., (2000). Multi-period performance persistence analysis of hedge funds.
Journal of Financial and Quantitative Analysis 53, 327-342.
Amenc, N., and L. Martellini (2003). The Alpha and Omega of Hedge Fund Performance
Measurement. Edhec.
Aswath. D (2011). Luck versus skill: How can you tell?. Available at:
<http://aswathdamodaran.blogspot.fi/2011/03/luck-versus-skill-how-can-you-tell.html>. cited
12.3.2011.
Avramov, D., R. Kosowski, N. Y. Naik, and M. Teo (2010). Hedge Funds, Manageria Skill, and
Macroeconomi Variables, Journal of Financial Economics 99 (3), 672-692
Avramov, D., L. Barras, and R. Kosowski (2013). Hedge Fund Predictability under the Magnifying
Glass, Journal of Financial and Quantitative Analysis 48, No. 4
Baquero, G., Horst, J., Verbeek, M., (2005). Survival, look-ahead bias, and the persistence in hedge
fund performance. Journal of financial and quantitative analysis, 40 (3), 493-517.
Barras, L., O. Scaillet, and R. Wermers (2009). False Discoveries in Mutual Fund Performance:
Measuring Luck in Estimated Alpha. Journal of Finance, Forthcoming.
Berk, J., Green, R., (2004). Mutual fund flows and performance in rational markets. Journal of
Political Economy 112, 1269-1295.
Berk, J., J. Binsbergen (2013).Measuring Skill in the Mutual Fund Industry, Stanford University and
NBER.
Brown, S., Goetzmann W., Ibbotson, R., (1999). Offshore hedge funds: survival and performance,
1989-95. Journal of Business 72, 91-117.
Chen, L., A. Adams, and R. Taffler (2010). What Style-Timing Skills do Mutual Fund Stars Possess?,
Center for Finance and Investment, School of Management and Languages, Heriot- Watt University.
Cuthberson, K., D. Nitzsche, and N. O’Sullivan (2011). Discovering False Discoveries in UK Mutual
Fund Performance. Cass Business School of London.
Cuthbertson, K., and D. Nitzsche (2010). Performance, Stock Selection and Market Timing of the
German Equity Mutual Fund Industry. Cass Business School of London.
De Figueiredo, R., and R. Evan (2011). Skill, Luck and the Multiproduct Firm: Evidence from Hedge
Funds. Management Science 57(11): 1963-1978.
Demissie, M. (2008). Multiple hypothesis testing in small microarray experiments, Licentiate thesis,
Department of Statistics, Orbero University.

52
Dudoit, S, Shaer, J.P. and J.C. Boldrick (2003). Multiple hypotheses testing in microarray
experiments. Statistical Science, 18, 71-103.
Fama, E., French, K., (1989). Business conditions and expected returns on stocks and bonds. Journal
of Financial Economics 19, 3-29.
Fama, E., French, K., (1993). Common risk factors in the returns on stocks and bonds. Journal
of Financial Economics 33, 3-56.
Fama, E. F. and K. R. French (2010). Luck versus Skill in the Cross-Section of Mutual Fund Returns.
Journal of Finance 65: 1915–1947.
Fung, W., Hsieh, D., (1997). Is mean variance analysis applicable to hedge funds? Economic Letters
62, 53-58.
Fung, W., Hsieh, D., (1999). A primer on hedge funds. Journal of Empirical Finance 6, 309-331.
Fung, W., Hsieh, D., (2000). Performance characteristics of hedge funds and CTA funds: natural
versus spurious biases. Journal of Financial and Quantitative Analysis 35, 291-307.
Fung, W., Hsieh, D., (2001). The risk in hedge fund strategies: theory and evidence from trend
followers. Review of Financial Studies 14, 313-341.
Fung, W., Hsieh, D., (2004). Hedge fund benchmarks: a risk based approach. Financial Analysts
Journal 60, 65-80.
Fung, W., Hsieh, D., (2009). Measurement biases in hedge fund performance data: an update.
Financial Analysts Journal 65, 36-38.
Fung, W., Hsieh, D., Naik, N., Ramadorai, T., (2008). Hedge funds: performance, risk, and capital
formation. Journal of Finance 63, 1777-1803.
Guimond, J. (2006). Do Mutual Fund Managers Have Superior Skills? An Analysis of the Portfolio
Deviations from a Benchmark. Working Paper.
Ibbotson, R., p. Chen, and K. Zhu (2011). The ABCs of Hedge Funds: Alphas, Betas, and Costs,
Financial Analyst Journal 67, 15-25.
Jame, R. (2012). How Skilled are Hedge Funds? Evidence from Their Daily Trades. School of
Banking and Finance, University of New South Wales, Sydney.
Joenväärä J., R. Kosowski, and P. Tolonen (2012). Revisiting Stylized Facts about Hedge Funds
Insights from a Novel Aggregation of the Main Hedge Fund Databases. University of Oulu, Oulu.
Joenväärä J., R. Kosowski, and P. Tolonen (2013). Hedge Fund Performance: What Do We Know.
Forthcoming.
Joenväärä, J. (2010). Predicting Hedge Fund Performance Using Optimal Portfolio Weights. Master
Thesis, Oulu Business School, University of Oulu, Oulu.
Kosowski, R., A. Timmermann, R. Wermers, and H. White (2006). Can Mutual Fund “Stars” Really
Pick Stocks? New Evidence from a Bootstrap Analysis. Journal of Finance 61: 2551–2595.
Kosowski, R., N.Y. Naik, M. Teo (2007). Do Hedge funds Deliver Alpha? A Bayesian and Bootstrap
Analysis, Journal of Financial Economics, 84(1), 229-264.
Lucean, (2013). Alpha: A Measure of Fund Management Skill. Available at:
<https://lucenaresearch.com/alpha-a-measure-of-fund-management-skill/>.cited 22.4.2013.

53
Man I. L. (2012). An Introduction to Hedge Fund. Financial Services Authority. available at:
<https://www.man.com/FI/what-is-a-hedge-fund>. Cited 3.12.2013
Patton, A.J. and Ramadorai, T., (2011). On the Dynamics of Hedge Fund Risk Exposures.
Forthcoming in Journal of Finance.
Patton, A. J., Ramadorai, T., Streatfield M., (2011). The reliability of voluntary disclosures: Evidence
from hedge funds. Duke University and University of Oxford.
Saxon, B (2013). Hedge funds Consistency Index, Alpha. Barclay Trading Group. Available at:
<http://www.hedgefund-index.com/d_alpha.asp#TOP>. Cited 15.12.2013.
Sen, S. (2006). Multiple comparisons and the false discovery rate. University of California San
Fransisco.
Storey, J. D. (2002). A direct approach to false discovery rates. Journal of the Royal Statistical Society
64, 479–498.
Storey, J. D. (2003). The positive false discovery rate: A Bayesian interpretation and the q-value.
Annals of Statistics 31, 2013–2035.
Storey, J. D., E. T. Jonathan, and S. David (2004). Strong control, conservative point estimation and
simultaneous conservative consistency of false discovery rates: A unified Approach. Journal of the
Royal Statistical Society 66, 187–205.

Mahdi-Thesis

More Related Content

What's hot (20)

Viewers also liked (16)

Similar to Mahdi-Thesis (20)

Mahdi-Thesis