Hedge Funds Investing:

A Quantitative Look Inside the Black Box

François-Serge Lhabitant*
August 2001

Abstract: There is an increasing amount of evidence that shows the benefits of considering
hedge funds as an asset class at the strategic asset allocation level. The investors’ greatest
challenge remains the identification of desirable investment vehicles, since very little formal
quantitative analysis of hedge funds has been done in the past. In this paper, we suggest an
innovative approach to hedge fund investing, which is valid at the individual fund level as well
as at the aggregate portfolio level (e.g. portfolio of hedge funds). This approach only relies on
hedge funds historical returns. We provide several illustrations, including static and dynamic
style analysis, benchmark construction, performance assessment, and value at risk calculations.

1 Introduction

The consequences of globalization have dramatically eroded traditional asset allocation

dogmas. International assets move more and more in lockstep fashion and fail to fulfill their
diversification promises, particularly in turbulent times. Consequently, investors have
broadened their investment horizons and started seeking alternative asset classes, among them,
hedge funds.

Investors commonly view hedge funds as the phenomenon of the 1990s. From a few hundred
players ten years ago, the hedge fund industry has expanded to several thousand today with
double-digit annual growth. With capital under management under $100 million, most of the
newcomers are small, but a few dozen funds had a capital base larger than $1 billion at the end
of 2000.

Hedge funds have traditionally managed capital for high net worth individuals, family offices,
and endowments. More recently, a broad range of institutional investors (including pension
funds in the Netherlands, Switzerland, Scandinavia as well as US funds) has also expressed an
increased interest in hedge funds. It is likely that this interest will continue in the near future,
increasing the role of hedge funds within traditional asset-allocation models.

*
François-Serge Lhabitant, Ph.D., is the Head of Quantitative Risk Management at Union Bancaire Privée
(Geneva), Professor of Risk Management at H.E.C. University of Lausanne and an Assistant Professor of
Finance at Thunderbird, the American Graduate School of International Management (Glendale). The
opinions expressed in this paper reflect solely the views of the author and do not engage any institution he
may belong to. Contact: [email protected] or Russel 11B CH-1205 St-Sulpice (Switzerland). Tel/fax:
+41 79 4381753.
Surprisingly, there is still no universally accepted definition of what constitutes a hedge fund.
Functionally, hedge funds and proprietary trading desks pursue similar goals: hiring
professional investment managers, rewarding them by performance-linked fee and
implementing a large diversity of strategies often involving leverage, derivatives, hedging and
short positions to exploit market inefficiencies. Organizationally, however, there are substantial
differences: hedge funds are typically private pooled investment vehicles with high minimum
investments and infrequent redemption opportunities.

Another key characteristic of hedge funds is that they dislike disclosing information about their
investment process and/or market bets. The reasons are twofold. First, hedge funds are often
registered as offshore companies or limited partnerships to avoid the restrictive registration and
investment requirements of mutual funds. Consequently, publishing performance figures,
portfolio compositions, or other informative statistics could be considered by regulators as
public solicitation and therefore illegal. Second, hedge funds are often taking concentrated and
leveraged positions in illiquid markets or securities. Disclosing them publicly could wipe out
future expected gains or harden the ability of unwinding these positions when necessary.

However, given the sector’s growing significance and the increasing turbulence of markets,
this situation is hardly sustainable. The collapse of Long Term Capital Management LP in 1998
raised serious concerns for credit providers, trading counterparties and market regulators.
Investors, chastened by the 1997 Asian crisis, the 1998 Russian crisis, the 1999 Brazil crisis
and the year 2000 Internet bubble burst now want to closely monitor their assets. Last, but not
least, banks, investment consultants and funds of hedge fund managers require having a clear
view of hedge funds’ risk and return profiles before including them on any recommended list.

Given the increase in investors’ interest, there is an urgent need of quantitative methods of
analyzing hedge funds and provide meaningful information while respecting their privacy. Most
hedge funds are now providing their net asset value on a regular and voluntary basis (e.g.
monthly) to commercial data vendors such as TASS, Altvest, Hedge Fund Research, Managed
Account Reports, etc. Although subject to a survivorship bias, these net asset values are
probably the most reliable source of data publicly available today on the hedge fund industry.
They incorporate what hedge funds actually do, not just what they say they do.

Our goal in this paper is therefore to propose an alternative approach to hedge funds analysis,
which can be summarized as “let the data speak”. It relies essentially on an extension of
Sharpe’s (1992) return-based style analysis technique for mutual funds. This extension takes
into account the particularities of hedge funds, such as the ability to go short and to use
leverage. It is easy to implement and only requires a small amount of data. Last, but not least,
its results are easy to understand, even by non-quantitative investors.

The structure of this paper is the following. Section 2 discusses the quantitative aspects of our
model and its assumptions. Section 3 illustrates several potential applications such as static and
dynamic style analysis, benchmark construction and performance assessment. Section 4
proposes an interesting and straightforward extension to compute the value at risk of a hedge
fund. Section 5 summarizes our findings and opens the door to further developments.

2
2 The hedge fund analysis model

As mentioned already, our analysis of hedge funds uses a time series of net asset values as a
raw input. Let us denote by NAVt the net asset value of a hedge fund at time t. From the
fund’s net asset values, returns are derived as follows:

Rt = (NAVt – NAVt-1) / NAVt-1 (1)

Next, we postulate a multi-factor model to explain the time-series of returns. If we denote Fk,t
as the value of factor number k at time t and we use N distinct factors, our multi-factor model
can be expressed as:

N
R t = α + ∑ β k Fkt + ε t (2)
k =1

We can interpret the intercept term of the model as the portion of return unexplained by the
factor model.

Multi-factor models are useful tools to describe the return structure of investment funds. The
major difficulty in implementing them is the identification of common and relevant factors.
Two approaches compete for this task. The first one consists in deriving endogenously a set of
implicit factors using principal components analysis. This ensures an optimal fitting of the
model, but the difficulty then lies in the economic interpretation of the implied factors. The
second approach consists in specifying exogenously a set of factors that we believe or know to
be relevant. The economic interpretation is then straightforward, which makes it the preferred
method for investment analysis.

Which factors should be considered? The answer is relatively straightforward for traditional
money managers, given their limited investment flexibility. For instance, Sharpe (1992) used
fifteen global stock and bond indices to describe the cross sectional returns of U.S. equity
funds, and Elton, Gruber, and Blake (1995) preferred a series of fundamental economic
variables for U.S. bond funds. However, the situation is different for alternative money
managers. First, they have a larger set of potential investments available, may use leverage or
short selling, and can invest in illiquid assets since they do not offer daily redemption. Second,
they attempt to provide investors exposure to unique and uncorrelated sources of return, i.e.
returns driven by factors not easily explained by the factors common to traditional investments.
Third, the benchmarks they use are of little help for this task, because they are hurdle rates
(such as T-Bill rates) for absolute performers rather than benchmarks for trackers.

This explains why academic research on the sources of hedge fund returns is much less
developed than for traditional funds. As an illustration of the recent approaches proposed and
without attempting to be comprehensive, let us mention the following papers. On the single
factor side, Irwin et al. (1994) used a simple managed futures benchmark and McCarthy et al.
(1997) suggested a single index with Bayesian adjustment for leverage. On the multiple factor
side, Mitev (1995) suggested using five implied factors, identified as trend following strategies,
surprise or stop-loss control models, agricultural markets, spread-strategies (primarily interest
rate) and fundamental economic factors or global markets. Fund and Hsieh (1996)
demonstrated the role of dynamic trading and investment approaches (distressed, global/macro,

3
systems, systems/opportunistic, and value). Liang (2000) relied on eight asset classes (S&P
500, MSCI world and MSCI emerging for equity, Salomon Brothers world government bond
and Salomon Brothers world corporate bond indices for bonds, Federal Reserve Bank trade-
weighted dollar index for currency, gold price for commodities, and one-month Eurodollar
deposit for cash): Finally, Schneeweis and Spurgin (1998) combined a very large number of
hedge fund indices, CTA indices, and traditional asset class indices.

All these models rely on regression-like analysis. Our approach is somehow different and
motivated by the following considerations: a) Unlike equity or bond mutual funds, the lack of a
single factor that describes the return process of hedge funds highlights the emphasis that
should be placed upon investment style rather than on the general return process. b) Static buy
and hold portfolios, such as equity and bond indices, fail to capture the dynamic asset
allocation and derivatives usage of hedge fund managers. We therefore start from an asset
allocation perspective, namely, Sharpe’s (1992) return-based style analysis model, which is
equivalent to equation (2) plus some constraints.

Table 1: Definition of hedge fund investment styles according to CSFB/Tremont

Category Strategy

Convertible Invest in the convertible securities of a company. A typical

arbitrage investment is to be long in convertible bond and short in stock of the
same company.
Dedicated short- Maintain consistent net short (or pure short) exposures to the
bias underlying market.
Emerging markets Equity or fixed income investing in emerging markets around the
world
Market neutral Exploit equity market inefficiencies by being simultaneously long and
short matched equity portfolios of the same size within a country.
Event-driven Equity-oriented investing designed to capture price movement
generated by an anticipated corporate event (merger, acquisition,
distressed securities, etc.)
Fixed-income Profit from price anomalies between related interest rate securities.
arbitrage
Global macro Leveraged views on overall market direction as influenced by major
economic trends and/or events.
Long/short equity Equity-oriented investing on both the long and short sides of the
market, with an objective different from being market neutral.
Managed futures Systematic or discretionary trading in listed financial and commodity
futures markets and currency markets around the world
Source: CSFB/Tremont

4
Sharpe’s model uses as factors returns on representative asset classes. These asset classes must
be 1) mutually exclusive; 2) exhaustive with respect to the investment universe; and 3) have
returns that differ. Since traditional asset classes are not really representative of hedge funds
behaviour, we are using hedge fund indices. Several broad-based hedge fund indices are now
publicly available to benchmark the performance of the industry. However, given the
substantial diversity in hedge funds strategies, we rather considered hedge fund style indices,
which group funds based on their managers’ self-described strategies. In the following analysis,
we are using the nine CSFB/Tremont style indices family (see Table 1), mostly because of their
quality, transparency, and adequate weighting scheme1. However, we could repeat our analysis
using any other family of hedge fund indices.

Thus, our model becomes:

9
R t = α + ∑ β i ⋅ I i,t + ε t (3)
i =1

where
I1 = return on the CSFB Tremont Convertible Arbitrage index
I2 = return on the CSFB Tremont Dedicated Short Bias index
I3 = return on the CSFB Tremont Event Driven index
I4 = return on the CSFB Tremont Global Macro index
I5 = return on the CSFB Tremont Long Short Equity index
I6 = return on the CSFB Tremont Emerging Markets index
I7 = return on the CSFB Tremont Fixed Income Arbitrage index
I8 = return on the CSFB Tremont Market Neutral index
I9 = return on the CSFB Tremont Managed Futures index

Sharpe’s model also imposes two restrictions on the beta coefficients in order to interpret them
as asset allocation weights: all betas must be positive (i.e. no short positions) and their sum
should equal one (i.e. the portfolio is fully invested). In our case, we kept the first constraint 2,
but relaxed the second one to account for the relative leverage possibility. Therefore, our
model can be seen as a constrained regression and needs to be solved by quadratic
programming3.

1
CSFB/Tremont hedge fund indices publicly disclose their calculation methodology and the index
constituents. The selection rules are simple: at least $10 million of assets under management and audited
statements. Surprisingly, less than 300 hedge funds out of the 3000 in the TASS database fulfill them.
CSFB/Tremont hedge fund indices are also the only asset-weighted indices of the hedge fund industry.
This very desirable property corresponds to a momentum strategy, i.e. winners (losers) naturally increase
(decrease) their relative weight in the index. In contrast, all other hedge fund indices are equally weighted,
which corresponds to a contrarian allocation, i.e. selling winners and buying losers at regular time
intervals.
2
Being short an investment style is rather hard to justify economically. For instance, what would mean
being short the “Dedicated Short Bias” style?
3
This raises some technical difficulties if one needs to obtain the asymptotic distribution of the beta
coefficients, for instance to build confidence intervals.

5
3 Applications

We now suggest several applications for our model at both the individual hedge fund and the
overall hedge fund portfolio level.

Clearly, our model provides useful insights to understand the investment strategy of a
particular hedge fund. If one considers the CSFB/Tremont indices as purely synthetic4 hedge
fund portfolios whose objective is to deliver a particular set of risk/return characteristics, the
betas of equation (3) can be interpreted as exposures to each of these investment styles. As an
illustration, Figure 1 represents graphically the nine betas that we obtained for two existing
hedge funds. The shapes of these graphs (that we call “hedge funds radars”) allow for an
immediate style classification. The first fund is a pure convertible arbitrage hedge fund
(β1=0.41, all other betas below 0.04). The second fund is a fund of hedge funds that appears to
be diversified across styles. Its highest exposures are in the convertible arbitrage (β1=0.22),
market neutral (β8=0.22) and global macro styles (β4=0.19), and moderate exposures in long-
short equity (β5=0.15), emerging markets (β6=0.13) and event-driven (β3=0.07, all other betas
below 0.01).

Thus, using hedge fund radars, it suddenly becomes easy to monitor what managers actually do
regardless of what they claim to be doing. It also helps in identifying hedge funds with a “pure”
investment style and hedge funds with a diversified style approach. This is specifically useful
for funds of hedge funds and multi-manager portfolios, which offer investors an efficient way
of diversifying risks across a number of managers and strategies. Are they effectively
diversified across styles? A single fund radar will help answer this question. Moreover, for
those who prefer a quantitative indicator, a useful indicator of style concentration is the
following Herfindahl-Hirschman-like concentration index, computed by adding the squared
(adjusted) style exposures:

2
 
 
9 
β 
HHI = ∑  9 i 
i =1  
(4)
∑ βj 

 j =1 

Adjustment is necessary here, because the sum of the exposures can be significantly different
between two funds, since they do not sum to 100%. As an illustration, if we compute the
concentration indices for the two hedge funds of our previous example, we obtain 0.66 for the
convertible arbitrage hedge fund, and 0.18 for the fund of funds.

4
Although CSFB/Tremont indices consider a relatively small number of hedge funds with respect to
concurrent indices, they are not easily investable. First, the high minimum investment in each fund
suggests that a substantial amount of capital would be required to track the index. Second, the liquidity and
redemption policies of individual hedge funds would not allow for an unambiguous rebalancing scheme,
leading to an increase of the tracking error.

6
 Figure 1: Hedge fund style radars

The figure shows the hedge fund radars obtained for a convertible arbitrage fund (top) and a
fund of hedge fund (bottom). The sensitivities (i.e. style-beta coefficients) are estimated using
three years of historical data.

Global Macro
0.45
0.4
Convertible Arbitrage 0.35 Fixed Income Arbitrage
0.3
0.25
0.2
0.15
0.1
Long Short Equity Market Neutral
0.05
0

Dedicated Short Bias Managed Futures

Emerging Event Driven

Dedicated Short Bias

0.25

Convertible Arbitrage 0.2 Fixed Income Arbitrage

0.15

0.1

Market Neutral 0.05 Managed Futures

Global Macro Event Driven

Long Short Equity Emerging

7
Hedge fund radars also open new doors to benchmarking hedge fund performance. The asset
allocation implied by the radar provides a perfect natural benchmark against which the fund
returns should be compared. Of course, it does not necessarily correspond to the tactical asset
allocation of the fund. However, over the considered period, what one can affirm is that the
fund behaved “as if” it effectively had this asset allocation. And the quality of this
approximation is easily verified using the r-square indicator obtained when estimating equation
(3).

Var(ε)
R 2 = 1− (5)
Var (R )

However, a fund radar only provides a static picture of a hedge fund’s behavior. Our model
also allows for a dynamic assessment of a hedge fund’s behavior. By using a rolling
observation period, it is possible to build up a series of radars that represent the evolution of
the fund’s exposures over time. These can also easily be represented graphically, allowing for
verifying the stability of the fund’s behavior.

As an illustration, Figure 2 shows the dynamic behavior of the two funds considered
previously. For our first fund, the exposure to convertible arbitrage appears only since the first
quarter 1999. Before this, the fund was clearly exposed to the long-short equity, event driven
and market neutral styles. This evidences that the track record of this fund will not be the one
of a pure convertible arbitrage fund, although the fund is now almost only exposed to this
style. By contrast, the fund of hedge funds appears diversified since it was created.

4 Risk management and value at risk

Another key application of our model is in the domain of risk management. Although the
quantitative measurement of risk is essential, given the wide investment latitude of hedge fund
managers, there is no real consensus on risk measures for hedge funds. Some funds simply
claim to focus on absolute returns and do not even consider producing risk figures. Others
provide a plethora of risk indicators, including volatility around the mean (although their return
distributions are non-symmetric), drawdowns (related to the length and depth of a decline in
portfolio value), downside risk measures (related to the statistical probability of not achieving a
certain return), and other inadequate information5.

Most hedge funds are still reluctant to provide value at risk (VaR) figures -- that is, their
expected maximum loss during a specified period at a given level of probability. This is
surprising, for at least two reasons. First, the world’s leading financial institutions and
proprietary trading desks have now adopted VaR, with the strong support of rating agencies
and market regulators. Second, VaR would be an invaluable tool for monitoring risks at both
overall portfolio and individual hedge fund level, allowing fund managers to communicate
intuitive and easy to understand risk measures without going into the intricate details of
complicated or proprietary trading strategies.

5
See Lhabitant (2000) for an illustration.

8
 Figure 2: A dynamic view on hedge fund style

The figure shows the historical evolution of the sensitivities (i.e. style-beta coefficients) for a
convertible arbitrage fund (top) and a fund of hedge fund (bottom).

0.8

0.7
Managed Futures

0.6 Market Neutral

Fixed Income Arbitrage

0.5

Emerging
0.4
Long Short Equity

0.3
Global Macro

0.2 Event Driven

Dedicated Short Bias

0.1

Convertible Arbitrage
0
98/01 99/01 00/01

1.8
Managed Futures
1.6
Market Neutral
1.4
Fixed Income Arbitrage
1.2
Emerging
1
Long Short Equity
0.8
Global Macro
0.6
Event Driven
0.4
Dedicated Short Bias
0.2
Convertible Arbitrage
0
98/01 99/01 00/01

9
Our model helps fill in this gap. From the style exposure of a hedge fund, one can easily apply
a two-step procedure to obtain its VaR. Let us assume for instance that we want a confidence
level of 99 percent for the one-month VaR6. The first step consists of “pushing” each
individual risk factor in the most disadvantageous direction (at the one percentile value of its
historical distribution) and estimating the overall impact on the fund, accounting for the risk
factors correlation. This gives us the VaR due to market moves (i.e. style indices moves), that
we call “value at market risk”. Mathematically,

9 9
VaMR = ∑ ∑ρ
i =1 j=1
i, j ⋅ β i ⋅ Fi* ⋅ β j ⋅ Fj* (6)

where ρi,j is the correlation between monthly returns of hedge fund indices i and j, and Fi* is the
one percentile extreme return of style index i returns over one month.

The second step estimates specific risk. We simply define specific risk ( σ2ε ) as the difference
between total risk (the observed fund variance σ 2P ) and systematic risk (the variance due to the
market, i.e. the hedge fund style indices). We have:

9 9
σ ε2 = σ 2P − ∑∑ ρ i , j ⋅ β i ⋅ σ i ⋅ β j ⋅ σ j (7)
i =1 j =1

By construction, the error terms εt of equation (3) are non-correlated and normally distributed
with zero mean. Since we want a 99 percent confidence level, we can apply a factor push of
2.33 times σε (corresponding to a 99% confidence level for a normal variable) to obtain the
specific risk of a hedge fund. Mathematically, the value at specific risk (VaSR) is given by

VaSR = 2.33 × σ ε (6)

To obtain the total VaR figure, simply add up market and specific risk figures, accounting for
their zero correlation:

VaR = (Value at Market Risk ) 2 + (Value at Specific Risk ) 2 (7)

Note that the procedure can easily be adapted to implement stress-scenario analysis, such as
breakdown in correlations. If we apply this model to the two hedge funds that we considered
previously, we obtain the following results. The convertible arbitrage fund has a VaR equal to
7.73% of its net asset value, which we can split into a VaMR of 2.11% and a VaSR of 7.42%.
The fund of hedge funds has a VaR equal to 7.88% of its net asset value, which we can split
into a VaMR of 5.83% and a VaSR of 5.31%. As expected, the proportion of specific risk is
larger in the pure-style hedge fund than in the diversified one.

6
The choice of the holding period depends upon the time needed to liquidate a portfolio. It typically varies
from one day for a trading portfolio to thirty days for a fund with a monthly redemption policy. The choice
of the confidence interval is also relatively subjective. Banks typically use 95% to 99% as confidence
intervals for measuring and reporting VaR.

10
How trustworthy is our approach? We back-tested this VaR model on a sample of 2934 hedge
funds from January 1994 to October 2000. Using a three-year observation period, we
computed funds’ exposures, the market parameters, and the funds’ VaRs (one month, 99%
confidence level). Then, we compare the VaR of each fund with the profit and loss of the
following month. If the loss is greater than the VaR figure, this is recorded as an exception. On
a total of 96,549 three-year observation periods, 1’026 exceptions were observed. This gives
an exception rate slightly higher than expected (1.06% versus 1%). However, things have to be
taken with precaution, since 614 of these exceptions occurred in August 1998 immediately
after the LTCM crisis, which can be qualified as an abnormal market situation. If we exclude
this month from our sample, the exception rate falls to 0.43%, a rather conservative figure7.

7
See Lhabitant (2001) for detailed back-testing results.

11
Table 2: Correlation, extreme moves and volatility figures for Tremont/CSFB hedge fund indices

Convertible Dedicated Event Global Long Emerging Fixed Market Managed

Arbitrage Short Bias Driven Macro Short Markets Income Neutral Futures
Equity Arbitrage
Convertible Arbitrage 1.00 -0.30 0.60 0.36 0.21 0.45 0.75 0.31 -0.61
Dedicated Short Bias -0.30 1.00 -0.73 -0.14 -0.77 -0.72 -0.06 -0.56 0.30
Event Driven 0.60 -0.73 1.00 0.40 0.66 0.80 0.41 0.52 -0.53
Global Macro 0.36 -0.14 0.40 1.00 0.52 0.46 0.58 -0.05 -0.04
Long Short Equity 0.21 -0.77 0.66 0.52 1.00 0.75 0.23 0.30 -0.15
Emerging 0.45 -0.72 0.80 0.46 0.75 1.00 0.35 0.46 -0.38
Fixed Income arbitrage 0.75 -0.06 0.41 0.58 0.23 0.35 1.00 -0.01 -0.32
Market neutral 0.31 -0.56 0.52 -0.05 0.30 0.46 -0.01 1.00 -0.17
Managed Futures -0.61 0.30 -0.53 -0.04 -0.15 -0.38 -0.32 -0.17 1.00

Index Volatility 1.83 6.41 2.54 4.20 4.76 6.67 1.66 0.76 2.89
Index extreme move -4.67 -8.68 -8.69 -9.38 -10.04 -18.39 -5.83 -0.69 -4.54

This table shows the correlation, volatility and extreme move figures for the hedge funds style indices at the end of September 2000. All values are
computed from a historical sample of 36 months. Volatility and Extreme moves are expressed on a monthly basis.
5 Conclusions

For a long time, due to the lack of transparency, hedge fund investing was more an art than a
science, mostly based on qualitative analysis and due diligence. The collapse of numerous large
players (LTCM, Granite funds, etc.) and the increasing turbulence of financial markets has
evidenced that qualitative approaches, although critical and necessary, were not sufficient.
Given the complexities involved with hedge fund strategies, there is a need to introduce new
quantitative tools.

Our model attempts to fill in this gap. Relying only on net asset values, it determines the
exposures of a hedge fund to several investment styles. This makes it a useful tool to assist
investors and fund managers in monitoring their portfolios, determining whether the benefits
are in line with expectations and assumptions made in asset allocation, verifying the
diversification benefits or highlighting the risks concentrated with any hedge fund manager or
investment strategy. A simple extension will transform our model into a powerful risk
management tool able to compute the value at risk of a hedge fund as well as the marginal risk
contributions of each investment style an investor is exposed to.

Of course, some may argue that historical data may not be representative of the future. Our
answer will be very pragmatic. First, we successfully back-tested our model on a large sample
of hedge funds, from January 1994 to October 2000. Second, historical data will always be
useful to assess past performance and to understand what were favorable and unfavorable
outcomes. Third, over the years, our discussions with several hedge fund managers led us to
believe that the vast majority of their funds may have unpredictable performance (because their
returns depend on too many variables and the number of past observations is too small to make
any inference), but have some form of partially predictable behavior (e.g. investment
philosophy, stop-loss, etc.). This is why past information, knowledgeably and intelligently
interpreted, will always be better than no information.
6 References

• Ackerman C., R. McNally and D. Ravenscraft (1999), The Performance of Hedge

Funds: Risk, Return, and Incentive, Journal of Finance, 54, 833-874.
• Chen S.J. and B. Jordan (1993), Some empirical tests in the arbitrage pricing theory:
macro variables versus derived factors, Journal of Banking and Finance, 17, 65-89.
• Edwards F. and A.C. Ma (1988), Commodity fund performance: is the information
contained in fund prospectuses useful?, Journal of Futures Markets, 8 (5), 589-616.
• Elton E.J., M.J. Gruber and C. Blake (1995), Fundamental economic variables, expected
returns, and bond fund performance, Journal of Finance, 50 (4), 1229-1256.
• Fung W. and D.A. Hsieh (1996), Performance attribution and style analysis: from mutual
funds to hedge funds, Working Paper # 9609, Duke University.
• Fung W. and D.A. Hsieh (1997a), Empirical characteristics of dynamic trading
strategies: the case of hedge funds, The Review of Financial Studies, 10 (2), 275-302.
• Fung W. and D.A. Hsieh. (1997b), Investment style and survivorship bias in the returns
of CTAs, The Journal of Portfolio Management, 24 (1), Fall, 30-41
• Jorion Ph. (1997), Value at Risk: the new benchmark for controlling market risk,
University of California, Irvine: McGraw-Hill
• Lhabitant F.S. (2000), Derivatives in portfolio management: why beating the market is
easy, Derivatives Quartely, vol. 7 (2), pp. 37-46, 2000
• Lhabitant F.S. (2001), Assessing market risk for hedge funds and hedge funds
portfolios, FAME working paper
• Liang B. (2000), Hedge funds: the living and the dead, working paper, Case Western
Reserve University
• McCarthy D., Th. Schneeweis and R. Spurgin (1997), Informational content in
historical CTA performance, Journal of Futures Markets, May, 317-340
• Mitev T. (1995), Classification of commodity trading advisors using maximum likelihood
factor analysis, working paper, University of Massachusetts.
• Schneeweis Th. and R. Spurgin (1998), Multifactor analysis of hedge funds, managed
futures and mutual fund return and risk characteristics, Journal of Alternative
Investments, 1, 1-24.
• Sharpe W. (1992), Asset allocation: management style and performance measurement,
Journal of Portfolio Management, 18 (2), 7-19.

14