Incentives and risk taking in hedge funds

Roy Kouwenberg
a
, William T. Ziemba
b,
*
a
Mahidol University, Bangkok, Thailand, and Erasmus University Rotterdam, The Netherlands
b
Sauder School of Business, UBC, Vancouver, BC, Canada V6T 1Z2
Available online 7 April 2007
Abstract
We study how incentive fees and managers own investment in the fund aect the investment
strategy of hedge fund managers. We nd that loss averse managers increase the risk of the funds
investment strategy with higher incentive fees. However, risk taking is greatly reduced if a substantial
amount of the managers own money (at least 30%) is in the fund. Using the Zurich hedge fund uni-
verse, we test the relation between risk taking and incentive fees empirically. Hedge funds with incen-
tive fees have signicantly lower mean returns (net of fees), while downside risk is positively related
to the incentive fee level. Fund of funds charging large incentive fees achieve relatively high mean
returns, but with signicantly higher risk as well.
2007 Elsevier B.V. All rights reserved.
JEL classication: G10; G29
Keywords: Hedge funds; Incentive fees; Optimal portfolio choice
1. Introduction
We present a theoretical analysis of risk and incentives in the behavioural framework of
prospect theory, followed by an empirical study of risk taking by a large number of hedge
funds and funds of hedge funds. Hedge fund managers typically receive a xed manage-
ment fee of about 1% of the assets under management. Most managers also receive
20% of the funds return in excess of a given benchmark. This incentive fee, gives the
0378-4266/$ - see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.jbankn.2007.04.003
*
Corresponding author.
E-mail addresses: roy.k@cmmu.net (R. Kouwenberg), ziemba@interchange.ubc.ca (W.T. Ziemba).
Available online at www.sciencedirect.com
Journal of Banking & Finance 31 (2007) 32913310
www.elsevier.com/locate/jbf
manager a strong incentive to deliver a return above the benchmark. As incentive
fee arrangements are widespread in the hedge fund industry, it is important to under-
stand the impact of incentive fees on the behaviour and investment strategy of fund
managers.
We use the value function of prospect theory, developed by Kahneman and Tversky
(1979), to describe the preferences of the fund manager. Investors evaluate gains and losses
relative to a given reference point and are loss averse, i.e. the pain of a loss is felt more
strongly than the positive feeling associated with an equivalent gain. Investors are risk
averse when dealing with gains, but they are risk seeking when dealing with losses. This
type of preferences is well suited to model fund manager behaviour under incentive fees.
First, the funds benchmark is a natural candidate for the reference point that separates
gains from losses in the mind of the manager. When the funds return exceeds this level
the manager receives the incentive fee, which can amount to a large sum of money. Once
the fee is in-the-money, the manager becomes risk averse to protect his gains.
Returns below the benchmark, on the other hand, will not be appreciated by the funds
investors and can lead to outows and loss of reputation for the manager. Additionally,
returns below the benchmark lead to regret over the lost income from incentive fees,
and therefore the pain of a loss is felt relatively strong. Risk seeking behaviour in the
face of losses means that the manager tries to maximize the probability of beating the
benchmark to avoid the negative consequences of a loss, while putting less emphasis
on a potential further decrease of fund value.
Related to our work is a paper by Carpenter (2000) that analyses the eect of incentive
fees on the optimal investment strategy of a risk averse fund manager with preferences
described by the class of utility functions with hyperbolic absolute risk aversion (HARA).
Carpenter (2000) nds that a manager with an incentive fee increases the risk of the funds
investment strategy if the funds return is below the hurdle rate. If the funds return is
above the benchmark the manager reduces volatility, in some cases even below the optimal
volatility level of a fund without incentive fees. Carpenter (2000) also proves that the man-
agers optimal response to an increase of the incentive fee level (e.g. a change from 10% to
20%) is to reduce the funds risk. One of the contributions of this paper is to demonstrate
that these conclusions do not hold for loss averse fund managers that maximize the value
function of prospect theory. We show that loss averse fund managers increase risk in
response to incentive fees, regardless whether the funds return is above or below the hur-
dle rate. Moreover, the fund managers reaction to an increase of the incentive fee level is
to invest a higher proportion of the fund in risky assets.
Apart from the fund managers preferences, we extend the model of Carpenter (2000)
along two other lines: we incorporate management fees and investments of the manager
in the fund. Most fund managers charge a xed proportion of the fund value as manage-
ment fee, to cover expenses and to make a living. One would expect that management fees
moderate risk taking, as negative investment returns will reduce the future stream of
income from management fees. A second important aspect of the hedge fund industry that
aects risk taking, is that most fund managers invest their own money in the fund. This
practice, known as eating your own cooking, helps to realign the motivation of the fund
manager with the objectives of the other investors in the fund. The fact that hedge
fund managers typically risk both their career and their own money while managing a fund
is a positive sign to outside investors. The personal involvement of the manager, combined
with a good and veriable track record, could explain why outside investors are willing to
3292 R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310
invest their money in hedge funds, even though investors typically receive very limited
information about hedge fund investment strategies and also possibly face poor liquidity
due to lock-up periods. We expect that the hedge fund managers own stake in the fund
is an essential factor inuencing the relation between incentives and risk taking.
Another paper related to our work is Goetzmann et al. (2003). Goetzmann et al. (2003)
study the eect of high-water mark provisions and incentives fees in a continuous-time
framework, under the assumption that the fund managers investment strategy is xed
(not optimized) and that fees are earned continuously (instead of periodically, as in
practice).
In the second part of the paper we complement the theoretical analysis with an empir-
ical study of the risk taking behaviour of a large number of hedge funds in the Zurich
Hedge Fund Universe (formerly known as the MAR database). One interesting and test-
able implication of our theoretical analysis of loss averse fund managers is the positive
relation between risk taking and incentive fees. For fund managers with HARA-utility
the relation between incentive fees and risk taking is negative (Carpenter, 2000), so an
empirical test provides information about manager preferences. Apart from volatility
we use alternative measures of risk such as downside deviation, maximum drawdown,
skewness and kurtosis to capture the non-normal shape of hedge fund return distributions.
The empirical relation between risk taking and incentives in hedge funds has been stud-
ied by Ackermann et al. (1999), Brown et al. (2001), and Agarwal et al. (2004). However
these papers only consider volatility as a risk measure, while hedge fund returns are known
to be non-normal (Fung and Hsieh, 1997, 2001; Getmansky et al., 2004; Gupta and Liang,
2005; Lo, 2001). This paper is the rst that tries to account for the highly skewed and
thick-tailed distribution of risk measures such as volatility in the fund cross-section, by
applying log-transformations and estimating a regression model with a skewed Student-
t error distribution (based on Fernandez and Steel, 1998). Without the latter adjustments
statistical inference can be erroneous, even in large samples.
Section 2 develops a continuous-time framework for deriving the fund managers opti-
mal investment strategy in the framework of prospect theory, taking into account manage-
ment fees and the managers own investment in the fund. The value of the managers
incentive fee contract is derived. Section 3 investigates the empirical relation between risk
taking and incentives using a large database of hedge funds and funds of hedge funds. Sec-
tion 4 concludes.
2. Theoretical analysis of incentives and risk taking
2.1. Model setup
We consider the optimal portfolio problem of a hedge fund manager with initial wealth
W(0). The initial size of the hedge fund is Y(0). The manager owns a fraction m of the fund,
with 0 6 m 6 1, while outside investors own the remaining portion of the assets (1 m).
The management fee equals a proportion a P0 of fund value (1 m)Y(T) at the end of
the evaluation period T (i.e. at the beginning of the next evaluation period). The incentive
fee is a percentage b P0 of the funds performance in excess of the benchmark B(T) at the
end of the evaluation period T: (1 m) bmax{Y(T) B(T), 0}. Similar to Carpenter (2000)
we assume that the fund manager does not hedge his exposure to the funds value with his
private portfolio, i.e. his wealth outside of the fund. For ease of exposition we set the rate
R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310 3293
of return on the private portfolio equal to the riskless rate R(0). The portfolio managers
wealth at the end of the period T is
W T mY T a1 mY T b1 m maxfY T BT; 0g
1 R0W 0 mY 0: 1
Risky assets with prices S
k
(0) for k = 1, . . . , K and a riskless asset with price S
0
(0) are
available as potential investments for the hedge fund manager. The risky asset prices fol-
low Ito processes with drift rate l
k
(t) and volatility r
k
(t), where t is between 0 and T, while
the riskless asset has a drift rate of r(t) and volatility of zero
dS
0
t rtS
0
t dt; 2
dS
k
t l
k
tS
k
t dt r
k
tS
k
t dBt; k 1; . . . ; K; 3
where the interest rate r(t), the vector of drift rates l(t) and the volatility matrix r(t) are
adapted processes (possibly path-dependent).
The fund manager selects a dynamic investment strategy, determined by the weights
w
k
(t) of risky assets k = 1, . . . , K in the fund, and the weight of the riskless asset w
0
(t),
at any time t in the continuous interval between 0 and T. For any self-nancing vector
of portfolio weights w(t) at time t, the fund value Y(t) then follows the stochastic process
(using vector notation)
dY t rtY t dt lt irt
0
wtY t dt rt
0
wtY t dBt; 4
where w
0
t 1

k
w
k
t has been substituted and i denotes a (K 1) vector of ones.
2.2. The fund managers preferences and reference point
The hedge fund manager evaluates wealth at the end of the period T with the value
function of prospect theory (Kahneman and Tversky, 1979)
V W T
AhT W T
c
1
; if W T 6 hT;
W T hT
c
2
; if W T > hT:
_
5
The fund manager has a reference point h(T) > 0 for separating gains and losses. The
parameters 0 < c
1
6 1 and 0 < c
2
6 1 determine the curvature of the value function over
losses and gains, respectively. The value function is convex below the reference point
and concave above. The parameter A > 0 is the level of loss aversion of the hedge fund
manager. In prospect theory it is assumed that A > 1, i.e. the pain of a loss exceeds the
positive feeling associated with an equivalent gain. In the hedge fund context the man-
agers loss does not necessarily have to be a monetary loss, but can also represent a loss
of time (opportunity costs) and a loss of reputation. Especially a deterioration of the man-
agers track record can have adverse consequences, as it can reduce the managers long
term potential to attract new fund ows and to generate income from fees.
The hedge fund manager maximizes the expectation of the value function at the end of
the evaluation period T, by choosing an optimal investment strategy for the fund:
3294 R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310
max
wt
EV W T
s:t: W T mY T a1 mY T b1 m maxfY T BT; 0g ZT;
dY t rtY t dt lt irt
0
wtY t dt rt
0
wtY t dBt;
Y t P0; for all 0 6 t 6 T;
6
where Z(T) = (1 + R(0))(W(0) mY(0)) denotes the fund managers wealth outside of the
fund at T. An additional aspect of prospect theory is that the decision maker evaluates the
expectation in (6) with a subjectively distorted probability distribution. We do not assume
that fund managers systematically distort probabilities, as we would like to focus solely on
the impact of loss aversion and other properties of the value function on risk taking under
incentives fees.
Before we proceed with the analysis we rst specify the fund managers personal threshold
h(T), separating gains from losses in the value function. The hedge fund manager will only
earn incentives fees if the fund value Y(T) exceeds the benchmark value B(T) at the end of the
evaluation period. Therefore, it seems natural that the fund value Y(T) = B(T) is the main
point of focus for the manager, separating failure from success. Just achieving the bench-
mark B(T) would leave the manager with the following amount of personal wealth at the
end of the year, W(T) = mB(T) + a(1 m)B(T) + Z(T). We assume that this amount of per-
sonal wealth is the threshold that mentally separates gains from losses for the fund manager
hT mBT a1 mBT ZT: 7
Given the threshold specication in Eq. (7), we can demonstrate that the condition
W(T) 6 h(T) is equivalent to Y(T) 6 B(T) (see Appendix for the proof). The manager will
consider fund performance below the benchmark as a loss (failure) with no incentive fees
for the time and eort extended, while other likely negative consequences are loss of rep-
utation and fund outows. For these reasons the pain of a loss is felt relatively strong.
Once the fund value drops below the benchmark the managers tries to maximize the prob-
ability of beating the hurdle rate in order to avoid the negative consequences of a loss,
i.e. the manager becomes risk seeking. On the other hand, when the fund value exceeds the
benchmark, the manager expects to receive additional income from incentive fees and
becomes risk averse to protect his gains.
2.3. The eect of incentive fees on implicit loss aversion
We analyse the eect of incentive fees on risk taking by examining the value function
V(W(T)) of the fund manager at the end of the evaluation period. Substituting the expres-
sion for W(T) in Eq. (1) into the value function V(W(T)) yields
V W T
AhT m a1 mY T ZT
c
1
; if W T 6 hT;
m a1 mY T b1 mY T BT ZT hT
c
2
;
if W T > hT:
_

_
8
Using the fact that W(T) 6 h(T) is equivalent to Y(T) 6 B(T) and substituting Eq. (7) for
h(T) into (8) yields the following expression for the managers value function, after mul-
tiplying the objective function by m a b1 m
c
2
:
R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310 3295
V

Y T

ABT Y T
c
1
; if Y T 6 BT;
Y T BT
c
2
; if Y T > BT;
_
9
where

A Am a1 m
c
1
=m a b1 m
c
2
is the implicit level of loss aversion rel-
evant for the optimal portfolio choice problem of the fund manager.
Hence, under the relatively mild assumption that the managers personal threshold for
separating gains and losses hinges on the hedge funds critical level B(T) for earning incen-
tive fees, the managers objective can be reduced back to the standard prospect theory
specication as a function of fund value Y(T), with B(T) as the threshold separating gains
from losses and

A as the implicit level of loss aversion. To investigate the eect of incentive
fees on risk taking, we now examine the expression for the implicit level of loss aversion

A
in (9).
Proposition 1. Given 0 6 m < 1, a P0 and m + a > 0, the implicit level of loss aversion

A of
the hedge fund manager strictly decreases as a function of the incentive fee b, i.e. d

A=db < 0.
Proposition 1 shows that an increase in the incentive fee will reduce the implicit level of
loss aversion of the hedge fund managers optimal portfolio choice problem. Hence, the
manager of a hedge fund with a large incentive fee should care less about investment losses
than a manager without such a fee, if the fund manager is trying to maximize the expec-
tation of the value function of prospect theory.
1
Proofs of all propositions are in
Appendix.
Proposition 2 considers the impact of the managers own stake in the fund on the impli-
cit level of loss aversion.
Proposition 2. Given 0 6 m 6 1 and b > 0, the condition c
1
= c
2
is sufcient for a strictly
positive relation between the managers own stake in the fund m and the implicit level of loss
aversion

A, i.e. d

A=dm > 0.
Given c
1
= c
2
, Proposition 2 states that a manager with a large own stake in the fund
should optimally care more about losses than a manager without such a stake. The su-
cient condition c
1
= c
2
means that the value function has the same curvature over gains as
over losses. Tversky and Kahneman (1992) have estimated the parameters of the value
function of prospect theory from the observed decisions made under uncertainty by a large
group of people. The estimates of Tversky and Kahneman (1992) are A = 2.25 for the
average level of loss aversion and c
1
= c
2
= 0.88 for the curvature of the value function.
As Tversky and Kahneman (1992) did not nd a signicant dierence between c
1
and
c
2
, the condition c
1
= c
2
of Proposition 2 seems plausible.
Given these estimated preference parameters, Fig. 1 displays the implicit level of loss
aversion

A as a function of the incentive fee for three dierent levels of the managers stake
1
Proposition 1 excludes the special case m = a = 0 in which the manager does not have a stake in the fund and
does not receive a management fee. In that special case

A 0 and d

A=db 0, so the level of the incentive fee does

not aect risk taking. In our opinion this special case is not very important and can be discarded, as only 4% of
the 1242 funds in our database do not charge a management fee. We do not have data on the managers stake in
the fund. Even in the few cases with m = a = 0, we would still expect the manager to behave as if some of his own
money is in the fund, as the managers reputation and future potential to earn incentive fees are implicitly at
stake.
3296 R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310
in the fund (m = 5%, m = 20% and m = 50%). Fig. 1 demonstrates that the managers impli-
cit level of loss aversion is equal to 2.25 without incentive fees (b = 0). As the incentive fee
increases, the implicit level of loss aversion of the fund manager starts to decrease, indicat-
ing that the manager should optimally care less about losses and more about gains due to
the convex compensation structure. The negative impact of incentive fees on implicit loss
aversion is mitigated if the manager owns a substantial part the fund.
2.4. The optimal investment strategy with incentive fees
We now study the impact of incentive fees on the investment strategy of the fund man-
ager. In the previous section we reduced the value function of the fund manager back to
standard format V
*
(Y(T)), as a function of terminal fund value Y(T). Hence, the optimal
portfolio choice problem (6) is equivalent to
max
wt
EV

Y T
s:t: dY t rtY t dt lt irt
0
wtY t dt rt
0
wtY t dBt;
Y t P0; for all 0 6 t 6 T:
10
To facilitate the solution of the optimal portfolio choice problem we assume that mar-
kets are dynamically complete. Market completeness implies the existence of a unique
state price density n(t), also known as a pricing kernel, dened as
dnt rtnt dt jtnt dBt; n0 1; 11
where j(t) = r(t)
1
(t)(l(t) ir(t)) denotes the market price of risk.
Under the assumption of complete markets, Berkelaar et al. (2004) solve the optimal
portfolio choice problem of a loss averse investor with the martingale methodology, using
the general approach for portfolio choice with non-concave and non-dierentiable objec-
tives of Basak and Shapiro (2001). The solution is derived in two steps. First, assuming
0.0
1.0
2.0
3.0
0.0% 5.0% 10.0% 15.0% 20.0% 25.0%
Incentive fee ( )
I
m
p
l
i
c
i
t

l
e
v
e
l

o
f

l
o
s
s

a
v
e
r
s
i
o
n
= 5%
= 50%
= 20%

Fig. 1. Implicit level of loss aversion as a function of incentive fee, with xed fee of a = 1%, and with separate
lines for dierent levels of the managers stake in the fund (m).
R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310 3297
general Ito processes for the asset prices, the optimal fund value Y
*
(T) for the manager at
the end of the evaluation period T is derived as a function of the pricing kernel n(T)
Y

T
BT
ynT
c
2
_ _
1=c
2
1
; if nT < n

;
0; if nT Pn

;
_
_
_
12
where n
*
solves f(n) = 0 with
f x
1 c
2
c
2
1
yx
_ _
c
2
=1c
2

c
2

1=1c
2

BTyx

ABT
c
1
13
and y P0 satises E[n(T)Y(T)] = n(0)Y(0). Second, an expression for the dynamic invest-
ment strategy w(t) that replicates these fund values Y
*
(T) can be derived explicitly under
the assumption that the asset prices follow Geometric Brownian motions and the riskless
rate is constant.
To analyze the eect of incentive fees on the investment strategy of the fund manager,
we use the fact that the implicit level of loss aversion

A of the fund manager decreases as a
function of the incentive fee level (see Proposition 1). The next proposition shows how a
decrease of

A aects the optimal fund values Y
*
(T) at the evaluation date T.
Proposition 3. Given 0 6 m < 1, an increase of the hedge fund managers incentive fee b will
lead to a decrease in the breakpoint n
*
of the optimal fund value function Y
*
(T) and a
decrease of the Lagrange multiplier y.
Proposition 3 shows that an increase of the incentive fee makes the manager seek more
payos in good states of the world with a low pricing kernel (due to the decrease of y) and
less in bad states (due to the decrease of n
*
). Under the quite general assumptions of a
complete market and Ito processes for the asset prices, we can conclude that a loss averse
managers optimal reaction to an increase of the incentive fee level is to seek a more risky
distribution of the fund value at the end of the evaluation period.
We now illustrate the general result of Proposition 3 in the more specic case that the
risky asset prices follow a Geometric Brownian motion and the riskless rate is constant.
See Berkelaar et al. (2004) for an expression for the optimal dynamic investment strategy
of a loss averse agent in this case. The eect of an increase of the incentive fee on the opti-
mal investment strategy is illustrated in Fig. 2. For ease of exposition, we assume that
there is only one risky asset, representing equity, with a Sharpe ratio of j = 0.10 and a vol-
atility of r = 20%, and a riskless asset with r
0
= 4%. The evaluation period is 1 year
(T = 1) and the fund manager has the standard preference parameters for the value func-
tion (A = 2.25, c
1
= c
2
= 0.88). The initial fund value is Y(0) = 1, the threshold for the
incentive fee is B(T) = 1, the management fee is a = 1% and the managers own stake in
the fund is m = 20%. Given these parameters, Fig. 2 shows the optimal weight of risky
assets in the fund w
*
(t), as a function of fund value Y(t) at time t = 0.5. Each line in
Fig. 2 represents a dierent level of incentive fee b, ranging from 0 to 30%.
Fig. 2 shows that the fund manager takes more risk in response to an increasing incen-
tive fee. The increase in risk is more pronounced when the fund value drops below the
benchmark B(T). Due to the structure of the value function of prospect theory, a fund
manager without an incentive fee will increase risk at low fund values as well; incentive
fees amplify this behaviour. Fig. 3 shows the eect on the optimal investment strategy
3298 R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310
of changing the managers own stake in the fund m, given an incentive fee of b = 20%.
Fig. 3 demonstrates that an increase of the managers share in the fund can completely
change risk taking. With a stake of 10% or less, the manager behaves extremely risk seek-
ing as a result of the incentive fee. However, with a stake of 30% or more, the investment
strategy is very similar to the base case of 100% ownership (without an incentive fee).
2.5. Discussion of the results and comparison with the literature
Carpenter (2000) studies the impact of incentive fees on the investment strategy of risk
averse fund managers with preferences described by the class of utility functions with
hyperbolic absolute risk aversion (HARA). Carpenter (2000) nds that a manager with
= 10%
0%
200%
400%
600%
0.750 0.875 1.000 1.125 1.250
Fund value Y (t ) at time t = 0.5
O
p
t
i
m
a
l

w
e
i
g
h
t

o
f

s
t
o
c
k
s

a
t

t

=

0
.
5
= 100%
= 30%
= 20%
= 15%
Fig. 3. Optimal weight of stocks as a function of fund value at time t, with incentive fee of b = 20%, and with
separate lines for dierent levels of the managers stake in the fund (m).
0%
100%
200%
300%
400%
0.750 0.875 1.000 1.125 1.250
Fund value Y (t ) at time t = 0.5
O
p
t
i
m
a
l

w
e
i
g
h
t

o
f

s
t
o
c
k
s

a
t

t

=

0
.
5
= 30%
= 25%
= 20%
= 15%
= 0%
Fig. 2. Optimal weight of stocks as a function of fund value at time t, with managers stake in the fund of
m = 20%, and with separate lines for dierent levels of incentive fee (b).
R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310 3299
an incentive fee increases the risk of the funds investment strategy if the funds value is
below the benchmark. For loss averse investors we also nd increased risk taking as the
funds value drops below the benchmark, but this strategy is common for all loss averse
investors due to their risk seeking behaviour in the face of losses. Incentive fees simply
make the pattern stronger, as illustrated in Fig. 2.
Carpenter (2000) nds that if the fund value is above the benchmark, a manager with
HARA preferences reduces volatility, in some cases even below the optimal volatility level
of a fund without incentive fees. Loss averse investors, on the other hand, always take
more risk with incentive fees than without such fees (see Propositions 1 and 2, and Figs.
2 and 3 for illustrations).
Carpenter (2000) also proves that the optimal response of a manager with HARA pref-
erences to an increase of the incentive fee level (e.g. a change from 10% to 20%) is to
reduce the funds risk. Proposition 3 shows that these conclusions do not hold for loss
averse fund managers that maximize the value function of prospect theory. Loss averse
investors seek more extreme fund value distributions, and hence take more risk, in
response to an increase of the incentive fee level.
2.6. The value of the managers incentive fee option
Atypical hedge fund charges an incentive fee of 20%. For hedge fund investors it is worth-
while to knowwhat the cost of such a fee arrangement is. The incentive fee can be considered
as a call option on fund value with exercise price B(T), granted by the investors to the fund
manager as additional management compensation. In a complete market, any European
option with payo X(T) at time T can be priced as follows with the pricing kernel n(T)
X0 n0
1
EnTXT; 14
where X(0) is the initial value of the contingent claim. The pay o of the incentive fee at time
T under the managers optimal strategy is X(T) = (1 m)bmax{Y
*
(T) B(T), 0}. Hence,
we can nd the value of the incentive fee at time 0 by evaluating the expectation in (14).
Proposition 4. Under the assumption of a complete market, Geometric Brownian motions for
the risky asset prices and a constant interest rate r(t) = r
0
, the initial value of the incentive fee
X(0), given the fund managers optimal investment strategy, is
X0 1 mb
c
2
yn0
_ _
1=1c
2

e
C0
Nd
2
n

; 15
C0
1 c
2
c
2
r
1
2
j
2
_ _
T
1
2
1 c
2
c
2
_ _
2
j
2
T; 16
d
1
x
lnx=n0 r 0:5j
2
T
kjk

T
p and d
2
x d
1
x
kjk

T
p
1 c
2
; 17
where N() is the standard normal cumulative distribution and n
*
is dened by (13) as a non-
linear function of a, b, m, c
1
, c
2
,

A and B(T).
Fig. 4 plots the value of a 20% incentive fee as a function of the managers stake in the
fund, using the same set of parameters as in Fig. 2 (j = 0.10, r = 20%, r
0
= 4%, T = 1,
Y(0) = 1, B(T) = 1, a = 1% and A = 2.25, c
1
= c
2
= 0.88). Fig. 4 shows that the value of
3300 R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310
the 20% incentive fee ranges from 0% to 17% of the initial fund value, depending on the
managers own stake in the fund. If the managers stake in the fund is 100%, the manager
does not care about the incentive fee and manages the fund conservatively since it is a per-
sonal account. However, as the managers stake in the fund goes to zero, the manager
starts to increase the riskiness of the investment strategy in order to reap more prots from
the incentive fee contract and the value of the option increases greatly.
3. Empirical analysis of incentives and risk taking in hedge funds
One interesting and testable implication of our theoretical analysis of loss averse hedge
fund managers is the positive relation between risk taking and incentive fees. In the
HARA-utility setting of Carpenter (2000) the relation between incentive fees and risk tak-
ing is negative, so an empirical test could provide interesting information about manager
preferences. While testing the relation it is important to control for other fund character-
istics and for this reason we run a cross-sectional regression of volatility on the level of the
hedge funds incentive fee and a number of other fund characteristics such as investment
style, the funds management fee and the amount of assets under management.
Hedge fund returns are well known to be non-normal due to the dynamic investment
strategies of the funds (see Fung and Hsieh, 1997, 2001; Getmansky et al., 2004; Gupta
and Liang, 2005; Lo, 2001; Mitchell and Pulvino, 2001). Still, empirical studies of the rela-
tion between risk taking and incentives in hedge funds only consider volatility as a risk
measure (Ackermann et al., 1999; Brown et al., 2001; Agarwal and Naik, 2004; Agarwal
et al., 2004) even though volatility cannot fully capture the non-normal shape of hedge
fund return distributions. To address this problem, we focus on non-symmetrical risk mea-
sures, namely the rst downside moment and maximum drawdown, as well as the skew-
ness and kurtosis of hedge fund returns. The rst downside (upside) moment is dened
as the conditional expectation of the fund returns below (above) the risk free rate. Maxi-
mum drawdown is dened as the worst performance among all runs of consecutive nega-
tive returns.
0.00
0.05
0.10
0.15
0.20
0% 20% 40% 60% 80% 100%
Manager's stake in the fund ( )
V
a
l
u
e

o
f

2
0

%

i
n
c
e
n
t
i
v
e

f
e
e

(
f
u
n
d

v
a
l
u
e

=

1
)

Fig. 4. Option value of a 20% incentive fee, as a function of the managers stake in the fund.
R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310 3301
3.1. Cross-sectional regression model
As the dependent variable in the regression we use risk or return measure j, e.g. vola-
tility, adjusted for dierences in investment style across hedge fund style groups. Eq. (18)
denes ^a
i;j
as the adjusted risk/return measure j for fund i, measured relative to the average
of fund is investment style group.
^a
i;j
a
i;j

G
g1
d
i;g
a
j;g
for fund i 1; . . . ; I and investment style groups g 1; . . . ; G;
18
where a
i,j
denotes hedge fund risk/return measure j = 1, . . . , 9, for fund i = 1, . . . , I, and a
j;g
is the average of measure j among all funds in investment style group g = 1, . . . , G. The
dummy d
i,g
is equal to one if fund i belongs to investment style group g and zero otherwise.
As control variables in the cross-sectional regression we use the funds management fee
(mf
i
), the logarithm of the funds assets under management (size
i
), the number of monthly
observations on the fund available in the database (age
i
) and a dummy variable equal to
one if a fund drops from the database prematurely during the sample period (dead
i
):
^a
i;j
c
0j
c
1j
if
i
c
2j
mf
i
c
3j
lnsize
i
c
4j
age
i
c
5j
dead
i
e
i
;
for funds i 1; . . . ; I and e
i
N0; g
j
independently normally distributed;
19
where if
i
is the incentive fee of fund i, c
0j
, c
1j
, c
2j
, c
3j
, c
4j
and c
5j
are coecients and g
j
is the
standard deviation of the residuals in the regression for risk/return measure j.
After estimating the regression model in (19) with ordinary least squares (OLS), we
checked for violations of the OLS assumptions. As the regression residuals show signs
of heteroscedasticity, i.e. systematic variation in g
j
across funds, we use White heteroske-
dasticity-consistent standard errors. Furthermore, the distribution of the residuals is
highly non-normal. For example, if we use volatility as the dependent variable in regres-
sion (19), then the distribution of the OLS residuals has a skewness of 2.55 and kurtosis of
13.91. As a rst step to address this problem, we transform all non-negative risk and
return measures with the logarithmic function (i.e. the measures volatility, rst downside
moment, rst upside moment, maximum drawdown and kurtosis). After the applying log-
transformation we correct for dierences in investment style, as in Eq. (18). The log-trans-
formation greatly reduces the non-normality of the non-negative measures, e.g. in the case
of volatility the distribution of OLS residuals has a skewness of 0.18 and kurtosis of 3.63
after log-transformation.
As a second step to deal with non-normality, we change the distribution of the error term
in (19) to a skewed Student t-distribution, dened by Fernandez and Steel (1998). The
skewed t-distribution has three parameters: the skewness parameter s > 0, the degrees of
freedom df > 2 and volatility g > 0. The error distribution is symmetric if s = 1, negatively
skewed for 0 < s < 1 and positively skewed for s > 1. The number of degrees of freedom df is
constrained to be larger than two in order to guarantee the existence of the variance of the
residuals. As the number of degrees of freedom df goes to innity, the tails become thinner
and converge to the normal case. We test the normality of the tails with the hypothesis H
0
:
1/df = 0 versus H
a
: 1/df > 0, while we test for the presence of skewness with the hypothesis
H
0
: s = 1 versus H
a
: s 51. We test the suitability of a normal error distribution with a like-
lihood ratio test (LR-test) of both restrictions, i.e. H
0
: 1/df = 0 and s = 1.
3302 R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310
3.2. Hedge fund data
We use the Zurich Hedge Fund Universe, formerly known as the MAR hedge fund
database, provided by Zurich Capital Markets. The database includes a large number
of funds that have disappeared over the years, which reduces the impact of survivorship
bias. The data starts in January 1977 and ends in November 2000. There are 2078 hedge
funds in the database and 536 fund of funds. We will analyse the hedge fund data from
January 1995 to November 2000 since the database keeps track of funds that disappear
starting January 1995. The return data is net of management fees and net of incentive fees.
The hedge funds in the database are classied into eight dierent investment styles by
the provider: Event-Driven, Market Neutral, Global Macro, Global International, Global
Emerging, Global Established, Sector and Short-Sellers. Unfortunately, the database does
not provide any information regarding the investment strategy of the fund of funds. We
suspect that some fund of funds only aim to provide investors due diligence service and
diversication benets, while other funds additionally try to add value with an active
hedge fund allocation strategy. Due to the absence of information regarding the invest-
ment style of fund of funds, we can not distinguish between these two groups.
Table 1 provides some descriptive statistics of the funds in the database. The table dis-
tinguishes between funds that were still in the database in November 2000 (alive) and
funds that dropped out (dead) and between individual hedge funds and fund of funds.
Table 1
Characteristics of the hedge fund data
Hedge funds Fund of funds
Alive Dead Alive Dead
Incentive fee Mean 18.2 17.9 16.7 13.5
Median 20.0 20.0 20.0 15.0
Std. Dev. 6.2 7.0 5.2 7.4
Management fee Mean 1.22 1.30 1.04 1.12
Median 1.00 1.00 1.20 1.00
Std. Dev. 0.52 0.64 0.75 0.70
Assets under management (in million USD) Mean 98.6 75.8 64.7 37.4
Median 28.1 15.7 19.1 8.8
Std. Dev. 300.9 376.8 213.8 113.2
Length of timeseries (in years) Mean 3.81 2.64 4.17 2.58
Median 3.92 2.42 4.58 2.42
Std. Dev. 1.81 1.60 1.85 1.53
Number of funds with incentive fee (if) of if = 0 61 44 3 25
0 < if < 20% 90 66 102 54
if = 20% 569 318 198 63
if > 20% 52 42 4 2
Total number of funds 772 470 307 144
Descriptive statistics of the hedge funds in the database are displayed. The cross-sectional mean, median and
standard deviation of the incentive fee, the management fee, assets under management and the length of the
timeseries of return observations per fund are listed respectively. The sample period is January 1995November
2000. Alive funds were still in the database in November 2000, while dead funds dropped out before this
date. We exclude funds that do not report fees.
R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310 3303
The median incentive fee for hedge funds is 20%. An incentive fee of 20% is the industry
standard, as 71.4% of the funds use it. Only 8.5% of all hedge funds do not charge an
incentive fee. The median management fee is 1%. The majority of funds (71.5%) charge
a fee between 0.5% and 1.5%, while only 4.2% of the funds do not charge a management
fee. A fund of funds investor also has to pay fees to the fund of fund manager. On average,
fund of funds charge slightly lower fees than individual hedge funds, although the median
incentive fee is still 20% (dead and alive funds combined). Only 6.2% of fund of funds do
not charge an incentive fee. The median management fee of fund of funds is 1%.
Table 1 shows that the hedge funds had an average net asset value of 98.6 million US
dollars during the period under consideration (75.8 million for dead funds). The net asset
value distribution is very positively skewed: the top 25% funds according to size manage
about 80% of the total asset value. The database contains 15 hedge funds and 2 fund of
funds with an average net asset value of more than 1 billion US dollar. The number of
years since January 1995 that funds are in the database is relatively short, with an average
of 4 years for living funds and about 2.5 years for dead funds (same for hedge funds and
fund of funds). The short average length of the time series has to do with the high fund
attrition rate and the large number of new funds that enters the database each year (the
hedge fund industry was growing rapidly during the sample period). For a study of the
performance of the funds in the database over this period, see Kouwenberg (2003).
3.3. Incentives and risk taking in hedge funds: Empirical results
Table 2 reports the cross-sectional regression results. For each risk and return measure,
the rst row of results is based on the OLS regression (19), before applying the log-trans-
formation. These results are reported to facilitate judging the size and economic signi-
cance of the estimated coecients. The second row of results is based on maximum
likelihood estimation of the regression model with a skewed Student-t error distribution,
after rst applying the log-transformation to all non-negative dependent variables. The
second row of results also shows the estimated parameters s and df of the skewed t-distri-
bution, and the LR-test statistic for the null hypothesis of a normal error distribution. The
p-value reported below the estimated value of s refers to the hypothesis H
0
: s 1 = 0,
while the p-value reported below df refers to the hypothesis H
0
: 1/df = 0.
The results in Table 2 show that the LR-test rejects a normal error distribution strongly
in all cases, so we will base our conclusions on the results of the regressions with a skewed
Student-t error distribution (reported in the second row for each measure). Concentrating
on the results for the incentive fee variable, we nd that funds with higher fees earn sig-
nicantly lower mean returns. Further, a change in the incentive fee from 0% to 20%
reduces of the Sharpe ratio by 0.16 (signicant at the 5% level). The eect of incentive fees
on risk taking is positive for the measures rst downside moment and maximum draw-
down, with p-values of 0.06 and 0.09, respectively. The impact of higher incentive fees
on the risk measure volatility is insignicant.
Considering the control variables, funds dropping out of the database (dead
i
) have sig-
nicantly lower returns, higher risk, more negatively skewed returns and lower risk-
adjusted performance. Fund size is also an important factor, as larger funds have signif-
icantly higher returns and lower risk. As the number of observations on a fund (age
i
)
increases, downside risk and maximum drawdown increase signicantly, while the return
3304 R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310
Table 2
Cross-sectional regressions of hedge funds risk and returns
if mf ln(size) age dead s 1 df R
2
/LR
Average return 4.14
*
0.42 6.66
*
0.56 5.65
*
6.6%
(0.03) (0.65) (0.00) (0.13) (0.00)
2.93
*
0.67 3.85
*
0.42 5.25
*
0.13
*
2.22
*
442.9
*
(0.00) (0.24) (0.00) (0.06) (0.00) (0.01) (0.00) (0.00)
First upside moment
a
0.74 1.90 2.88 1.17
*
3.42
*
1.9%
(0.76) (0.12) (0.06) (0.01) (0.02)
0.06 0.06 0.10
*
0.02 0.13
*
0.07 8.08
*
34.2
*
(0.30) (0.06) (0.01) (0.22) (0.00) (0.11) (0.00) (0.00)
Volatility
a
1.54 2.48
*
5.25
*
0.33 0.72 4.7%
(0.28) (0.01) (0.00) (0.29) (0.50)
0.00 0.10
*
0.23
*
0.00 0.07 0.09
*
17.01
*
12.7
*
(0.95) (0.00) (0.00) (0.88) (0.11) (0.02) (0.01) (0.00)
First downside moment
a
3.38
*
1.47 9.54
*
0.65 2.05 7.7%
(0.03) (0.19) (0.00) (0.12) (0.15)
0.13 0.02 0.43
*
0.01 0.22
*
0.25
*
5.37
*
134.6
*
(0.09) (0.64) (0.00) (0.65) (0.00) (0.00) (0.00) (0.00)
Maximum drawdown
a
3.51
*
3.67
*
7.15
*
2.39
*
4.96
*
5.5%
(0.04) (0.02) (0.00) (0.00) (0.00)
0.18 0.09 0.29
*
0.14
*
0.33
*
0.26
*
4.98
*
154.0
*
(0.06) (0.05) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Skewness 0.03 0.03 0.19
*
0.16
*
0.45
*
4.8%
(0.80) (0.81) (0.02) (0.00) (0.00)
0.01 0.11 0.02 0.16
*
0.33
*
0.18
*
3.25
*
188.4
*
(0.93) (0.07) (0.77) (0.00) (0.00) (0.00) (0.00) (0.00)
Kurtosis
a
0.16 1.33
*
1.05
*
0.99
*
1.07
*
8.4%
(0.77) (0.01) (0.01) (0.00) (0.00)
0.02 0.07
*
0.02 0.12
*
0.05 0.90
*
11.96 209.3
*
(0.64) (0.00) (0.38) (0.00) (0.07) (0.00) (0.06) (0.00)
Sharpe ratio 0.15 0.04 0.45
*
0.08
*
0.50
*
7.0%
(0.19) (0.59) (0.00) (0.00) (0.00)
0.16
*
0.05 0.41
*
0.01 0.38
*
0.14
*
2.90
*
488.6
*
(0.04) (0.24) (0.00) (0.67) (0.00) (0.01) (0.00) (0.00)
Alpha 0.25 0.05 0.66
*
0.09
*
0.63
*
8.3%
(0.09) (0.56) (0.00) (0.00) (0.00)
0.03 0.13
*
0.33
*
0.10
*
0.55
*
0.03 2.00
*
529.7
*
(0.74) (0.01) (0.00) (0.00) (0.00) (0.53) (0.00) (0.00)
The results of a cross-sectional regression are displayed of the risk and return measures of 1114 hedge funds in the
database on a constant, the level of the funds incentive fee (if), management fee (mf), the natural logarithm of the
assets under management (size), the number of monthly observations (age) and a dummy indicating that the fund
dropped out of the database (dead). The risk and return measures are dierences relative to the average within
each of the eight hedge fund style groups, to control for dierences in investment style. The explanatory variables
if, mf and ln(size) are scaled by the median value, while age is measured in years. The rst row of results is based
on normally distributed residuals (OLS), while the second row is based on a skewed Student-t distribution. The
sample period is January 1995November 2000. Only funds with at least 12 monthly observations are included.
White heteroskedasticity-consistent p-values are reported below the estimates. Below the estimated number of
degrees of freedom df we report the p-value for the test of the null-hypothesis 1/df = 0, i.e. normal tails. LR is the
likelihood-ratio test of 1/df = 0 and s = 1, i.e. normality of the residuals.
a
Log-transformation of the dependent variable has been applied for the second row of results.
*
Denotes signicance at the 5% level.
R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310 3305
Table 3
Cross-sectional regressions of fund of funds risk and returns
if mf ln(size) age dead s 1 df R
2
/LR
Average return 1.41 1.88
*
1.30 0.58 5.56
*
9.3%
(0.40) (0.01) (0.11) (0.09) (0.00)
2.87
*
1.29
*
0.53 0.61
*
3.34
*
0.07 2.40
*
228.7
*
(0.01) (0.00) (0.33) (0.00) (0.00) (0.25) (0.00) (0.00)
First upside moment
a
6.15
*
0.38 4.21
*
0.17 1.27 6.4%
(0.00) (0.60) (0.00) (0.62) (0.31)
0.34
*
0.01 0.21
*
0.01 0.05 0.25
*
52.56 12.3
*
(0.00) (0.75) (0.00) (0.54) (0.45) (0.00) (0.73) (0.00)
Volatility
a
4.50
*
1.42
*
3.92
*
0.46
*
1.43 10.4%
(0.00) (0.01) (0.00) (0.04) (0.14)
0.36
*
0.12
*
0.32
*
0.05
*
0.13 0.24
*
131.00 12.4
*
(0.01) (0.01) (0.00) (0.03) (0.10) (0.00) (1.00) (0.00)
First downside moment
a
4.68
*
2.26
*
5.50
*
0.70
*
4.16
*
12.1%
(0.01) (0.00) (0.00) (0.03) (0.00)
0.35
*
0.18
*
0.44
*
0.09
*
0.29
*
0.36
*
9.47
*
46.4
*
(0.02) (0.00) (0.00) (0.00) (0.01) (0.00) (0.04) (0.00)
Maximum drawdown
a
6.64
*
4.19
*
6.10
*
2.65
*
5.06
*
12.0%
(0.00) (0.00) (0.00) (0.00) (0.01)
0.44
*
0.19
*
0.41
*
0.26
*
0.26
*
0.35
*
3.65
*
88.1
*
(0.00) (0.00) (0.00) (0.00) (0.01) (0.00) (0.00) (0.00)
Skewness 0.20 0.13 0.08 0.17
*
0.48
*
6.4%
(0.33) (0.20) (0.43) (0.00) (0.00)
0.49
*
0.03 0.10 0.16
*
0.45
*
0.37
*
3.67
*
67.6
*
(0.00) (0.57) (0.27) (0.00) (0.00) (0.00) (0.00) (0.00)
Kurtosis
a
0.08 0.43 0.42 0.73
*
0.21 8.5%
(0.92) (0.22) (0.32) (0.00) (0.66)
0.00 0.02 0.03 0.11
*
0.01 1.32
*
175.15 78.7
*
(0.95) (0.39) (0.43) (0.00) (0.86) (0.00) (1.00) (0.00)
Sharpe ratio 0.19 0.26
*
0.28
*
0.11
*
0.58
*
12.3%
(0.26) (0.00) (0.02) (0.01) (0.00)
0.06 0.23
*
0.28
*
0.05 0.45
*
0.35
*
2.74
*
129.0
*
(0.63) (0.00) (0.00) (0.05) (0.00) (0.00) (0.00) (0.00)
Alpha 0.02 0.20
*
0.13 0.08
*
0.70
*
11.6%
(0.88) (0.01) (0.10) (0.01) (0.00)
0.13 0.12
*
0.05 0.09
*
0.40
*
0.14
*
2.00
*
383.8
*
(0.13) (0.00) (0.30) (0.00) (0.00) (0.01) (0.00) (0.00)
The results of a cross-sectional regression are displayed of risk and return measures of 403 fund of funds in the
database on a constant, the level of the funds incentive fee (if), management fee (mf), the natural logarithm of the
assets under management (size), the number of monthly observations on the fund (age) and a dummy indicating
that the fund dropped out of the database (dead). The explanatory variables if, mf and ln(size) are scaled by the
median value of the series, while age is measured in years. The rst row of results is based on normally distributed
residuals (OLS), while the second row is based on a skewed Student-t distribution. The sample period is January
1995November 2000. Only funds with at least 12 monthly observations are included. White heteroskedasticity-
consistent p-values are reported below the estimates. Below the estimated number of degrees of freedom df we
report the p-value for the test of the null-hypothesis 1/df = 0, i.e. normal tails. LR is the likelihood-ratio test of
1/df = 0 and s = 1, i.e. normality of the residuals.
a
Log-transformation of the dependent variable has been applied for the second row of results.
*
Denotes signicance at the 5% level.
3306 R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310
distribution becomes more negatively skewed. Higher management fees are positively
related to volatility and kurtosis, indicating increased risk taking.
3.4. Incentives and risk taking in fund of funds: Empirical results
Table 3 presents the estimation results of the cross-sectional regression model (19) for
fund of funds. Again normality of the regression errors is strongly rejected, so we focus on
the results for the skewed Student-t regression, displayed in the second row for each mea-
sure. The coecient of the incentive fee variable is signicantly positive in the cross-sec-
tional regression on the average return, rst upside moment, volatility, rst downside
moment, maximum drawdown and skewness. Hence, for the fund of funds in the data-
base, higher incentive fees are linked to increased risk-taking, higher returns and more
upside potential, while the impact on the risk-adjusted performance measures is insignif-
icant. A potential explanation for these results is that loss averse fund of fund managers
with incentive fees opt for a more risky basket of hedge funds to increase the value of their
call option on fund value.
2
In the case of management fees, Table 3 shows that they are a drag on performance:
higher management fees lead to signicantly lower returns, increased risk taking and lower
risk-adjusted performance. Larger funds tend to take less risk and have higher Sharpe
ratios. As the number of observations in the database on a fund of fund increases, the
average return drops, risk goes up and risk-adjusted performance deteriorates. Fund of
funds that drop out of the database have lower average returns, more risky return distri-
butions and very poor risk-adjusted-performance.
4. Summary and conclusions
We analyse the relation between incentives and risk taking in the hedge fund industry.
The value function of prospect theory is used to model the hedge fund managers prefer-
ences, while taking into account incentive fees, management fees and the managers own
stake in the fund. We prove that incentive fees reduce the managers implicit level of loss
aversion, leading to increased risk taking. However, if the managers own stake in the fund
is substantial (e.g. >30%), risk taking is reduced considerably. We also derive an expres-
sion for the option value of the incentive fee arrangement, taking into account the man-
agers optimal investment strategy.
One of the testable implications of our theoretical analysis is that higher incentive fees
lead to increased risk taking, all other things equal. Our result for loss averse fund man-
agers contrasts with the negative relation between risk taking and incentive fees that holds
for risk averse managers with HARA utility (see Carpenter, 2000). In the second part of
the paper we test empirically whether hedge fund managers with incentive fees indeed take
more risk in practice, using the Zurich Hedge Fund Universe (formerly known as the
MAR database). We apply a cross-sectional regression of risk on the incentive fee level
and four control variables. Apart from volatility, we employ alternative risk measures
such as maximum drawdown and the rst downside moment as the dependent variable.
2
Another explanation could be that fund of funds with high incentive fees try to add value with an active hedge
fund allocation strategy, while low fee funds tend to focus on oering due diligence service and diversication. It
is dicult to disentangle the two explanations, as the investment strategy might depend on the incentive fee level.
R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310 3307
Furthermore, we correct for the highly non-normal cross-sectional distribution of the risk
measures by using log-transformations and a skewed Student-t error distribution.
The cross-sectional analysis shows that hedge funds with incentive fees have signi-
cantly lower mean returns (net of fees). The risk measures maximum drawdown and rst
downside moment are positively related to the incentive fee level, with p-values of 0.09 and
0.06, respectively. There is no signicant relation between volatility and incentive fees.
Among funds of funds there is a signicantly positive relation between incentive fees
and all three risk measures at the 5% level. Fund of funds with higher incentive fees also
have higher average returns and more upside return potential.
Based on our empirical results, investors should evaluate individual hedge funds with
incentive fee arrangements critically. The presence of a 20% incentive fee reduces a funds
average after-fee return by 2.93% (absolute reduction) and the Sharpe ratio by 0.16, com-
pared to a fund without incentive fee and assuming other things equal. Among fund of
funds, the impact of incentive fees appears to be a shift upward in both risk and return,
without adverse consequences for risk-adjusted performance. The presence of a 20% incen-
tive fee increases the average after-fee return by 2.87% and volatility by about 4.5%, while
the funds alpha and Sharpe ratio are not aected signicantly.
An interesting question for further research is why the relation between incentive fees
and risk taking is relatively mild for individual hedge funds, but quite strong among fund
of funds. A potential explanation tting in the theoretical framework of this paper is
that managers of individual hedge funds tend to have a larger stake in their own fund than
fund of fund managers. Alternative explanations might be that peer group pressure is
stronger among managers of individual hedge funds, or dierences in high-water mark
provisions.
Acknowledgements
This research was supported by Inquire Europe, the Swiss Banking Institute, University
of Zurich and the Social Sciences and Humanities Research Council of Canada. An earlier
version of this paper was presented at the Monaco meeting of the European Financial
Modelers, November 2003, to the Swiss Finance Society, April 2004 and the Workshop
on New Methods in Risk Theory, Florence, October 2005. We would like to thank Cees
Dert for helpful comments.
Appendix A
Proof of the equivalence of W(T) <h(T) and Y(T) <B(T)
Substituting (1) for W(T) and (7) for h(T), W(T) 6 h(T) is equivalent to
mY T a1 mY T b1 m maxfY T BT; 0g ZT 6 mBT
a1 mBT ZT: A:1
Rearranging terms to the left-hand-side, it follows that (A.1) is equivalent to (A.2)
mY T BT a1 mY T BT b1 m maxfY T BT; 0g 6 0:
A:2
3308 R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310
Finally, (A.2) holds if and only if Y(T) 6 B(T).
Proof of Proposition 1. Dierentiating

A with respect to b yields
d

A=db c
2
1 mAm a1 m
c
1
=m a b1 m
c
2
1
: A:3
Given 0 6 m < 1, A > 0, a P0, b P0, a + m > 0 and 0 < c
2
6 1, the inequality d

A=db < 0
holds.
Proof of Proposition 2. Dierentiating

A with respect to m yields
d

A=dm

Ac
1
1 am a1 m
1
c
2
1 a bm a b1 m
1
: A:4
Note that d

A=dm > 0 is equivalent to

c
1
1 am a b1 m > c
2
1 a bm a1 m: A:5
Given 0 6 m 6 1, a P0, b > 0, the condition c
1
= c
2
is sufcient for d

A=dm > 0.
Proof of Proposition 3. An increase of the managers incentive fee m leads to a reduction of
the implicit level of loss aversion

A (see Proposition 1). Berkelaar et al. (2004) prove that a
decrease of

A leads to a decrease in the breakpoint n
*
of the optimal fund value function
Y
*
(T) and a decrease of the Lagrange multiplier y.
Proof of Proposition 4. We substitute the pay o of the incentive fee contract X(T) into
Eq. (14) to yield (A.6):
X0 n0
1
EnT1 mbmaxfY

T BT; 0g: A:6

Eq. (A.6) can be simplied by substituting the managers optimal fund value function
Y
*
(T)
X0
_
n

0
nT1 mb
ynT
c
2
_ _
1=c
2
1
dnT: A:7
Under the assumption of a complete market, a constant risk free rate and Geometric
Brownian motions for the asset prices, the pricing kernel n(T) at time T is log-normally
distributed. Using this fact, the expression in (15) for X(0) follows from solving the integral
(A.7).
References
Ackermann, C., McEnally, R., Ravenscraft, D., 1999. The performance of hedge funds: Risk, return and
incentives. Journal of Finance 54, 833874.
Agarwal, V., Naik, N.Y., 2004. Risk and portfolio decisions involving hedge funds. Review of Financial Studies
17 (1), 833874.
Agarwal, V., Daniel, N.D., Naik, N.Y., 2004. Flows, Performance, and managerial incentives in hedge funds,
Working paper, London Business School.
Basak, S., Shapiro, A., 2001. Value-at-risk based risk management: Optimal policies and asset prices. Review of
Financial Studies 14, 371405.
R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310 3309
Berkelaar, A.B., Kouwenberg, R., Post, G.T., 2004. Optimal portfolio choice under loss aversion. Review of
Economics and Statistics 86, 973987.
Brown, S.J., Goetzmann, W.N., Park, J., 2001. Careers and survival: Competition and risk in the hedge fund and
CTA industry. Journal of Finance 56, 18691886.
Carpenter, J., 2000. Does option compensation increase managerial risk appetite? Journal of Finance 55, 2311
2331.
Fernandez, C., Steel, M., 1998. On Bayesian modelling of fat tails and skewness. Journal of the American
Statistical Association 93, 359371.
Fung, W., Hsieh, D.A., 1997. Empirical characteristics of dynamic trading strategies: The case of hedge funds.
Review of Financial Studies 10, 275302.
Fung, W., Hsieh, D.A., 2001. The risk in hedge fund strategies: Theory and evidence from trend followers.
Review of Financial Studies 14, 313341.
Getmansky, N., Lo, A.W., Makarov, I., 2004. An econometric model of serial correlation and illiquidity in hedge
fund returns. Journal of Financial Economics 74 (3), 529609.
Goetzmann, W.N., Ingersoll, J.E., Ross, S.A., 2003. High-water marks and hedge fund management contracts.
Journal of Finance 58, 16851718.
Gupta, A., Liang, B., 2005. Do hedge funds have enough capital: A value-at-risk approach. Journal of Financial
Economics 77, 219253.
Kahneman, D.H., Tversky, A., 1979. Prospect theory: An analysis of decision making under risk. Econometrica
47, 263290.
Kouwenberg, R., 2003. Do hedge funds add value to a passive portfolio: Correcting for non-normal returns and
disappearing funds. Journal of Asset Management 3, 361382.
Lo, A.W., 2001. Risk management for hedge funds: Introduction and overview. Financial Analysts Journal 57
(6), 1633.
Mitchell, M., Pulvino, T., 2001. Characteristics of risk and return in risk arbitrage. Journal of Finance 56, 2135
2175.
Tversky, A., Kahneman, D.H., 1992. Advances in prospect theory: Cumulative representation of uncertainty.
Journal of Risk and Uncertainty 5, 297323.
3310 R. Kouwenberg, W.T. Ziemba / Journal of Banking & Finance 31 (2007) 32913310