Eectiveness of CPPI Strategies

under DiscreteTime Trading

Sven Balder
Michael Brandl
Antje Mahayni
Department of Banking and Finance
University of Bonn
This version: October 5, 2005
Abstract.
The paper analyzes the eectiveness of the constant proportion portfolio
insurance (CPPI) method under trading restrictions. If the CPPI method
is applied in continuous time, the CPPI strategies provide a value above a
oor level unless the price dynamic of the risky asset permits jumps. The
risk of violating the oor protection is called gap risk. In practice, it is
caused by liquidity constraints and price jumps. Both can be modelled
in a setup where the price dynamic of the risky asset is described by a
continuoustime stochastic process but trading is restricted to discrete time.
We propose a discretetime version of the continuoustime CPPI strategies
which satises three conditions. The resulting strategies are selfnancing,
the asset exposure is nonnegative and the value process converges. We
determine risk measures such as the shortfall probability and the expected
shortfall and discuss criteria which ensure that the gap risk does not increase
to a level which contradicts the original intention of portfolio insurance.
JEL: G11, G12
Keywords: Portfolio insurance, discretetime trading, return guarantees,
gap risk, volatility risk.
1
Corresponding Author: Antje Mahayni, Department of Banking and Finance, University of Bonn,
Adenauerallee 24-42, D-53113 Bonn, Germany, phone: +49 (228) 736103, fax: +49 (228) 735050, e-mail:
[email protected]
The authors would like to thank Klaus Sandmann for helpful discussions.
EFFECTIVENESS OF CPPI STRATEGIES UNDER DISCRETETIME
TRADING
Abstract. The paper analyzes the eectiveness of the constant proportion portfolio
insurance (CPPI) method under trading restrictions. If the CPPI method is applied in
continuous time, the CPPI strategies provide a value above a oor level unless the price
dynamic of the risky asset permits jumps. The risk of violating the oor protection
is called gap risk. In practice, it is caused by liquidity constraints and price jumps.
Both can be modelled in a setup where the price dynamic of the risky asset is described
by a continuoustime stochastic process but trading is restricted to discrete time. We
propose a discretetime version of the continuoustime CPPI strategies which satises
three conditions. The resulting strategies are selfnancing, the asset exposure is non
negative and the value process converges. We determine risk measures such as the
shortfall probability and the expected shortfall and discuss criteria which ensure that
the gap risk does not increase to a level which contradicts the original intention of
portfolio insurance.
1. Introduction
Financial strategies which are designed to limit downside risk and at the same time to
prot from rising markets are summarized in the class of portfolio insurance strategies.
Among others, Grossman and Villa (1989) and Basak (2002) dene a portfolio insurance
trading strategy as a strategy which guarantees a minimum level of wealth at a specied
time horizon, but also participates in the potential gains of a reference portfolio. The
most prominent examples of dynamic versions are the constant proportion portfolio in-
surance (CPPI) strategies and optionbased portfolio insurance (OBPI) strategies with
synthetic puts.
1
Here, synthetic is understood in the sense of a trading strategy in basic
(traded) assets which creates the put. In a complete nancial market model, there exists a
perfect hedge, i.e. a selfnancing and duplicating strategy. In contrast, the introduction
of market incompleteness impedes the concept of perfect hedging.
In this paper, the incompleteness is caused by trading restrictions. The price process
of the benchmark index, i.e. the risky asset, is driven by a continuoustime process, while
trading is restricted to discrete time. Therefore, the eectiveness of the OBPI approach is
given by the eectiveness of a discretetime option hedge. The error of timediscretizing
a continuoustime hedging strategy for a put (or call) is extensively studied in the litera-
ture. Discretely adjusted option hedges, rst analyzed in Boyle and Emanuel (1980), are
also treated in Bertsimas, Kogan, and Lo (1998) and more recently in Mahayni (2003),
1
Optionbased portfolio insurance (OBPI) with synthetic puts is introduced in Leland and Rubinstein
(1976), constant proportion portfolio insurance (CPPI) in Black and Jones (1987). For the basic procedure
of the CPPI see also Merton (1971).
1
Talay and Zheng (2003) and Hayashi and Mykland (2005).
2
While the implications of
discretetime option hedges for portfolio protection are interesting in themselves, the main
focus of this paper is on the eects of timediscretizing the CPPI strategies which has,
to our knowledge, not been done yet. However, we keep in mind that the OBPI is one
alternative to the CPPI.
The optimality of an investment strategy depends on the risk prole of the investor.
In order to determine the optimal rule, one has to solve for the strategy which maxi-
mizes the expected utility. Thus, portfolio insurers can be modelled by utility maximizers
where the maximization problem is given under the additional constraint that the value
of the strategy is above a specied wealth level. Without postulating completeness, we
refer to the works of Cox and Huang (1989), Brennan and Schwartz (1989), Grossman
and Villa (1989), Grossman and Zhou (1993, 1996), Basak (1995), Cvitanic and Karatzas
(1995, 1999), Browne (1999), Tepla (2000, 2001) and El Karoui, Jeanblanc, and Lacoste
(2005). Mostly, the solution of the maximization problem is given by the unconstrained
problem including a put option. Obviously, this is in the spirit of the OPBI method. The
introduction of various sources of market incompleteness in terms of stochastic volatility
and trading restrictions makes the determination of an optimal investment rule under
minimum wealth constraints quite dicult if not impossible. For example, if the payo
of a put (or call) option is not attainable, the OBPI approach is not a viable method
in the above setup.
3
Another problem is posed by model risk. This is generated by the
possible inconsistency between the unknown true model and the model the risk man-
ager relies on in order to determine the hedging strategy. That is, one has to use some
(educated) assumptions about the data-generating processes. However, strategies which
are based on an optimality criterion with respect to one particular model, fail to be op-
timal if the true asset price dynamics deviate from the assumed ones. Summing up,
one alternative to the maximization approach, either based on utility or other optimality
criteria, is given by a more general analysis of robustness properties of a stylized strategy.
4
For the reasons given above, we follow an approach where the analysis is already based on
stylized portfolio strategies, i.e. we take the CPPI rule as given. Because of its simplicity
and the possibility to customize it to the preferences of an investor, the CPPI has become
very popular with practitioners.
In Black and Perold (1992), it is shown that in a complete market, the CPPI can be char-
acterized as expected utility maximizing when the utility function is piecewise HARA and
the guaranteed level is growing with the riskless interest rate. Obviously, this argument
loses it validity if an additional incompleteness is introduced by trading restrictions. The
properties of continuoustime CPPI strategies are studied extensively in the literature,
c.f. Bookstaber and Langsam (2000) or Black and Perold (1992). A comparison of OBPI
2
Transaction costs can naturally explain the reason for discretetime hedging. The implication of
transaction costs, conducted by Leland (1985), is studied in Bensaid, Lesne, and Scheinkman (1992),
Boyle and Vorst (1992), Avellaneda and Paras (1994), Grannan and Swindle (1996) and Toft (1996).
3
Hedging strategies in incomplete markets depend on some dynamic risk measure that has to be
minimized. For a discussion, see e.g. Schweizer (2001).
4
With respect to the robustness of option hedges we refer the reader to Avellaneda, Levy, and Paras
(1995), Lyons (1995), Bergman, Grundy, and Wiener (1996), El Karoui, Jeanblanc-Picque, and Shreve
(1998), Hobson (1998), Dudenhausen, Schlogl, and Schlogl (1998) and Mahayni (2003).
2
and CPPI (in continuous time) is given in Bertrand and Prigent (2002a). An empirical
investigation of both methods can, for example, be found in Do (2002) who simulates
the performance of these strategies using Australian data. The literature also deals with
the eects of jump processes, stochastic volatility models and extreme value approaches
on the CPPI method, c.f. Bertrand and Prigent (2002b), Bertrand and Prigent (2003).
In contrast to this, we consider the risk resulting from trading restrictions. We propose
a discretetime version of a simple CPPI strategy which satises three conditions. The
strategy is selfnancing, the asset exposure is nonnegative and the value process con-
verges. Assuming that the underlying price process is given by a geometric Brownian
motion, trading restrictions in the sense of discretetime trading are sucient to model
the possibility of a oor violation. The advantage of a model setup along the lines of
Black and Scholes (1973) is that risk measures, such as the shortfall probability and the
expected shortfall which are implied by the discretetime CPPI method can be given in
closed form. It is shown that the same holds for the price of the gap risk. Of course, this
is only possible because of the proportional structure of the CPPI strategies. A CPPI
investor species two parameters, a constant multiplier and the oor (or guarantee). Then
the amount which is invested in a risky asset is determined by the product of the multiplier
and the excess of the portfolio value over the oor. The remaining part, i.e. the dierence
of the portfolio value and the asset exposure is invested in a riskless asset. This implies
that the strategy is selfnancing. If the price process of the risky asset does not permit
jumps, the continuoustime application of the CPPI ensures that the portfolio value does
not fall below the oor. The strategy outperforms the prescribed oor unless there is a
sudden drop in market prices such that the investor is not able to rebalance his portfolio
adequately. With respect to trading restrictions, the eectiveness of the discretetime
strategies should improve with an increasing trading frequency which, of course, is also
true for a discretetime option hedge. However, a synthetic put can only be represented
by a stochastic multiplier. Intuitively, this explains why the risk measures can be given
in closed form in the case of a discretetime CPPI but not in the case of a discretetime
option based strategy.
In our setup, once the risk measures are determined, the gap risk can be priced eas-
ily. However, the main focus is not the pricing. Instead, the relevant risk measures are
used to discuss criteria which must be satised such that the CPPI strategy is still ef-
fective if applied in discrete time.
5
For example, it turns out that for a small number of
rehedges, the shortfall probability, i.e. the probability that the strategy falls below the
oor at the terminal date, may as well rst increase in the trading frequency before it
decreases. However, after a critical number of rehedges, the shortfall probability is always
decreasing in the number of rehedges. The change in monotonicity can be interpreted in
terms of a minimal number of rehedges which is necessary such that a portfolio protection
can be achieved by applying the CPPI technique in discrete time. Obviously, the critical
number of rehedges depends on the model parameters, the drift and the volatility, and the
strategy parameters, in particular the multiplier. The same is true for the risk measures.
We discuss criteria which ensure that the CPPI method is eective in a discretetime
5
It is worth mentioning that while arbitrage free pricing is based on the expectation under the mar-
tingale measure, the risk measures must be determined with respect to the real world measure.
3
setup. If the volatility is known, it is, for example, possible to specify the strategy pa-
rameters of a CPPI, i.e. the multiplier and the oor, such that the shortfall probability
is bounded above by a condence level. The eects of model and strategy parameters on
the risk measures are illustrated by examples.
The outline of the paper is as follows. Sec. 2 gives the model setup, motivates the CPPI
method and reviews the structure and the properties of continuoustime CPPI strate-
gies. A discretetime version of a CPPI strategy where the asset exposure is restricted
to be nonnegative is dened in Sec. 3. The properties of the discretetime version are
derived in analogy to the continuoustime version. The assumption that the asset price
increments are independent and identically distributed yields a closedform solution for
the shortfall probability and the expected shortfall. The calculations are given in Sec.
4 which also includes a sensitivity analysis of the risk measures with respect to model
and strategy parameters. Besides, it is shown that the value process of the discretetime
version converges to the value process of the continuoustime strategy in distribution if
the trading restrictions vanish. Sec. 5 illustrates the results and discusses criteria which
ensure that the discretetime strategy is eective, i.e. the portfolio protection is still valid
in discrete time. Sec. 6 concludes the paper.
2. Model Setup
All stochastic processes are dened on a stochastic basis (, F, (F
t
)
t[0,T

]
, P) which sat-
ises the usual hypotheses.
We consider two investment possibilities: a risky asset S and a riskless bond B which
grows with constant interest rate r, i.e. dB
t
= B
t
r dt where B
0
= b. The evolution of the
risky asset S, a stock or benchmark index, is given by a geometric Brownian motion, i.e.
d S
t
= S
t
(dt + dW
t
) , S
0
= s, (1)
where W = (W
t
)
0tT
denotes a standard Brownian motion with respect to the real
world measure P. and are constants and we assume that > r 0 and >
0. A continuoustime investment strategy or saving plan for the interval [0, T] can be
represented by a predictable process (
t
)
0tT
.
t
denotes the fraction of the portfolio
value at time t which is invested in the risky asset S. If there are no additional borrowing
restrictions, we can, w.l.o.g., restrict ourselves to strategies which are selfnancing, i.e.
strategies where money is neither injected nor withdrawn during the trading period ]0, T[.
Thus, the amount which is invested at date t in the riskless bond B is given in terms of
the fraction 1
t
. V = (V
t
)
0tT
denotes the portfolio value process which is associated
with the strategy , i.e. V
t
is the solution of
dV
t
() = V
t
_

t
dS
t
S
t
+ (1
t
)
dB
t
B
t
_
, where V
0
= x. (2)
Notice that there are alternative possibilities for portfolio insurance. Let T denote the
terminal trading date. For example, one might think of T as the retirement day. The
minimal wealth which must be obtained is denoted by G. The guaranteed amount is
assumed to be less than the terminal value of a pure bond investment, i.e. we assume
G < e
rT
V
0
. Besides a pure bond investment, a trivial possibility is given by a static
trading strategy where at the initial time t = 0 the present value of the guarantee, i.e.
4
Ge
rT
is invested in the bond B and the remaining part, i.e. the surplus V
0
e
rT
G, is
invested in the risky asset S. Thus, although
t
=
(V
0
e
rT
G)
V
t
S
t
S
0
is stochastic, the strategy
is static in the sense that there are no rebalancing decisions involved during the interval
]0, T]. Abstracting from stochastic interest rates, the above strategy honors the guarantee
G independent of the stochastic process generating the asset prices. Another example of
portfolio insurance is given by a stoplossstrategy which is represented by a portfolio
fraction
t
= 1
{
V
t
>e
r(Tt)
G
}
. Here, everything is invested in the asset until the cush-
ion (or surplus) V
t
e
r(Tt)
G is exhausted. This means that the strategy is eective
with respect to the guarantee if continuoustime monitoring (trading) is possible and the
asset price process does not permit jumps. Together, the above strategies can be used
to explain the basic idea of the constant proportion portfolio insurance. A combination
of continuoustime monitoring and keeping the cushion under control yields the CPPI
approach.
However, in a complete market there is a second possibility, the option based portfo-
lio insurance approach. The completeness implies that there is a selfnancing and du-
plicating strategy in S and B for any claim with payo h(S
T
) at T. Notice that for
h(S
T
) =
_
S
T
+
_
G

S
T

+
_
= G +
_
S
T

G

+
and > 0 it holds h(S
T
) G. Buy-
ing assets and putoptions with strike
G

enables a portfolio insurance, too.

6
If the
associated options are not traded, they must be synthesized by a hedging strategy in S
and B. If the concept of perfect hedging is impeded by market incompleteness, the OBPI
and the CPPI can both violate the purpose of portfolio insurance. In terms of model
risk, i.e. the problem that one does not know which process can describe the true data
generating process adequately, the OBPI approach causes more problems than the CPPI
technique. The composition of the CPPI strategy is model independent. In contrast to
this, it is necessary to incorporate a volatility guess in order to implement the OBPI
approach with synthetic options. Thus, there is an additional error introduced by using
the wrong hedging model.
In the following, we concentrate on the CPPI approach. It is worth mentioning that
even without an utility based justication, the CPPI is an important strategy in prac-
tice.
7
We x the notation and review the basic form and properties of continuoustime
CPPI strategies. Recall that the basic idea of the CPPI approach is to invest the amount
of portfolio value which is above the present value of the guarantee in the risky asset
S. Normally, the symbol F is used to denote the present value of the guarantee G, i.e.
F
t
:= exp {r(T t)} G. This is equivalent to
dF
t
= F
t
r dt with F
0
= exp {rT} G.
The surplus is called cushion C, i.e. C
t
:= V
t
F
t
. If the cushion is monitored in
continuous time, it is even possible to invest a multiple of the cushion in the risky asset.
6
Or buying call options with strike
G

and a riskless investment of Ge

rT
.
7
Besides the importance of CPPI strategies in the context of hedge funds, the CPPI technique has
recently been extended to the credit derivatives market, c.f. Fletcher (2005). ABN Amro created the
rst credit CPPI product in April 2004. It is called Rente Booster.
5
Let m denote the multiplier, then the fraction of a CPPI strategy is given by
8

t
:=
mC
t
V
t
.
Notice that there are various extensions to the CPPI. For example, besides borrowing
constraints it is also possible to permit only a maximal fraction of wealth to be invested
in the risky asset. Furthermore, one might think of a oor adjustment which allows to
protect the gains in the case of a favorable asset performance. However, all these exten-
sions prohibit closedform solutions. For this reason, we call a continuoustime CPPI
strategy which satises the above form simple. Notice that a simple CPPIstrategy is
given in terms of the guarantee G and the multiplier m 0. Normally, a CPPI even
implies that m 2. In addition to the protection feature this ensures that the value of
the CPPI strategy is convex in the asset price, at least in a continuoustime setup with
continuous asset paths. Throughout the paper, the guarantee is given exogenously, i.e. it
is the minimal value of wealth which is needed at T. In contrast to the OBPI approach,
the CPPI includes an additional degree of freedom which is introduced by the multiplier
m. While CPPI strategies are protective with respect to the guarantee for all m 0, this
is not true if trading is restricted to discrete time. Heuristically, this is easily explained
by the static case where a protection is only possible for m 1.
For the sake of completeness, we review some basic properties of the continuoustime
CPPI technique. First, consider the cushion process (C
t
)
0tT
.
Lemma 2.1. If the asset price dynamic is lognormal, i.e. if it satises Equation (1), the
cushion process (C
t
)
0tT
of a simple CPPI is lognormal, too. In particular, it holds
dC
t
= C
t
((r + m( r) dt + mdW
t
) .
Proof: Notice that C
t
:= V
t
F
t
implies
dC
t
= d (V
t
F
t
)
= V
t
_
mC
t
V
t
dS
t
S
t
+
_
1
mC
t
V
t
_
dB
t
B
t
_
F
t
dB
t
B
t
= C
t
_
m
dS
t
S
t
(m1)r dt
_
.
The rest of the proof follows with Equation (1).
Proposition 2.2. The tvalue of the a simple CPPI with parameter m and G is
V
t
= Ge
r(Tt)
+
V
0
Ge
rT
S
m
0
exp
__
r m
_
r
1
2

2
_
m
2

2
2
_
t
_
S
m
t
. (3)
Proof: Notice that Equation (3) can also be represented as follows
V
t
= F
t
+
V
0
F
0
S
m
0
exp
__
r m
_
r
1
2

2
_
m
2

2
2
_
t
_
S
m
t
.
8
For simplicity, we abstract from borrowing constraints which can be modelled by
t
=
min{m(V
t
F
t
),pV
t
}
V
t
with p 0.
6
The proof of this equation is well known, c.f. for example Bertrand and Prigent (2002a).
Together with
S
t
= S
0
e
(

1
2

2
)
t+W
t
it follows that
V
t
F
t
= (V
0
F
0
)e
(
r+m(r)
1
2
m
2

2
)
t+mW
t
(4)
which matches the result of Lemma 2.1.
Equation (3) illustrates the basic property of a simple CPPI. The tvalue of the strategy
consists of the present value of the guarantee G, i.e. the oor at t, and a nonnegative
part which is proportional to
_
S
t
S
0
_
m
. Thus, the value process of a simple CPPI strategy is
path independent.
9
The payo above the guarantee is linear for m = 1 and it is convex for
m 2. In nancial terms, the payo of a CPPI strategy with m 2 can be interpreted as
a power claim. The portfolio protection is ecient with probability one, i.e. the terminal
value of the strategy is higher than the guarantee with probability one.
The expected value and the variance of a simple CPPI are easily calculated as follows.
Lemma 2.3.
E [V
t
] = F
t
+ (V
0
F
0
) exp {(r + m( r)) t}
V ar [V
t
] = (V
0
F
0
)
2
exp {2 (r + m( r)) t}
_
exp
_
m
2

2
t
_
1
_
.
Proof: With Equation (4) it follows
E [ln (V
t
F
t
)] = ln (V
0
F
0
) +
_
r +m( r)
1
2
m
2

2
_
t
V ar [ln (V
t
F
t
)] =
2
m
2
t.
Notice that for X N (
X
,
X
) we have
E
_
e
X

= e

X
+
1
2

2
X
, V ar
_
e
X

= e
2
X
e

2
X
_
e

2
X
1
_
.
It is worth mentioning that the expected terminal value of a simple CPPI strategy is in-
dependent of the volatility . In contrast, the standard deviation increases exponentially
in the volatility of the asset S, c.f. Figure 1 and 2. Intuitively, this property explains
that the eectiveness of a CPPI strategy with respect to various sources of market incom-
pleteness does not only depend on the asset price drift but even more importantly on the
volatility of the underlying asset. In particular, this is the case for a rather high value of
the multiplier.
9
Notice that this is not true if one deviates from the concept of a simple CPPI.
7
Expectation and Standard Deviation of a simple CPPI
0 5 10 15 20
multiplier
1100
1200
1300
1400
e
x
p
e
c
t
a
t
i
o
n
Figure 1. Expected nal
value of a simple CPPI with
V
0
= 1000, G = 800, T = 1
and varying m for = 0.1,
= 0.1 and r = 0.05.
0 0.05 0.1 0.15 0.2 0.25 0.3
volatility
0
200
400
600
800
1000
1200
1400
s
t
a
n
d
a
r
d
d
e
v
i
a
t
i
o
n
Figure 2. Standard devia-
tion of the nal value of a sim-
ple CPPI with V
0
= 1000, G =
800, T = 1 and varying for
= 0.1, r = 0.05 and m = 2
(m = 4, m = 8 respectively).
3. Trading restrictions
We assume now that trading is restricted to a discrete set of dates and dene a discrete
time version of the simple CPPI strategy satisfying the following three conditions. Firstly,
the value process of the discretetime version converges in distribution to the value pro-
cess of the continuoustime simple CPPI strategy. Secondly, the discretetime version is a
selfnancing strategy. This means, that after the initial investment V
0
= x, there are no
in or outow of funds. Thirdly, the strategy does not allow for a negative asset exposure.
Notice that the rst condition implies that the cushion process of the discretetime ver-
sion converges to a lognormal process in distribution. However, the cushion process with
respect to a discretetime set of trading dates may also be negative. Therefore, to avoid
a negative asset exposure, this must be captured by the denition of the discretetime
version.
Let
n
denote a sequence of equidistant renements of the interval [0, T], i.e.

n
=
_
t
n
0
= 0 < t
n
1
< < t
n
n1
< t
n
n
= T
_
,
where t
n
k+1
t
n
k
=
T
n
for k = 0, , n1. To simplify the notation, we drop the superscript
n and denote the set of trading dates with instead of
n
. The restriction that trading
is only possible immediately after t
k
implies that the number of shares held in the
risky asset is constant on the intervals ]t
i
, t
i+1
] for i = 0, . . . , n1. However, the fractions
of wealth which are invested in the assets change as assets prices uctuate. Thus, it is
necessary to consider the number of shares held in the risky asset and the number of
bonds , i.e. the tupel = (, ). With respect to the continuoustime simple CPPI
8
strategies, it holds

t
=

t
V
t
S
t
=
mC
t
S
t
,

t
=
(1
t
)V
t
B
t
=
V
t
mC
t
B
t
.
The following argumentation illustrates that a timediscretized strategy

which is de-
ned by

t
:=
t
k
for t ]t
k
, t
k+1
], k = 0, . . . , n 1
is in general not selfnancing. The value process V

:= V (; ) which is associated with

the discretetime version of , i.e. with

, is dened by V

0
:= V
0
and
V
t
(; ) :=
t
k
S
t
+
t
k
B
t
for t ]t
k
, t
k+1
]
= V
t
() (
t

t
k
)S
t
(
t

t
k
)B
t
for t ]t
k
, t
k+1
],
where
V
t
() :=
t
S
t
+
t
B
t
.
If is selfnancing, this is not necessarily true for

. Notice that

is selfnancing i

t
k
S
t
k+1
+
t
k
B
t
k+1
=
t
k+1
S
t
k+1
+
t
k+1
B
t
k+1
for all k = 0, . . . , n 1
V
t
k+1
(; ) = V
t
k+1
() for all k = 0, . . . , n 1.
Obviously, this is only true in the limit, i.e. for n . It is worth mentioning that it
is not even clear whether the above timediscretized version is meanselfnancing with
respect to the real world measure, c.f. for example Mahayni (2003). In order to specify a
meaningful discretetime version of a simple CPPI strategy, it is necessary to admit only
selfnancing strategies. This is equal to the condition that

t
=
1
B
t
k
_
V

t
k

t
S
t
k
_
for t ]t
k
, t
k+1
]. (5)
Finally, recall that constant proportion portfolio insurance means that the fraction of
wealth which is invested in the risky asset is given proportionally to the dierence of
the portfolio value and the oor, i.e. the cushion. Let C

denote the discretetime version

of the cushion process C, then
C

t
:= V

t
F
t
.
In addition, we do not allow for short positions in the risky asset, i.e. the asset exposure is
bounded below by zero. Thus, it is necessary to consider the positive part of the cushion.
The above reasoning gives the following denition.
Definition 3.1 (DiscreteTime CPPI). A strategy

= (

) where for t ]t
k
, t
k+1
]
and k = 0, . . . , n 1

t
:= max
_
m C

t
k
S
t
k
, 0
_

t
:=
1
B
t
k
_
V

t
k

t
S
t
k
_
9
is called simple discretetime CPPI.
Proposition 3.2 (Discretetime cushion process). Dene
t
s
:= min
_
t
k
|V

t
k
() F
t
k
0
_
with t
s
= if the minimum is not attained. It holds
V

t
k+1
F
t
k+1
= e
r(t
k+1
min{t
s
,t
k+1
})
_
V

t
0
F
t
0
_
min{s,k+1}

i=1
_
m
S
t
i
S
t
i1
(m1)e
r
T
n
_
.
Proof: Notice that
V

t
k+1
= max
_
mC

t
k
S
t
k
, 0
_
S
t
k+1
+
_
V

t
k
max
_
mC

t
k
S
t
k
, 0
_
S
t
k
_
B
t
k+1
B
t
k
=
_
_
_
F
t
k
B
t
k+1
B
t
k
+
_
V

t
k
F
t
k
_
_
m
S
t
k+1
S
t
k
(m1)
B
t
k+1
B
t
k
_
for V

t
k
F
t
k
> 0
V

t
k
B
t
k+1
B
t
k
for V

t
k
F
t
k
0.
Together with F
t
k
B
t
k+1
B
t
k
= F
t
k+1
it follows
V

t
k+1
F
t
k+1
=
_
_
V

t
k
F
t
k
_
_
m
S
t
k+1
S
t
k
(m1)e
r
T
n
_
for V

t
k
F
t
k
> 0
_
V

t
k
F
t
k
_
e
r
T
n
for V

t
k
F
t
k
0,
for all k = 0, . . . , n 1. In particular, we have
V

T
=
_
V

t
s
e
r(Tt
s
)
for t
s
t
n1
G
T
+
_
V

t
n1
F
t
n1
_
_
m
S
t
n
S
t
n1
(m1)
B
t
n
B
t
n1
_
for t
s
t
n
.
Notice that the value process V

converges in distribution to the value process V if the

trading restrictions vanish, i.e. if n . The proof is based on the convergence of
the corresponding expectation and variance. Therefore, it is postponed to the end of the
following section where we calculate the moments and risk measures of the discretetime
CPPI.
4. Risk Measures of DiscreteTime CPPI
Recall that the basic idea of a CPPI strategy is a portfolio protection. Heuristically, the
usage of these strategies is explained by an investor who wants to participate in bullish
markets but does not want the terminal value of the strategy to end up below a guar-
anteed amount G. Thus, the investor is completely risk averse for values below the oor
(or guarantee). As motivated in the previous sections, as soon as a source of market
incompleteness is considered, i.e. a restriction on the set of trading dates, the concept of
a perfect portfolio protection is impeded, in particular for dynamic strategies. With the
exception of static portfolio insurance strategies, there is a positive probability that the
terminal value is below the guaranteed amount. In particular, this is true for CPPI and
OBPI strategies which include a synthetic put. The use of such constrained strategies
or strategies which include a gap risk can be explained as follows. On the one hand,
one might think of an investor who accepts, because of market incompleteness, a strategy
which gives the guaranteed amount with a certain success probability. On the other hand,
10
one might think of retail products which are based on the CPPI method and are thus also
hedged by a CPPI strategy. Normally, the buyer of such a product gets the guaranteed
amount even in the case that the strategy fails to provide it. Here, the issuer takes the gap
risk and considers this in his product pricing. In both cases, the risk prole of the CPPI is
of great interest. It is necessary to compute risk measures which allow a characterization
if the constrained CPPI is still eective in terms of portfolio insurance.
In the following, we take the view of an investor who uses the CPPI as a savings plan with
portfolio protection. A CPPI strategy contradicts the original idea of the portfolio insur-
ance if it results in a very high gap risk, i.e. if the shortfall probability and the expected
shortfall are prohibitively high. The investor has to decide whether this additional risk is
not too high in terms of a portfolio insurance. In addition to the expected nal value and
its standard deviation, we consider the shortfall probability and the expected shortfall
given default as the risk measures which determine the eectiveness of the discretetime
CPPI strategy.
10
The shortfall probability is the probability that the nal value of the
discretetime CPPI strategy is less or equal to the guaranteed amount G. Intuitively,
one can also dene a local shortfall probability (given that no prior shortfall happened
before). Additionally, we use the expected shortfall given default to describe the amount
which is lost if a shortfall occurs.
Definition 4.1 (Risk measures).
P
SF
:= P (V

T
G) = P (V

T
F
T
) shortfall probability
P
LSF
t
i
,t
i+1
:= P
_
V

t
i+1
F
t
i+1
|V

t
i
> F
t
i
_
local shortfall probability
ES := E [GV

T
|V

T
G] expected shortfall given default.
It turns out that, in contrast to a discretetime option based strategy with synthetic put,
the calculation of the shortfall probability implied by a CPPI strategy is very simple.
This is easily explained if one observes that the shortfall event is equivalent to the event
that the stopping time which is dened in Proposition 3.2 is prior to the terminal date.
It is convenient to consider the following lemma.
Lemma 4.2. Let A
k
:=
_
S
t
k
S
t
k1
>
m1
m
e
r
T
n
_
for k = 1, . . . , n, then it holds
{t
s
> t
i
} =
i

j=1
A
j
and {t
s
= t
i
} = A
C
i

_
i1

j=1
A
j
_
for i = 1, . . . , n.
10
Notice that the shortfall probability is not a coherent risk measure, i.e. it is not subadditive. In
contrast, the expected shortfall given default is a coherent risk measure. We remain within the class of
stylized strategies, i.e. the CPPI strategies. Thus, it is in fact not a problem even if the eectiveness of
the strategies is analyzed by using a risk measure which is not subadditive. For details on coherent risk
measures we refer to the work of Artzner, Delbaen, Eber, and Heath (1999).
11
Shortfall probability
0 10 20 30 40 50
number of rehedges
0
0.05
0.1
0.15
0.2
0.25
0.3
S
F
P
Figure 3. V
0
= 1000, G
T
=
1000, m = 12 (15 and 18 re-
spectively), = 0.085, r =
0.05 and = 0.1.
0 10 20 30 40 50
number of rehedges
0.6
0.7
0.8
0.9
1
S
F
P
Figure 4. V
0
= 1000, G
T
=
1000, m = 12 (15 and 18 re-
spectively), = 0.085, r =
0.05 and = 0.3.
Proof: According to the poof of Proposition 3.2 it holds
V

t
k+1
F
t
k+1
=
_
_
V

t
k
F
t
k
_
_
m
S
t
k+1
S
t
k
(m1)e
r
T
n
_
for V

t
k
F
t
k
> 0
_
V

t
k
F
t
k
_
e
r
T
n
for V

t
k
F
t
k
0.
The rest of the proof follows immediately with the denition of the stopping time t
s
and
m
S
t
k+1
S
t
k
(m1)e
r
T
n
> 0
S
t
k+1
S
t
k
>
m1
m
e
r
T
n
.
Lemma 4.3. The local shortfall probability is independent of t
i
and t
i+1
, i.e.
P
LSF
t
i
,t
i+1
= P
LSF
= N (d
2
) (6)
where d
2
:=
ln
m
m1
+ ( r)
T
n

1
2

2 T
n

_
T
n
. (7)
Proof: Notice that
P
LSF
t
i
,t
i+1
= P
_
V

t
i+1
F
t
i+1
|V

t
i
> F
t
i
_
= P (t
s
= t
i+1
|t
s
> t
i
) = P
_
S
t
1
S
t
0

m1
m
e
r
T
n
_
,
where the last equality follows with Lemma 4.2 and the assumption that the asset price
increments are independent and identically distributed (iid).
Proposition 4.4. The shortfall probability P
SF
is given in terms of the local shortfall
probability P
LSF
, i.e.
P
SF
= 1
_
1 P
LSF
_
n
.
Proof: The above lemma is a direct consequence of Lemma 4.2 and the independence
of asset price increments.
P
SF
= 1 P(t
s
= ) = 1
_
1 P
LSF
_
n
.
12
It can be shown, c.f. Lemma C.1 of the appendix, that the shortfall probability converges
to zero if we approach continuoustime trading, i.e. lim
n
P
SF
= 0. At rst glance, it
might be tempting to think that the shortfall probability is monotonically decreasing in
the hedging frequency, i.e. the number of rehedges n. In general, this is only true after
a suciently high n is reached. The eect that the shortfall probability is increasing for
small n is more pronounced for high volatilities and high multipliers, c.f. Figure 3 and
Figure 4.
11
Let n

denote the number of rehedges such that the shortfall probability is

increasing in n for n n

and decreasing for n n

. The critical level n

is to be inter-
preted as a minimal number of rehedges which is necessary such that the CPPI method
is eective for m 2 in discrete time. Consider for example a guaranteed amount G
given by G = e
rT m1
m
V
0
such that
0
= 1, i.e. the initial exposure in the risky asset
coincides with the initial investment. If in addition n is chosen to be one, i.e. there is no
rehedge until T, the discretetime CPPI strategy coincides with a pure asset investment.
Obviously, the CPPI method can not be eective for n = 1, i.e. a pure asset investment is
not in the spirit of the CPPI method. Thus, it is intuitively clear that a minimal number
of rehedges becomes necessary such that the CPPI method applies if trading is restricted
to discrete time. The critical level n

and its implications are further discussed in Sec. 5

where the eectiveness of the discretetime CPPI method is studied in detail.
If a shortfall is possible, one should also consider the amount of the shortfall or a risk mea-
sure which describes the amount of the shortfall. One possibility is given by the expected
shortfall ES which is introduced in Denition 4.1. It turns out that in order to determine
the expected shortfall, it is convenient to decompose the expected terminal payo into
two parts. One part is given by the expected terminal value if a shortfall occurs and the
other by the expectation on the set where the terminal value is above the guarantee.
Proposition 4.5 (Expected nal value). It holds
E [V

T
] = G + (V
0
F
0
)
_
E
n
1
+e
r
T
n
E
2
e
rT
E
n
1
1 E
1
e
r
T
n
_
where E
1
:= me

T
n
N (d
1
) e
r
T
n
(m1)N (d
2
)
E
2
:= e
r
T
n
_
1 + m
_
e
(r)
T
n
1
__
E
1
.
d
2
is the same as in Lemma 4.3 and d
1
:= d
2
+
_
T
n
.
Proof: Notice that
E [V

T
] = E
_
V

T
1
{t
s
=}

+E
_
V

T
1
{t
s
t
n
}

.
With Lemma 4.2 and Lemma A.1 of Appendix A it follows
E
_
V

T
1
{t
s
=}

= E
_
F
T
n

i=1
1
_
S
t
i
S
t
i1
>
m1
m
e
r
T
n
_
_
+ E
_
(V

T
F
T
)
n

i=1
1
_
S
t
i
S
t
i1
>
m1
m
e
r
T
n
_
_
= G P (t
s
= ) + (V
0
F
0
) E
n
1
= G
_
1 P
SF
_
+ (V
0
F
0
) E
n
1
.
11
It is straightforward to show that the shortfall probability is monotonically increasing in m and .
13
For the second expectation, observe that
E
_
V

T
1
{t
s
t
n
}

=
n

i=1
E
_
V

T
1
{t
s
=t
i
}

.
The remaining part of the proof follows with Lemma A.2 of Appendix A and
n

i=1
e
r(Tt
i
)
_
F
t
i
P
LSF
(1 P
LSF
)
i1
+ E
2
E
i1
1
(V
0
F
0
)
_
= G P
SF
+ (V
0
F
0
)e
r
T
n
E
2
e
rT
E
n
1
1 E
1
e
r
T
n
.
The calculation of the expected shortfall ES is now straightforward.
12
Corollary 4.6 (Expected Shortfall). The expected shortfall ES which is dened as in
Denition 4.1 is given by
ES = G
(V
0
F
0
)e
r
T
n
E
2
e
rT
E
n
1
1E
1
e
r
T
n
P
SF
.
Proof: According to the denition, it holds
ES = E [GV

T
|t
s
< ] = G
E
_
V

T
1
{t
s
t
n
}

P
SF
.
The proof is completed by inserting the result given in the proof of Proposition 4.5.
Proposition 4.7 (Variance of nal value). It holds
V ar [V

T
] = (V
0
F
0
)
2
_

E
n
1
+ e
2r
T
n

E
2
e
2rT

E
n
1
1 e
2r
T
n

E
1
_
(E[V

T
] G)
2
where

E
1
:= m
2
e
(2+
2
)
T
n
N (d
3
) 2m(m1)e
(+r)
T
n
N (d
1
) + (m1)
2
e
2r
T
n
N (d
2
) ,

E
2
:= m
2
e
(2+
2
)
T
n
2m(m1)e
(+r)
T
n
+ (m1)
2
e
2r
T
n

E
1
.
d
1
, d
2
are dened as above and
d
3
:=
2 ln
m
m1
+ 2( r)
T
n
+ 3
2 T
n
4
_
T
n
.
Proof: Notice that
V ar [V

T
] = V ar [V

T
F
T
] = E
_
(V

T
F
T
)
2

(E [V

T
] F
T
)
2
where E
_
(V

T
F
T
)
2

= E[(V

T
F
T
)
2
1
{t
s
=}
] +
n

i=1
E
_
(V

T
F
T
)
2
1
{t
s
=t
i
}

.
12
The same is true for the price of the associated gap risk, i.e. the price of an option where the
payo at T is given by (GV

T
)
+
. Notice that by standard nancial theory, the t
0
price is given by the
expected value of the discounted payo under the martingale measure. However, the risk measures which
are considered here must be given with respect to the real world measure.
14
Sensitivity of risk measures
Risk measures Strategy parameter Model parameter
G m
Mean
Stdv.
P
SF

ESF
Table 1. Sensitivity analysis of risk measures. We use the symbol for mono-
tonically increasing and for monotonically decreasing.
Analogously to the proof of Proposition 4.5, it follows with Lemma B.1 and Lemma B.2
of the appendix that
E
_
(V

T
F
T
)
2

= (V
0
F
0
)
2

E
n
1
+
n

i=1
(V
0
F
0
)
2
e
2r(Tt
i
)

E
2

E
i1
1
.
The remaining part of the proof follows with
n

i=1
e
2r(Tt
i
)

E
i1
1
= e
2r
T
n
e
2rT

E
n
1
1 e
2r
T
n

E
1
.
The calculation of the expectation and variance of the discretetime CPPI strategy can
now be used to prove the convergence, i.e.
Proposition 4.8 (Convergence). For n , the value process V

converges to the
value process V in distribution, i.e.
V

L
V.
In particular, it holds
lim
n
E [V

T
] = G + (V
0
F
0
) exp {(r + m( r)) T}
lim
n
V ar [V

T
] = (V
0
F
0
)
2
exp {2 (r + m( r)) T}
_
exp
_
m
2

2
T
_
1
_
.
Proof: The proof is given in Appendix C.
Before we study the eectiveness of the timediscretized CPPI in detail, we end this sec-
tion with a sensitivity analysis of the risk measures. In order to avoid a lengthy discussion
of all possible sensitivities, we summarize the main results in Table 1. The corresponding
proofs are straightforward. Notice that the shortfall probability is independent of G, c.f.
Proposition 4.4. Partial dierentiation immediately yields that the shortfall probability
is increasing in and m but decreasing in . In contrast, the sensitivity analysis of the
other risk measures is tedious. For example, the monotonicity of the expected terminal
value, i.e. E[V

T
], in is shown in Appendix D. Similar arguments to the ones presented
here can also be used to show that the expected terminal payo is also increasing in and
m. Monotonicity in G and V is immanent. With respect to the standard deviation, it is
intuitively clear that the the volatility has a positive eect on the standard deviation,
so does m. It is worth mentioning that both the shortfall probability and the expected
15
Moments of continuoustime CPPI
multiplier m Mean Stdv. Dev.
12 1078.03 (1078.03) 140.04 (1387.90)
15 1086.67 (1086.67) 252.51 (7801.45)
18 1096.27 (1096.27) 476.83 (62763.30)
Table 2. Moments.
shortfall are increasing in m and . This implies that a discretetime CPPI is not eective
in discrete time if either the standard deviation is too high in comparison to the multiplier
or vice versa.
5. Effectiveness of the discretetime CPPI method
As shown above, the eectiveness of the discretetime CPPI method depends on the strat-
egy parameters, i.e. the multiplier m, the number of rehedges n and the guarantee G,
as well as the model parameters and . The most important inuences are caused by
the multiplier m and the volatility . Therefore, all examples are considered for varying
multipliers and volatilities. If not mentioned otherwise, we consider a model scenario
where = 0.085, = 0.1 (0.2 or 0.3, respectively) and r = 0.05. The time to maturity
of the CPPI strategy is equal to one year (T = 1), the initial investment coincides with
the guarantee, i.e. V
0
= G = 1000. Thus, the goal of the strategies under consideration
is to ensure 100% of the initial capital. This is in accordance to guaranteed fund man-
agement.
13
For the multiplier m we consider the values 12, 15 and 18. Here, the initial
asset exposure m
_
V
0
e
rT
G
_
is 585.247 for m = 12, 731.559 for m = 15 and 877.870 for
m = 18 such that the relative initial asset investment varies between 0.585 and 0.88. A
high multiplier is convenient in order to emphasize all eects and to highlight the eect
of a small change in volatility.
First, we consider the question whether the discretetime CPPI method gives a good
approximation of the continuoustime CPPI for a nite number of rehedges n. Recall
that the value process of the discretetime CPPI converges to the value process of the
continuoustime CPPI in distribution, c.f. Proposition 4.8. Since the cushion process of
the continuoustime CPPI is lognormal, the payo distribution of the continuoustime
CPPI is described by its mean and its standard deviation. These numbers are summa-
rized in Table 2. In comparison, Table 3 summarizes the moments and risk measures for
various numbers of rehedges n.
Now consider the shortfall probability. Observe, that in the case where = 0.1, a monthly
CPPIstrategy (n = 12) with a multiplier m = 12 implies a shortfall probability of only
0.01. In contrast, a volatility of = 0.2 gives a shortfall probability of more than 0.5.
Thus, the monthly CPPI strategy ensures a signicant protection level for = 0.1 while
13
It is worth mentioning that the probability that the CPPI portfolio value is higher than the OBPI
value increases in the percentage of the insured initial investment, c.f. Bertrand and Prigent (2003).
Recall that V
OBPI
T
= G+[S
T
G]
+
. Thus, the above eect is intuitively explained by observing that the
probability of exercising the embedded call option is decreasing in the strike.
16
Moments of discretetime CPPI
n m Mean Stdv. Dev. SFP ESF
12 12 1077.53 (1080.23) 125.04 (703.03) 0.0115 (0.5430) 5.463 (25.933)
24 12 1077.77 (1078.60) 132.01 (948.79) 0.0002 (0.3195) 2.981 (12.296)
48 12 1077.90 (1077.98) 135.88 (1133.36) 0.0000 (0.0580) 1.574 (5.802)
96 12 1077.97 (1077.97) 137.92 (1249.06) 0.0000 (0.0009) 0.000 (3.037)
n m Mean Stdv. Dev. SFP ESF
12 15 1085.94 (1074.28) 206.30 (1874.59) 0.0767 (0.7592) 8.901 (57.01)
24 15 1086.22 (1090.92) 226.81 (3361.17) 0.0069 (0.6610) 4.836 (27.86)
48 15 1086.44 (1087.43) 238.86 (4936.18) 0.0000 (0.3258) 2.597 (11.03)
96 15 1086.56 (1086.60) 245.46 (6130.89) 0.0000 (0.0333) 1.364 (5.02)
n m Mean Stdv. Dev. SFP ESF
12 18 1095.70 (1120.63) 339.07 (4924.65) 0.2094 (0.8691) 13.911 (118.32)
24 18 1095.65 (1111.58) 396.37 (12759.40) 0.0494 (0.8593) 7.296 (64.66)
48 18 1095.90 (1101.08) 432.75 (25691.30) 0.0015 (0.6767) 3.908 (23.70)
96 18 1096.08 (1096.68) 453.66 (39053.60) 0.0000 (0.2131) 2.067 (8.30)
Table 3. The time horizon is T = 1 year and the guarantee G is equal to the
initial investment V
t
0
= 1000. The model parameters are given by = 0.085,
r = 0.05 and = 0.1 ( = 0.2 respectively).
the concept of portfolio insurance is already impeded for = 0.2. Here (for = 0.2),
even a weekly rehedging, i.e. n = 48 is not enough to achieve a shortfall probability of less
than 0.05. This illustrates that the eectiveness of the discretetime CPPI method is very
sensitive to the volatility of the asset price process. Besides, the higher the multiplier,
the more pronounced the eect is. For example, notice that the shortfall probability for
a CPPIstrategy with n = 24 and m = 18 is 0.049 for = 0.1 but 0.86 for = 0.2.
Recall that the shortfall probability is not necessarily monotonically decreasing in the
number of rehedges. A very large shortfall probability implies that the number of re-
hedges is still too low to achieve an eective portfolio protection. For example, one might
think of the extreme case that n = 1, i.e. the case where the portfolio is held constantly
on the trading period [0, T]. Obviously, a portfolio protection can only be achieved if only
the surplus is invested in the risky asset. One can argue that the CPPI method is not
eective if the number of rehedges n is still in a region where the shortfall probability is
increasing in n. Thus, it is convenient to determine the minimal number n

such that
an increase in the number of portfolio rebalancing dates is able to reduce the shortfall
probability. For dierent combinations of and m, the critical number n

is illustrated
in Table 4.
14
However, n

can only be used as a number which is at least necessary to

achieve an eective portfolio insurance.
One solution to ensure the eectiveness of the discretetime CPPI method is given by
the possibility to determine the contract parameters such that the probability of falling
below the guarantee is bounded above by a condence level , for example = 0.99 (or
14
Compare also the remarks in the last section referring to Figure 3 and Figure 4.
17
Minimal number of rehedges
m n

m n

m n

12 0.1 2.00 15 0.1 3.08 18 0.1 4.40

12 0.2 7.00 15 0.2 11.09 18 0.2 16.11
12 0.3 15.35 15 0.3 24.44 18 0.3 35.64
Table 4. Minimal number n

of rehedges such that the shortfall probability is

decreasing in n.
= 0.95). This can be explained by an investor who is aware of market incompleteness
and accepts a small shortfall probability with respect to the guarantee. Again, we con-
sider the same model scenario where T = 1, = 0.085, r = 0.05, V
0
= G = 1000 and
distinguish between = 0.1 and = 0.2. For illustration, we determine (n, m)tupels
which give a shortfall probability of 0.01 and 0.05. The resulting values as well as the
corresponding other risk measures are given in Table 5. For example, observe that in
the case of = 0.1, the CPPI method with monthly rehedging and a multiplier of 11.84
ensures that the capital is maintained with a probability of 0.99. At the same time the
expected payo and the variance of the payo are similar to the ones obtained by a direct
investment in S, i.e. for the expectation compare 1077 to 1088 and for the standard de-
viation compare 121.75 to 109.14.
15
Therefore, in the case where = 0.1, even a monthly
rehedging is enough to give a high success probability if the multiplier is chosen appro-
priately.
16
However, in case of a volatility scenario where = 0.2, the multiplier is to be
chosen much more conservatively. Finally, it is worth mentioning that it is sucient to
control the shortfall probability if one also wants to control the expected shortfall which
is unarguably a more convincing risk measure. In the above example, it is approximately
the same if one keeps the shortfall probability on a 0.01 level or if one keeps the expected
shortfall at a level of 5.2.
6. conclusion
The introduction of market incompleteness and model risk impedes the concept of dynamic
portfolio insurance, i.e. the technique of constant portfolio insurance. The introduction
15
A direct investment of V
0
in the asset S gives for = 0.1 ( = 0.2 respectively)
E
_
V
0
S
T
S
0
_
= V
0
e
T
= 1088.72 (1088.72)

Var
_
V
0
S
T
S
0
_
= V
0
_
e
(2+
2
)T
e
2T
= 109.144 (219.939)
P
_
V
0
S
T
S
0
G
_
= 0.212 (0.373).
16
Again, it is worth mentioning that although a multiplier of approximately 12 seems to be fairly large,
it is to be interpreted in combination with the low volatility. In particular, a multiplier of m = 11.843
implies that for a guarantee G = V
0
= 1000 the initial amount invested in S is given by
V
0
= m(V
0
F
0
) = 11.843(1000 e
0.05
1000) = 577.59.
18
Risk prole for discretetime CPPI strategies with a shortfall probability of
0.01 (0,05).
= 0.1
n m Mean Stdv. ES
12 11.843 (14.124) 1077.118 (1083.377) 121.752 (178.420) 5.313 (7.770)
24 15.446 (18.024) 1087.558 (1095.730) 246.087 (398.225) 5.157 (7.319)
36 18.146 (20.956) 1096.273 (1106.154) 432.362 (774.426) 5.149 (7.217)
48 20.386 (23.389) 1104.150 (1115.646) 717.129 (1419.070) 5.186 (7.219)
60 22.336 (25.507) 1111.528 (1124.588) 1152.310 (2511.390) 5.243 (7.267)
= 0.2
n m Mean Stdv. ES
12 6.065 (7.152) 1063.302 (1065.747) 107.138 (150.350) 4.478 (6.432)
24 7.879 (9.128) 1067.464 (1070.485) 204.334 (316.650) 4.275 (5.931)
36 9.234 (10.605) 1070.748 (1074.241) 345.136 (591.266) 4.190 (5.720)
48 10.358 (11.829) 1073.591 (1077.500) 554.966 (1048.690) 4.145 (5.605)
60 11.335 (12.893) 1076.156 (1080.449) 868.650 (1804.760) 4.121 (5.535)
Table 5. For a given discretization in terms of n, the multiplier is determined
such that the implied shortfall probability is 0.01 (0.05 respectively).
of tradings restrictions is one possibility to model a gap risk in the sense that a CPPI
strategy can not be adjusted adequately. Measuring the risk that the value of a CPPI
strategy is less than the oor (or guaranteed amount) is of practical importance for at
least two reasons. On the one hand, CPPI strategies are common in hedge funds and
retail products. Often, a CPPI strategy is prespecied in the term sheet of hedge funds.
In addition, it is combined with a guarantee for the investor. Thus, an additional option
is written. The option is exercised if the value of the CPPI strategy is below the oor.
On the other hand, CPPI strategies can be used to protect return guarantees which are
embedded in unitlinked life insurance contracts. The terminal date T is interpreted as
the time of retirement and the guarantee is interpreted as the amount which is at least
needed by the insured. The assumption that the insurer wants to back up the guarantee
by a simple and discretetime investment strategy highlights some advantages in favor of
the CPPI method. Firstly, it is computationally very simple and it can easily be applied in
discrete time. Secondly, the composition of a CPPI strategy is independent of the model
assumption of the investor or insurer who might use a misspecied model. Thirdly, the
riskiness in terms of commonly used risk measures which is induced by trading restrictions
can be given in closed form. In particular, this is also true for the price of an additional
option which is normally included in CPPIbased products.
The analysis of the risk measures of a discretetime CPPI strategy poses various problems
which are to be considered. Basically, it is necessary to check the associated risk measures
and to determine whether the strategy is still eective in terms of portfolio protection.
For example, the protection feature is violated if the shortfall probability of the CPPI
strategy under consideration exceeds the shortfall probability of a pure asset investment.
Formally, the last one can be interpreted as a static CPPI. Intuitively, this explains the
19
result that the shortfall probability of a discretetime CPPI is only decreasing in the
hedge frequency after a suciently high number of rehedges. Below this critical number,
the shortfall probability increases such that additional adjustments of the strategy yield
a shortfall probability which is even higher than the one of a pure asset investment. This
eect is even more pronounced for high asset price volatilities and high multipliers. Thus,
if one restricts the set of admissible strategies to those strategies which satisfy a condence
level of protection, the choice of the CPPImultiplier is naturally restricted. A similar
reasoning is applied to other risk measures such as the expected shortfall.
Appendix A. Proof of Proposition 4.5
Lemma A.1. Let d
1
, d
2
and E
1
be dened as in Proposition 4.5, then it holds
E
_
_
V

t
i
F
t
i
_
i

j=1
1
_
S
t
j
S
t
j1
>
m1
m
e
r
T
n
_
_
= (V
0
F
0
) (E
1
)
i
for all i = 1, . . . , n.
Proof: The following calculations are based on Proposition 3.2, Lemma 4.2 and the
assumption that the asset price increments are independent and identically distributed
(iid).
E
_
_
V

t
i
F
t
i
_
i

j=1
1
_
S
t
j
S
t
j1
>
m1
m
e
r
T
n
_
_
= (V

0
F
0
) E
_
i

j=1
_
m
S
t
j
S
t
j1
(m1)e
r
T
n
_
1
_
S
t
j
S
t
j1
>
m1
m
e
r
T
n
_
_
= (V

0
F
0
)
i

j=1
E
_
_
m
S
t
j
S
t
j1
(m1)e
r
T
n
_
1
_
S
t
j
S
t
j1
>
m1
m
e
r
T
n
_
_
= (V

0
F
0
)
_
E
_
_
m
S
t
1
S
0
(m1)e
r
T
n
_
1
_
S
t
1
S
0

m1
m
e
r
T
n
_
__
i
.
Finally, notice that the last expectation matches the denition of E
1
, i.e. it holds
E
1
=
m
S
0
e

T
n
E
_
e

T
n
_
S
t
1

m1
m
S
0
e
r
T
n
_
+
_
= me

T
n
N (d
1
) e
r
T
n
(m1)N (d
2
) .
Lemma A.2. Let E
1
and E
2
be dened as in Proposition 4.5, then it holds
E
_
V

T
1
{t
s
=t
i
}

= G P
LSF
(1 P
LSF
)
i
+ e
r(Tt
i
)
E
2
E
i1
1
(V
0
F
0
).
Proof: With Proposition 3.2 and Lemma 4.2 it follows
E
_
V

T
1
{t
s
=t
i
}

= E
_
e
r(Tt
i
)
V

t
i
1
{
V

t
i
F
t
i
}
i1

j=1
1
_
V

t
j
>F
t
j
_
_
= e
r(Tt
i
)
E
_
V

t
i
1
_
S
t
i
S
t
i1

m1
m
e
r
T
n
_
i1

j=1
1
_
S
t
j
S
t
j1
>
m1
m
e
r
T
n
_
_
.
20
Notice that if there is no shortfall until t
i1
it holds, c.f. Proposition 3.2
V

t
i
= F
t
i
+
_
V

t
i1
F
t
i1
_
_
m
S
t
i
S
t
i1
(m1)e
r
T
n
_
such that
E
_
V

T
1
{t
s
=t
i
}

= e
r(Tt
i
)
E
_
F
t
i
1
_
S
t
i
S
t
i1

m1
m
e
r
T
n
_

i1
j=1
1
_
S
t
j
S
t
j1
>
m1
m
e
r
T
n
_
_
+
e
r(Tt
i
)
E
_
_
V

t
i1
F
t
i1
__
m
S
t
i
S
t
i1
(m1)e
r
T
n
_
1
_
S
t
i
S
t
i1

m1
m
e
r
T
n
_

i1
j=1
1
_
S
t
j
S
t
j1
>
m1
m
e
r
T
n
_
_
.
With Lemma A.1 and the assumption that the asset price increments are iid, it follows
E
_
V

T
1
{t
s
=t
i
}

= e
r(Tt
i
)
F
t
i
P
LSF
_
1 P
LSF
_
i1
+e
r(Tt
i
)
(V
0
F
0
) E
i1
1
E
_
_
m
S
t
1
S
0
(m1)e
r
T
n
_
1
_
S
t
1
S
0

m1
m
e
r
T
n
_
_
.
Finally, it is straightforward to check that the above expectation satises the denition
of E
2
.
Appendix B. Variance of DiscreteTime CPPI
Analogously to Appendix A, we consider the following two lemmas.
Lemma B.1. Let

E
1
be dened as in Proposition 4.7, then it holds
E
_
_
V

t
i
F
t
i
_
2
i

j=1
1
_
S
t
j
S
t
j1
>
m1
m
e
r
T
n
_
_
= (V
0
F
0
)
2
_

E
1
_
i
for all i = 1, . . . , n.
Proof: Analogously to the proof of Lemma A.1 it is straightforward to show that
E
_
_
V

t
i
F
t
i
_
2
i

j=1
1
_
S
t
j
S
t
j1
>
m1
m
e
r
T
n
_
_
= (V
0
F
0
)
2
_
E
_
_
m
S
t
1
S
0
(m1)e
r
T
n
_
2
1
_
S
t
1
S
0
>
m1
m
e
r
T
n
_
__
i
Lemma B.2. Let

E
1
and

E
2
be dened as in Proposition 4.7, then it holds
E
_
(V

T
F
T
)
2
1
{t
s
=t
i
}

= e
2r(Tt
i
)
(V
0
F
0
)
2

E
2

E
i1
1
.
Analogously to the proof of Lemma A.1, observe that
E[(V

T
F
T
)
2
1
{t
s
=t
i
}
] = e
2r(Tt
i
)
E
_
(V

t
i
F
t
i
)
2
1
{t
s
=t
i
}

.
21
An application of Lemma B.1 gives
E
_
(V

t
i
F
t
i
)
2
1
{t
s
=t
i
}

= E
_
(V

t
i
F
t
i
)
2
1
_
S
t
i
S
t
i1

m1
m
e
r
T
n
_
i1

j=1
1
_
S
t
j
S
t
j1
>
m1
m
e
r
T
n
_
_
= E
_
_
V

t
i1
F
t
i1
_
2
_
m
S
t
i
S
t
i1
(m1)e
r
T
n
_
2
1
_
S
t
i
S
t
i1

m1
m
e
r
T
n
_
i1

j=1
1
_
S
t
j
S
t
j1
>
m1
m
e
r
T
n
_
_
= E
_
_
m
S
t
1
S
0
(m1)e
r
T
n
_
2
1
_
S
t
1
S
0

m1
m
e
r
T
n
_
_
(V
0
F
0
)
2

E
i1
1
.
Finally, it is straightforward to check that the above expectation satises the denition
of

E
2
.
Appendix C. Convergence (Proof of Proposition 4.8)
First, we consider the convergence of the shortfall probability, the expected value and the
variance of the value.
Lemma C.1. It holds
lim
n
P
SF
= 0,
i.e. the shortfall probability converges to zero if the trading restrictions vanish.
Proof: Let f C
1
(R) such that lim
x
f(x) = 1. With
lim
x
(f(x))
x
= lim
x
_
1 +
x(f(x) 1)
x
_
x
= e
lim
x
x(f(x)1)
together with an application of LHopitals rule, i.e.
lim
x
x(f(x) 1) = lim
x
f(x) 1
x
1
= lim
x
x
2
f
x
(x),
it follows lim
x
(f(x))
x
= e
lim
x
x
2 f
x
(x)
if lim
x
x
2 f
x
(x) exists. Thus,
lim
n
P
SF
= lim
n
1 (1 P
LSF
)
n
= 1 lim
n
N(d
2
)
n
= 1 e
lim
n
n
2
N(d
2
)
n
= 1 e
lim
n
n
2
N

(d
2
)
d
2
n
.
The rest of the proof follows immediately with the denition of d
2
, c.f. Proposition 4.5,
and
lim
n
e
n
n
k
= 0 for all k N.
Lemma C.2. It holds
lim
n
E [V

T
] = G + (V
0
F
0
)e
(r+m(r))T
,
i.e. the expected value of the discretetime CPPI converges to the expected value of a
simple CPPI if the trading restrictions vanish.
22
Proof: According to Proposition 4.5 it holds
E[V

T
] = G + (V
0
F
0
)E
n
1
+ (V
0
F
0
)e
r
T
n
E
2
e
rT
E
n
1
1 E
1
e
r
T
n
. (8)
First, we consider the limit of E
n
1
. Using the denition of E
1
, c.f. Proposition 4.5, it is
straightforward to show that lim
n
E
1
= 1. According to the proof of Lemma C.1 it
holds
lim
n
E
n
1
= e
lim
n
n
2
E
1
n
where
n
2
E
1
n
= mTe

T
n
N(d
1
) (m1)rTe
r
T
n
N(d
2
)
me

T
n
n
2
N

(d
1
)
d
1
n
+ (m1)e
r
T
n
n
2
N

(d
2
)
d
2
n
.
Notice that the last two terms on the right-hand side vanish for n . Besides, with
the denitions of d
1
and d
2
, c.f. Proposition 4.5, it immediately follows
lim
n
N(d
1
) = 1, lim
n
N(d
2
) = 1
such that
lim
n
(V
0
F
0
)E
n
1
= (V
0
F
0
)e
mT(m1)rT
.
Thus, it is still to show that the last term of the right hand side of Equation (8) converges
to zero. Inserting E
2
according to its denition, c.f. Proposition 4.5, the relevant term is
(V
0
F
0
)(e
rT
E
n
1
)
_
_
1 +
m
_
e
(r)
T
n
1
_
1 E
1
e
r
T
n
_
_
.
Notice that lim
n
m
_
e
(r)
T
n
1
_
= 0 and lim
n
1 E
1
e
r
T
n
= 0. With the rule of
LHopital and similar arguments as above it follows
lim
n
m
_
e
(r)
T
n
1
_
1 E
1
e
r
T
n
= lim
n
_
_

m(r)T
n
2
e
(r)
T
n

E
1
rTe
r
T
n
n
2
e
r
T
n
E
1
n
_
_
=
m( r)T
rT (mT (m1)rT)
= 1.
Lemma C.3. It holds
lim
n
V ar [V

T
] = (V
0
F
0
)
2
e
2(r+m(r))T
(e
m
2

2
T
1),
i.e. the variance of the discretetime CPPI converges to the variance of a simple CPPI
if the trading restrictions vanish.
Proof: Recall that according to Proposition 4.7 it holds
V ar[V

T
] = (V
0
F
0
)
2

E
n
1
+ (V
0
F
0
)
2
e
2r
T
n

E
2
e
2rT

E
n
1
1 e
2r
T
n

E
1
(E[V

T
] G)
2
.
23
Analogously to the proof of Lemma C.2, it can be shown that

E
2
e
2rT

E
n
1
1e
2r
T
n

E
1
n
0. For
the convergence of

E
n
1
we use again that lim
n

E
n
1
= e
lim
n
n
2


E
1
n
. Notice that
n
2

E
1
n
= m
2
e
(2+
2
)
T
n
(2 +
2
)TN(d
3
) 2m(m1)e
(+r)
T
n
( +r)TN(d
1
)
+(m1)
2
e
2r
T
n
2rTN(d
2
) m
2
e
(2+
2
)
T
n
n
2
N

(d
3
)
d
3
n
+2m(m1)e
(+r)
T
n
n
2
N

(d
1
)
d
1
n
(m1)
2
e
2r
T
n
n
2
N

(d
2
)
d
2
n
.
Similar arguments to the ones given in the proofs of Lemma C.2 and Lemma C.1 imply
lim
n

E
n
1
= e
m
2
(2+
2
)T2m(m1)(+r)T+(m1)
2
2rT
= e
2(r+m(r))T
e
m
2

2
T
.
Finally, Lemma C.2 immediately gives
lim
n
(E [V

T
] G)
2
= (V
0
F
0
)
2
e
2(r+m(r))T
.
In order to prove Proposition 4.8 it is remains to show that, for n , the limiting
distribution of ln(V

T
) is a normal distribution . Let

n
:=
n

i=1
ln
_
m
S
t
i
S
t
i+1
(m1)e
r
T
n
_
. .
=:
i,n
.
With Lemma C.1, i.e. lim
n
P
SF
= 0, it follows that it is sucient to show that the
limiting distribution of
n
is a normal distribution. Applying the results for rowwise
independent arrays of Gnedenko and Kolmogorov (1954), c.f. in particular Theorem 1 in
Ch.5 26, it remains to show that
n

i=1
P(|
i,n
| > )
n
0 for all > 0.
Using the independency, one only needs to show that
nP(|
1,n
| > )
n
0 for all > 0.
This proof is straightforward, i.e. it is given by arguments which are similar to the ones
used in the proofs of the above lemmas.
Appendix D. Expected terminal value of discretetime CPPI is
monotonically increasing in the volatility
With Proposition 4.5 and the denition of E
2
it follows that
E[V

T
] = V
0
e
rT
+ (V
0
F
0
)m
_
e
(r)
T
n
1
_
e
rT
E
n
1
1 E
1
e
r
T
n
.
It is straightforward to show that E
1
> e
r
T
n
. For > r, the expected terminal value of
the discrete CPPI strategy is always larger than the investment in the riskless asset. This
24
is quite intuitive.
Now, consider the derivative with respect to , i.e.
E[V

T
]

= m(V
0
F
0
)
_
e
(r)
T
n
1
_ nE
n1
1
E
1

_
1 E
1
e
r
T
n
_

_
E
n
1
e
rT
_
E
1

e
r
T
n
_
1 E
1
e
r
T
n
_
2
.
For > r, the leading factors are positive. Besides, we have
E
1

> 0.
The proof of the above inequality is omitted here. In particular, analogous calculations as
for the determination of the socalled vega of a calloption price in a Black/Scholestype
model are needed. Finally, it is to show that
nE
n1
1
_
1 E
1
e
r
T
n
_

_
E
n
1
e
rT
_
e
r
T
n
0.
An application of Bernoullis inequality gives
nE
n1
1
_
E
1
e
r
T
n
1
_
e
r
T
n
_
E
n
1
e
rT
_
= e
r
T
n
_
nE
n1
1
_
E
1
e
r
T
n
_
E
n
1
+ E
n
1
_
1 +
e
r
T
n
E
1
E
1
_
n
_
e
r
T
n
_
nE
n1
1
_
E
1
e
r
T
n
_
E
n
1
+ E
n
1
_
1 + n
e
r
T
n
E
1
E
1
__
= 0.
Notice that because E
1
> e
r
T
n
, the above inequality is also strict.
25
References
Artzner, P., Delbaen, F., Eber, J., Heath, D., 1999. Coherent measures of risk. Mathe-
matical Finance 9, 203228.
Avellaneda, M., Levy, A., Paras, A., 1995. Pricing and Hedging Derivative Securities in
Markets with Uncertain Volatilities. Applied Mathematical Finance 2, 7388.
Avellaneda, M., Paras, A., 1994. Dynamic Hedging Portfolios for Derivative Securities in
the Presence of Large Transaction Costs. Applied Mathematical Finance 1, 165193.
Basak, S., 1995. A General Equilibrium Model of Portfolio Insurance. Review of Financial
Studies 8, 10591090.
Basak, S., 2002. A Comparative Study of Portfolio Insurance. The Journal of Economic
Dynamics and Control 26, 12171241.
Bensaid, B., Lesne, J., Scheinkman, J., 1992. Derivative Asset Pricing with Transaction
Costs. Mathematical Finance 2, 6386.
Bergman, Y., Grundy, B., Wiener, Z., 1996. General Properties of Option Prices. Journal
of Finance 51, 15731610.
Bertrand, P., Prigent, J.-L., 2002a. Portfolio Insurance Strategies: OBPI versus CPPI.
Discussion paper, GREQAM and Universite Montpellier1.
Bertrand, P., Prigent, J.-L., 2002b. Portfolio Insurance: The Extreme Value Approach to
the CPPI. Finance 23.
Bertrand, P., Prigent, J.-L., 2003. Portfolio Insurance Strategies: A Comparison of Stan-
dard Methods When the Volatility of the Stoch is Stochastic. International Journal of
Business 8, 1531.
Bertsimas, D., Kogan, L., Lo, A., 1998. When is Time Continuous. Discussion paper,
Massachusetts Institute of Technology (MIT).
Black, F., Jones, R., 1987. Simplifying Portfolio Insurance. The Journal of Portfolio Man-
agement 14, 4851.
Black, F., Perold, A., 1992. Theory of Constant Proportion Portfolio Insurance. The
Journal of Economic Dynamics and Control 16, 403426.
Black, F., Scholes, M., 1973. The Pricing of Options and Corporate Liabilities. Journal
of Political Economics 81, 637654.
Bookstaber, R., Langsam, J., 2000. Portfolio Insurance Trading Rules. The Journal of
Futures Markets 8, 1531.
Boyle, P., Emanuel, D., 1980. Discretely Adjusted Option Hedges. Journal of Financial
Economics 8, 259282.
Boyle, P., Vorst, T., 1992. Option Replication in Discrete Time with Transaction Costs.
Journal of Finance 47, 271294.
Brennan, M., Schwartz, E., 1989. Portfolio Insurance and Financial Market Equilibrium.
Journal of Business 62, 455472.
Browne, S., 1999. Beating a Moving Target: Optimal Portfolio Strategies for Outperform-
ing a Stochastic Benchmark. Finance and Stochastics 3, 275294.
Cox, J., Huang, C.-F., 1989. Optimal Consumption and Portfolio Policies when the Asset
Price follows a Diusion Process. Journal of Economic Theory 49, 3383.
Cvitanic, J., Karatzas, I., 1996. On Portfolio Optimization under Drawdown Constraints.
Mathematical Finance 65, 3545.
Cvitanic, J., Karatzas, I., 1999. On dynamic measures of risk. Finance and Stochastics 3,
451482.
26
Do, B., 2002. Do, B.H.. Accounting and Finance 42, 279296.
Dudenhausen, A., Schlogl, E., Schlogl, L., 1998. Robustness of Gaussian Hedges under Pa-
rameter and Model Misspecication. Discussion paper, University of Bonn, Department
of Statistics.
El Karoui, N., Jeanblanc, M., Lacoste, V., 2005. Optimal Portfolio Management with
American Capital Guarantee. Journal of Economic Dynamics and Control 29, 449468.
El Karoui, N., Jeanblanc-Picque, M., Shreve, S., 1998. Robustness of the Black and
Scholes Formula. Mathematical Finance 8, 93126.
Fletcher, P., 2005. Guaranteed to take the credit. Risk Magazine 18.
Gnedenko, B., Kolmogorov, A., 1954. Limit distributions for sums of independent random
variables, Addison-Wesley, New York.
Grannan, E., Swindle, G., 1996. Minimizing Transaction Costs of Option Hedging Strate-
gies. Mathematical Finance 6, 341364.
Grossman, S., Villa, J., 1989. Portfolio Insurance in Complete Markets: A Note. Journal
of Business 62, 473476.
Grossman, S., Zhou, J., 1993. Optimal Investment Strategies for Controlling Drawdowns.
Mathematical Finance 3, 241276.
Grossman, S., Zhou, J., 1996. Equilibrium Analysis of Portfolio Insurance. Journal of
Finance 51, 13791403.
Hayashi, T., Mykland, P., 2005. Evaluating Hedging Errors: An Asymptotic Approach.
Mathematical Finance 15, 309343.
Hobson, D., 1998. Volatility Misspecication, Option Pricing and Superreplication via
Coupling. Annals of Applied Probability 8, 193205.
Leland, H., 1985. Option Pricing and Replication with Transaction Costs. Journal of
Finance 40, 12831301.
Leland, H., Rubinstein, M., 1976. The Evolution of Potfolio Insurance, in: D.L. Luskin,
ed., Portfolio Insurance: A guide to Dynamic Hedging, Wiley.
Lyons, T., 1995. Uncertain Volatility and the Risk-free Synthesis of Derivatives. Applied
Mathematical Finance 2, 117133.
Mahayni, A., 2003. Eectiveness of Hedging Strategies under Model Misspecication and
Trading Restrictions. International Journal of Theoretical and Applied Finance 6, 521
552.
Merton, R., 1971. Optimal Consumption and Portfolio Rules in a Continuous Time Model.
Journal of Economic Theory 3, 373413.
Schweizer, M., 2001. A Guided Tour on Quadratic Hedging Approaches, in: Jouini, E.,
Cvitanic, J., Musiela, M. (Eds.), Option Pricing, Interest Rates and Risk Management,
Cambridge University Press, pp. 538574.
Talay, D., Zheng, Z., 2003. Quantiles of the Euler Scheme for Diusion Processes anf
Financial Applications. Mathematical Finance 13, 187199.
Tepla, L., 2000. Optimal portfolio policies with borrowing and shortsale constraints. The
Journal of Economic Dynamics and Control 24, 16231639.
Tepla, L., 2001. Optimal Investment with Minimum Performance Constraints. The Jour-
nal of Economic Dynamics and Control 25, 16291645.
Toft, K. B., 1996. On the Mean-Variance Tradeo in Option Replication with Transaction
Costs. Journal of Financial and Quantitative Analysis 31, 233263.
27