Constant Proportion Portfolio Insurance Strategies under

Cumulative Prospect Theory with Reference Point

Adaptation

Anil Khuman1 , Nick Constantinou∗2 and Steve Phelps1

1 Centre for Computational Finance and Economic Agents (CCFEA), University of
Essex, Colchester CO4 3SQ, UK
2 Essex Business School (EBS), University of Essex, Colchester CO4 3SQ, UK

Abstract
Constant Proportion Portfolio Insurance (CPPI) is a significant and highly popular investment strat-
egy within the structured product market. This has led to recent work which attempts to explain the
popularity of CPPI by showing that it is compatible with Cumulative Prospect Theory (CPT). We
demonstrate that this cannot explain the popularity of ratcheted CPPI products which lock-in gains
during strong growth in the portfolio. In this paper we conjecture that CPPI investors not only fol-
low CPT, but crucially that they also adapt their reference point over time. This important distinction
explains investors preference for ratcheted products.
JEL: G11; G13.
Keywords: Constant Proportion Portfolio Insurance; Ratchets; Cumulative Prospect Theory; Adaptive
Reference Point.

1 Introduction
In the current financial climate investors have a greater concern about risk. Naturally this impacts their
choice of investment product as they have become more wary of losing money. Portfolio insurance (PI)
offers a solution to this problem by giving investors a guarantee on the minimum payoff they will receive
at maturity. Typically the guarantee is on the initial amount invested, providing investors with peace of
mind that they will at the very least not lose their original capital.
There are several different approaches to providing portfolio insurance. A common approach is Con-
stant Proportion Portfolio Insurance (CPPI) (Black and Jones, 1987; Perold and Sharpe, 1988). CPPI is
a portfolio insurance strategy that dynamically manages capital between a risky and risk-free asset with
the goal of providing a minimum guarantee at maturity. Although the strategy was developed many years
ago, it is still very popular with investors, and together with its variants it accounts for a significant pro-
portion of the structured fund market whose importance has come to the attention of regulators (Pain,
2008).
The original CPPI model is formulated in continuous time and assumes instantaneous trading and
smooth price changes. However, in reality these assumptions are violated. This introduces the notion
∗ Corresponding author. Tel.: ++ 44 (0) 1206 873919, Fax: ++ 44 (0) 1206 873429, E-mail: [email protected].

Electronic copy available at: http://ssrn.com/abstract=2109551

of gap risk - the risk that the portfolio value will not meet the guarantee at maturity. Discontinuities in
the price of the risky asset, trading frictions and a lack of liquidity all contribute to gap risk. To this
end Cont and Tankov (2009) examine the impact of price jumps using historical parameter estimates and
find that although there is some gap risk, it is relatively low. Bertrand and Prigent (2002) apply extreme
value theory to allow higher multiplier values when a quantile hedging approach is taken. Maringer and
Ramtohul (2011) employ simulated data in the form of a GARCH risky asset price process, while Dichtl
and Drobetz (2011) and Do (2002) use historical data from large indices. Do and Faff (2004) successfully
apply CPPI to a futures market. Balder et al. (2009) investigate discrete trading models in the presence of
transaction costs by using a calendar rebalancing approach. Alternative strategies whereby rebalancing is
triggered by movements in the risky asset are investigated by Jessen (2010), Dichtl and Drobetz (2011),
Do (2002) and Maringer and Ramtohul (2011).
The control parameter of the CPPI, the multiplier, governs the amount of exposure the portfolio has
to the risky asset. It can therefore be considered a proxy for the risk aversion of the investor. Although
values as a high as 17 have been recommended by Bertrand and Prigent (2002), typically multiplier values
of between 2 and 5 are used (Hamidi et al., 2009; Jessen, 2010). Chen et al. (2008) and Ben Ameur (2010)
use dynamic multiplier approaches, demonstrating their potential to outperform standard static multiplier
models.
Although the basic CPPI assumes a floor that grows at a constant risk-free rate, the implementation of
more complex floor dynamics are investigated by Boulier and Kanniganti (1995), Ben Ameur and Prigent
(2011) and Lee et al. (2010). Boulier and Kanniganti (1995) find that the addition of a ratchet increases
the performance of the CPPI in comparison to a leverage constrained implementation. The increasing
of the floor conditional on increases in the portfolio value is commonly termed ratcheting. The concept
is similar to the Time Invariant Portfolio Protection (TIPP) introduced by Estep and Kritzman (1988);
Choie and Seff (1989). In TIPP an initial floor level is set equal to the guarantee which is below the initial
portfolio value. The floor/guarantee value is then continuously revised upwards on gains in the portfolio
value. Because the floor is not discounted back from maturity a constant risk-free rate is not required. A
benefit of this is that the maturity of the investment does not have to be defined in advance. However, this
means that the strategy is unable to ensure 100% of the initial portfolio value.
The popularity of CPPI investments can be explained by the preferences of investors under certain
utility functions with some restrictions. It has been shown that the CPPI strategy is utility maximising
for the piecewise Hyperbolic Absolute Risk Aversion (HARA) utility function. Although with the in-
troduction of leverage and trading constraints this is no longer true (Black and Perold, 1992). When
the guaranteed amount is considered as a subsistence level the CPPI is optimal under Constant Relative
Risk Aversion (CRRA) (Basak, 2002; Branger et al., 2010). However, more recent research indicates that
the desire for protected products, including the CPPI, is compatible with the Cumulative Prospect The-
ory (CPT) framework. Indeed Dichtl and Drobetz (2011), Dierkes et al. (2010) and Vrecko and Branger
(2009) all provide evidence that a CPT investor generally favours portfolio insurance over other strategies,
e.g. constant mix and buy-and-hold.
It is typically assumed that investors evaluate their utility based on a function of their terminal wealth.
Alternatively, under prospect theory it is the difference between this terminal wealth and some fixed
reference level. However, research suggests that investors’ actual evaluation of investment choices is
far more complex; Arkes et al. (2008) provide evidence that investors adapt their reference point in
relation to changes in investment value. A much more rapid increase in the reference point is seen
for gains than a decrease for losses. Lin et al. (2006) investigate the effect of regret on the perception
of an investor’s performance when giving the choice between three stocks. By framing other possible
investments as missed opportunities they find that an investor’s regret is ‘”most influenced by what their
outcomes might have been had they not invested, by their expected outcomes and by the best-performing
unchosen stocks”‘.
This paper aims to explain investors’ preference for ratcheted guarantee investment products by posit-

Electronic copy available at: http://ssrn.com/abstract=2109551

ing that they use cumulative prospect theory with an adaptive reference point. We demonstrate that a
dynamic reference point, increasing in relation to the maximum portfolio value, is crucial for an investor
to select a ratcheted CPPI strategy over the standard unratcheted one.
The structure of this paper is as follows. In Section 2 we provide a detailed description of the CPPI
strategies. Section 3 describes GJR-GARCH model used to simulate the risky asset price dynamics. In
Section 4 we describe the cumulative prospect theory model. The results are presented and discussed in
Section 5, and finally Section 6 concludes.

2 Background
2.0.1 Standard Model
As stated in the introduction, the canonical CPPI model, introduced by Black and Jones (1987); Perold
and Sharpe (1988), makes a number of simplifications and assumptions that renders it unsuitable for
practical implementation. Firstly, it is not possible to trade instantaneously and the risky asset does not
exhibit continuous and smooth price changes. Secondly, the assumption of the risky asset following a
geometric Brownian motion (GBM) is contrary to empirical evidence. Thirdly, shorting of the risky asset
is often not possible and there is a finite limit on the amount of capital that may be borrowed. Finally, it
is unlikely that the risk-free rate would remain constant during the entire investment period. Aside from
this last point, all the other assumptions are relaxed in this paper. Keeping the risk-free rate constant is
considered acceptable since it contributes a small amount of risk to the strategy in comparison to the risky
asset. The following subsections presents the discrete-timed CPPI model used and its extensions.

2.1 CPPI in Discrete Time

In this subsection we present a discrete time CPPI model where the following convention is adopted: a
horizon T and n + 1 equidistant points:
0 = t0 < t1 ... < tn−1 < tn = T , such that tk+1 − tk = Tn for k = 0, ..., n − 1. Where n is the number of times
the price of the risky asset is observed after the initial construction of the portfolio.
Every period the CPPI model rebalances capital between a risky asset S and a risk-free asset that
grows at the constant rate r. At any time tk , a floor Ftk is calculated by discounting back from maturity
T the guarantee amount Gtk = gtk V0 using the risk-free rate r, where V0 is the initial capital and g0 the
percentage of that amount initially guaranteed i.e. g0 = 100% guarantees the initial investment. The
cushion Ctk is defined as the difference between the portfolio value Vtk and the floor. The investment in
the risky asset Et , termed the exposure, is defined as a constant multiple m of the cushion, whereby a
higher m results in a greater exposure. The remainder of the capital Btk is invested in the risk-free asset.
The above can be summarised in the following set of equations:

Vtk = Ftk +Ctk (1a)

−r(T −tk )
Ftk = Gtk e (1b)
Gtk = gtk V0 (1c)
Etk = mCtk (1d)
Btk = Vtk − Etk . (1e)

The time subscript on the guarantee percentage g and the guaranteed amount G is required for instances
where the value of the guarantee and floor are conditional on some process. This is discussed later.
The progression of the CPPI is driven by changes in the risky asset price Stk . When Stk increases at a
rate greater than r, then more capital is invested in the risky asset by selling some of the risk-free asset.

Electronic copy available at: http://ssrn.com/abstract=2109551

If Stk grows at a rate less than r then some of the risky asset is sold and the proceeds are invested in the
risk-free asset. In both cases the amount bought or sold is such that Equation (1d) is always satisfied.
Thus Vtk can be expressed in terms of Stk as (see e.g. Balder et al. (2009) for derivation)
  
F + (V − F ) m Stk − (m − 1)er Tn if Vtk−1 > Ftk−1
tk tk−1 tk−1 Stk−1
Vtk = (2)
Vt er Tn if Vt ≤ Ft .
k−1 k−1 k−1

Note that Equation (2) exhibits clear path dependence of the value of the CPPI in discrete time, provided
the floor has not been hit. Also note that the definition is independent of the price process governing S.
In the following subsection we describe some common extensions to the basic CPPI framework.

2.2 Model Extensions

2.2.1 Shorting and Leverage Constraints
In the standard CPPI implementation there are no restrictions on either shorting the risky asset or on
borrowing additional funds at the risk free rate. In practice however, restrictions on these actions are in
place. In the extensions to the model described in this section, constraints are imposed on the CPPI to
prevent shorting of the risky asset. This is achieved by ensuring that the exposure never becomes negative:

Ctk = max[(Vtk − Ftk ), 0]. (3)

Additionally, constraints are placed on the amount of leverage (borrowing) that may be used. Defining l
as a multiple limiting the maximum exposure allowed, the value of the exposure is redefined as
Etk = min[mCtk , lVtk ]. (4)
Therefore l = 1 implies that the portfolio is self-financing whilst l = 2 allows a maximum leverage of
200% i.e. up to 100% of the portfolio’s current value may be borrowed at the risk-free rate r. The
maximum leverage permitted may be set by regulatory law, but regardless a maximum leverage of 200%
is typically used in the industry e.g. Pain (2008).

2.2.2 Ratchets
In the original CPPI model, the floor is discounted back from maturity at the risk-free rate and is indepen-
dent of the portfolio’s performance. Due to the length of a typical CPPI investment being several years,
previous gains in the portfolio value during strong periods of growth will be eroded in subsequent periods
of decline. To help retain previous gains a ratchet or profit lock-in is commonly applied to the CPPI
strategy. This serves to increase the terminal guarantee by increasing the value of the floor in reaction to
increases in the portfolio value.
A ratcheting mechanism can be applied to the CPPI, allowing 100% of the initial investment value
to be guaranteed with a floor that grows continuously at the risk-free rate. Equation (5) describes such a
ratchet whereby the guarantee is increased by ξ G % every time the portfolio value increases by a further
ξ V % step.
(Vtk /V0 ) − 1
   
Λtk = max floor , Λtk−1 (5a)
ξV
floor(y) = max{z ∈ Z|z < y} (5b)
G
gtk = g0 + Λtk ξ (5c)
Gtk = gtk V0 (5d)
−r(T −tk )
Ftk = Gtk e . (5e)

4
The number of “clicks” of the ratchet applied up to time tk is denoted by Λtk with Λ0 = 0. Since the
multiplier value remains the same, the probability of the floor being hit remains the same.
The floor ratchet mechanism is illustrated with the following example. Consider the initial guarantee
is equal to the initial investment which is 100 (g0 = 100%, G0 = V0 = 100). The ratchet portfolio trigger
ξ V % is set to 10% and the guarantee increase value ξ G % is set at 5%. If during the investment period
the portfolio value increases to 122, then the ratchet will be clicked twice (floor((122/100 − 1)/0.1) = 2)
and the new guarantee value be 110.

3 Methodology
We assess the performance of different CPPI strategies under a CPT utility function by evaluating their
performance using Monte Carlo simulation of the price process employing a GJR-GARCH model (Glosten
et al., 1993). The GJR model extends the standard GARCH model with an additional term that captures
and assigns extra weight to negative returns. The GJR-GARCH(P,Q) model is defined as
Stk = Stk−1 eRtk (6a)
Rtk = K + εtk (6b)
εtk = σtk ηtk (6c)
Q Q P
σt2k = ω + ∑ αq εt2k−q + ∑ ψq εt2k−q 1{εt <0} + ∑ β p σt2k−p (6d)
k−q
q=1 q=1 p=1

under the conditions

P Q Q
1
ω ≥ 0, αq , β p ≥ 0, αq + ψq ≥ 0, ∑ β p + ∑ αq + 2 ∑ ψq < 1, (7)
p=1 q=1 q=1

where Rtk is the daily log-return, εtk is the unexpected return and σt2k is the conditional variance at time
tk . ηtk is a random variable drawn from a Student’s t distribution with ν degrees of freedom, zero mean
and unit variance. The constant K alters the expected value of the process. The model has been fitted to
FTSE 100 data as described in A.

4 Cumulative Prospect Theory

Attempts to capture the preferences of investors, under uncertainty, had previously been formulated under
the expected utility theory (EUT) framework. Under EUT investors are generally considered to be risk
averse and have a concave value function. The expected utility value being determined through the linear
probability weighting of utility values based on absolute values of wealth. However, EUT fails to capture
the more complex behaviours that are apparent. For example, individuals exhibit loss aversion, non-
linear preferences and can be risk-seeking as well as risk-averse depending on the probability and value
of outcomes (Kahneman and Tversky, 1979).
Prospect theory was introduced by Kahneman and Tversky (1979) to remedy apparent failures in EUT
and was further developed as cumulative prospect theory (CPT) (Tversky and Kahneman, 1992). Prospect
theory differs from expected utility theory in that it uses a reference point to distinguish between losses
and gains. The value function is convex in the losses, exhibiting loss aversion and concave in the gains as
in EUT. Additionally, actual probabilities are transformed, with the overweighting of small probabilities,
to capture insurance and lottery type effects.
The same functional forms of CPT as used by Dichtl and Drobetz (2011) are also adopted here.
However, it is defined as in the manner of Schmidt et al. (2008) where the reference point may be a

5
stochastic value dependent on a realised state. Originally the reference point was assumed to be fixed and
represent the status quo value. The value of an investment strategy f dependent on a state si is f [si ]. The
reference point h generated under this state is h[si ]. The outcome of a state is x[si ] = f [si ] − h[si ]. The
number of states is defined as I = I + + I − , with I + being the number of positive or neutral outcomes and
I − the number of negative outcomes. Outcomes are ordered in ascending order such that i > j if and only
if x[si ] > x[s j ]. The cumulative weighted prospect value is given as
I+
CPV (h) = ∑ v(x[si ])W (si ; f , h), (8)
i=−I −
i6=0

where v(x[si ]) is the relative value under each state and W (si ; f , h) is the weight applied to each state.
As described earlier, the relative value function is a monotonically increasing function that is convex in
losses and concave in gains, such that
(
xα if x ≥ 0
v(x) = (9)
−λ (−x)β if x < 0.

The ranked weights are formed from the following cumulative weighting scheme
 +
w (πi ) if i = I + ,
 + I+

+ I+
w (∑ j=i π j ) − w (∑ j=i+1 π j ) if 1 ≤ i ≤ I + − 1,
W (si ; f , h) = − i − i−1 − (10)
w (∑ j=−I − π j ) − w (∑ j=−I − π j ) if − I + 1 ≤ i ≤ −1,


 −
w (πi ) if i = −I − ,

using the weight function (Lattimore et al., 1992)

δ pγ
w(p) = , (11)
δ pγ + (1 − p)γ
where δ , γ = δ + , γ + |x ≥ 0 and δ , γ = δ − , γ − |x < 0. The parameter values used are: α = β = 0.88;
λ = 2.25; δ + = 0.65; δ − = 0.84; γ + = 0.6; γ − = 0.65, which are the same values as used by Dichtl and
Drobetz (2011).
In this application to portfolio insurance the investment strategy f is VTG with the outcome in (8)
defined as:
x[si ] = VTG [si ] − h[si ] (12)
where VTG is the terminal portfolio value from the perspective of the buyer, with the guarantee in place i.e.
VTG = max(VT , GT ). The determination of the reference point h is discussed in the following subsection.

4.1 Investor Reference Point

As described in the previous subsection, a reference point is required to calculate the cumulative prospect
value (CPV). The reference point is often assumed to be static and equal to the initial capital V0 e.g.
see Dichtl and Drobetz (2011). However, Arkes et al. (2008) suggests that this is not the case and in
fact the investor is likely to adjust their reference point depending on stochastic outcomes of some bench-
mark. Furthermore, it is possible that the investor evaluates their performance based on multiple reference
points. Koop and Johnson (2010) suggest three reference points representing the minimum acceptable
value, the standard CPT reference point and the goal.
For the portfolio insurance investor, potential reference points include: the initial investment value V0 ;
the guarantee amount GT (equal to V0 when no ratcheting occurs); the pure risk-free investment VTr f i.e.

6
V0 erT or a CPPI strategy with m = 0; the gapless portfolio VTm=1 ; the pure risky asset ST ; the maximum
portfolio value achieved during the investment period max[V0..T ]; the maximum risky asset value achieved
during the investment period max[S0..T ]. The first two points represent minimum acceptable values to the
investor. The following three are simple alternative strategies in the investor’s opportunity set that were
not taken, but which the investor wants to outperform. The final two points attempt to capture past highs
that the investor wishes to retain. The ratcheted floor can be considered an attempt to retain some amount
of the value max[V0..T ].
As discussed in the introductions using multiple criteria to assess an investment is a complex area of
research. In particular even if a suitable functional form is found, there is still the problem of eliciting
parameter values. We apply a straightforward method whereby multiple reference points h j are combined
into a single overall point h∗ . In the simplest case J reference points can be combined using a linear
weighted sum
J
h∗ = ∑ w jh j, (13)
j=1

where ∑Jj=1 w j = 1, 0 ≤ w j . Using this model it is possible for the investor to express a concrete prefer-
ence (e.g. 50% V0 , 50% max[V0..T ]) before investment, but for an actual stochastic reference point value
to be realised only at maturity.
There are two alternative formulations for combining the characteristics of multiple reference points
without modifying the original reference point values. Firstly, the weighted linear sum can be applied to
the independent values of the cumulative prospect value e.g. wCPV (h1 ) + (1 − w)CPV (h2 ). Secondly,
a multi-objective Pareto optimal approach could be taken whereby non-dominated solutions are located.
However, the creation of a single new reference point is the most parsimonious model.

5 Results
Unless stated otherwise the results are given without leverage i.e. l = 100%. The risk-free rate of return
r is taken to be 4%, the investment period is 5 years (T = 5) and daily trading (n = 1260) is assumed.

5.1 Standard CPPI

Figure 1 plots the cumulative prospect value of the CPPI, risky asset and risk-free portfolio with a ref-
erence point of V0 . It is shown that the risky asset is preferred over the riskless asset and that the CPPI
is preferred for multiplier values greater than or equal to 2. The CPT value is increasing with the multi-
plier value until m = 6 and then declines slightly. The leverage constraint is responsible for limiting the
increase in the CPT with m.
The effect of increasing levels of leverage are shown in Figure 2. As the maximum amount of lever-
age allowed increases, the increase in cumulative prospect value is substantial. This illustrates that the
leverage constraint is a key factor in limiting the cumulative prospect value for multiplier values greater
than 2.

5.2 Ratcheting
Given the popularity of ratcheting type mechanisms, an explanation is sought under the assumption that
the investor uses cumulative prospect theory to value investments. A comparison of the standard and
ratcheted CPPI under CPT is plotted in Figure 3. The graph clearly shows that using V0 as the reference
point, ratcheting is not favoured for all multiplier values. The greater the ratcheting effect i.e. the greater
ξ G , the worse the performance. This results suggests, for the parameter values considered at least, that
ratcheting of the floor is not favoured by a CPT investor. This is likely due to the ratchet reducing

7
 26

24

22
CPV (V0 )

20

18

16

14
1 2 3 4 5 6 7 8 9 10
m

VT ST
VTrf

Figure 1: Cumulative prospect theory value with initial wealth V0 as the reference point, for various
multiplier values.

45

40

35
CPV (V0 )

30

25

20

15
1 2 3 4 5 6 7 8 9 10
m

l = 100% l = 150% l = 200%

l = 250% ST VTrf

Figure 2: Cumulative prospect value with initial wealth V0 as the reference point. Given for various
multiplier values with maximum leverage values (l) of 100% (no additional leverage), 150%, 200% and
250%.

8
exposure to the risky asset as it rises, limiting growth. Additionally, the greater guarantee value protects
values larger than the initial investment, but since these values are already on the gain side of the CPT
value function this provides little benefit.

ξ V =10%
26

24

22
CPV (V0 )

20

18

16

14
1 2 3 4 5 6 7 8 9 10
m

No ratchet ξ G =2.5% ξ G =5%

ξ G =7.5% ST VTrf

Figure 3: Cumulative prospect value with initial wealth V0 as the reference point. Given for various
multiplier values with a ratcheted floor with ξ V = 10% and ξ G = 2.5%, ξ G = 5% and ξ G = 7.5%.

Since one aspect of the ratchet’s poor performance is due to it allocating capital to the risk-free asset
when the risky asset rises, allowing extra leverage may improve its performance. Figure 4 plots the
cumulative prospect value for two ratchet parameter sets with ξ V = 10% and ξ G = 2.5%, ξ G = 5% with
maximum leverages of 150% and 200%. Compared to the ratchet with no additional leverage in Figure 3
it is seen that leverage provides an increase in the cumulative prospect value. However, compared to the
standard levered CPPI in Figure 2 the levered ratcheted strategy still performs significantly worse.
Previously the reference point has been assumed to be equal to the initial portfolio value. However,
given the long duration of portfolio insurance investments which are typically around 5 years, it is unlikely
that the investor’s reference point remains at their initial wealth. Since the investor of portfolio insurance
forgoes the certainty of the pure risk-free investment for the chance of greater returns from the risky
asset, the risk-free portfolio value is an intuitive reference point. Additionally, since it grows with time it
scales depending on the maturity of the investment. Figure 5 plots the ratcheted CPPI using the terminal
risk-free portfolio value VTr f . For m ≥ 5 the ξ G = 2.5% and ξ G = 5% ratchets slightly outperform the
standard CPPI. This result indicates that a growing reference point may better describe the preference of
an investor.
Given that the guarantee at the initial wealth protects a CPT investor from encountering the loss part
of the value function, it would be intuitive to assume that the reference point adapts to equal the guarantee.
Figure 6 plots the cumulative prospect value under a reference point of GT for ξ V = 10% and ξ G = 2.5%,
ξ G = 5% and ξ G = 7.5%. Under ξ V = 10% and ξ G = 2.5% the ratchet portfolio value is increasing in

9
 ξ V =10%

35

30
CPV (V0 )

25

20

15
1 2 3 4 5 6 7 8 9 10
m

ξ G =2.5%, l = 150% ξ G =5%, l = 150%

ξ G =2.5%, l = 200% ξ G =5%, l = 200%
ST VTrf

Figure 4: Cumulative prospect value with initial wealth V0 as the reference point for various multiplier
values with maximum leverage values (l) of 150% and 200%. Using a ratcheted floor with values ξ V =
10% and ξ G = 2.5%, ξ G = 5%

10
 ξ V =10%
6

4
CPV (VTrf )

0
1 2 3 4 5 6 7 8 9 10
m

No ratchet ξ G =2.5% ξ G =5%

ξ G =7.5% ST VTrf

Figure 5: Cumulative prospect value with the risk-free portfolio value VTr f as the reference point. Given
for various multiplier values with a ratcheted floor with ξ V = 10% and ξ G = 2.5%, ξ G = 5% and ξ G =
7.5%.

11
m, while with ξ V = 10% and ξ G = 5% it is only increasing slightly and when ξ V = 10% and ξ G = 7.5%
it decreases slightly. The limited increase (ξ G = 2.5%, ξ G = 5%) or decrease (ξ G = 7.5%) in CPV is
partly due to the leverage constraint as described in the previous subsection. The other factor is the effect
of the ratchet simultaneously increasing the guarantee, and hence the reference point, and reducing the
exposure to the risky asset. This effect is most prominent when ξ G = 7.5%. The standard CPPI performs
better than the ratcheted strategies for lower m, but then declines as m increases. The better performance
for lower multiplier values is due to the ratchet only coming into effect in very high growth scenarios for
the risky asset. These scenarios benefit the standard CPPI more because their growth is not limited by the
ratchet reducing exposure. As m increases, many scenarios provide sufficient growth for the ratchets to
click and subsequent drops in the risky asset value are protected against. For all of the ratchets, both the
risk-free portfolio and risky asset are declining with m.
It has been implicitly assumed that the reference point of the investor follows the guarantee values as
dictated by the ratcheted strategy. However, it should be noted that it is biased to compare the standard
and ratcheted strategies using the ratcheted strategies’ guarantee value as the reference point. Since it is
assumed that the reference point follows the guarantee value as dictated by the strategy itself, then the
standard CPPI can only be evaluated with a reference point of GT = V0 (since g0..T = 100%). In the
following subsection it is argued that ratchets attempt to capture a more general behaviour of investors
and under this generalisation they are preferable to investors.

5.3 Investor Preferences and Reference Point

As the previous results have shown, a CPT investor would not choose to use a ratcheted CPPI over the
unratcheted strategy if their reference point is fixed at their initial wealth (V0 ). However, this behaviour
is inconsistent with the popularity of ratchets leading to the proposition of an adaptive reference point.
Furthermore, since the ratchet seeks to retain the past gains of the portfolio, this can be generalised to
describe the behaviour of the investor as seeking to retain the best ever portfolio value max[V0..T ]. The
reference point h∗ is then defined as

h∗ = (1 − w)V0 + w max[V0..T ] (14)

where 0 ≤ w ≤ 1 is the percentage weight preference an investor has towards the maximum portfolio
value observed of a particular strategy i.e. h1 = V0 and h2 = max[V0..T ] in Equation (13). When h2 =
max[V0..T ] it is assumed that the investor strictly compares the performance of the strategy to its current
maximum observed value, and is ignorant of the performance of other strategies including the risk-free
and risky asset. In this subsection a weight preference of w = 50% has been chosen to demonstrate an
investor who has a reference point located at the midpoint between their initial wealth and the maximum
observed portfolio value. Three sets of ratchet parameters: ξ V = 5%, ξ G = 2.5%; ξ V = 10%, ξ G = 5%;
ξ V = 15%, ξ G = 7.5%, have been selected to reflect that 50% of the wealth should be protected. This is
in line with the premise that it is beneficial for the guarantee to equal the reference point, as proposed in
the previous subsection.
Figure 7 compares the CPV of the unratcheted CPPI, ratcheted CPPI, risk-free portfolio and risky
asset. The risky asset performs the worst, reflecting its higher volatility which allows it to attain high
maximum values, but not retain them. The unratcheted CPPI also performs poorly due to it not possessing
any mechanism to retain past maximums. This effect is more pronounced for higher multiplier values
where higher maximum values are achieved. Since the risk-free portfolio grows at a constant rate to
maturity it always achieves its maximum value giving it a positive CPV for w < 100%. The ξ V =
5%, ξ G = 2.5% ratchet is the best performing strategy followed by the ξ V = 10%, ξ G = 5% and ξ V =
15%, ξ G = 7.5% ratchets. The more sensitive the ratchet the better it performs because it more rapidly
protects each new maximum level of wealth. The CPV values of the ratcheted strategies peak between

12
 ξ V =10%, ξ G =2.5%
20

15
CPV (GT )

10

0
1 2 3 4 5 6 7 8 9 10
m

ξ V =10%, ξ G =5%

10
CPV (GT )

−10

−20
1 2 3 4 5 6 7 8 9 10
m

ξ V =10%, ξ G =7.5%
10

0
CPV (GT )

−10

−20

−30
1 2 3 4 5 6 7 8 9 10
m

No ratchet Ratchet ST VTrf

Figure 6: Cumulative prospect value with the guarantee level GT , as determined by the ratcheted model,
as the reference point. Given for various multiplier values with a ratcheted floor with ξ V = 10% and
ξ G = 2.5%, ξ G = 5% and ξ G = 7.5%.

13
m = 4 and m = 5 and then decline for higher multiplier values. This decline is caused by growth in the
portfolio increasing the maximum value before the ratchet can be clicked.

10

8
C P V (h ∗ )

1 2 3 4 5 6 7 8 9 10
m

No Ratchet ξ V =5%, ξ G =2.5%

ξ V =10%, ξ G =5% ξ V =15%, ξ G =7.5%
ST V Trf

Figure 7: Cumulative prospect value using a reference point of w = 50%, h1 = V0 and h2 = max[V0..T ].
Given for various multiplier values and ratcheted and unratcheted CPPI.

The impact of 200% maximum leverage is shown in Figure 8. Compared to Figure 7 where no
additional leverage is allowed, the unratcheted CPPI is seen to perform even worse owing to even greater
maximum values obtained. The ratcheted strategies all perform better, benefiting from the additional
growth provided by the leverage with higher multiplier values.
The previous two figures assumed that the investor used the maximum observed value of a strategy
to evaluate that particular strategy. However, as evident from Equation (14), this leads to an issue in
strategies with lower terminal portfolio values, but also lower maximums, potentially resulting in higher
payoffs i.e. the ratchet allows the terminal portfolio to get closer to the maximum because it limits
growth resulting in a lower maximum than what the unratcheted CPPI would achieve. Given that we seek
to investigate whether augmenting the standard CPPI with a ratchet is actually beneficial, the reference
point h∗ can be redefined as
h∗ = (1 − w)V0 + w max[V0..T
stand
], (15)
where max[V0..Tstand ] is the maximum observed value of the standard unratcheted strategy i.e. the CPPI

strategy with the constant rate floor. We can then fairly assess if it is beneficial for the investor to switch
to the ratcheted variant of the CPPI.

14
 12

10

8
C P V (h ∗ )

−2

1 2 3 4 5 6 7 8 9 10
m

No Ratchet ξ V =5%, ξ G =2.5%

ξ V =10%, ξ G =5% ξ V =15%, ξ G =7.5%
ST V Trf

Figure 8: Cumulative prospect value using a reference point of w = 50%, h1 = V0 and h2 = max[V0..T ].
Given for various multiplier values and ratcheted and unratcheted CPPI with 200% maximum leverage
(l = 200%).

15
 Figure 9 compares the different strategies for multiplier values of 1 to 10. In contrast to all previous
figures the results cannot be compared across multiplier values, but only between strategies for a particular
multiplier value. In essence it is assumed that the multiplier value has already been chosen and the
decision to be made is which strategy to select. The figure shows that the risk-free portfolio performs
very poorly since it cannot adjust to the maximum value achieved by the CPPI. The risky asset performs
well for lower multiplier values, but less so for m > 4. When m is low the maximum values are likely to
be low so there would be many instances of ST > h∗ . However, for higher multiplier values this would not
be the case. The ratcheted strategies perform best for m ≥ 4. For m < 4 the unratcheted CPPI performs
better because the ratchets greatly restrict growth, whereas when m ≥ 4 there is sufficient exposure to
grow and retain past gains.

10

0
C P V (h ∗ )

−5

−10

−15

−20
1 2 3 4 5 6 7 8 9 10
m

No Ratchet ξ V =5%, ξ G =2.5%

ξ V =10%, ξ G =5% ξ V =15%, ξ G =7.5%
ST V Trf

stand ].
Figure 9: Cumulative prospect value using a reference point of w = 50%, h1 = V0 and h2 = max[V0..T
Given for various multiplier values and ratcheted and unratcheted CPPI.

In Figure (10) a maximum leverage of 200% is permitted. The impact of this is that the ratcheted
strategies are unable to outperform the standard CPPI for any of the multiplier values tested. This is due
to the ratchets limiting growth too greatly to match the higher maximum values attained by the unratcheted
CPPI. Further evidence of this is seen in the ordering of the ratcheted strategies. The ξ V = 15%, ξ G =
7.5% ratchet performs best followed by the ξ V = 10%, ξ G = 5% and the ξ V = 5%, ξ G = 2.5%. Since
the ξ V = 15%, ξ G = 7.5% ratchet clicks less often it inhibits growth less and therefore performs better.
However, given that the reference point is dependent on the standard CPPI strategy, the ratchet parameters
tested may perform best when w 6= 50%. This effect is more apparent when leverage is permitted, as

16
demonstrated later. Figure 11 plots the strategies for w = 40%. When m ≥ 7 the ratcheted strategies are
able to outperform the standard CPPI.

10

−10
C P V (h ∗ )

−20

−30

−40

−50

1 2 3 4 5 6 7 8 9 10
m

No Ratchet ξ V =5%, ξ G =2.5%

ξ V =10%, ξ G =5% ξ V =15%, ξ G =7.5%
ST V Trf

stand ].
Figure 10: Cumulative prospect value using a reference point of w = 50%, h1 = V0 and h2 = max[V0..T
Given for various multiplier values and ratcheted and unratcheted CPPI with 200% maximum leverage
(l = 200%).

Figures 10 and 11 demonstrated that a ratchet designed to protect 50% of the maximum wealth may
not specifically suit an investor with a w = 50%. In the following two figures: Figure 12 and Figure 13,
the difference in CPV between the standard CPPI and a ratcheted ξ V = 10%, ξ G = 5% strategy is shown
for w values from 0-100%. Figure 12 clearly shows that the ratcheted CPPI is preferred for a broad area
around m = 6 and a weighting of 50%. The effect of 200% maximum leverage is shown in Figure 13.
Compared to Figure 12, the area where the ratcheted CPPI is preferred is shifted down and to the right.
The difference is also significantly smaller. A ξ V = 10%, ξ G = 5% ratchet would benefit an investor
with w = 40% more than one with w = 50% when leverage is permitted because the unratcheted strategy
attains higher maximums. The ratcheted strategy reduces exposure and growth resulting in it protecting
only 40% rather than 50% of the unratcheted CPPI’s maximum.
When there is no leverage a trough is present for multiplier values between 2 and 3. With a maximum
leverage of 200% there is a trough between 3 and 4. In both cases there is a larger negative value when
there is a higher preference weighting. This feature can be explained as follows. The CPV increases
most under a CPPI strategy when there is strong growth and therefore a high payoff. However, for higher
multiplier values the leverage constraint prevents increased investment in the risky asset resulting in 100%

17
 10

0
C P V (h ∗ )

−10

−20

−30

−40
1 2 3 4 5 6 7 8 9 10
m

No Ratchet ξ V =5%, ξ G =2.5%

ξ V =10%, ξ G =5% ξ V =15%, ξ G =7.5%
ST V Trf

stand ].
Figure 11: Cumulative prospect value using a reference point of w = 40%, h1 = V0 and h2 = max[V0..T
Given for various multiplier values and ratcheted and unratcheted CPPI with 200% maximum leverage
(l = 200%).

18
 100

90 2

80
1
70
0
60

∆CPV (h∗ )
−1
w%

50

40 −2

30 −3

20
−4
10
−5
0
0 1 2 3 4 5 6 7 8 9 10
m

Figure 12: Difference in cumulative prospect value between ratcheted (ξ V = 10%, ξ G = 5%) and un-
stand ].
ratcheted CPPI for an investor with h1 = V0 and h2 = max[V0..T

19
 100

90 0

80

70

60 −5

∆CPV (h∗ )
w%

50

40

30 −10

20

10
−15
0
0 1 2 3 4 5 6 7 8 9 10
m

Figure 13: Difference in cumulative prospect value between ratcheted (ξ V = 10%, ξ G = 5%) and un-
stand ] with 200% leverage (l = 200%).
ratcheted CPPI for an investor with h1V0 to h2 max[V0..T

20
(or 200% for the levered approach) investment in S. As demonstrated in Subsection 5.1, increasing the
leverage causes a large increase in the CPV so the leverage constraint is a big limiting factor. This affects
both the standard and ratcheted strategies. However, for the lower multiplier values, where the troughs
occur, the leverage constraint does not have much impact. Instead the ratcheting of the floor is the biggest
inhibitor of growth because capital is allocated away from the risky asset to the risk-free asset. The result
is that the standard CPPI outperforms the ratcheted CPPI.

6 Conclusion
Constant proportion portfolio insurance is a widely used investment strategy. Dichtl and Drobetz (2010,
2011) support the view that investors of CPPI have a preference structure adequately described by cumu-
lative prospect theory. However, a conflict occurs with the introduction of a ratcheted floor. Ratcheted
floor products (including time-invariant portfolio protection) are extensions (alternatives) to the CPPI that
are popular with investors. Under CPT however, our results have show that a ratcheted floor is not pre-
ferred over the standard CPPI. This is because the ratchet protects increasing levels of wealth while the
reference point remains static at the initial wealth value. Since the ratchet limits upside growth in order
to protect greater wealth levels it is not rewarded under the standard fixed reference point cumulative
prospect theory.
In this paper we have proposed an extension to CPT accommodate both the ideas that investors use
CPT, but also like ratcheted investments. We have proposed that the investor does not have a fixed
reference point, but rather it adapts over time. Specifically, the investor uses a linear combination of their
initial wealth V0 and the current observed maximum risky price value max[V0..T ]. The evidence in this
paper supports our proposed framework with the investor with an adaptive reference point preferring the
ratcheted CPPI over the plain model in some cases. More generally our results suggest that the CPT
investor prefers to have the guarantee minimum payoff match their reference point. This allows the loss
part of the value function to be completely ignored producing a higher prospect theory value.

References
Arkes, H., Hirshleifer, D., Jiang, D., Lim, S., Jan. 2008. Reference point adaptation: Tests in the domain
of security trading. Organizational Behavior and Human Decision Processes 105 (1), 67–81.
Balder, S., Brandl, M., Mahayni, A., 2009. Effectiveness of CPPI strategies under discrete-time trading.
Journal of Economic Dynamics & Control 33 (1), 204–220.
Basak, S., Jul. 2002. A comparative study of portfolio insurance. Journal of Economic Dynamics and
Control 26 (7-8), 1217–1241.
Ben Ameur, H., 2010. GARCH Models with CPPI Application. In: Jawadi, F., Barnett, W. A. (Eds.),
Nonlinear Modeling of Economic and Financial Time-Series. (International Symposia in Economic
Theory and Econometrics, Volume 20. Emerald Group Publishing Limited, pp. 187–205.
Ben Ameur, H., Prigent, J.-l., 2011. CPPI Method with a Conditional Floor. International Journal of
Business 16 (3), 218–229.
Bertrand, P., Prigent, J.-l., 2002. Portfolio Insurance the extreme value approach to the CPPI method.
Finance 23 (2), 69–86.
Black, F., Jones, R. C., 1987. Simplifying portfolio insurance. Journal of Portfolio Management 14 (1),
48–51.

21
Black, F., Perold, A. F., 1992. Theory of constant proportion portfolio insurance. Journal of Economic
Dynamics & Control 16 (3-4), 403–426.
Boulier, J.-F., Kanniganti, A., 1995. Expected Performance and Risks of Various Portfolio Insurance
Strategies. In: 5th AFIR Colloquium. Brussels, Belgium, pp. 1093–1124.
Branger, N., Mahayni, A., Schneider, J. C., Jun. 2010. On the optimal design of insurance contracts with
guarantees. Insurance: Mathematics and Economics 46 (3), 485–492.
Chen, J.-S., Chang, C.-L., Hou, J.-L., Lin, Y.-T., 2008. Dynamic proportion portfolio insurance using
genetic programming with principal component analysis. Expert Systems with Applications 35, 273–
278.
Choie, K. S., Seff, E. J., 1989. TIPP: Insurance without complexity: Comment. The Journal of Portfolio
Management 16 (1), 107–108.
Cont, R., Tankov, P., Jul. 2009. Constant Proportion Portfolio Insurance in the Presence of Jumps in Asset
Prices. Mathematical Finance 19 (3), 379–401.
Dichtl, H., Drobetz, W., 2010. On the Popularity of the CPPI Strategy: A Behavioral-Finance-Based
Explanation and Design Recommendations. The Journal of Wealth Management 13 (2), 41–54.
Dichtl, H., Drobetz, W., Dec. 2011. Portfolio insurance and prospect theory investors: Popularity and
optimal design of capital protected financial products. Journal of Banking & Finance 35 (7), 1683–
1697.
Dierkes, M., Erner, C., Zeisberger, S., May 2010. Investment horizon and the attractiveness of investment
strategies: A behavioral approach. Journal of Banking & Finance 34 (5), 1032–1046.
Do, B. H., 2002. Relative Performance of Dynamic Portfolio Insurance Strategies: Australian Evidence.
Accounting and Finance 42 (3), 279–296.
Do, B. H., Faff, R. W., Jun. 2004. Do Futures-Based Strategies Enhance Dynamic Portfolio Insurance?
Journal of Futures Markets 24 (6), 591–608.
Estep, T., Kritzman, M., 1988. TIPP: Insurance without complexity. The Journal of Portfolio Management
14 (4), 38–42.
Glosten, L. R., Jagannathan, R., Runkle, D. E., 1993. On the Relation Between the Expected Value and
the Volatility of the Nominal Excess Return on Stocks. Journal of Finance 48 (5), 1779–1801.
Hamidi, B., Jurczenko, E., Maillet, B., 2009. A CAViaR Time-Varying Proportion Portfolio Insurance.
Bankers, Markets & Investors September- (102), 4–21.
Jessen, C., 2010. Constant Proportion Portfolio Insurance : Discrete-time Trading and Gap Risk Cov-
erage. In: 23rd Australasian Finance and Banking Conference 2010. Copenhagen Business School -
Department of Finance, pp. 1–24.
Kahneman, D., Tversky, A., Mar. 1979. Prospect Theory: An Analysis of Decision under Risk. Econo-
metrica 47 (2), 263.
Koop, G. J., Johnson, J. G., 2010. The Use of Multiple Reference Points in Risky Decision Making.
Journal of Behavioral Decision Making 25 (1), 49–62.
Lattimore, P. K., Baker, J. R., Witte, A. D., 1992. The influence of probability on risky choice: A para-
metric examination. Journal of Economic Behavior & Organization 17 (3), 377–400.

22
Lee, H.-I., Hsu, H., Chiang, M.-H., Nov. 2010. Portfolio insurance with a dynamic floor. Journal of
Derivatives & Hedge Funds 16 (3), 219–230.
Lin, C., Huang, W., Zeelenberg, M., Dec. 2006. Multiple reference points in investor regret. Journal of
Economic Psychology 27 (6), 781–792.
Maringer, D., Ramtohul, T., 2011. GP-based rebalancing triggers for the CPPI. In: Computational Intel-
ligence for Financial Engineering and Economics (CIFEr), 2011 IEEE Symposium on. pp. 1–8.
Pain, D., 2008. Recent developments in portfolio insurance. Bank of England Quarterly Bulletin Q1 (Re-
search and Analysis), 37–46.
Perold, A. F., Sharpe, W. F., 1988. Dynamic Strategies for Asset Allocation. Financial Analysts Jour-
nal (February), 16–27.
Schmidt, U., Starmer, C., Sugden, R., May 2008. Third-generation prospect theory. Journal of Risk and
Uncertainty 36 (3), 203–223.
Tversky, A., Kahneman, D., 1992. Advances in Prospect Theory: Cumulative Representation of uncer-
tainty. Journal of Risk and Uncertainty 5 (4), 297–323.
Vrecko, D., Branger, N., 2009. Why is portfolio insurance attractive to investors?
http://ssrn.com/abstract=1519344.

A Time Series Statistics

A.1 Model Parameters
A GJR-GARCH(1,1) model with Student-t distributed innovations was fitted to daily log-return data from
the FTSE 100 index, from 8nd March 1990 to 15th December 2010 inclusively. Parameter estimation and
simulations were undertaken using the MATLAB Garch Toolbox. Simulations were of daily log-returns
for 5 years (1260 realisations) across 106 paths. Table 1 gives the model parameter values and their errors.
Test statistics for the innovations are given in Table 2, confirming the null hypothesis that the innovations
are uncorrelated for 30 lags and are t-distributed with 13.291 degrees of freedom. The process yields an
expected annual return and annual volatility of approximately 8% and 16% respectively.

Table 1: Fitted parameter values, standard errors and t-statistics.

Parameter Value Standard Error t-Statistic
K 2.7084e-4 1.1308e-004 2.3952
ω 1.1744e-006 2.0216e-7 5.8096
α1 0.0111 0.0065 1.7141
β1 0.9250 0.0068 135.5325
ψ1 0.1047 0.0108 9.6616
DoF 13.2910 1.7828 7.4551

A.2 Summary Statistics

The summary statistics for the log-returns from the FTSE 100 index from 2nd April 1984 to 15th Decem-
ber 2010 inclusively is given in Table 3.

23
Table 2: Ljung-Box and Kolmogorov-Smirnov statistics at 5% confidence interval.
Ljung-Box KS
(30 lags)
p-value 0.5690 0.7568
Critical value 43.7730 0.0186
Q-statistic 28.0276 -
KS Statistic - 0.0092

Table 3: Summary statistics for FTSE 100 daily log-returns.

Quantiles:
Mean 2.4741e-4 1% -0.0311
Median 6.0529e-4 5% -0.0168
Std. dev. 0.0112 10% -0.0118
Skewness -0.3868 25% -0.0054
Kurtosis 11.7876 75% 0.0064
Min -0.1303 90% 0.0118
Max 0.0938 95% 0.0163
99% 0.0286

24