Documente Academic
Documente Profesional
Documente Cultură
Richard J. Haber
Mathematical Institute, 24-29 St Giles, Oxford OX1 3LB
Philipp J. Schönbucher
University of Bonn, Statistical Department,
Adenaueralle 24-42, 53113 Bonn, Germany
Paul Wilmott
Mathematical Institute, 24-29 St Giles, Oxford OX1 3LB
July 1997
1 Introduction
Parisian options are barrier options for which the knock-in/knock-out fea-
ture is only activated after the price process has spent a certain prescribed,
consecutive time beyond the barrier. This specification has two motivations:
First, there is the need to make the option more robust against short-term
movements of the share price. This is achieved in Parisian options where it
is ensured that a single outlier cannot trigger the barrier. In particular, it
is far harder to affect the triggering of the barrier by manipulation of the
underlying (see Taleb [4]). Second, classical barrier options present hedg-
ing problems close to the barrier because their Gamma becomes very large.
To some extent, these problems are reduced, or at least ‘smoothed’, in the
Parisian contract.
We present a flexible approach to valuing such options using the numerical
solution of a partial differential equation. This approach can price a variety
of modifications of the basic Parisian contract including Parasian options
(activation of the barrier conditional on the total time spent above the bar-
rier), American early exercise rights, discrete sampling of the barrier and
1
more general payoffs. The approach readily accommodates features, such as
early exercise, that render the traditional Monte Carlo approach impractical.
To demonstrate the flexibility of this method we have written a program for
valuing many types of the Parisian option contract. This program may be
downloaded free of charge from http://www.wilmott.com.
In January 1997 Risk Magazine published an article by Chesney, et al.[1] on
the pricing of Parisian contracts by Laplace transform methods. That article
provided the motivation for the present work since we believe our method to
be superior for at least four reasons:
The crucial point to note about Parisian options is that they are strongly path
dependent; not only does the payoff depend on the value of the underlying
at expiry but also on the path taken to get there. Yet this path dependence
is perfectly manageable within the partial differential equation framework:
We do not need to know all the details of the path taken. The only further
information we require is the value of the new state variable τ , the barrier
time. The barrier time is defined as the length of time the asset price has
been beyond the barrier in the current excursion
τ := t − sup t0 ≤ t | S(t0 ) ≤ S .
(1)
This expression gives the difference between the current time t and the last
time the share price S was below the barrier S. This is the correct definition
of the barrier time for an ’up’ option. If the share price is currently below
the barrier, τ is zero.
The analogous expression for a ‘down’ barrier is obtained by reversing the
second inequality in relation (1)
τ := t − sup t0 ≤ t | S(t0 ) ≥ S .
(2)
2
The dynamics of τ , for the ‘up’ barrier, are given by the simple expression
dt if S(t) > S,
dτ (t) = −τ (t−) if S(t) = S, (3)
0 if S(t) < S,
where τ (t−) is the left limit of τ . The barrier time increases at the same rate
as t if the share price S(t) is beyond the ‘up’ barrier S(t) > S, it is reset to
zero if the share price hits the barrier S(t) = S, and it does not change if the
share price is below the barrier S(t) < S.
The new state variable τ can be viewed as a clock that starts ticking as soon
as the share price crosses the barrier level, S, and is immediately reset when
the share price returns below S. The knock-in/knock-out feature is only
activated if the length of the current excursion exceeds a given level τ ≥ T ,
where T is the barrier-time triggering parameter. This specification exactly
covers the covenants in the Parisian option’s payoff.
The other two state variables needed are the share price S, and time t.
We assume that the share pays dividends at rate D, and its price follows a
lognormal Brownian motion given by
dS = µS dt + σS dW (4)
3
S
t
-
S
0 τ
Exhibit 1: Characteristics for an ‘up’ Parisian option. The τ -axis describes
the time spent beyond the barrier so far (in this excursion), time runs along
the t-axis and the current share price level is denoted on the S-axis. As long
as the barrier is not exceeded (S < S̄) the sample path moves along the
τ = const-plane, after the share price crosses the barrier S the barrier time
τ and time t increase at the same rate.
4
unchanged and we must solve the basic Black–Scholes equation with a con-
tinuous dividend rate D
∂V σ2S 2 ∂ 2V ∂V
+ + (r − D)S − rV = 0. (5)
∂t 2 ∂S 2 ∂S
The second case occurs when the asset rises above the barrier, S > S, and
so the clock, τ , is ticking. Here we must solve
∂V σ2S 2 ∂ 2V ∂V ∂V
+ 2
+ (r − D)S − rV + = 0, (6)
∂t 2 ∂S ∂S ∂τ
where the new state variable τ gives rise to a modified form of the Black–
Scholes equation.
We will show below how the solutions are linked in these two regions. Both
regions are shown in exhibit 2. The region where equation (5) governs the
price is the plane denoted by C in exhibit 2. This is the area where the
barrier has not been passed yet (S ≤ S), hence τ = 0 here. The region
where the modified Black-Scholes equation (6) has to be satisfied is the area
where S has exceeded the barrier (S ≥ S). In the exhibit this is the ’box’
bounded by the rectangles A and B. Here the barrier time τ can take values
between 0 and T . For Parisian ’down’ options the positions of the regions
are reversed.
In a standard Parisian option the clock variable τ is reset to zero once the
share price moves below S. This will be the case even if the excursion lasts
for a very short time. For example, in the case of an ‘up’ barrier the share
price can reside above the barrier for almost the entire time without trig-
gering the option, provided that it returns below S often enough. However,
the constraint on the time τ being consecutive may defeat one of the original
intentions of the Parisian option which was to make the triggering of the
knock-in/knock-out less susceptible to one-off outliers and very short-term
movements of the share price (or even manipulation). With the given speci-
fication the triggering of the barrier is fairly robust, but the resetting of τ is
still very much subject to short-term price movements.
In the light of this it seems natural to introduce a variation of the Parisian
contract, an option where τ is not reset and where the knock-in/knock-out
feature is only activated if the cumulative time spent beyond S exceeds some
5
D
S
A
t
t
-
S
C T
0 τ
Exhibit 2: The domain for a Parisian ’up’ option. The τ -axis describes the
time spent beyond the barrier so far (in this excursion), time runs along
the t-axis and the current share price level is denoted on the S-axis. In the
lower box (with the rectangle denoted by C as left side) equation (5) must be
satisfied. In the upper box (bounded by the rectangles A and B) the modified
Black-Scholes equation (6) has to be satisfied. (This box is unbounded in
the S → ∞-direction.)
6
prescribed value. This aggregation feature resembles closely the averaging
feature of an Asian option which is why we call this contract the Parasian
option. Note that in all cases that the Parasian contract knocks in/out, the
equivalent Parisian will have done so too.
This variation requires only a minor modification in the model, we just have
to change the definition (3) of the dynamics of τ so that it does not reset at
the boundary S = S:
dt if S ≥ S,
dτa = (7)
0 if S < S.
4 Boundary Conditions
However, in the case of a Parasian option V does not jump at S̄, thus
invalidating condition (8).
The exact specification of the option enters our model via the boundary con-
ditions that we specify. These conditions are to be applied at the boundaries
of the state space as shown in exhibit 2: at expiry t = T , at trigger of the
barrier τ = T and at the limit values for the share price S = 0, S → ∞. In
its most general form a Parisian-type option is specified as follows:
7
time the share price spent beyond the barrier. This option would then
have a payoff of F (S, τ ) = cτ . In exhibit 2 the function F prescribes
the option values on the rectangle A and the line D.
All common variations of the option can be accommodated within this frame-
work, e.g.
In this framework ‘ins’ and ‘outs’ are treated the same. The boundary con-
ditions to specify are
5 An American in Paris
Since we can exercise whenever we want, we should act to maximise the value
of the contract to us. This optimality amounts to insisting that ∂V /∂S, the
8
option delta, is continuous. In solving the US-style contract numerically by
an explicit finite-difference scheme, discussed below, all we need do is to add
to our code a line representing (10).
6 Numerical Algorithm
i2 σ 2 i(r − D)
Lni,j = n n n n n
− rVijn .
Vi+1,j − 2Vi,j + Vi−1,j + Vi+1,j − Vi−1,j
2 2
(11)
There are two regions to consider (see exhibit 2): The area where the ‘clock’
τ is running (i.e. beyond the barrier) and the area where the ‘clock’ τ stands
still. For an up barrier the barrier domain is defined by i > ī and the time
stepping is accomplished by the difference equation
n+1 n
Vi,j+1 = Vi,j + ∆t · Lni,j ∀i > ī. (12)
Note that in equation (12) we had to increase j since the ‘clock’ τ is ticking.
The system can be visualized as diffusively
√ √ propagating in time along a di-
agonal plane, with normal vector (0, 2 , − 22 ) in (S, t, τ ) space, as shown in
2
exhibit 1.
1
The stability condition ∆t < 1/(i2max σ 2 ) must be satisfied, where imax is the largest i
index in the scheme. It is advisable to discretise x := ln S instead of S which will make
the stability condition less stringent. See Wilmott et al. [5] for details. Here we chose the
direct setup for its greater simplicity.
9
In the second region, i < ī, the ‘clock’ τ does not change. Hence the system
evolves along the vertical plane, with normal vector (0, 0, 1), (the rectangle
C in exhibit 2) in accordance to the time-stepping scheme
n+1 n
Vi,j = Vi,j + ∆t · Lni,j ∀i < ī. (13)
Until now we have neglected the dynamics at the the boundary i = ī which
link the upper and lower regions of the state space. In fact, it is this condition
that differentiates between the Parisian and Parasian option.
A Parisian option requires the resetting of the ‘clock’ at i = ī. We first use
(13) to calculate the values of Vi,n+1
j̄
for i ≤ ī of the option at or below the
barrier S = ī∆S and for barrier time zero (τ = 0 or j = j̄).
Vi,n+1
j̄
= Vi,nj̄ + ∆t · Lni,j̄ ∀i ≤ ī. (14)
n+1
Next we set the remaining values Vī,j of the option at the boundary S̄ =
ī∆S. These values are set for all positive barrier times (j < j) to the values
at barrier time zero (j = j) i.e.
n+1
Vi,j = Vi,n+1
j̄
∀j < j̄. (15)
n+1
Then we proceed to calculate Vi,j for i > ī according to the scheme (12).
This stepping means that all values of V at or below the barrier are given by
the values of V for τ = 0, the values of V on the rectangle C in exhibit 2.
In contrast, for a Parasian option we simply apply (13) over the domain i ≤ ī,
and then use (12) for i > ī without resetting the values at the boundary. This
means that we need to solve and calculate values of V for all possible values
of τa even if S is below the barrier S̄, in figure 2 the whole box with l.h.s. C
is relevant, not only the rectangle C.
The payoff at knockin or knockout enters the scheme through the specification
of the value of the payoff at j = 0 ⇔ τ = T̄ by way of boundary condition
n
Vi,0 = Ĝ(i∆S, T − n∆t) ∀i, (16)
Finally, for the case of an early exercise feature we have to check before
n+1
updating Vi,j , if the newly determined value is larger than the early exercise
10
Inputs Outputs
Underlying Barrier type
Spot In/Out Option value
Volatility Call/Put Delta
Dividend yield Strike Gamma
Interest rate Expiry Theta
Reset (Y/N)
Trigger time
payoff A(i∆S, T − (n + 1)∆t, T̄ − j∆t). If not, we simply set the two equal
n+1
by way of the conditional relation Vi,j = A(i∆S, T − (n + 1)∆t, T̄ − j∆t).
The program, which may be downloaded from http://www.wilmott.com,
solves for the price and hedging variables by way of the aforementioned ex-
plicit finite-difference scheme. Although explicit finite-difference schemes are
similar in spirit to the binomial numerical method they are more general
and thus more flexible. The downloadable program is fast but certainly not
optimal. Had speed had been our prime concern we would have used an
implicit method such as Crank-Nicolson. This, along with other methods,
is discussed by Dewynne and Wilmott [3] and in great depth (with sam-
ple code) in Wilmott et al. [5]. The finite-difference solution of financial
partial differential equations is becoming accepted as the most time-efficient
method of pricing and hedging certain types of contract. The method is
time efficient because it is extremely easy to program and the programs
run very quickly. It is suitable for many types of contract including most
common path-dependent derivatives and is trivially – with one extra line
of code – extended to American-style early exercise. It easily outperforms
Monte Carlo simulation, the other popular choice for path-dependent con-
tracts. The downloadable program has the inputs and outputs as given in
exhibit 3. The outputs are arrays (against the underlying).
All presented results are for options on a dividend-free share S with volatility
σ = 20% and a risk-free instantaneous interest rate r = 5%. The numerical
scheme is identical to the one presented in the previous section, with the only
exception that the share price was discretised using equal steps in ln S and
11
not in S.
We will first consider the case of a Parisian up-and-out European Call option
with exercise price E = 10, barrier level S̄ = 12, barrier time T̄ = 0.1 and
time to expiry T = 1. Exhibit 4 depicts the option value V versus share
price S and time τ beyond the barrier, exhibit 5 depicts the associated delta
of the option.
The option value shows the typical up-and-out knockout shape: first (for low
S) the Call option feature dominates and the option value increases with
S but from a certain level of S onwards (here this is around S = 10.18)
the approaching knockout barrier becomes stronger and the option price
decreases with increasing S.
Because the option is of the Parisian type, the option price for share prices
below the barrier (S < S̄ = 12) is constant: Here the barrier time τ will
immediately be re-set to τ = 0 as described in the numerical algorithm. For
share prices larger than the barrier the option prices start to differ depending
on how long the share has spent beyond the barrier so far. If the share price
has just crossed the barrier from below (i.e. at τ = 0) the smoothing effect
of the Parisian specification can be seen, while for τ close to the barrier
activation time (τ = 0.1) the option price approaches zero.
Looking at the delta of the option in this area shows that the Parisian spec-
ification only provides a smooth transition for small τ (or τ = 0 which is
the case when the barrier is first crossed from below). Crossing backwards
from above will result in large changes in the delta of the option because τ
is reset to zero when the barrier is touched. Consequently, the hedging of
Parisian options will provide difficulties if the share price returns close to the
reset barrier S̄ after having spent some time beyond it. Exhibit 6 highlights
this effect: for τ = 0 the smooth transition at S = 12 can be seen, while for
τ = 0.095 (i.e. close to knockout) crossing back below the activation barrier
S = 12 will have a significant effect. The numerical values for exhibit 6 are
given in exhibit 7.
Exhibits 8 and 9 show the value and delta of the equivalent Parasian up-
and-out European Call option, again with the same parameter values as for
the previous option.
The Parasian option exhibits the same up-and-out shape (increasing, then
decreasing) as the Parisian but because the ‘clock’ τ for the Parasian option
is not reset after crossing back below S̄, the option price depends on τ even
for S < S̄. This can be seen clearly in the price plot in exhibit 8. For large
12
0 ,3 5
0 ,3
0 ,25
0 ,2
0 ,1 5
0 ,1
0 ,0 5
ta u= 0
0
ta u= .03
6,95
7,31
7,68
8,07
8,48
8,92
ta u= .0 6
9,37
9,85
10 ,3 5
10 ,88
11,43
12,02
ta u= .09
1 2,63
1 3,27
13,95
S
14,66
Exhibit 4: Value V versus barrier time τ and asset price S for a Parisian
up–and-out European Call option with exercise price E = 10, barrier level
S̄ = 12, barrier time T̄ = 0.1 and time to expiry T = 1. The underlying
asset S has volatility σ = 20% and no dividends, the risk-free instantaneous
interest rate is r = 5%.
13
0 ,2
0,1
-0 ,1
D e lt a
- 0,2
-0 ,3 tau = 0
tau = .02
- 0,4
ta u= .04
tau= .0 8
-0,6
7,07
7,31
7,55
8 ,0 7
7,81
8,34
8,63
9 ,2 2
9 ,53
8,92
1 0,18
1 0,52
9,85
11 ,2 4
1 1,62
10,88
12 ,0 2
1 2,42
1 2,84
13 ,2 7
1 3,72
1 4,66
14,18
14
0 ,3 00 0
0 ,2 50 0
0 ,2 00 0
ta u= 0
ta u= .0 5
ta u= .0 95
0 ,1 50 0
V
ta u= 0
ta u= .0 5
ta u= .0 95
0 ,1 00 0
0 ,0 50 0
0 ,0 00 0
1 1,00 11 ,5 0 1 2,00 12 ,5 0 1 3,00 1 3,50 14,00
15
Parisian Parasian
S tau= 0 tau= 0.05 tau= 0.095 tau= 0 tau= .05 tau= .095
11.06 0.2783 0.2783 0.2783 0.1917 0.1366 0.0710
11.24 0.2566 0.2566 0.2566 0.1696 0.1181 0.0577
11.43 0.2314 0.2314 0.2314 0.1450 0.0980 0.0437
11.62 0.2030 0.2030 0.2030 0.1185 0.0766 0.0292
11.82 0.1719 0.1719 0.1719 0.0904 0.0543 0.0146
12.02 0.1387 0.1242 0.0622 0.0613 0.0316 0.0000
12.22 0.1062 0.0814 0.0000 0.0399 0.0168 0.0000
12.42 0.0771 0.0478 0.0000 0.0249 0.0080 0.0000
12.63 0.0529 0.0249 0.0000 0.0148 0.0034 0.0000
12.84 0.0342 0.0113 0.0000 0.0083 0.0012 0.0000
13.05 0.0208 0.0044 0.0000 0.0044 0.0004 0.0000
13.27 0.0118 0.0014 0.0000 0.0022 0.0001 0.0000
13.49 0.0063 0.0004 0.0000 0.0010 0.0000 0.0000
13.72 0.0031 0.0001 0.0000 0.0004 0.0000 0.0000
13.95 0.0014 0.0000 0.0000 0.0002 0.0000 0.0000
Exhibit 7: The data to exhibit 6. Value V versus asset price S for a Parisian
and a Parasian up–and-out European Call options for different barrier times
τ . (Parameters: Exercise price E = 10, barrier level S̄ = 12, knockout barrier
time T̄ = 0.1, time to expiry T = 1. Underlying asset volatility σ = 20%
zero dividends, risk-free instantaneous interest rate r = 5%.)
16
0,3
0 ,2 5
0,2
V 0 ,1 5
0 ,1
0,05
ta u= 0
0
ta u= .03
6,95
7,31
7,68
8,07
8,48
8,92
ta u= .0 6
9,37
9,85
10 ,3 5
10 ,88
11,43
12,02
ta u= .09
1 2,63
1 3,27
13,95
S
14,66
Exhibit 8: Value V versus barrier time τ and asset price S for a Parasian
up–and-out European Call option with exercise price E = 10, barrier level
S̄ = 12, barrier time T̄ = 0.1 and time to expiry T = 1. The underlying
asset S has volatility σ = 20% and no dividends, the risk-free instantaneous
interest rate is r = 5%.
17
0,15
0,1
0,05
D e lta
tau= 0
-0,05
tau= .015
tau= .03
tau= .0 45
-0,1
tau= .06
tau= .075
7,55
8,0 7
7,31
7,81
8,34
8,92
8,63
9,53
9,22
9,85
10,18
10,88
10,52
11,24
11,62
12,42
12,02
12,84
13,72
13,27
14,18
14,66
Exhibit 9: Delta ∆ = ∂V ∂S
versus barrier time τ and asset price S for a
Parasian up–and-out European Call option with exercise price E = 10, bar-
rier level S̄ = 12, barrier time T̄ = 0.1 and time to expiry T = 1. The
underlying asset S has volatility σ = 20% and no dividends, the risk-free
instantaneous interest rate is r = 5%.
18
barrier times τ → T̄ the Parasian option price approaches the price of a plain
vanilla knockout option. The problem with the excessive deltas of Parisian
options when S leaves the barrier region does not exist for the Parasian, as
can be seen from exhibit 9. The deltas of the Parasian option remain within
a much smaller range.
Exhibit 6 and table 7 directly contrast the values of the Parisian and the
Parasian version of the option. At any given τ , the Parisian option domi-
nates its Parasian counterpart for all share prices S. Intuitively this makes
goods sense since the cumulative effect on τ of the Parasian amplifies the
‘out’ feature of the option, the Parasian is knocked out more often than the
Parisian.
In conclusion, out method provides an extremely flexible, fast, easy to pro-
gram method for evaluating more sophisticated variations of traditional bar-
rier options such as the Parisian and Parasian options under various exercise
and payoff structures. This work clearly demonstrates the versatility and
ease of utilizing the partial differential equation framework for the practical
implementation of exotic option pricing.
19
References
[1] Chesney, M, Cornwall, J, Jeanblanc-Picqué, M, Kentwell, G & Yor, M
1997 Parisian Pricing, Risk Magazine, January 1997.
[3] Dewynne, JN & Wilmott, P 1993 Partial to the Exotic, Risk Magazine,
March
20
8 About the Authors
21