Advanced Topics in Derivative Pricing Models

Topic 2 - Lookback style derivatives

2.1 Product nature of lookback options
2.2 Pricing formulas of European lookback options
Floating strike lookback options
Fixed strike lookback options
2.3 Rollover strategy and strike bonus premium
2.4 Dierential equation formulation
2.5 Multistate lookback options
2.6 Dynamic fund protection
Fixed number of resets
1
2.1 Product nature of lookback options
The payo of a lookback option depends on the minimum or maxi-
mum price of the underlying asset attained during certain period of
the life of the option.
Let T denote the time of expiration of the option and [T
0
, T] be the
lookback period. We denote the minimum value and maximum value
of the asset price realized from T
0
to the current time t (T
0
t T)
by
m
t
T
0
= min
T
0
t
S

and
M
t
T
0
= max
T
0
t
S

2
A oating strike lookback call gives the holder the right to buy
at the lowest realized price while a oating strike lookback put
allows the holder to sell at the highest realized price over the
lookback period.
Since S
T
m
T
T
0
and M
T
T
0
S
T
so that the holder of a oating
strike lookback option always exercise the option.
Hence, the respective terminal payo of the lookback call and
put are given by S
T
m
T
T
0
and M
T
T
0
S
T
.
A xed strike lookback call (put) is a call (put) option on the
maximum (minimum) realized price. The respective terminal
payo of the xed strike lookback call and put are max(M
T
T
0

X, 0) and max(X m
T
T
0
, 0), where X is the strike price.
3
An interesting example is the Russian option, which is in fact a
perpetual American lookback option. The owner of a Russian
option on a stock receives the historical maximum value of the
asset price when the option is exercised and the option has no
pre-set expiration date.
Under the risk neutral measure, the stochastic price process of the
underlying asset is governed by
dS
t
S
t
= r dt + dZ
t
or d
_
ln
S
t
S
0
_
= dU
t
=
_
r

2
2
_
dt + dZ
t
,
where U
t
= ln
S
t
S
0
and = r

2
2
.
4
2.2 Pricing formulas of European lookback options
We dene the following stochastic variables
y
T
= ln
m
T
t
S
= min{U

, [t, T]}
Y
T
= ln
M
T
t
S
= max{U

, [t, T]},
and write = T t. Here, S is the asset price at the current time t
(dropping the subscript t for brevity).
Downstream barrier
For y 0 and y u, we can deduce the following joint distribution
function of U
T
and y
T
from the transition density function of the
Brownian motion with the presence of a downstream barrier
P[U
T
u, y
T
y] = N
_
u +

_
e
2y

2
N
_
u +2y +

_
.
5
Illustration of [U
T
u, Y
T
y]
U

= ln
S

S
0
is visualized as the restricted Brownian motion with
constant drift rate and downstream absorbing barrier y.
y y
T

downstream
barrier
T U u
T

u y
T
y U

6
Upstream barrier
For y 0 and y u, the corresponding joint distribution function
of U
T
and Y
T
is given by
P[U
T
u, Y
T
y] = N
_
u

_
e
2y

2
N
_
u 2y

_
.
By taking y = u in the above two joint distribution functions, we
obtain the respective distribution function for y
T
and Y
T
P(y
T
y) = N
_
y +

_
e
2y

2
N
_
y +

_
, y 0,
P(Y
T
y) = N
_
y

_
e
2y

2
N
_
y

_
, y 0.
The density function of y
T
and Y
T
can be obtained by dierentiating
the above distribution functions.
7
Illustration of [U
T
u, Y
T
y]
Y y
T

Upstream
barrier
T
U u
T

u
Y
T
y U

Time frame
S
t
S
T
T
0
t
T
M
T0
t
M
T0
T
m
T0
t
m
T0
T
For convenience, we write
S = S
t
, M = M
t
T
0
and m = m
t
T
0
.
8
European xed strike lookback options
Consider a European xed strike lookback call option whose terminal
payo is max(M
T
T
0
X, 0). The value of this lookback call option at
the current time t is given by
c
fix
(S, M, t) = e
r
E
_
max(max(M, M
T
t
) X, 0)
_
,
where S
t
= S, M
t
T
0
= M and = T t, and the expectation is
taken under the risk neutral measure. The payo function can be
simplied into the following forms, depending on M X or M > X:
(i) M X
max(max(M, M
T
t
) X, 0) = max(M
T
t
X, 0)
(ii) M > X
max(max(M, M
T
t
) X, 0) = (M X) +max(M
T
t
M, 0).
9
Dene the function H by
H(S, ; K) = e
r
E[max(M
T
t
K, 0)],
where K is a positive constant. Once H(S, ; K) is determined, then
c
fix
(S, M, ) =
_
H(S, ; X) if M X
e
r
(M X) +H(S, ; M) if M > X
= e
r
max(M X, 0) +H(S, ; max(M, X)).
c
fix
(S, M, ) is independent of M when M X because the
terminal payo is independent of M when M X.
When M > X, the terminal payo is guaranteed to have the
oor value M X. If we subtract the present value of this
guaranteed oor value, then the remaining value of the xed
strike call option is equal to a new xed strike call but with the
strike being increased from X to M.
10
Recall that when X is a non-negative random variable, we have
E[X] =

0
[1 F
X
(t)] dt, if X is continuous.
Since max(M
T
t
K, 0) is a non-negative random variable, its ex-
pected value is given by the integral of the tail probabilities where
H(S, ; K)
= e
r
E[max(M
T
t
K, 0)]
= e
r

0
P[M
T
t
K x] dx
= e
r

K
P
_
ln
M
T
t
S
ln
z
S
_
dz, z = x +K
= e
r

ln
K
S
Se
y
P[Y
T
y] dy, y = ln
z
S
_
dy =
1
z
dz, z = Se
y
_
= e
r

ln
K
S
Se
y
_
N
_
y +

_
+e
2y

2
N
_
y

__
dy
11
= SN(d) e
r
KN(d

)
+ e
r

2
2r
S
_
_
e
r
N(d)
_
S
K
_

2r

2
N
_
d
2r

_
_
_
,
where
d =
ln
S
K
+
_
r +

2
2
_

.
The European xed strike lookback put option with terminal payo
max(X m
T
T
0
, 0) can be priced in a similar manner. Write m = m
t
T
0
and dene the function
h(S, ; K) = e
r
E[max(K m
T
t
, 0)].
12
The value of this lookback put can be expressed as
p
fix
(S, m, ) = e
r
max(X m, 0) +h(S, ; min(m, X)),
where
h(S, ; K) = e
r

0
P[max(K m
T
t
, 0) x] dx
= e
r

K
0
P[K m
T
t
x] dx 0 max(K m
T
t
, 0) K
= e
r

K
0
P[m
T
t
z] dz, z = K x
= e
r

ln
K
S
0
Se
y
P[y
T
y] dy, y = ln
z
S
= e
r

ln
K
S
0
Se
y
_
N
_
y

_
+e
2y

2
N
_
y +

__
dy
= e
r
KN(d +

) SN(d) +e
r

2
2r
S
_
_
_
S
K
_
2r/
2
N
_
d +
2r

_
e
r
N(d)
_
_
.
13
European oating strike lookback options
By exploring the pricing relations between the xed and oating
lookback options, we can deduce the price functions of oating
strike lookback options from those of xed strike options. Consider a
European oating strike lookback call option whose terminal payo
is S
T
m
T
T
0
, the present value of this call option is given by
c
f
(S, m, ) = e
r
E[S
T
min(m, m
T
t
)]
= e
r
E[(S
T
m) +max(mm
T
t
, 0)]
= S me
r
+h(S, ; m)
= SN(d
m
) e
r
mN(d
m

) +e
r

2
2r
S
_
_
_
S
m
_

2r

2
N
_
d
m
+
2r

_
e
r
N(d
m
)
_
_
,
where
d
m
=
ln
S
m
+
_
r +

2
2
_

.
14
Consider a European oating strike lookback put option whose ter-
minal payo is M
T
T
0
S
T
, the present value of this put option is given
by
p
f
(S, M, ) = e
r
E[max(M, M
T
t
) S
T
]
= e
r
E[max(M
T
t
M, 0) (S
T
M)]
= H(S, ; M) (S Me
r
)
= e
r
MN(d
M
+

) SN(d
M
) + e
r

2
2r
S
_
_
e
r
N(d
M
)
_
S
M
_

2r

2
N
_
d
M

2r

_
_
_
,
where
d
M
=
ln
S
M
+
_
r +

2
2
_

.
Remark
Through H(S, ; M) and H(S, ; X), we can deduce the xed-oating
relation between lookback call and put; the form of which is depen-
dent on either M X or M > X.
15
Boundary condition at S = m
Consider the scenario when S = m, that is, the current asset price
happens to be at the minimum value realized so far. The probability
that the current minimum value remains to be the realized minimum
value at expiration is seen to be zero. In other words, the probability
that S
t
touches the minimum value m once (one-touch) and remains
above m at all subsequent times is zero.
Recall the distribution formula for m
T
t
:
P[m
T
t
m] = N
_
ln
m
S
+

t
_

_
S
m
_
1
2r

2
N
_
ln
m
S
+

t
_
so that P[m
T
t
m] = 0 when S = m.
16
Insensitivity of lookback option price to m when S = m
We can argue that the value of the oating strike lookback call
should be insensitive to innitesimal changes in m since the change
in option value with respect to marginal changes in m is proportional
to the probability that m will be the realized minimum at expiry
c
f
m
(S, m, )

S=m
= 0.
Alternatively, we may argue that the future updating of the realized
minimum value does not require the current realized minimum value
m. Hence, the oating strike lookback call is insensitive to m when
S = m.
17
Lookback options for market entry
Suppose an investor has a view that the asset price will rise
substantially in the next 12 months and he buys a call option
on the asset with the strike price set equal to the current asset
price.
Suppose the asset price drops a few percent within a few weeks
after the purchase, though it does rise up strongly to a high
level at expiration, the investor should have a better return if he
had bought the option a few weeks later.
Timing for market entry is always dicult to be decided. The
investor could have avoided the above diculty if he has pur-
chased a limited period oating strike lookback call option
whose lookback period only covers the early part of the options
life.
It would cause the investor too much if a full period oating
strike lookback call were purchased instead.
18
Let [T
0
, T
1
] denote the lookback period where T
1
< T, T is the
expiration time, and let the current time t [T
0
, T
1
]. The terminal
payo function of the limited period lookback call is max(S
T

m
T
1
T
0
, 0).
We write S
t
= S, m
t
T
0
= m and = T t. The value of this lookback
call is given by
c(S, m, ) = e
r
E
Q
[max(S
T
m
T
1
T
0
, 0)]
= e
r
E
Q
[max(S
T
m, 0)1
{mm
T
1
t
}
]
+e
r
E
Q
[max(S
T
m
T
1
t
, 0)1
{m>m
T
1
t
}
]
= e
r
E
Q
[S
T
1
{S
T
>m,mm
T
1
t
}
]
e
r
mE
Q
_
1
{S
T
>m,mm
T
1
t
}
_
+e
r
E
Q
[S
T
1
{S
T
>m
T
1
t
,m>m
T
1
t
}
]
e
r
E
Q
[m
T
1
t
1
{S
T
>m
T
1
t
,m>m
T
1
t
}
], t < T
1
,
where the expectation is taken under the risk neutral measure Q.
19
For the rst term, the expectation can be expressed as
E
Q
[S
T
1
{S
T
>m,mm
T
1
t
}
=

ln
m
S

ln
m
S
x
Se
xz
k(z)h(x, y) dz dxdy,
where k(z) is the density function for z = ln
S
T
S
T
1
and h(x, y) is the
bivariate density function for x = ln
S
T
1
S
and y = ln
m
T
1
t
S
.
The third and fourth terms can be expressed as
E
Q
[S
T
1
{S
T
>m
T
1
t
,m>m
T
1
t
}
=

ln
m
S

yx
Se
xz
k(z)h(x, y) dz dxdy
and
E
Q
[m
T
1
t
1
{S
T
>m
T
1
t
,m>m
T
1
t
}
] =

ln
m
S

yx
Se
y
k(z)h(x, y) dz dxdy.
20
The price formula of the limited-period lookback call is found to
be
c(S, m, )
= SN(d
1
) me
r
N(d
2
) +SN
2
_
_
d
1
, e
1
;

T T
1
T t
_
_
+e
r
mN
2
_
_
f
2
, d
2
;

T
1
t
T t
_
_
+e
r

2
2r
S
_
_
_
S
m
_

2r

2
N
2
_
_
f
1
+
2r

T
1
t, d
1
+
2r

T
1
t
T t
_
_
e
r
N
2
_
_
d
1
, e
1
;

T T
1
T t
_
_
_
_
+e
r(TT
1
)
_
1 +

2
2r
_
SN(e
2
)N(f
1
), t < T
1
,
21
where
d
1
=
ln
S
m
+
_
r+

2
2
_

, d
2
= d
1

,
e
1
=
_
r+

2
2
_
(TT
1
)

TT
1
, e
2
= e
1

T T
1
,
f
1
=
ln
S
m
+
_
r+

2
2
_
(T
1
t)

T
1
t
, f
2
= f
1

T
1
t.
One can check easily that when T
1
= T (full lookback period), the
above price formula reduces to standard oating strike lookback call
price formula.
Suppose the current time passes beyond the lookback period, t > T
1
,
the realized minimum value m
T
1
T
0
is now a known quantity. This
limited period lookback call option then becomes a European
vanilla call option with the known strike price m
T
1
T
0
.
22
2.3 Rollover strategy and strike bonus premium
The sum of the rst two terms in c
f
can be seen as the price
function of a European vanilla call with strike price m, while the
third term can be interpreted as the strike bonus premium.
Rollover strategy
At any time, we hold a European vanilla call with the strike price set
at the current realized minimum asset value. In order to replicate
the payo of the oating strike lookback call at expiry, whenever a
new realized minimum value of the asset price is established at a
later time, one should sell the original call option and buy a new call
with the same expiration date but with the strike price set equal to
the newly established minimum value.
23
Strike bonus premium
Since the call with a lower strike is always more expensive, an extra
premium is required to adopt the rollover strategy. The sum of these
expected costs of rollover is termed the strike bonus premium.
The strike bonus premium can be shown to be obtained by inte-
grating a joint probability distribution function involving m
T
t
and
S
T
. Firstly, we observe
strike bonus premium = h(S, ; m) +S me
r
c
E
(S, ; m)
= h(S, ; m) p
E
(S, ; m),
where c
E
(S, , m) and p
E
(S, ; m) are the price functions of European
vanilla call and put, respectively. The last result is due to put-call
parity relation.
24
Recall
h(S, ; m) = e
r

m
0
P[m
T
t
] d
and
p
E
(S, ; m) = e
r

0
P[max(mS
T
, 0) x] dx
= e
r

m
0
P[S
T
] d.
Since the two stochastic state variables satises 0 m
T
t
S
T
, we
have
P[m
T
t
] P[S
T
] = P[m
T
t
< S
T
]
so that
strike bonus premium = e
r

m
0
P[m
T
t
< S
T
] d.
25
Sub-replication and replenishing premium
Replenishing premium of a European put option
put value = e
r

K
0
P[S
T
] d
We divide the interval [0, K] into n subintervals, each of equal width
so that n = K. The put can be decomposed into the sum
of n portfolios, the j
th
portfolio consists of long holding a put with
strike j and short selling a put with strike (j1), j = 1, 2, , n,
where all puts have the same maturity date T.
Potential liabilities occur when S
T
falls within [0, K] (the put expires
in-the-money). Hedge the exposure over successive n intervals:
n = K and
j
= j
26
For the j
th
portfolio:
hold one put with strike j (more expensive)
short one put with strike (j 1) (less expensive)
Present value of this j
th
portfolio
= e
r
{E[(j S
T
)1
{S
T

j
}
] E[((j 1) S
T
)1
{S
T

j1
}
]}
e
r
P[S
T

j
] (to leading order in ).
In the limit n , we obtain
put value = e
r
lim
n
n

j=1
P[S
T

j
]
= e
r

K
0
P[S
T
] d.
These n portfolios can be visualized as the appropriate replenishment
to the sub-replicating portfolio in order that the writer of the put
option is immunized from possible loss at the maturity of the option.
The sub-replicating portfolio is taken to be the null portfolio in the
current example.
27
With the addition of the n
th
portfolio [long a put with strike K
and short a put with strike (K )] into the sub-replicating
portfolio, the writer faces a loss only when S
T
falls below K.
Deductively, the protection over the interval [(j 1), j]
in the out-of-the-money region of the put is secured with the
addition of the j
th
portfolio.
One then proceeds one by one from the n
th
portfolio down to
the 1
st
portfolio so that the protection over the whole interval
[0, K] is achieved.
With the acquisition of all these replenishing portfolios, the
writer of the put option is immunized from any possible loss at
options maturity even the put expires out-of-the-money. The
cost of acquiring all these n portfolios is the replenishing pre-
mium.
28
Replenishing premium of a European call option
Potential liabilities occur when S
T
falls within [K, ) (the call expires
in-the-money). Hedge the exposure over successive n intervals:
jth
interval
(j-1)
j
K
) [
K+ K+

For the j
th
portfolio:
hold one call with strike K +(j 1) (more expensive)
short one call with strike K +j (less expensive)
29
Present value of this j
th
portfolio
= e
r
{E[{S
T
[K +(j 1)]}1
{S
T
K+(j1)}
]
E[[S
T
(K +j)]1
{S
T
K+j}
]}
e
r
P[S
T
K +j] (to leading order in ).
In the limit n , we obtain
call value = e
r
lim
n
n

j=1
P[S
T
K +j]
= e
r

K
P[S
T
] d.
Since the sub-replicating portfolio has been chosen to be the null
portfolio, the call value is then equal to the replenishing premium.
The above result is distribution free.
30
Put-call parity relations of continuously monitored oating
strike and xed strike lookback options
We let [T
0
, T] be the continuously monitored period for the mini-
mum value of the asset price process. The current time t is within
the monitoring period so that T
0
< t < T, and that the period of
monitoring ends with the maturity of the lookback call option.
The terminal payo of the continuously monitored oating strike
lookback call option is given by
c
f
(S
T
, T) = S
T
m
T
T
0
= S
T
min(m, m
T
t
).
Here, m
T
t
is a stochastic state variable with dependence on S
u
, u
[t, T].
31
(i) forward as the sub-replicating instrument
Suppose we choose the sub-replicating instrument to be a forward
with the same maturity and delivery price m. The terminal payo
of the sub-replicating instrument is below that of the forward only
when m
T
t
< m; otherwise, their terminal payos are equal. Here,
m
T
t
is the random variable that determines the occurrence of under
replication.
Potential liabilities occur when m
T
t
falls within [0, m] with payo
mm
T
t
. Similar to the put with payo mS
T
when S
T
falls within
[0, m], the replenishing premium is (replacing S
T
in put by m
T
t
)
e
r

m
0
P[m
T
t
] d.
The replenishing premium can be visualized as the value of a Euro-
pean xed strike lookback put option with xed strike m and whose
terminal payo is
_
mm
T
t
_
+
=
_
mmin(m, m
T
t
)
_
+
=
_
mm
T
T
0
_
+
.
32
Let p
fix
(S, t; K) denote the value of a xed strike lookback put with
strike K, whose terminal payo is max(K m, 0). This gives the
following put-call parity relation for lookback options:
c
f
(S, t; m) = S e
r
m+e
r

m
0
P[m
T
t
] d
= S e
r
m+p
fix
(S, t; m),
where S is the current asset price and Se
r
m is the present value
of the forward with delivery price m and maturity date T.
The probability distribution P[m
T
t
] is given by the distribution
function for the restricted asset price process with the down barrier
over the interval [t, T].
33
(ii) European call option as the sub-replicating instrument
Suppose we change the sub-replication portfolio to be a European
call option whose terminal payo is (S
T
m)
+
. Comparing S
T
m
T
t
with (S
T
m)
+
, there are two sources of risks. One is the realization
of lower minimum value while the other is that S
T
may stay below
m. The terminal payo c
f
(S
T
, T) is decomposed as
S
T
m
T
t
= (mm
T
t
)
. .
(mS
T
)
. .
xed strike vanilla put
lookback with strike m
put with
strike m
When S
T
m and m
T
t
m, both of the above puts are in-the-money.
Replenishing premium
= e
r

m
0
_
P[m
T
t
] P[S
T
]
_
d
= e
r

m
0
P[m
T
t
< S
T
] d
= strike bonus premium = h(S, m, t) p
E
(S, t; m).
34
Remarks
There are 3 other possible cases of the relative position of m
T
t
and
S
T
with respect to m.
1. Since m
T
t
S
T
, we rule out S
T
< m and m
T
t
> m.
2. The full replication is achieved when S
T
> m and m
T
t
> m.
3. When S
T
> m and m
T
t
< m, the amount of sub-replication is
mm
T
t
. The corresponding replenishing premium is reduced to
e
r

m
0
P[m
T
t
] d
since P[S
T
] = 0 for [0, m].
The integral formulation of the replenishing premium on the last
page covers all these 4 cases.
35
2.4 Partial dierential equation formulation
We would like to illustrate how to derive the governing partial dif-
ferential equation and the associated auxiliary conditions for the
European oating strike lookback put option. First, we dene the
quantity
M
n
=
_

t
T
0
(S

)
n
d
_
1/n
, t > T
0
,
the derivative of which is given by
dM
n
=
1
n
S
n
(M
n
)
n1
dt
so that dM
n
is deterministic. Taking the limit n , we obtain
M = lim
n
M
n
= max
T
0
t
S

,
giving the realized maximum value of the asset price process over
the lookback period [T
0
, t].
36
We attempt to construct a hedged portfolio which contains one
unit of a put option whose payo depends on M
n
and units of
the underlying asset. Again, we choose so that the stochastic
components associated with the option and the underlying asset
cancel.
Let p(S, M
n
, t) denote the value of the lookback put option and
let denote the value of the above portfolio. We then have
= p(S, M
n
, t) S.
The dynamics of the portfolio value is given by
d =
p
t
dt +
1
n
S
n
(M
n
)
n1
p
M
n
dt +
p
S
dS +

2
2
S
2

2
p
S
2
dt dS
by virtue of Itos lemma. Again, we choose =
p
S
so that the
stochastic terms cancel.
37
Using the usual no-arbitrage argument, the non-stochastic port-
folio should earn the riskless interest rate so that
d = rdt,
where r is the riskless interest rate. Putting all equations to-
gether, we have
p
t
+
1
n
S
n
(M
n
)
n1
p
M
n
+

2
2
S
2

2
p
S
2
+rS
p
S
rp = 0.
Next, we take the limit n and note that S M. When
S < M, lim
n
1
n
S
n
(M
n
)
n1
= 0; and when S = M,
p
M
= 0. Hence,
the second term becomes zero as n .
The governing equation for the oating strike lookback put is
given by
p
t
+

2
2
S
2

2
p
S
2
+rS
p
S
rp = 0, 0 < S < M, t > T
0
.
38
The domain of the pricing model has an upper bound M on S. The
variable M does not appear in the equation, though M appears as
a parameter in the auxiliary conditions. The nal condition is
p(S, M, T) = M S.
In this European oating strike lookback put option, the boundary
conditions are applied at S = 0 and S = M. Once S becomes zero,
it stays at the zero value at all subsequent times and the payo at
expiry is certain to be M.
Discounting at the riskless interest rate, the lookback put value at
the current time t is
p(0, M, t) = e
r(Tt)
M.
The boundary condition at the other end S = M is given by
p
M
= 0 at S = M.
39
Partial dierential equation formulation of the lookback option price
function
Let S denote the stock price variable and M denote the realized
maximum of the stock price recorded from the initial time of the
lookback period to the current time. Let t denote the calendar time
variable, T be the maturity date of the lookback option and = T t
be the time to expiry.
The formulation for the price function V (S, M, ) of the one-asset
European lookback option model with terminal payo V
T
(S, M) is
given by
V

=

2
2
S
2

2
V
S
2
+rS
V
S
rV, 0 < S < M, > 0,
V
M

S=M
= 0, > 0,
V (S, M, 0) = V
T
(S, M), (1)
where r is the riskless interest rate and is the volatility of the
stock price.
40
The price function is essentially two-dimensional with state vari-
ables S and M. However, the dierential equation exhibits the
degenerate nature in the sense that it does not involve the look-
back variable M.
M only occurs in the Neumann boundary condition
V
M

S=M
= 0
and the terminal payo function. The Neumann boundary con-
dition signies that if the current stock price equals the value of
the current realized maximum then the option price is insensitive
to M.
We reformulate the pricing model (1) using the following new
set of variables:
x = ln
M
S
, y = lnM.
Note that
V
M
=
1
M
_
V
x
+
V
y
_
since both x and y have depen-
dence on M.
41
The lookback pricing model formulation can be rewritten as
V

=

2
2

2
V
x
2

_
r

2
2
_
V
x
rV, x > 0, < y < , > 0,
_
V
x
+
V
y
_

x=0
= 0, > 0,
V (x, y, 0) = V
T
(e
yx
, e
y
). (2)
The triangular wedge shape of the original domain of denition
D = {(S, M) : 0 < S < M} is now transformed into a new domain
which is the semi-innite two-dimensional plane

D = {(x, y) : x > 0 and < y < }.

42
However, the boundary condition along x = 0 involves the function
V
x
+
V
y
.
M
S
V
M
=0
V
x
=0
y
x
+
V
y
x=0
43
Floating strike lookback options
We consider the valuation of lookback options with payo of the
form Sf
_
M
S
_
, which includes the oating strike payo as a special
example. By taking V
T
(S, M) = Sf
_
M
S
_
and applying the trans-
formations of variables: x = ln
M
S
and U(x, ) =
V (S, M, )
S
to the
pricing formulation (1), we obtain
U

=

2
2

2
U
x
2

_
r +

2
2
_
U
x
, x > 0, > 0,
U
x

x=0
= 0, > 0
U(x, 0) = f(e
x
).
Once the terminal condition is free of y (namely, lnM), the depen-
dence on y of the price function disappears.
44
The Neumann boundary condition at x = 0 indicates that x = 0
is a reecting barrier for the system particle hitting the reecting
barrier will be reected, unlike an absorbing barrier which removes
the particle from the system.
To resolve the diculty of dealing with the Neumann boundary
condition along x = 0, we extend the domain of denition from
the semi-innite domain to the full innite domain.
This is achieved by performing continuation of the initial condi-
tion to the domain x < 0 such that the price function can satisfy
the Neumann boundary condition.
45
Due to the presence of the drift term in the dierential equation, the
simple odd-even extension is not applicable. For the oating strike
payo M S, we have U(x, 0) = e
x
1, x > 0. The continuation of
the initial condition to the domain x < 0 is found to be
U(x, 0) =
1 e
(2 1)x
2 1
, x < 0, where =
r

2
+
1
2
.
To show the claim, we set
U(x, ) =

U(x, )e
x+

,
where =
r

2
+
1
2
and

=
1
2
2
_
r +

2
2
_
2
. The transformation
on U is equivalent to applying the change of measure to make the
underlying price process to be drift free.
46
Now

U(x, ) is governed by

=

2
2

2
U
x
2
, x > 0, > 0
_

U
x
+

U
_

x=0
= 0, > 0, (Robin condition)

U(x, 0) = e
x
f(e
x
) = h
+
(x), x > 0.
The fundamental solution is the free space Green function:
(x, ; ) =
1

2
2

exp
_

(x )
2
2
2

_
, < x < , > 0.
47
Let h

(x) denote the continuation of the initial condition for x < 0,

U(x, ) can then be formally represented by

U(x, ) =

(x , )h

() d +

0
(x , )h
+
() d.
The function h

(x) is determined by enforcing the satisfaction of

the Robin boundary condition by the solution

U(x, ).
We then obtain the following ordinary dierential equation for h

(x):
h

(x) + h

(x) +h

+
(x) + h
+
(x) = 0,
with matching condition:
h
+
(0) = h

(0).
For example, suppose f(e
x
) = e
x
1, then h
+
(x) = e
x
(e
x
1).
48
By solving the above equation, we obtain
h

(x) =
e
x
e
( 1)x
2 1
.
In general, the solution is found to be
h

(x) = h
+
(x) +2 e
x

x
0
e

h() d.
We obtain the integral price formula of lookback option with payo
Sf
_
M
S
_
as follows:
V (S, M, ) = S
_
M
S
_

1
_

_
ln
M
S
+ln,
_
+
_
ln
M
S
ln,
_
+2

_
ln
M
S
+ln,
__

_
1
d
_
_
f()

+2
d,
where

=
1
2
2
_
r +

2
2
_
2
.
49
For the oating strike lookback option, we have f() = 1. The
corresponding price function is found to be
V
f
(S, M, ) = Me
r
_
_
N(d +

2
2r
_
M
S
_
2r/
2
N
_
d
2r

_
_
_
S
_
N(d)

2
2r
N(d)
_
,
where
d =
ln
M
S

_
r

2
2
_

.
50
2.5 Multistate lookback options
Two-asset semi-lookback option
The terminal payo of a semi-lookback option depends on the ex-
treme value of the price of one asset and the terminal values of
the prices of other assets. Let V
2
semi
(S
1
, S
2
, t; S
2
[T
0
, t]) denote the
value of the two-asset semi-lookback option whose terminal payo
is given by (S
2
[T
0
, T] S
1,T
K)
+
. We write S
2
[T
0
, t] = M
2
and
S
2
[t, T] = (M
2
)
T
t
, and let the terminal payo be expressed as
(max(M
2
K, (M
2
)
T
t
K) S
1,T
)
+
.
We choose the sub-replicating instrument to be the put option on
S
1
with strike M
2
K, where the corresponding terminal payo is
((M
2
K) S
1,T
)
+
.
51
Similarly, there are two sources of risks: (i) realization of higher
maximum asset value, where (M
2
)
T
t
> M
2
, (ii) S
1,T
> M
2
K and
(M
2
)
T
t
K > S
1,T
M
2
< S
1,T
+K < (M
2
)
T
t
.
The rst scenario leads to an increase in the strike of the semi-
lookback option from M
2
K to (M
2
)
T
t
K.
The second scenario leads to positive terminal payo in the semi-
lookback option but zero terminal payo in the vanilla put.
The two cases can be combined into M
2
< S
1,T
+K < (M
2
)
T
t
. When
such scenario occurs, the under replication at maturity is
(M
2
)
T
t
K S
1,T
= [(M
2
)
T
t
M
2
] [(S
1,T
+K) M
2
].
52
The above terminal payo is equivalent to the sum of those of the
xed strike lookback call on (M
2
)
T
t
and vanilla call on S
1,T
+K, both
with the same strike M
2
.
The required replenishing premium is given by
e
r
_

M
2
P[(M
2
)
T
t
> ] d

M
2
P[S
1,T
+K > ] d
_
= e
r

M
2
P[(M
2
)
T
t
> S
1,T
+K] d.
The value of the two-asset semi-lookback option is given by
V
2
semi
(S
1
, S
2
, t; S
2
[T
0
, t]) = p(S
1
, t; M
2
K)
+ e
r

M
2
P[(M
2
)
T
t
> S
1,T
+K] d.
53
Discretely monitored oating strike lookback call options
Suppose the monitoring of the minimum value of the asset price
takes place only at discrete time instant t
j
, j = 1, 2, , n, where
t
n
is on or before the maturity date of the lookback call option.
Suppose the current time is taken to be within [t
k
, t
k+1
). The
terminal payo of the discretely monitored oating strike lookback
call option is given by
c
dis
f
(S
T
, T) = S
T
min(S
t
1
, S
t
2
, , S
t
n
).
We use the notation S[i, j] to denote min(S
t
i
, S
t
i+1
, , S
t
j
), j > i.
At the current time, S[1, k] = min(S
t
1
, S
t
2
, , S
t
k
) is already known.
Similar to the continuously monitored case, we choose the sub-
replicating instrument to be a forward with the same maturity date
T and delivery price S[1, k].
54
Potential liabilities occur when S[k + 1, n] falls below S[1, k]. This
is similar to a put with the stochastic state variable S[k +1, n] and
strike S[1, k].
The replenishing premium required to compensate for under repli-
cation is given by
replenishing premium = e
r

S[1,k]
0
P[S[k +1, n] ] d.
The present value of the discretely monitored European oating
strike lookback call option is then given by
c
dis
f
(S, t; S[1, k]) = S e
r
S[1, k] +e
r

S[1,k]
0
P[S[k +1, n] ] d.
Remark Following the strike bonus approach, we can also obtain
c
dis
f
(S, t; S[1, k]) = c
E
(S, t; S[1, k])+e
r

S[1,k]
0
P[S[k+1, n] < S
T
] d.
55
The distribution function P[S[k +1, n] ] can be expressed as
P[S[k +1, n] ] =
n

j=k+1
E1
{S
t
j
,S
t
j
/S
t
i
1 for all i=j,k+1in}
,
where the indicator function in the jth term corresponds to the
event that S
t
j
is taken be the minima among S
t
k+1
, , S
t
n
; and j
runs from k +1 to n.
Suppose the asset price follows the Geometric Brownian motion,
then S
t
j
and S
t
j
/S
t
i
, i = j, k + 1 i n, are all lognormally dis-
tributed. The expectation values in above equation can be expressed
in terms of multi-variate cumulative normal distribution functions.
56
One-asset lookback spread option
The terminal payo of an one-asset lookback spread option is given
by
c
sp
(S
T
, T; K) =
_
M
T
T
0
m
T
T
0
K
_
+
.
Choice of sub-replicating portfolio:
long holding of one unit of European lookback call and one unit
of lookback put, both of oating strike;
short holding of a riskless bond of par value K.
All these instruments have the same maturity as that of the lookback
spread option.
Terminal payo of this sub-replicating portfolio
= (M
T
T
0
S
T
) +(S
T
m
T
T
0
) K = M
T
T
0
m
T
T
0
K.
57
Note that
M
T
T
0
m
T
T
0
K =max(M, M
T
t
) min(m, m
T
t
) K
M mK,
so if the lookback spread is currently in-the-money, then it will expire
in-the-money. The sub-replication is a full replication when the
lookback spread option is currently in-the-money.
On the other hand, if the lookback spread option is currently out-of-
the-money, the terminal payo of the sub-replicating portfolio would
be less than that of the lookback spread option if the lookback
spread option expires out-of-the-money. That is,
max(M, M
T
t
) min(m, m
T
t
) K < 0.
The sources of risks include (i) M
T
t
> M and m
T
t
< m, (ii) M
T
t

m
T
t
K < 0 m
T
t
+K > M
T
t
. These cases can be combined into
M < M
T
t
< m
T
t
+K < m+K.
58
We hedge the exposure over successive subintervals within [M, m+
K]: (M +(j 1), M +j) and write
j
= M +j.
The potential liability of has to be replenished when both of the
following 2 events occur together:
M
T
t
<
j
and m
T
t
+K >
j
.
The expected replenishing premium over the j
th
interval is
e
r
P[M
T
t
<
j
< m
T
t
+K].
Summing over all intervals, we obtain
e
r

m+K
M
P[M
T
t
< < m
T
t
+K] d.
59
In summary
(i) M mK 0 (currently in-the-money or at-the-money)
c
sp
(S, t; M, m) = c
f
(S, t; m) +p
f
(S, t; M) Ke
r
;
(ii) M mK < 0 (currently out-of-the-money)
c
sp
(S, t; M, m)
= c
f
(S, t; m) +p
f
(S, t; M) Ke
r
+ e
r

m+K
M
P[M
T
t
< < m
T
t
+K] d.
60
The distribution function P[M < m+K] can be deduced from
P[X x, X y]
=

n=
e
[2n(yx)]/
2
__
N
_
y t 2n(y x)

t
_
N
_
x t 2n(y x)

t
__
e
2x/
2
_
N
_
y t 2n(y x) 2x

t
_
N
_
x t 2n(y x) 2x

t
___
.
61
Two-asset lookback spread option
Let S
1,u
and S
2,u
denote the price process of asset 1 and asset
2, respectively. Similarly, we write S
1
[t
1
, t
2
] and S
2
[t
1
, t
2
] as the
realized maximum value of S
1,u
and realized minimum value of S
2,u
over the period [t
1
, t
2
], respectively. The terminal payo of a two-
asset lookback spread option is given by
c
sp
(S
1,T
, S
2,T
, T; K) = (S
1
[T
0
, T] S
2
[T
0
, T] K)
+
,
where K is the strike price.
62
Replicating portfolio
Since we can express S
1
[T
0
, T] S
2
[T
0
, T] K as
(S
1
[T
0
, T] S
1,T
) +(S
2,T
S
2
[T
0
, T]) +S
1,T
S
2,T
K,
a natural choice of the sub-replicating portfolio would consist of
long holding of one European oating strike lookback put on asset
1, one European oating strike lookback call on asset 2, one unit
of asset 1 and short holding of one unit of asset 2 and a riskless
bond of par value K. All instruments in the portfolio have the same
maturity as that of the two-asset lookback spread option.
Similar to the one-asset counterpart, the two-asset lookback spread
option is guaranteed to expire in-the-money if it is currently in-the-
money.
The sub-replicating portfolio will expire with a terminal payo below
that of the lookback spread option if the lookback spread option
expires ont-of-the-money.
63
The current value of the two-asset European lookback spread option
is given by
(i) S
1
[T
0
, t] S
2
[T
0
, t] K 0
c
sp
(S
1
, S
2
, t; S
1
[T
0
, t], S
2
[T
0
, t]) = p
f
(S
1
, t; S
1
[T
0
, t]) +c
f
(S
2
, t; S
2
[T
0
, t])
+S
1
S
2
Ke
r
;
(ii) S
1
[T
0
, t] S
2
[T
0
, t] K < 0
c
sp
(S
1
, S
2
, t; S
1
[T
0
, t], S
2
[T
0
, t]) = p
f
(S
1
, t; S
1
[T
0
, t]) +c
f
(S
2
, t; S
2
[T
0
, t])
+S
1
S
2
Ke
r
+ e
r

S
2
[T
0
,t]+K
S
1
[T
0
,t]
P(S
1
[t, T] < S
2
[t, T] +K) d.
The joint distribution of maximum of one asset and minimum of
another asset cannot be deduced using the method of images.
One has to resort to eigenfunction expansion technique, and the
resulting expression involves the modied Bessel functions.
64
Lookbacks on two assets
Double Maxima: call or put on the dierence between the max-
imum of S
1
and the maximum of S
2
:
max[0, (aS
1
(T) bS
2
(T)) K]
max[0, K (aS
1
(T) bS
2
(T))]
where a > 0 and b > 0 are parameters to be chosen by investors.
In practice, it may make sense to pick a and b such that aS
1
(0) =
bS
2
(0). For example, a = 1/S
1
(0) and b = 1/S
2
(0).
With a and b chosen in this way, the double maxima represents
an option on the dierence between the maximum returns of
the two stocks over a given period. When K = 0, the double
maxima call is equivlent to an option to buy the maximum of
S
1
at the maximum of S
2
.
65
Double Minima: call or put on the dierence between the mini-
mum of S
1
and the minimum of S
2
:
max[0, (aS
1
(T) bS
2
(T)) K]
max[0, K (aS
1
(T) bS
2
(T))].
When K = 0, the double minima call is equivalent to an option
to sell the minimum of S
1
for the minimum of S
2
.
Double Lookback Spread: call or put on the spread between the
maximum S
1
and the minimum of S
2
:
max[0, (aS
1
(T) bS
2
(T)) K]
max[0, K (aS
1
(T) bS
2
(T))].
66
Uses of these double-asset lookback options
Double maxima/minima represent options on the dierence be-
tween the maximum/ minimum returns of two stocks over a
given period. They provide investors with a special vehicle to
take a view on how these two stocks will perform relative to
each other.
Similarly, double lookback spreads capture the dierence be-
tween the maximum upside of one stock and the maximum
downside of another stock. This type of options can be an
aggressive play on the volatilities of the two stocks as well as
on the correlation of the two stocks.
V
Dmax
(x
1
, x
2
) = max
_
0, aS
1
(0)e
max(M
1
,x
1
)
bS
2
(0)e
max(M
2
,x
2
)
K
_
V
Dmin
(x
1
, x
2
) = max
_
0, aS
1
(0)e
min(m
1
,x
1
)
bS
2
(0)e
min(m
2
,x
2
)
K
_
V
DLS
(x
1
, x
2
) = max
_
0, aS
1
(0)e
max(M
1
,x
1
)
bS
2
(0)e
min(m
1
,x
2
)
K
_
67
Recall
X
i
(t) =
i
t+
i
Z
i
(t), i = 1, 2; cov(Z
1
(t), Z
2
(t)) = t; is a constant.
X
i
(t) = min
0st
X
i
(s), X
i
(t) = max
0st
X
i
(t).
The call prices C
Dmax
, C
Dmin
, and C
DLS
, respectively, for double
maxima, double minima, and double lookback spread options are
determined as follows:
C
Dmax
(x
1
, x
2
) = e
rT

0
dx
1

0
dx
2
V
Dmax
(x
1
, x
2
)

2
P(X
1
(t) x
1
, X
2
(t) x
2
)
x
1
x
2
C
Dmin
(x
1
, x
2
) = e
rT

dx
1

dx
2
V
Dmin
(x
1
, x
2
)

2
P(X
1
(t) x
1
, X
2
(t) x
2
)
x
1
x
2
C
DLS
(x
1
, x
2
) = e
rT

dx
1

0
dx
2
V
DLS
(x
1
, x
2
)

2
P(X
1
(t) x
1
, X
2
(t) x
2
)
x
1
x
2
.
68
Probability density / distribution functions of the extreme values of
two correlated Brownian motions.
P(X
1
(t) dx
1
, X
2
(t) dx
2
, X
1
(t) m
1
, X
2
m
2
)
=p(x
1
, x
2
, t; m
1
, m
2
,
1
,
2
,
1
,
2
, ) dx
1
dx
2
(i) For x
1
m
1
, x
2
m
2
, where m
1
0, m
2
0,
p(x
1
, x
2
, t; m
1
, m
2
,
1
,
2
,
1
,
2
, )
=
e
a
1
x
1
+a
2
x
2
+bt

1
2
h(x
1
, x
2
, t; m
1
, m
2
,
1
,
2
,
1
,
2
, ),
where
h(x
1
, x
2
, t; m
1
, m
2
,
1
,
2
,
1
,
2
, )
=
2
t

n=1
e
(r
2
+r
2
0
)/2t
sin
n
0

sin
n

I
(n)/
_
rr
0
t
_
.
69
The parameter values are given by
a
1
=

1
(1
2
)
2
1

2
, a
2
=

2

2
(1
2
)
1

2
2
,
b =
1
a
1

2
a
2
+
1
2

2
1
a
2
1
+
1

2
a
1
a
2
+
1
2

2
2
a
2
2
,
tan =

1
2

, [0, ],
z
1
=
1

1
2
__
x
1
m
1

1
_

_
x
2
m
2

2
__
, z
2
=
_
x
2
m
2

2
_
,
z
10
=
1

1
2
_

m
1

1
+
m
2

2
_
, z
20
=
m
2

2
,
r =

z
2
1
+z
2
2
, tan =
z
2
z
1
, = [0, ],
r
0
=

z
2
10
+z
2
20
, tan
0
=
z
20
z
10
,
0
[0, ].
70
The distribution function p satises the Fokker-Planck equation
p
t
=
1
p
x
1

2
p
x
2
+
1
2

2
1

2
p
x
2
1
+
1

2
p
x
1
x
2
+
1
2

2
2

2
p
x
2
2
,
t > 0, m
1
< x
1
< , m
2
< x
2
< ;
with the following initial condition:
p(x
1
, x
2
, t = 0) = (x
1
)(x
2
)
and absorbing boundary conditions
p(x
1
= m
1
, x
2
, t) = 0
p(x
1
, x
2
= m
2
, t) = 0.
71
To get rid of the drift terms, we dene
p(x
1
, x
2
, t) = e
a
1
x
1
+a
2
x
2
+bt
q(x
1
, x
2
, t),
where a
1
, a
2
, and b are dened as above. Then q(x
1
, x
2
, t) satises
q
t
=
1
2

2
1

2
q
x
2
1
+
1

2
q
x
1
x
2
+
1
2

2
2

2
p
x
2
2
with auxiliary conditions:
q(x
1
, x
2
, t = 0) = (x
1
)(x
2
)
q(x
1
= m
1
, x
2
, t) = 0
q(x
1
, x
2
= m
2
, t) = 0.
72
This PDE can be simplied by a suitable transformation of coordi-
nates, eliminate the cross-partial derivative and normalize the Brow-
nian motions. Explicitly, if we dene the set of new coordinates z
1
and z
2
, where
z
1
=
1

1
2
_
x
1
m
1

x
2
m
2

2
_
and z
2
=
x
2
m
2

2
, and write
q(x
1
, x
2
, t) =
h(z
1
, z
2
, t)

1
2
.
73
This procedure is similar to dene two uncorrelated Brownian mo-
tions from two given correlated Brownian motions. Here, h(z
1
, z
2
, t)
satises the following standard diusion equation (without the cross-
derivative term):
h
t
=
1
2
_

2
h
z
2
1
+

2
h
z
2
2
_
with boundary condition:
h(z
1
, z
2
, t) = (z
1
z
10
)(z
2
z
20
)
h(L
1
, t) = h(L
2
, t) = 0,
where z
10
=
1

1
2
_

m
1

1
+
m
2

2
_
and z
20
=
m
2

2
; and
L
1
= {(z
1
, z
2
) : z
2
= 0}, L
2
=
_

_
(z
1
, z
2
) : z
2
=

1
2

z
1
_

_
.
74
x
1
x
2
m
1
m
2
L :x =m
2 1 1
L :x =m
1 2 2
.
z
2
z
1

.
(z ,z )
10 20
L ':z =0
1 2
0 r < , 0 ;
tan =

1
2

.
75
These boundary conditions along L
1
and L
2
are more conveniently
expressed in polar coordinates. Introducing polar coordinates (r, )
corresponding to (z
1
, z
2
) as dened above, we obtain
h
t
=
1
2
_

2
h
r
2
+
1
r
h
r
+
1
r
2

2
h

2
_
, r
2
= z
2
1
+z
2
2
and tan =
z
2
z
1
;
with boundary conditions: h(r, , t = 0) =
1
r
0
(r r
0
)(
0
),
h(r, = 0, t) = 0, h(r, = , t) = 0.
Note that
1
r
0
arises from the Jacobian of transformation from (z
1
, z
2
)
to (r, ).
76
To solve this PDE for h(r, , t), we look for separable solutions of
the form R(r)()T(t). Plugging this into the PDE, we nd
T

T
=
1
2
_
R

R
+
1
r
R

R
+
1
r
2

_
=
2
/2,
where the separation constant is negative because the solutions
must decay as t .
Hence, we have the eigenfunction for T(t) to be an exponential
function:
T(t) e

2
t/2
and
k
2
k
2
..
_
r
2
R

R
+r
R

R
+
2
r
2
_
+
..
_

_
= 0.
Dening

/ = k
2
, we obtain
() Asink +Bcos k.
77
The boundary conditions require that (0) = () = 0, and hence
k must be real, B = 0, and sink = 0.
This last requirement restricts k to discrete values of the form:
k
n
=
n

, n = 1, 2, . . ..
Thus the most general angular solution consistent with the boundary
conditions is
() sin
n

, n = 1, 2, . . . .
78
Finally, the radial part of the solution is
r
2
R

+rR

+(
2
r
2
k
2
n
)R = 0.
Dening y = r, we can rewrite this in the standard form
y
2
d
2
R
dy
2
+y
dR
dy
+(y
2
k
2
n
)R = 0.
This is the Bessel equation, with the well known fundamental solu-
tions J
k
n
(y) and I
k
n
(y). Since I
k
n
(0) diverges and we require R(0)
to be well-behaved, the I
k
n
(x) solution is not permitted. Hence the
general radial solution is R(r) J
k
n
(r).
79
In summary, the most general solution to the PDE for h(r, , t) con-
sistent with the absorbing boundary conditions:
h(r, 0, t) = h(r, , t) = 0,
is given by
h(r, , t) =

n=1
c
n
()e

2
t/2
sin
_
n

_
J
n/
( r) d.
Note that T
n
(t) has a continuum of eigenvalues .
The next step is to nd the coecients c
n
() which t the initial
condition:
h(r, , 0) = r
1
0
(r r
0
)(
0
).
80
Setting t = 0, we have
1
r
0
(r r
0
)(
0
) = h(r, , 0) =

n=1
c
n
() sin
n

Jn

(r) d.
Next, we multiply the previous equation at t = 0 by sin(m/)
and integrate over the interval [0, ] in . From the orthogonality
relation, we obtain
r
1
0
(r r
0
) sin
_
m
0

_
=

2

0
c
m
()J
m/
(r) d.
Using the following relation:

0
xJ
v
(ax)J
v
(bx)dx = a
1
(a b),
and comparing the equations, we ontain
c
m
(

) =
2

sin
_
m
0

_
J
m/
(

r
0
).
81
Putting all the results together, we nally obtain
h(r, , t) =

0
_
2

n=1
e

2
t/2
sin
_
n
0

_
sin
_
n

_
J
n/
(r
0
)J
n/
(r) d.
The integration with respect to can be performed explicitly using
the relation:

0
xe
c
2
x
2
J
v
(ax)J
v
(bx) dx =
1
2c
2
e
(a
2
+b
2
)/4c
2
I
v
_
ab
2c
2
_
.
We then obtain the nal expression, where
h(r, , t) =
2
t

n=1
e
(r
2
+r
2
0
)/2t
sin
_
n
0

_
sin
_
n

_
I
n/
_
rr
0
t
_
.
82
Based on the symmetry relations, we obtain
(ii) For x
1
m
1
, x
2
M
2
, where m
1
0, M
2
0, we have
P(X
1
(t) dx
1
, X
2
(t) dx
2
, X
1
(t) m
1
, X
2
m
2
)
=p(x
1
, x
2
, t; m
1
, M
2
,
1
,
2
,
1
,
2
, ) dx
1
dx
2
.
(iii) For x
1
M
1
, x
2
M
2
, where M
1
0, M
2
0, we have
P(X
1
(t) dx
1
, X
2
(t) dx
2
, X
1
(t) m
1
, X
2
m
2
)
=p(x
1
, x
2
, t; M
1
, M
2
,
1
,
2
,
1
,
2
, ) dx
1
dx
2
.
83
From the joint density function of correlated restricted Brownian
motions, we derive the distribution function of joint maximum of
the two-dimensional correlated Brownian motions as follows:
Integrating over the density functions and applying a change of
polar-coordinates, we obtain the distribution function:
P(X
1
(t) x
1
, X
2
(t) x
2
) = e
a
1
x
1
+a
2
x
2
+bt
f(r

, t),
where
f(r

, t) =
2

n=1
sin
_
n

_
e
r
2
/2t

0
sin
_
n

_
g
n
() d,
84
with
g
n
() =

0
re
r
2
/2t
e
b
1
r cos()b
2
r sin()
I
n/
_
rr

t
_
dr,
r

=
1

1
2
_
x
2
1

2
1

2x
1
x
2

2
+
x
2
2

2
2
_
1/2

= +, with cos =
x
1

1
r

tan =

1
2
,

= +

2
,
b
1
=a
1

1
+a
2

2
and b
2
= a
2

1
2
.
Similar expressions can be derived for
P(X
1
(T) x
1
, X
2
(t) x
2
) and P(X
1
(t) x
1
, X
2
(t) x
2
).
85
Corollary
When the correlation can take on only the special values

= cos

n
, n = 1, 2, . . . ,
then tan =

1 cos
2
n
cos

n
= tan

n
so that =

n
. In this case, the
density function p has the special form
p(x
1
, x
2
, t) =
e
a
1
x
1
+a
2
x
2
+bt

1
2
n
h(z
1
, z
2
, t),
where h is a nite sum of bivariate normal densities
h(z
1
, z
2
, t) =
n1

k=0
[g
+
k
(z
1
, z
2
, t) +g

k
(z
1
, z
2
, t)]
g

k
(z
1
, z
2
, t) =
1
2
exp
_

1
2
_
_
z
1
r
0
cos
_
2k
n

0
__
2
+
_
z
2
r
0
sin
_
2k
n

0
__
2
__
.
86
/3
A method of images solution for the two-dimensional case.
87
When
n
= cos(/n), the angles between the lines L
1
and L
2
take
the special values

n
= /n, n = 1, 2, . . . .
For these angles, a method of images solution to the PDE is pos-
sible. Note that
g

(z
1
, z
2
, t; a
1
, a
2
) =
1
2t
exp
_

1
2
[(z
1
a
1
)
2
+(z
2
a
2
)
2
]
_
satises the PDE with the initial condition
g

(z
1
, z
2
, 0; a
1
, a
2
) = (z
1
a
1
)(z
2
a
2
).
88
2.6 Dynamic fund protection
Provides an investor with a oor level of protection during the in-
vestment period.
This feature generalizes the concept of a put option, which
provides only a oor value at a particular time.
The dynamic fund protection ensures that the fund value is
upgraded if it ever falls below a certain threshold level.
89
References
1. Gerber, H.U. and G. Pafumi (2000), North American Actuarial
Journal, vol. 4(2) P.28-37.
2. Imai, J. and P.P. Boyle (2001), North American Actuarial Jour-
nal, vol. 5(3) P.31-51.
3. Chu, C.C. and Y.K. Kwok (2004), Insurance, Mathematics and
Economics, vol. 34 P.273-295.
90
Typical sample path of the fund unit values
upgraded
fund
primary fund
91
1.

F(0) = F(0)
2.

F(t) = F(t) max
_
1, max
0st
K
F(s)
_
Whenever

F(t) drops below K, just enough money will be added
so that the upgraded fund unit value does not fall below K.
Write M(t) = max
_
1,
K
min
0st
F(s)
_
, then

F(T) = F(T) max

_
M(t),
K
min
tsT
F(s)
_
.
Here, M(t) is the number of units of primary fund acquired
at time t. The lookback feature of the dynamic protection is
revealed by M(t).
92
Protection with reference to a stock index under nite number
of resets
F(t): value of the primary fund
I(t): value of the reference stock index

F(t): value of the protected fund

When the investor makes its reset decision, the sponsor of the pro-
tected fund has to purchase additional units of the primary fund so
that the protected fund value is upgraded to that of the reference
index.
93
F as the numeraire
At the reset instant
i
, i = 1, 2, . . ., we have

F(
i
) = I(
i
) = n(
i
)F(
i
),
where n(
i
) = I(
i
)/F(
i
) > 1 is the new number of units of the
primary fund in the investment fund.
It is obvious that n(
1
) < n(
2
) < . The normalized upgraded
fund value

F/F is related to I/F. This motivates us to use F as
the numeraire.
94
Fund value dynamics
Under the risk neutral valuation framework, we assume that the
primary fund value F(t) and the reference index value I(t) follow
the Geometric Brownian motions:
dF
F
= (r q
p
)dt +
p
dZ
p
,
dI
I
= (r q
i
)dt +
i
dZ
i
,
where r is the riskless interest rate, q
p
and q
i
are the dividend yield
of the primary fund and stock index, respectively,
p
and
i
are
the volatility of the primary fund value and reference index value,
respectively, and dZ
p
dZ
i
= dt. Here, is the correlation coecient
between the primary fund process and reference index process.
95
Pricing formulation of the protected fund with n resets
Let V
n
(F, I, t) denote the value of the investment fund with dy-
namic protection with respect to a reference stock index, where the
investor has n reset rights outstanding. We rst consider the sim-
pler case, where there has been no prior reset. That is, the number
of units of the primary fund is equal to one at current time t.
The dimension of the pricing model can be reduced by one if F is
chosen as the numeraire. We dene the stochastic state variable
x =
I
F
where F is the fund value at the grant date (same as the primary
fund value).
96
Note that x follows the Geometric Brownian motion
dx
x
= (q
p
q
i
)dt + dZ,
where
2
=
2
p
2
p

i
+
2
i
. Accordingly, we dene the normalized
fund value function with F as the numeraire by
W
n
(x, t) =
V
n
(F, I, t)
F
.
Whenever a reset has occurred, the upgraded fund value

F will be
used as the numeraire so that x =
I

F
. Note that F(t) and

F(t) have
the same dynamics equation since

F(t) is scalar multiple of F(t).
The investor should never reset when F(t) stays above I(t).
With only a nite number of reset rights, he does not reset
immediately when F(t) just hits the level of I(t).
97
With reference to the variable x, the investor resets when x reaches
some suciently high threshold value (denoted by x

n
). The value
of x

n
is not known in advance, but has to be solved as part of the
solution to the pricing model.
Upon reset at x = x

n
, the sponsor has to increase the number of
units of the primary fund so that the new value of the investment
fund equals I.
The corresponding number of units should then be x

n
, which is the
ratio of the reference index value to the primary fund value right
before the reset moment.
98
After the reset, we have
the number of resets outstanding is reduced by one,
the value of x becomes one since the ratio of the reference index
value to the newly upgraded fund value is one.
We then have
W
n
(x

n
, t) = x

n
W
n1
(1, t).
To show the claim, note that
W
n1
(1, t
+
) =
V
n1
(

F,

F, t
+
)

F(t
+
)
=
V
n
(

F(t

), x

n
(t

)

F(t

), t

)
x

n
(t

)

F(t

)
=
1
x

n
(t

)
W
n
(x, t

).
99
Smooth pasting condition
We should have the smooth pasting condition at x = x

n
, based on
optimality consideration:
W

n
(x

n
, t) = W
n1
(1, t).
This extra smooth pasting condition determines the value of x

n
such
that the investment fund value is maximized.
The terminal payo of the investment fund is simply equal to F, if
no reset has occurred throughout the life of the fund. This gives
W
n
(x, T) = 1.
100
Governing equation
In the continuation region, inside which the investor chooses not to
exercise the reset right, the value function V
n
(F, I, t) satises the
Black-Scholes equation with the two state variables, F and I.
In terms of x, the governing equation and the associated auxiliary
conditions for W
n
(x, t) are given by
W
n
t
+

2
2
x
2

2
W
n
x
2
+(q
p
q
i
)x
W
n
x
q
p
W
n
= 0, t < T, x < x

n
(t),
W
n
(x

n
, t) = x

n
W
n1
(1, t) and W

n
(x

n
, t) = W
n1
(1, t), W
n
(x, T) = 1,
where x

n
(t) is the time-dependent threshold value at which the
investor optimally exercises the reset right.
101
Free boundary value problem
The pricing model leads to a free boundary value problem with the
free boundary x

n
(t) separating the continuation region {(x, t) : x <
x

n
(t), t < T} and the stopping region {(x, t) : x x

n
(t), t < T}. The
free boundary is not known in advance but has to be determined as
part of the solution of the pricing model. This is a multiple optimal
stopping problem.
At times close to expiry, the investor should choose to reset even
when F(t) is only slightly below I(t), so we deduce that x

n
(T

) = 1.
The free boundary x

n
(t) is a monotonically decreasing function
of t since the holder should reset at a lower threshold fund
value as time is approaching maturity. At xed value of t, x

n
(t)
is a monotonically decreasing function of n since the investor
should become more conservative with less number of resets
outstanding.
102
The exercise payo is the value of the dynamic protected fund with
outstanding reset rights reduced by one. When n , x

n
() tends
to 1 since the investor chooses to reset whenever

F(t) falls to I(t).
As x

n
() > x

n
(t) > 1, so x

(t) = 1.
103
Simplied pricing model under limit of innite number of re-
sets - automatic reset
With x

(t) = 1 for all t < T. This gives W

(x

, t) = W

(1, t). For

convenience, we dene
W

(y, t) =
V

(

F, I, t)

F
, where y = lnx = ln
I

F
, = T t.
Note that the free boundary x

n
(t) becomes the xed boundary y =
lnx

(t) = ln1 = 0.
The governing equation and auxiliary conditions for W

(y, ) are
reduced to
W

=

2
2

2
W

y
2
+
W

y
q
p
W

, > 0, y < 0;
W

y
(0, ) = W

(0, ), W

(y, 0) = 1,
where = q
p
q
i

2
2
[Note that y < 0 F(t) > I(t)].
104
The Robin boundary condition:
W

y
(0, ) = W

(0, ) can be ex-

pressed as
V

F
(I,

F, t) = 0 at

F = I. If the index value is taken to
be the constant value K, then the Neumann boundary condition:
V

F
(

F, t)

F=K
= 0 is equivalent to the reecting boundary condi-
tion with respect to the upgraded fund

F placed at the guarantee
level K.
The Robin boundary condition at y = 0 leads to a slight complica-
tion in the solution procedure. The analytic representation of the
solution W

(y, ) admits dierent forms, depending on q

p
= q
i
or
q
p
= q
i
.
105
Let = 2(q
i
q
p
)/
2
and = +
2
, the price function V

(F, I, t)
is found to be
(i) q
p
= q
i
V

(F, I, t)
= Ie
q
i

_
1
1

_
N
_
ln(I/F) +

_
+
I

_
I
F
_

e
q
p

N
_
ln(I/F)

_
+ Fe
q
p

N
_
ln(I/F)

_
, F > I.
(ii) q
p
= q
i
(write the common dividend yield as q)
V

(F, I, t)
= Ie
q

n
_
ln(I/F) +(
2
/2)

_
+Ie
q
_
ln
I
F
+1 +

2

2
_
N
_
ln(I/F) +(
2
/2)

_
+ Fe
q
N
_
ln(I/F) +(
2
/2)

_
, F > I.
106
Derivation of W

(y, )
We perform the continuation of the initial condition to the whole
domain (, ), where
W

(y, 0) =
_
1 if y < 0,
(y) if y 0.
y
no upgrading occurs
until maturity
(y) to be determined
so as to satisfy the
Robin condition
107
The function (y) is determined such that the Robin boundary
condition is satised. Let g(y, ; ) denote the fundamental solution
to the governing dierential equation, where
g(y, ; ) =
e
q
p

2
2

exp
_

( y )
2
2
2

_
.
The following relations are useful in later derivation procedure.

g(y, ; ) d = e
q
p

_
1 N
_
y +

__
and

g(y, ; ) d = e
q
p

n
_
y +

_
= g(y, ; 0).
108
The solution to W

(y, ) can be formally expressed as

W

(y, ) =

(, 0)g(y, ; ) d
= e
q
p

_
1 N
_
y +

__
+

0
()g(y, ; ) d.
Performing the dierentiation with respect to y on both sides of the
above equation and integration by parts, we have
W

y
(y, ) = g(y, ; 0)

0
()
g

(y, ; ) d
=

()g(y, ; ) d.
109
Note that
W

(0, ) =

0
()g(0, ; ) d +

0
g(0, ; ) d.
We apply the Robin boundary condition:
W

y
(0, ) = W

(0, ) to
obtain

0
{[

() ()]g(0, ; ) g(0, ; )} d = 0
This relation works for any choice of g. In order that the Robin
boundary condition is satised, () and g(0, ; ) have to observe
the following relation:

() () =
g(0, ; )
g(0, ; )
= e
(+1)
, where =
2(q
i
q
p
)

2
.
Interestingly, g(0, ; )/g(0, ; ) has no dependence on . If other-
wise, this procedure does not work.
110
The auxiliary condition for () is obtained by observing continuity
of W(y, 0) at y = 0, giving (0) = 1. The solution of () depends
on = 0 or = 0, namely,
(i) when = 0,
() = e

_
e

+1
1

_
;
(ii) when = 0, () = e

_
1 + lim
0
e

_
= (1 +)e

.
111
By substituting the known solution of (), we obtain
(i) when = 0,
W

(y, ) = e
yq
i

_
1
1

_
N
_
y +

_
+
1

e
(1+)yq
p

N
_
y

_
+ e
q
p

N
_
y

_
, y < 0;
(ii) when = 0 (write q as the common dividend yield),
W

(y, ) = e
q
N
_
y +(
2
/2)

_
+e
yq

n
_
y +(
2
/2)

_
+ e
yq
_
y +1 +

2

2
_
N
_
y +(
2
/2)

_
, y < 0.
Remark
When q
p
= 0, the primary fund does not pay dividend. The dynamic
protection has no value, so we expect
lim
F
V

(F, I, t) = F.
112
Mid-contract valuation
Let M denote the path dependent state variable which represents
the realized maximum value of the state variable x from the grant-
date to the mid-contract time t, that is,
M = max
0ut
I(u)
F(u)
.
At any mid-contract time, the number of units of primary fund held
in the investment fund is given by
n(t) =
_
1 if M 1,
M if M > 1.
If the primary fund value has been staying above or at the reference
stock index so far (corresponding to M 1), then upgrade has never
occurred, so the number of units of primary fund remains at one.
Otherwise, the number of units is upgraded to M.
113
Governing dierential equation
Let V
mid
(F, I, M, t) denote the mid-contract investment fund value
at time t, with dependence on the state variable M. Like the govern-
ing equation for the lookback option price function, M only enters
into the auxiliary conditions but not the governing dierential equa-
tion. Both V
mid
(F, I, M, t) and V

(F, I, t) satisfy the same two-state

Black-Scholes equation, namely,
_

t
+L
F,I
_
V
mid
(F, I, M, t) = 0 and
_

t
+L
F,I
_
V

(F, I, t) = 0.
(A.1)
where
L
F,I
=

2
p
2
F
2

2
F
2
+
p

i
FI

2
FI
+

2
i
2
I
2

2
I
2
+(rq
p
)F

F
+(rq
i
)I

I
r.
114
Terminal payo
The terminal payo of the investment fund value at maturity T
is given by F max(M, 1), a payo structure that involves both F
and M. The valuation of the mid-contract value may seem to be
quite involved, but economic intuition gives the mid-contract value
V
mid
(F, I, M, t) in terms of the grant-date value V

(F, I, t).
Relationship between V
mid
(F, I, M, t) and V

(F, I, t)
When M > 1, the number of units of primary fund is increased to
M so that the investment fund is equivalent to one unit of new
primary fund having fund value MF. When M 1, V
mid
is insensitive
to M since the terminal payo value will not be dependent on the
current realized maximum value M. That is, V
mid
remains constant
at dierent values of M, for all M 1. By continuity of the price
function with respect to the variable M, V
mid
at M = 1 is equal to
the limiting value of V
mid
(corresponding to the regime: M > 1) as
M 1
+
.
115
Mid-contract value function
V
mid
(F, I, M, t) = V

(max(M, 1)F, I, t)
=
_
V

(F, I, t) M 1,
V

(MF, I, t) M > 1.
(A.2)
For M 1, V
mid
(F, I, M, t) = V

(F, I, t) is obvious since there

has been no upgrading of fund value occurs. It is only necessary
to show
V
mid
(F, I, M, t) = V

(MF, I, t) for M > 1.

116
Proof of the formula
When the current index value I equals MF, then V
mid
is insensitive
to M. We impose the usual auxiliary condition for a lookback option:
V
mid
M

M=I/F
= 0.
One can check that V
mid
(F, I, M, t) given in Eq.(A.2) satises the
governing dierential equation (A.1) since the multiplier M is can-
celed in the dierential equation when F is multiplied by M. It
suces to show that V
mid
(F, I, M, t) satises the terminal payo
condition and the above auxiliary condition.
Firstly, when M > 1, it is guaranteed that M at maturity must be
greater than one so the terminal payo becomes F max(1, M) =
MF. Since V

(F, I, T) = F so that V

(MF, I, T) = MF, hence the

terminal payo condition is satised.
117
Recall: W

_
ln
I
F
,
_
=
V

(F, I, t)
F
so that
V

(MF, I, t) = MFW

_
ln
I
MF
,
_
.
We perform dierentiation with respect to M to obtain
V

M
(MF, I, t) = F
_
W

(y, )
W

y
(y, )
_
,
where y = ln(I/MF). When M = I/F, we have y = 0 so that
V
mid
M
(F, I, M, t)

M=I/F
=
V

M
(MF, I, t)

M=I/F
= F
_
W

(0, )
W

y
(0, )
_
= 0
by virtue of the Robin boundary condition.
118
Cost to the sponsor
U
grant
(F, I, t) = cost to the sponsor at the grant date
U
mid
(F, I, M, t) = cost to the sponsor at mid-contract time
Note that the terminal payo U
mid
(F, I, M, T) is given by
U
mid
(F, I, M, T) = V
mid
(F, I, M, T) F = max(M 1, 0)F.
Relation between U
mid
and V
mid
U
mid
(F, I, M, t) = V
mid
(F, I, M, t) Fe
q
p
(Tt)
since both V
mid
(F, I, M, t) and Fe
q
p
(Tt)
satisfy the Black-Scholes
equation and the terminal payo condition.
The factor e
q
p
(Tt)
appears in front of F since the holder of the
protected fund does not receive the dividends paid by the primary
fund.
119
At the grant-date, we have F I so that M = I/F 1. We then
have
U
grant
(F, I, t) = U
mid
(F, I, M, t) = V

(F, I, t) Fe
q
p
(Tt)
.
The two cost functions U
mid
(F, I, M, t) and U
grant
(F, I, t) are related
by
U
mid
(F, I, M, t) = U
grant
(max(M, 1)F, I, t)+max(M1, 0)Fe
q
p
(Tt)
.
The last term in the above equation gives the present value of
additional units of primary fund supplied by the sponsor due to the
protection clause. The sponsor has to add M 1 units of primary
fund when M > 1, but supplements nothing when M 1.
120
Integral representation of the price function V
mid
(F, I, M, t)
Let

M
t

t
= max
_
max
tut

I(u)
F(u)
, 1
_
and at the current time t, the following quantity

M
t
0
= max
_
max
0ut
I(u)
F(u)
, 1
_
= max(M, 1)
is known. The terminal payo of the protected fund can be ex-
pressed as max(

M
t
0
,

M
T
t
)F
T
, and F is the current value of the primary
fund.
121
Rollover hedging strategy
We increase the number of units of fund to

M
u
t
e
q
p
(Tu)
whenever
a higher realized maximum value of

M
u
t
occurs at time u, where
t u T. This rollover strategy would guarantee that the number
of units of fund at maturity T is max(

M
t
0
,

M
T
t
). The corresponding
present value of the replenishment premium is
e
q
p
(Tt)
FE[max(

M
T
t


M
t
0
, 0)] = e
q
p
(Tt)
F

M
t
0
P[

M
T
t
] d.
The value of the protected fund is the sum of the value of the sub-
replicating portfolio and the replenishment premium. An alternative
analytic representation of the mid-contract price function is given
by
V
mid
(F, I, M, t) = max(M, 1)e
q
p
(Tt)
F
+ e
q
p
(Tt)
F

max(M,1)
P[

M
T
t
] d.
122
Note that max(M, 1) is the number of units of primary fund to be
held at time t, given the current value of M. In the remaining life
of the contract, the holder is not entitled to receive the dividend
yield, so the discount factor e
q
p
(Tt)
should be appended.
Distribution function
For 1, we have the following distribution function under the
Geometric Brownian motion:
P[

M
T
t
] = P
_
max
tuT
I(u)
F(u)

_
= e
2/
2
N
_

_
+N
_

_
,
where = q
p
q
i

2
2
and
2
=
2
p
2
p

i
+
2
i
.
123