Discussion Paper Series A:

No. 2007-184

Valuing Continuous-Installment Options

Toshikazu KIMURA

July 2007
(Revised: July 8, 2007)

Graduate School of Economics and

Business Administration
Hokkaido University
Nishi 7, Kita 9, Kita-ku
Sapporo 060-0809, Japan

Valuing Continuous-Installment Options

Toshikazu KIMURA
Graduate School of Economics and Business Administration, Hokkaido University
Kita 9, Nishi 7, Kita-ku, Sapporo 060-0809, Japan
Abstract
Installment options are path-dependent contingent claims in which the premium
is paid discretely or continuously in installments, instead of paying a lump sum at the
time of purchase. This paper deals with valuing European continuous-installment
options written on dividend-paying assets in the standard Black-Scholes-Merton
framework. The valuation of installment options can be formulated as a free boundary problem, due to the exibility of continuing or stopping to pay installments. On
the basis of a PDE for the initial premium, we derive an integral representation for
the initial premium, being expressed as a dierence of the corresponding European
vanilla value and the expected present value of installment payments along the optimal stopping boundary. Applying the Laplace transform approach to this PDE, we
obtain explicit Laplace transforms of the initial premium as well as its Greeks, which
include the transformed stopping boundary in a closed form. Abelian theorems of
Laplace transforms enable us to characterize asymptotic behaviors of the stopping
boundary close and at innite time to expiry. We show that numerical inversion of
these Laplace transforms works well for computing both the option value and the
optimal stopping boundary.

Keywords: continuous installments; European style; stopping boundary; integral

representation; Laplace transforms; asymptotic properties; numerical inversion
AMS Subject Classifications: 90A09, 91B28, 93E20
JEL Classifications: G12, G13

Introduction

Installment options or pay-as-you-go options are path-dependent claims in which a small

amount of up-front premium instead of a lump sum is paid at the time of purchase, and
then a sequence of installments are paid up to a xed maturity. The holder has the right
of stopping payments at any time, thereby terminating the option contract: If the option
is not worth the NPV (Net Present Value) of the remaining payments, she/he does not
have to continue to pay further installments. Hence, an optimal stopping problem arises
for the installment option even in European style. The option can be exercised only if all
installments are paid until maturity. Due to the additional right to terminate payments,
the total premium charged for an installment option is greater than that for a standard
1

option. An installment option with payments at pre-specied dates is usually referred to

as a discrete-installment option, whereas its continuous-time limit in which premium is
paid at a certain rate per unit time is referred to as a continuous-installment option. This
paper deals with a European-style continuous-installment option.
In actual markets, installment options have been traded actively, e.g., installment warrants on Australian stocks listed on the Australian Stock Exchange (ASX) (Ben-Ameur
et al., 2005, 2006), a 10-year warrant with 9 annual payments oered by Deutsche Bank
(Davis et al., 2001) and so on. Also, many life insurance contracts and capital investment projects can be thought of as installment options (Davis et al., 2004). For example,
Majd and Pindyck (1987) developed a model for optimal sequential investment, in which
a rm invests continuously until the project is completed, investment can be stopped
and later restarted without paying any additional costs. Their model can be considered
as a European-style continuous-installment option where the remaining expenditure required to complete the project is used as a state variable instead of time; see also Dixit
and Pindyck (1994, Chapter 10). However, there have been relatively few studies on
installment options: For European-style discrete-installment options, the case of two installments is the compound option, which is an option written on an option; see Geske
(1977, 1979). Davis et al. (2001, 2002) applied the concept of compound options and
NPV to obtain no-arbitrage bounds of the initial call premium in the (possibly more general) Black-Scholes-Merton framework, and then to examine dynamic and static hedging
strategies. They intuitively showed that holding an installment call option is equivalent to
holding an associated European call option with the same payo plus the right to sell this
option at any installment date at a price equal to the NPV of all future installments. The
latter can be understood as an American compound put option written on the vanilla
call option, where all the maturity dates are same. Griebsch et al. (2007) proved this
intuitive idea on premium decomposition to be correct; see also Wystup et al. (2004). For
American-style discrete-installment options, Ben-Ameur et al. (2006) developed a DP algorithm for computing the option value approximated by a piecewise-linear interpolation,
which is applied to valuing ASX installment warrants with dilution eects.
For continuous-installment options, Ciurlia and Roko (2005) analyzed the American
case approximately by applying the multipiece exponential function (MEF) method to
an integral representation of the initial premium. To check the accuracy of the MEF
approximation, they executed numerical comparisons with benchmark results obtained
by the nite dierence method as well as a Monte Carlo method. The MEF method
has been originated by Ju (1998) in the valuation of the standard American put option,
which generates a piecewise continuous, i.e., discontinuous optimal stopping and early
exercise boundaries. This discontinuity is a serious obstacle to decision-makings of the
option holder. As for the European case, Alobaidi et al. (2004) used an integral transform
to solve a free boundary problem due to the exibility of continuing or stopping to pay
installments, obtaining asymptotic properties of an optimal stopping boundary close to

maturity. However, their method is not appropriate for quantitative valuation, because
the integral transform adopted there is too special to invert it numerically. The target of
this paper is also a European continuous-installment option written on a dividend-paying
asset in the setup of the standard Black-Scholes-Merton framework, to which we apply a
Laplace transform approach.
This paper is organized as follows: In Section 2, on the basis of a partial dierential
equation (PDE) for the values of the continuous-installment call/put options, we derive
an integral representation for each initial premium, being expressed as a dierence of the
corresponding European vanilla value and the expected present value of installment payments along the optimal stopping boundary. In Section 3, applying the Laplace transform
approach to the PDE, we obtain explicit Laplace transforms of the initial premium as well
as its Greeks, which include the transformed stopping boundary in a closed form. We
prove that the Laplace transform of the initial premium can be decomposed into those
of the associated vanilla option and its American compound put option. Abelian theorems of Laplace transforms enable us to characterize asymptotic behaviors of the stopping
boundary. In Section 4, we show some computational results for particular cases with the
aid of numerical Laplace transform inversion. Finally, in Section 5, we conclude and give
further remarks as well as directions of future research.

Integral Representation

Let (St )t0 the price process of the underlying asset. Assume that (St )t0 is a riskneutralized diusion process described by the linear stochastic dierential equation (SDE)
dSt
= (r )dt + dWt ,
St

t 0,

(2.1)

where r > 0 is the risk-free rate of interest, 0 is the continuous dividend rate,
and > 0 is the volatility coecient of the asset price. In (2.1), W (Wt )t0 denotes a one-dimensional standard Brownian motion process on a ltered probability space
(, (Ft )t0 , F , P) where (Ft )t0 F is the natural ltration corresponding to W and the
probability measure P is chosen so that the stock has mean rate of return r. In addition,
let q > 0 be the continuous installment rate, which means the holder pays an amount q dt
in time dt, while the asset itself pays a continuous dividend in the amount of St dt to
the holder at the same time.
The initial premium V V (t, St ; q) of a continuous-installment option is a function
of the time t, the current asset value St S, and the continuous installment rate q. From
the standard argument of constructing the hedged portfolio consisting of one option and
an amount V
of the underlying asset, we see that the initial premium V satises an
S
inhomogeneous PDE
2V
V
V
+ 12 2 S 2 2 + (r )S
rV = q.
t
S
S
3

(2.2)

See Ciurlia and Roko (2005) for details. If q = 0, then the homogeneous equation agrees
with the so-called Black-Scholes-Merton PDE.

2.1

Call Case

Consider a European-style installment call option with maturity date T and strike price
K. The payo at the maturity is given by (ST K)+ , where (x)+ = max(x, 0). Let
c c(t, St ; q) denote the value of the continuous-installment call option at time t [0, T ].
In the absence of arbitrage opportunities, the value c(t, St ; q) is a solution of an optimal
stopping problem


 
q
r(T t)
+
r(c T t) 
(2.3)
1e
(ST K)
c(t, St ; q) = ess sup E 1{c T } e
 Ft
r
c [t,T ]
for t [0, T ], where a b = min{a, b}, c is a stopping time of the ltration F and the
conditional expectation is calculated under the risk-neutral probability measure P. The
random variable c [t, T ] is called an optimal stopping time if it gives the supremum
value of the right-hand side of (2.3). If q 0, then c = T (a.s.), i.e., it is not optimal to
stop paying installments before maturity.
Solving the optimal stopping problem (2.3) is equivalent to nding the points (t, St )
for which termination of the contract is optimal. Let D = [0, T ] [0, +), and S and
C denote the stopping region and continuation region, respectively. In terms of the value
function c(t, St ; q), the stopping region S is dened by
S = {(t, St ) D | c(t, St ; q) = 0} ,
for which the optimal stopping time c satises
c = inf{u [t, T ] | (u, Su) S}.
The continuation region C is the complement of S in D, i.e.,
C = {(t, St ) D | c(t, St ; q) > 0} .
The boundary that separates S from C is referred to as a stopping boundary (or a cancellation boundary), which is dened by
S t = inf {St [0, +) | c(t, St; q) > 0} ,

t [0, T ].

(2.4)

Since c(t, St ; q) is nondecreasing in St , the stopping boundary (S t )t[0,T ] is a lower critical

asset price below which it is advantageous to terminate the option contract by stopping
the payments, and it vanishes when q 0, i.e., S t 0 for t [0, T ].
In the continuation region C, the call value c(t, S; q) (S St for abbreviation) is
obtained by solving the inhomogeneous PDE
c
c 1 2 2 2 c
+ 2 S
rc = q,
+ (r )S
2
t
S
S
4

S > St,

(2.5)

with the boundary conditions


 lim c(t, S; q) = 0,
 SS t



 lim c = 0,
 SS t S



 lim c < .
 S S

(2.6)

The rst (value matching) condition implies that the initial premium is continuous across
the stopping boundary, and the second (smooth pasting) condition further implies that
the slope is continuous. The terminal condition is clearly given by
c(T, S; q) = (S K)+ .

(2.7)

Theorem 1 The value function of the continuous-installment call option has the integral
representation
 T


er(ut) d (St , S u , u t) du,
(2.8)
c(t, St ; q) = c(t, St ) q
t

where ( ) is the standard normal cumulative distribution function given by

 x
1
2
(x) =
ey /2 dy,
x R,
2
log(x/y) + (r 12 2 )

,
d (x, y, ) =

(2.9)

and c(t, St ) = c(t, St ; 0) is the value of the associated vanilla call option, i.e.,




c(t, St ) = St e(T t) d+ (St , K, T t) Ker(T t) d (St , K, T t) .

(2.10)

Proof. For u [t, T ], let f (u, Su ) er(ut) c(u, Su ; q) be the discounted option value
function dened in D. The function f (u, Su ) is convex in Su for all u and belongs to the
class C1,2 (D ). Applying Itos lemma to f (u, S) (S Su for abbreviation), we have

 T
 T
2
f
f
1 2 2 f
S
+
f (T, ST ) = f (t, St ) +
dS +
du,
2
S
S 2 u
t
t
which yields
er(T t) c(T, ST ; q) = c(t, St ; q)


 T
 T
2
c
r(ut) c
r(ut)
1 2 2 c
dS +
e
e
S
rc(u, S; q) +
+
du. (2.11)
2
S
S 2
u
t
t
Substituting c(T, ST ; q) = (ST K)+ and
c(t, St ; q) = 1{St >S t } c(t, St ; q) + 1{St S t } 0 = 1{St >S t } c(t, St ; q)

into (2.11), we obtain

r(T t)

(ST K) = c(t, St ; q) +

r(ut)

e
t

c
S1{S>S u } dWu + q
S

T
t

er(ut) 1{S>S u } du.

(2.12)

The conditional expectation of (2.12) with respect to Ft and the relation

c(t, St ) = E[er(T t) (ST K)+ | Ft ]
lead to


c(t, St ) = c(t, St ; q) + q

er(ut) P{Su > S u | Ft}du.


Since P{Su > S u | Ft } = P{log(Su /St ) > log(S u /St ) | Ft } and log(Su /St ) N (r

12 2 )(u t), 2 (u t) under P, we have P{Su > S u | Ft } = (d (St , S u , u t)) for
u [t, T ], which completes the proof.


The integral representation (2.8) expresses the initial premium of the continuousinstallment call option as a dierence of the corresponding vanilla call value and the
expected present value of installment payments along the optimal stopping boundary.
From (2.8), we immediately see that c(t, St ; q) c(t, St ) for t [0, T ], i.e., the payment
of installments makes the initial premium lower than the vanilla counterpart. Due to
the value matching condition, we also see that the optimal stopping boundary (S t )t[0,T ]
satises the integral equation
 T


er(ut) d (S t , S u , u t) du = 0,
(2.13)
c(t, S t ) q
t

which can be solved numerically for (S t )t[0,T ] , e.g., by the MEF method (Ju, 1998) just
as in Ciurlia and Roko (2005). In this paper, however, we use an alternative approach
based on Laplace transforms, which generates the transformed stopping boundary in a
closed form; see Equation (3.7) in Theorem 3.

2.2

Put Case

Now we consider a European-style installment put option with the same maturity date
T and strike price K as the call case. A notable dierence between call and put cases is
that there exists for each time t an upper asset price St above which it is advantageous to
terminate the option contract by stopping the payments. The continuous path (St )t[0,T ] is
also called the stopping boundary, which divides the whole domain D into a continuation
region C = {(t, St ) [0, T ] [0, St )} and a stopping region S = {(t, St ) [0, T ] [St , )}.
Let p p(t, St ; q) denote the value of the continuous-installment put option at time
t [0, T ]. In the continuation region C, the value p(t, S; q) can be obtained by solving the
inhomogeneous PDE
p
p 1 2 2 2 p
+ 2 S
rp = q,
+ (r )S
2
t
S
S
6

S < St ,

(2.14)

with the boundary conditions


 lim p(t, S; q) = 0,
 SSt



 lim p = 0,
 SSt S



 lim p < ,
 S0 S

(2.15)

and the terminal condition is given by

p(T, S; q) = (K S)+ .

(2.16)

Theorem 2 The value function of the continuous-installment put option has the integral
representation
 T


er(ut) d (St , Su , u t) du,
(2.17)
p(t, St ; q) = p(t, St ) q
t

where p(t, St ) = p(t, St ; 0) is the value of the associated vanilla put option, i.e.,




p(t, St ) = Ker(T t) d (St , K, T t) St e(T t) d+ (St , K, T t) ,

(2.18)

and d are dened in (2.9).

3
3.1

Valuation with Laplace-Carlson Transforms

Call Case

With the change of variables = T t, let

c(, S; q) = c(T , S; q) = c(t, St ; q) and
= S T = S t for 0. We refer to (S )0 T as the backward running process of
S
(St )t0 . For > 0, dene the Laplace-Carlson transforms (LCTs) of these time-reversed
quantities as

c (, S; q) = LC[
c(, S; q)]

and
]
S () = LC[S

e
c(, S; q)d,

d.
e S

No doubt, there is no essential dierence between the LCT and the Laplace transform
(LT) dened by


c(, S; q) = L[
c(, S; q)]
e
c(, S; q)d.
0

Clearly, we have c (, S; q) =

c(, S; q) for > 0. The principal reason why we prefer

LCTs to LTs is that LCTs generate relatively simpler formulas than LTs for option pricing
problems because constant values are invariant after taking transformation.
In the context of option pricing, LCTs have been rst adopted in the randomization of
Carr (1998) for valuing an American vanilla put option with an exponentially distributed
7

random maturity T . The idea of randomization gives us another interpretation that the
LCT c (, S; q) can be regarded as an exponentially weighted sum (integral) of the timereversed value
c(, S; q) for (innitely many) dierent values of the maturity T R+ , and
hence for R+ , which makes LCTs be well dened. Actually, Carrs randomization is
an algorithm of extracting
c(, S; q) for a specied value of > 0 from the sum c (, S; q)
by a sequence of Erlangian distributions converging to Diracs delta function concentrated
at .
Lemma 1 Let c (, S) = LC[
c(, S)] be the LCT of the associated vanilla call value for
the backward running process. Then,

S<K
1 (S),

c (, S) =
2 (S) + S K , S K,
+ +r
where for i = 1, 2,

K
i (S) =
1 2 +

   i

S
r
3i
,
1
+r
K

and the parameters 1 1 () > 1 and 2 2 () < 0 are two real roots of the quadratic
equation
1 2 2
+ (r 12 2 ) ( + r) = 0.
2
Proof. As c(t, S) = c(t, S; 0), it satises the PDE (2.5) with q = 0, the boundary conditions

 lim c(t, S) = 0
 S0

(3.1)

 lim dc < ,
 S dS
and the terminal condition c(T, S) = (S K)+ . Hence, the call value for the backward
running process can be obtained by solving the PDE

c 1 2 2 2
c

c
+ 2 S
r
c = 0,
+
(r

)S

S 2
S

S > 0,

(3.2)

with the conditions of the same form as (3.1) and the initial condition
c(0, S) = (S K)+ .
Taking the LCTs of (3.2) and the boundary conditions, we see that c (, S) satises the
ordinary dierential equation (ODE)
1 2 2
S
2

dc
dc
( + r)c + (S K)+ = 0,
+
(r

)S
dS 2
dS

with the boundary conditions


 lim c (, S) = 0
 S0



 lim dc < .
 S dS

S > 0,

(3.3)

(3.4)

It is straightforward to solve (3.3) with (3.4) as well as the continuity conditions of c (, S)

and its rst derivative at S = K. Assuming a general solution of the form
2
  S i

ai
,
S<K

K
i=1
(3.5)
c (, S) =
 i
2

S
K
S

, S K,
ai+2
+

K
+ +r
i=1
for unknown constants ai (i = 1, . . . , 4), we obtain the desired results.

Using the PDE (2.5) with the conditions (2.6) and (2.7) and Lemma 1, we obtain a
closed-form solution for the LCT c (, S; q) as follows:
Theorem 3 If S > S ,
1
q
c (, S; q) = c (, S) +
+ r 1 2

S
S

2

q
,
+r

(3.6)

otherwise c (, S; q) = 0, where S S () is given by



2( + )q
S () =
(1 2 )K 2

11
K.

(3.7)

Proof. For S [0, S ], the result is obvious because halfway cancellation is optimal in
this region. In a similar way in the proof of Lemma 1, we obtain the ODE for c (, S; q)
as
2
dc
1 2 2d c
( + r)c = q (S K)+ ,

S
+
(r

)S
S > S ,
(3.8)
2
dS 2
dS
together with the boundary conditions

 lim c (, S; q) = 0,
 SS


dc

= 0,
 lim
(3.9)
 SS dS


dc

< .
 lim
S dS
For a given S , it is straightforward to solve the ODE (3.8) with the last two boundary
conditions in (3.9) as well as the continuity conditions of c (, S; q) and its rst derivative
at S = K, assuming a general solution of the form
2
  S i

,
S < S < K,
ai

K
+r
i=1
c (, S; q) =
(3.10)
  i


S
S
K + q

ai+2
+

, SK

K
+
+r
i=1
9

which yields a3 = 0 and

a1 = 1 (K),

 1 2

a = (K) 1 S
,
2
1
2 K

 

1 S 1 2

.
a4 = 2 (K) 1 (K)
2 K

(3.11)

Rearranging the terms in the equations above and using Lemma 1, we obtain the nal
result (3.6) after somewhat cumbersome calculations. The LCT S can be obtained by
applying the rst (i.e., value matching) condition in (3.9) into (3.10), yielding


S
K

1
=

2 ( + )q
.
K { + r (r )2 }

Combining this and the relation + r (r ) = 12 2 ( 1), we obtain (3.7).

Remark 1 When the position is deep-in-the-money, we have

c (, S; q)

K + q
S

,
+
+r

which implies that the option value has the asymptote

c(t, S; q) Se(T t) Ker(T t)


q
1 er(T t) ,
r

for large S K. We see from Theorem 1 that this asymptote gives a lower bound of the
+

option value, because c(t, S) Se(T t) Ker(T t) and ( ) 1.
Griebsch et al. (2007) proved that the total premium has the decomposition
c(t, St ; q) + Kt = c(t, St ) + Pc (t, St ; q),

(3.12)

where


q
1 er(T t)
r
is the NPV of the future payment stream at time t, and for the set St,T of stopping times
of the ltration F with values in [t, T ] (a.s.),
Kt =

 

Pc (t, St ; q) = ess sup E er(st) (Ks c(s, Ss ))+  Ft
sSt,T

represents the value of an American compound put option maturing in time T written on
the vanilla call option. From (2.8), (3.12) and the relation (x) = 1 (x) (x R), we
obtain
 T


Pc (t, St ; q) = q
er(ut) d (St , S u , u t) du.
(3.13)
t

10

Also, by virtue of Theorem 3, the LCT of the time-reversal Pc (, S; q) LC[Pc (, S; q)]

is given by
  2
1
S
q

,
(3.14)
Pc (, S; q) =
+ r 1 2 S


because
LC




q
q
1 er(T t) = Kt .
eru du =
=q
+r
r
0

As a corollary of Theorem 3, we obtain a few Greeks of c(t, S; q), i.e., delta, gamma
and theta, in certain hybrid forms:
Corollary 1 For S > S t , we have
c(t,S;q) =

c
= c(t,S) + LC 1 [Pc ],
S

2c
= c(t,S) + LC 1 [Pc ],
S 2
c
= c(t,S) + qer + LC 1 [Pc ],
=

c(t,S;q) =
c(t,S;q)
where



c(t,S) = e d+ (S, K, ) ,
c(t,S) =
c(t,S)


e 
d+ (S, K, ) ,
S






Se 
= d+ (S, K, ) + Se d+ (S, K, ) rKer d (S, K, ) ,
2

and
Pc

1 2 1
q
=
+ r 1 2 S

S
S

 2
< 0,

  2
S
q 1 2 (2 1) 1
Pc =
> 0,
2
+ r 1 2 S
S
  2
S
1
q
Pc =
< 0.
+ r 1 2 S
Proof. The results for c(t,S;q) and c(t,S;q) can be obtained from the premium decomposition (3.12), the Black-Scholes formula (2.10) for c(t, S) and the relations

 2 

Pc
Pc
2 Pc
Pc

and Pc = LC
,
Pc = LC
=
=
S
S
S 2
S 2
while the result for c(t,S;q) follows from (3.12), (2.10) and




Pc
Pc = LC
= Pc (, S; q) P c (0, S; q) = Pc (, S; q),

11

because
P c (0, S; q) = Pc (T, S; q) = c(T, S; q) + KT c(T, S)
= (S K)+ (S K)+ = 0,

which completes the proof.

Remark 2 We see from (3.6) and (3.8) that there exists a parity relation among the
Greeks for Pc (, S; q) such that
1 2 2
S Pc
2

+ (r )SPc + Pc = rPc,

(3.15)

and hence that one of the Greeks can be computed by the other two Greeks and Pc .
Using the Abelian theorems of Laplace transforms, we can characterize asymptotic
behaviors of the stopping boundary at a time close to expiration and at innite time to
expiration, which is
) 0 , we have
Theorem 4 For the time-reversed stopping boundary (S

= lim S = K,
lim S

and

tT

>0
,
=
lim S
2q

, = 0.
2r + 2

at = 0 (i.e., t = T )
Proof. By virtue of the initial-value theorem of LTs, the value S

can be obtained by letting in S (). We rst rewrite (3.7) as

2( + )q
= (1 2 )
K 2

S
K

1
(3.16)

Replacing i () (i = 1, 2) in (3.16) by their large asymptotics, i.e., 1 () O( ) and

2 () O( ) (as ), we obtain

 

S ()
2q
, as .
(3.17)
O( ) exp O( ) log
2
K
K
Since S ()/K 1, if S ()/K
1 as , the right-hand side converges to 0, while
the left-hand side is positive. Hence, lim S ()/K = 1, which proves the rst half.
To prove the second half, we also use the expression (3.16) and the nal-value theorem
of LTs. If > 0, then the left-hand side of (3.16) diverges as 0, while the right-hand
side becomes


S (0+) 1

(1 2 )
,
K

12

where i = lim0 i (i = 1, 2). Hence,

= S (0+) = .
lim S

If = 0, then 1 = 1 and 2 = 2r/ 2 , so that (3.16) can be rewritten as



2r S (0+)
2q
,
= 1+ 2
K 2

K

which proves the result for = 0.

Remark 3 Consider the case where = 0 and = (i.e., T = ). Then, due to

lim0 1 (K) = K and lim0 2 (K) = 0, the coecients ai (i = 1, 2, 4) in (3.11) have the
limits

lim a1 = K,

0

1+ 2r2
(3.18)

2q
K 2

lim a2 = lim a4 =
,
0
0
2r
(2r + 2 )K
from which for S > S (0+) = 2q/(2r + 2 ) we obtain

 2r2

2q
q 2
q

.
+
S

lim c(, S; q) = c (0+, S; q) =

r 2r + 2 (2r + 2 )S
r

(3.19)

Note that c (0+, S; q) does not depend on the strike price K.

3.2

Put Case

For continuous-installment put options, we also prove theorems similar to Theorems 3

and 4 via the same PDE/LCT approach as above, although properties of the stopping
boundary for the put case are subtly dierent from the call case. We will provide the
corresponding results without proofs except Theorem 6.
( 0), dene the LCTs
For the time-reversed quantities p (, S; q), p (, S) and S

p (, S; q), p (, S) and S () ( > 0). Then, we have

Lemma 2 Let p (, S) = LC[
p(, S)] be the LCT of the associated vanilla put value for
the backward running process. Then,

1 (S) S + K ,
S<K
+ +r
p (, S) =

S K.
2 (S),
Theorem 5 If S [0, S ),
2
q
p (, S; q) = p (, S)
+ r 1 2

S
S

1

otherwise p (, S; q) = 0, where S S () is given by

21

2(
+
)q

S () =
K.
(1 1)K 2
13

q
,
+r

(3.20)

(3.21)

We see from Theorem 5 that the total premium for the put option also has the decomposition
p(t, St ; q) + Kt = p(t, St ) + Pp (t, St ; q)
(3.22)
where

 

Pp (t, St ; q) = ess sup E er(st) (Ks p(s, Ss ))+  Ft
sSt,T

is the value of an American compound put option maturing in time T written on the vanilla
put option. From the integral representation (2.17) and the premium decomposition
(3.22), we obtain
 T


Pp (t, St ; q) = q
er(ut) d (St , Su , u t) du.
(3.23)
t

and from (3.20) its LCT is given by

Pp(, S; q)

2
q
=
+ r 1 2

 1
.

(3.24)

Corollary 2 For S (0, St ), we have

p(t,S;q) =

p
= p(t,S) + LC 1 [Pp ],
S

2p
= p(t,S) + LC 1 [Pp ],
S 2
p
=
= p(t,S) + qer + LC 1 [Pp ],

p(t,S;q) =
p(t,S;q)
where

 


p(t,S) = e d+ (S, K, ) 1 ,

p(t,S) =
p(t,S)


e 
d+ (S, K, ) ,
S






Se 
= d+ (S, K, ) Se d+ (S, K, ) + rKer d (S, K, ) ,
2

and
Pp

1 2 1
q
=
+ r 1 2 S

S
S

1
> 0,

 1
S
q 1 2 (1 1) 1
Pp =
> 0,
+ r 1 2 S 2 S
 1
S
2
q
Pp =
< 0.
+ r 1 2 S
) , we have
Theorem 6 For the time-reversed stopping boundary (S
0
= lim S = K,
lim S

tT

14

(3.25)

and

= 0.
lim S

(3.26)

Proof. The asymptotic result (3.25) close to expiry can be proved in much the same way
as in Theorem 4 by the use of the relation
 2
K
2( + )q
=
.
(3.27)
2
(1 1)K
S
To prove (3.26) at innite time to expiry, we let 0 in (3.27). If > 0, then the
left-hand side of (3.27) diverges as 0 because 1 > 1 and 2 > 0, and hence
= S (0+) = 0.
lim S

On the other hand, if = 0, then letting 0 in (3.27), we also have S (0+) = 0

because lim0 1 () = 1 = 1 from above and 2 > 0. Hence, the asymptotic result
(3.26) holds independently of .

Computational Results

As we saw in Remark 1 and Theorems 4 and 6, the LCTs are useful to do asymptotic
analysis. However, the LCT are also useful for computing numerical values of the option
prices and stopping boundaries by numerical inversion. Since the LCTs obtained in this
paper are so complicated that they cannot be analytically inverted by manipulating tabled
formulas of special cases, numerical inversion is the only measure left for analyzing the
real-time behaviors.
Among many methods for performing numerical Laplace transform inversion, it has
been known that some variants of the Fourier-series method are easy to implement and
to do an error analysis for a specic class of functions; see Abate and Whitt (1992) for
a comprehensive survey. In particular, the Euler method, an alternating-series approach
for inverting Laplace transforms exploiting Euler summation, is easy to use, requiring
codes of less than 50 lines. In our experiments below, we will use the Euler method for
inverting the LCTs of the stopping boundaries S () and S (), given in (3.7) and (3.21),
respectively. However, we cannot directly apply the Euler method to the inversion of
the option values c (, S; q) and p (, S; q), since the algorithm of the Euler method is
based on an integral in complex domain C where must be treated as a complex number.
The option value c (, S; q) in (3.6) (p (, S; q) in (3.20)) has dierent representations
according to the values S, K and S () (S ()), raising a grave issue of selecting a valid
expression to be evaluated when is a complex number. To dodge this issue, we use the
integral representations in Theorems 1 and 2. That is, we can compute the option value
by the following three steps:
(i) compute the value of stopping boundary at time u [t, T ] via the Euler method;
15

(ii) compute the denite integral on the nite interval [t, T ] appeared in the integral
representation via numerical integration; and
(iii) compute the option value by using the values of this integral and the associated
vanilla option.
An alternative way for computing the option value is to use another inversion method
working with the transform evaluated only at real numbers, e.g, the Gaver-Stehfest
method (Gaver, 1966; Stehfest, 1970): Consider the LCT f () = LC[f ( )]() for a
given function f ( ) L1 (R+ ). Gaver (1966) developed an inversion algorithm based on
the asymptotic result
f ( ) = lim fn ( ),
0
n

(n)
fn ( )

(m)

where fn ( )
(n 1) is dened by using a sequence {fn ( ); n, m 1} generated by the recursion



log 2
(m)

,
n=0
f0 ( ) = f m




f (m) ( ) = 1 + m f (m) ( ) m f (m+1) ( ), n 1.

n
n n1
n n1
To accelerate the convergence of (fn ( ))n1 to f ( ), Stehfest (1970) proposed an extrapolation formula
n

(1)(nk) k n

fn ( ) =
fk ( ),
k!(n k)!
k=1
which has been known under an alias of the n-point Richardson extrapolation scheme in
the context of option pricing. The procedure for generating the n-th approximation fn ( )
is called the Gaver-Stehfest method; see Abate and Whitt (1992) for details.
With the the Gaver-Stehfest method, the case-switching issue above is resolved. However, the Gaver-Stehfest method has been known to be much less robust than the Euler
method. For many problems, it works very well, but for others it does not. In addition, we
need high-precision computation (e.g. more than 30-digit precision) in the Gaver-Stefest
method, because it is based on dierentiation instead of integration. Nevertheless, we use
the Gaver-Stehfest method in our experiments as a benchmark of checking the accuracy
of the Euler method. If the two methods agree to the prescribed accuracy, we can put
condence in the results. In general, using two dierent methods has the advantage of
helping to check shortcomings in each procedure.

To evaluate the negative premium due to paying installments, we compare the values
of the continuous-installment and its associated vanilla options. Figure 1(a) (1(b)) plots
the call (put) option values as functions of S as well as their asymptotes drawn in dashed
lines. Also, the gray thin line indicates the intrinsic value of the vanilla option. From
these gures, we observe that the dierence between the two curves is almost constant at
16

50

50

40

40

30

30

20

20

10

vanilla
60

80

10

installment
100
S

120

140

installment
60

(a) call case

vanilla
80

100
S

120

140

(b) put case

Figure 1: Values of continuous-installment and the associated vanilla options (t = 0,

T = 1, K = 100, q = 10, r = 0.05, = 0.04, = 0.2)
Table 1: Values of continuous-installment call options (t = 0, T = 1, K = 100, r = 0.03,
= 0.05, = 0.2)
q
1

S
95
105
115
95
105
115
95
105
115

Euler-based
3.7071
8.3994
14.8530
2.2280
6.6385
12.9687
0.6754
4.2745
10.2533

Gaver-Stehfest
3.6841
8.3871
14.8471
2.2039
6.6210
12.9585
0.6840
4.2725
10.2489

Lower Bound
3.6076
8.3560
14.8348
1.6373
6.3857
12.8645
0.
3.4303
9.9091

least when the position is in-the-money. This observation suggests us that lower bounds
given by

 +
q
r(T t)
c(t, St ; q) c(t, St )
1e
r
(4.1)

+
q
r(T t)
1e
p(t, St ; q) p(t, St )
r
seem to be tight and they can be used as in-the-money approximations for the option
values. The bounds in (4.1) can be readily obtained by substituting ( ) 1 into the
integral representations (2.8) and (2.17).
To see the accuracy of the lower bounds in (4.1), we compare them with the option
values computed by numerical Laplace transform inversion in Tables 1 and 2 for some call
and put cases, respectively. In these tables, Euler-based denotes the values generated
by the algorithm based on the Euler-method together with the integral representation
and Gaver-Stehfest denotes those by the Gaver-Stehfest method with the 8-point ex17

Table 2: Values of continuous-installment put options (t = 0, T = 1, K = 100, r = 0.03,

= 0.05, = 0.2)
q
1

S
85
95
105
85
95
105
85
95
105

Euler-based
16.9438
10.3046
5.5703
15.0001
8.4283
3.8486
12.1253
5.7647
1.7010

Gaver-Stehfest
16.9402
10.2929
5.5413
14.9908
8.3989
3.7813
12.1159
5.7435
1.6712

105

Lower Bound
16.9381
10.2854
5.5215
14.9678
8.3151
3.5512
12.0124
5.3596
0.5957

120

100

q=15

115

95

q=10

110

q=5

q=10
90

105
q=5

85

0.2

q=15

100

0.4

0.6

0.8

0.2

0.4

0.6

0.8

(a) call case

(b) put case

Figure 2: Stopping boundaries of continuous-installment options (T = 1, K = 100,

r = 0.05, = 0.04, = 0.2)
trapolation. We see from the tables that the bounds in (4.1) certainly perform well as the
simple in-the-money approximations especially good for put options. Furthermore, we see
that the Gaver-Stehfest method generates a little bit smaller values than the Euler-based
algorithm for most cases, though both are greater the lower bound. This tendency of
underestimation also has been observed in other numerical experiments.

Figure 2(a) (2(b)) illustrates some optimal stopping boundaries (S t )t[0,T ] ((St )t[0,T ] )
for q {5, 10, 15}. For the call case, we see that the boundary value is an increasing
function of q. This property is partly justied by the inequality
dS
K
=
dq
q1

2( + )q
(1 2 )K 2
18

11
> 0,

102

112

100

110

98

108

=0.08

=0.08
96

=0.04

106
=0.04

94

104

92

102

=0.00

=0.00
90

100
0

0.2

0.4

0.6

0.8

0.2

0.4

0.6

0.8

(a) call case

(b) put case

Figure 3: Stopping boundaries of continuous-installment options (T = 1, K = 100, q = 10,

r = 0.05, = 0.2)
which is a necessary condition that dS t /dq > 0 for t [0, T ]. Similarly, from (3.21), we
can prove a necessary condition that the put boundary value is a decreasing function of
q. In addition, we see from these gures that the optimal stopping boundaries are not
monotone functions of t, namely, the call (put) boundary is a convex (concave) function of
t. Figure 3(a) and 3(b) also illustrate the optimal stopping boundaries with the dividend
yield running in {0.00, 0.04, 0.08}, which show the property that the boundary value
for call (put) is an increasing (a decreasing) function of . Note that the boundary value
at maturity T = 1 in these gures agrees with the strike price K = 100, as proved in
Theorems 4 and 6. This is a certain evidence that the Euler method works well in such
an extreme situation.
To examine the performance of the numerical transform inversion for the LCT of
Greeks, we compute the hedged parameter . Figures 4(a) and 4(b) plot as functions
of S for the installment option with q {5, 10, 15} as well as that for the associated
vanilla option drawn in a dashed line. The plot range is restricted to the interval where
the position is in-the-money, because the Gaver-Stehfest method performs very poorly
if the position is out-of-the-money. We see from these gures that for call (put) is a
decreasing (an increasing) function of q.

Conclusion

In this paper, for European continuous-installment call/put options written on dividendpaying assets, we obtained integral representations for their initial premiums via the
PDE approach, showing structural relations between two options with/without paying
installments. Combining the PDE approach with LCTs, we also obtained the LCTs of
the stopping boundaries, option values and some hedging parameters of the continuousinstallment options. Applying the Abelian theorems to the LCTs of the stopping boundaries, we characterized their asymptotic behaviors at a time to close to expiration and at
19

0.8

-0.2

0.6

-0.4

q=15
q=10
q=5
q=5
-0.6

0.4
q=10
0.2

-0.8
q=15

100

110

120

130

140

150

50

60

70

80

90

100

(a) call case

(b) put case

Figure 4: The Greek of continuous-installment options in-the-money (T = 1, K = 100,

r = 0.05, = 0.04, = 0.2)
innite time to expiration. Alobaidi et al. (2004); Alobaidi and Mallier (2006) obtained
more detailed asymptotic properties near expiration, which are
 


S t K exp (T t)| log(T t)| ,

as t T .



S K exp (T t)| log(T t)| ,

t
For further research to this direction, a certain asymptotic expansion of the LCTs S ()
and S () at = + might be helpful. In this paper, we instead computed the whole
stopping boundaries numerically. Our numerical analysis showed that either the GaverStehfest method or the Euler method work well for inverting the LCTs of stopping boundaries. However, for option values and their Greeks, the Gaver-Stehfest method sometimes
generates unstable solutions around the smooth-pasting point where these values are almost zero. This defect seems to be harmless from a practical point of view.
There are at least two straightforward generalizations of the continuous-installment
options: One direction is to generalize the payo function at the maturity. To a class of
exotic options whose initial premium satises the inhomogeneous PDE in (2.2), the same
PDE/LCT approach as in this paper can be applied, though boundary conditions for
those options may be changed. Another direction is to generalize the stochastic process
for the underlying asset price, e.g., to a process with jumps (Petrella and Kou, 2004), a
process with stochastic volatility (Lewis, 2000), and so on. The PDE/LCT approach has
a potential power to deal with these and further generalizations, likewise nite-dierence
methods and lattice models.

Acknowledgments
I would like to thank Kazuaki Kikuchi of Sumitomo Mitsui Banking Corporation for
his contribution to computational experiments. This research was supported in part by
20

the Grant-in-Aid for Scientic Research (No. 16310104) of the Japan Society for the
Promotion of Science in 20042008.

References
Abate, J. and W. Whitt, The Fourier-series method for inverting transforms of probability distributions, Queueing Systems, 10 (1992) 588.
Alobaidi, G., R. Mallier and S. Deakin, Laplace transforms and installment options,
Mathematical Models and Methods in Applied Sciences, 14 (2004) 11671189.
Alobaidi, G. and R. Mallier, Installment options close to expiry, Journal of Applied
Mathematics and Stochastic Analysis, 2006, to appear.
Ben-Ameur, H., M. Breton and P. Francois, Pricing ASX installment warrants under
GARCH, Working Paper G-2005-42, GERAD, 2005.
Ben-Ameur, H., M. Breton and P. Francois, A dynamic programming approach to price
installment options, European Journal of Operational Research, 169 (2006) 667676.
Carr, P., Randomization and the American put, Review of Financial Studies, 11 (1998)
597626.
Ciurlia, P. and I. Roko, Valuation of American continuous-installment options, Computational Economics, 25 (2005) 143165.
Davis, M., W. Schachermayer, and R. Tompkins, Pricing, no-arbitrage bounds and robust
hedging of installment options, Quantitative Finance, 1 (2001) 597610.
Davis, M., W. Schachermayer, and R. Tompkins, Installment options and static hedging,
Journal of Risk Finance, 3 (2002) 4652.
Davis, M., W. Schachermayer, and R. Tompkins, The evaluation of venture capital as
an instalment option, Zeitschrift f
ur Betriebswirtschaft, 3 (2004) 7796.
Dixit, A.K. and R.S. Pindyck, Investment under Uncertainty, Princeton University Press,
Princeton NJ, 1994.
Gaver, D.P., Observing stochastic processes and approximate transform inversion, Operations Research, 14 (1966) 444459.
Geske, R., The valuation of corporate liabilities as compound options, Journal of Financial and Quantitative Analysis, 12 (1977) 541552.
Geske, R., The valuation of compound options, Journal of Financial Economics, 7
(1979) 6381.
Griebsch, S., U. Wystup and C. K
uhn, Instalment options: A closed-form solution and
the limiting case, Working Paper, Goethe University, 2007.
Ju, N., Pricing an American option by approximating its early exercise boundary as a
multipiece exponential function, Review of Financial Studies, 11 (1998) 627646.

21

Lewis, A.L., Option Valuation under Stochastic Volatility, Finance Press, Newport Beach
CA, 2000.
Majd, S. and R.S. Pindyck, Time to build, option value, and investment decisions,
Journal of Financial Economics, 18 (1987) 727.
Petrella, G. and S. Kou, Numerical pricing of discrete barrier and lookback options via
Laplace transforms, Journal of Computational Finance, 8 (2004), 137.
Stehfest, H., Algorithm 368: Numerical inversion of Laplace transforms, Communications of the ACM, 13 (1970) 4749.
Wystup, U., Griebsch, S. and K
uhn, C., FX instalment options, Working Paper, Goethe
University, 2004.

22