DISCUSSION PAPER B272

A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN

BARRIER OPTIONS
MATTHIAS REIMER AND KLAUS SANDMANN
Abstract. The extension of the BlackScholes option pricing theory to the valuation of barrier
options is reconsidered. Working in the binomial framework of CRR we show how various types
of barrier options can be priced either by backward induction or by closed binomial formulas. We
also consider analytically and numerically the convergence of the prices in discrete time to their
continuoustime limits. The arising numerical problems are solved by quadratic interpolation.
Furthermore, the case of American barrier options is analyzed in detail. For American barrier
call options, binomial formulae and their limit results are given. Finally, the binomial approach
is applied to contracts with local and partial barrier checks.
1. Introduction
Barrier options are very similar to standard call and put options. However, a nal payo can
only occur if during a monitoring period the price of the underlying asset has depending on the
specic contract under consideration either attained or failed to attain a prespecied upper or
lower level, called the barrier. Such contracts have indeed become the most popular types of
exotic options.
Merton [1973] and in particular Conze, Viswanathan [1991] have extended the BlackScholes
model to obtain closed formulas for the valuation of several types of barrier options in continuous
time. In general, approximate prices for options can be obtained with binomial models even in
cases where it is not possible to derive closed formulas. Here we show that prices for the whole
class of barrier options can be obtained within the binomial model, if the backward induction
algorithm is suitably adjusted. Fortunately, in many cases the application of the reection prin-
ciple allows us to obtain binomial formulas and hence to avoid backward induction .
Similar to the limit result by Cox, Ross, and Rubinstein [1979] (CRR hereafter) for standard
options we recover the wellknown continuous time formulas for the price of some barrier options
as limits of binomial formulas. The results can be seen as a justication for using a binomial
Date. This version: March 1995.
Key words and phrases. Arbitrage, Barrier Option, Option Pricing, Path dependent payo.
Address: K. Sandmann, Department of Finance and Banking, University of Bonn, Adenauer-Allee 24-42,
D-53113 Bonn, Germany. E-mail: k.sandmann@wiwi.uni-bonn.de. Phone: + 49 228 738227. Fax: +49 228 73.
The authors are grateful for helpful discussions with Kristian Miltersen, Markus Schafer, Erik Schlogl, Daniel
Sommer, and Dieter Sondermann. The usual disclaimers apply. Financial support by Deutsche Forschungsge-
meinschaft, Sonderforschungsbereich 303 at University of Bonn, is gratefully acknowledged. Document typeset in
L
A
T
E
X.
1
2 MATTHIAS REIMER AND KLAUS SANDMANN
model as a discrete approximation of the continuoustime setting. However, unfortunately sim-
ulations reveal that with an increasing renement of the binomial lattice option prices converge
in a very irregular manner. We explain and solve this problem using quadratic interpolation.
The pricing of American options continues to be of great interest to researchers. In the case
of barrier options early exercise can be optimal even for call options because losses from the
underlying hitting a knockout barrier can thus be avoided. Consequently, the earlyexercise
feature of such contracts is examined in detail. Exploiting special properties of the discretetime
setup, we succeed in constructing binomial formulas for American barrier calls. In particular,
a constant early exercise level can be derived in the discrete setup. In the limit, we recover the
formulas for European barrier call options with rebate at the barrier.
Finally, we briey extend the analysis to further contract variations. Special attention should
be paid to options where the barrier is not continuously but only temporarily or locally checked,
since such features, which occur frequently in practice, can result in considerable price dier-
ences.
The rst binomial option pricing model was developed simultaneously by Cox, Ross and Ru-
binstein [1979] and Rendleman and Bartter [1979 ]. CRR presented the fundamental economic
principles of option pricing by arbitrage considerations in the simplest manner. In addition, they
showed that their binomial option pricing formula for a European call yields the BlackScholes
formula as a continuoustime limit.
The pricing of downandout options dates back to Merton [1973 ]
1
. Cox and Rubinstein
[1985] explain how the pathdependence of a downandout call can be resolved in the binomial
model. However, they do not examine the more dicult case of inoptions and American options.
In a dierent but simular context, Sondermann [1988] imposes subjective price boundaries on the
price path of the underlying in a discretetime setup. Using the reection principle he obtains
a binomial formula for which a limit result is derived. Conze, Viswanathan [1991] dene several
barrier options and derive exact replication and valuation formulas using the reection principle
in continuous time. In addition they derive some results for the corresponding American type
options. Rubinstein, Reiner [1991] list continuoustime formulas for all the eight dierent bar-
rier options. Recently, Boyle and Sok Hoon Lau [1994] have pointed out the irregularities in the
convergence of prices of barrier options in binomial lattices which we mentioned above. They
solved this diculty by extracting a subset of renements of the binomial lattice such that con-
vergence is smooth. These ndings where independently put forth in Reimer, Sandmann [1993] .
However, we additionally propose a dierent method, because the method for computing tting
tree renements may fail. A quadratic interpolation method exhibits stable pricing results for
arbitrary barrier conditions and arbitrary, especially constant, tree renements.
2. The discrete time model
Let T = 0 = t
0
< t
1
< ... < t
N
= T be the equidistant discretization of the time axis. Suppose
that S(t
0
) is the initial asset value at time t
0
. The stochastic behaviour of the asset is then
1
cp. also Ingersoll in The New Palgrave
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 3
modeled by
(1) S(t
n
, i) = S(t
0
)u
i
d
ni
i = 1, ..., n; t
n
T
where S(t
n
, i) denotes the asset price at time t
n
after i upmovements and u > d > 0, with ud =
1, are the time and state independent proportional asset movements per period. Furthermore
assume that the interest rate is constant during the time interval [0, T] and let r be the interest
rate per period. The binomial model is arbitrage free i there exists a probability measure P
such that the discounted asset price process is a martingale under P. This socalled equivalent
martingale measure exists and is unique i u > 1 + r > d where the transition probability is
given by
(2) p := P[S(t
n+1
, .) = S u[S(t
n
, i) = S] =
1 +r d
u d
Since the market structure is complete, the price of an ArrowDebreusecurity (n, i) at t
0
,which pays one unit at time t
n
if the asset price is equal to S(t
0
)u
i
d
ni
and otherwise nothing
is equal to
(n, i) :=
_
1
1 +r
_
n
_
n
i
_
p
i
(1 p)
ni
(3)
The arbitrage price of any state dependent contingent claim G whose payments are only condi-
tioned on the asset price at time t
N
is therefore equal to
(4) (G) =
_
1
1 +r
_
N
E
P
[G(S
T
)] =
_
1
1 +r
_
N N

i=0
_
N
i
_
p
i
(1 p)
Ni
G
_
S(t
0
)u
i
d
Ni
_
where (.) is the unique arbitrage free price system. With barrier options, this general pricing
principle cannot be applied in a straightforward manner. Due to the barrier condition the payo
depends on the whole price path and not only on the nal asset price at time t
N
. To overcome
this problem, one method to calculate the arbitrage free price of European type barrier options
is given by a backward induction argument
2
. Consider the case of a downandout put or call
option with barrier H. Then the following recursive algorithm yields the arbitrage price of these
barrier options. Denote by G
T
(t
n
, i) the value of a down-and-out option issued at time t
n
T
and state i = 0, ..., n with xed maturity t
N
= T. Due to the contract specication at time
t
N
= T, the value of G
T
(t
N
, i) must be equal to the immediate payo for all states i = 0, ..., N :
(5) G
T
(t
N
, i) :=
_
[S(t
0
)u
i
d
Ni
K]
+
resp. [K S(t
0
)u
i
d
Ni
]
+
if S(t
0
)u
i
d
Ni
> H
0 if S(t
0
)u
i
d
Ni
H
and t
n
T t
N
and i = 0, ..., n
(6) G
T
(t
n
, i) :=
_
1
1+r
[pG
T
(t
n+1
, i + 1) + (1 p)G
T
(t
n+1
, i)] if S(t
0
)u
i
d
ni
> H
0 if S(t
0
)u
i
d
ni
H
The backward induction is based on the martingale property of the discounted price process
G
T
. A similar recursive algorithm can be applied to upandout put or call options. Due to
the close relationship between the dierent European barrier options and standard options the
backward induction method can be applied straightforward to compute the arbitrage price of
2
For the case of a downandoutcall this was already demonstrated by Cox, Rubinstein [1985].
4 MATTHIAS REIMER AND KLAUS SANDMANN
all these options. Furthermore this algorithm can be modied easily for American type down
andout resp. upandout put or call options. For example, consider the adjustment to (6) for
an American downandout call:
(6a)
G
T
(t
n
, i) :=
_

_
maxS(t
0
)u
i
d
ni
K;
1
1+r
[pG
T
(t
n+1
, i + 1) + (1 p)G
T
(t
n+1
, i)]
if S(t
0
)u
i
d
ni
> H
0 if S(t
0
)u
i
d
ni
H
Again, a similar algorithm can be applied to American upandout put or call options. Unfor-
tunately we cannot deduce the price of American type in options from those of the American
type out options. To obtain a backward induction algorithm for American type in options
it is worthwile to consider the European case more closely.
For example, consider the downandin put option in more detail. Let H be the lower barrier
and assume that H is a possible terminal realization of the asset at time t
N
= T. Let J
H
IIN
such that H = S(t
0
)u
J
H
d
NJ
H
. Furthermore, since u d = 1 the symmetry of the binomal
lattice implies that N 2J
H
is the minimum number of immediate downmovements such that
the asset reaches the barrier for the rst time. Since H is a lower barrier (i.e. H < S) we have
N 2J
H
> 0. Note that whenever S(t
N
) = H a downandin option issued at time t
n
with
xed maturity t
N
= T is equal to a standard European option issued at time t
n
. With the same
notation as before the following algorithm yields the arbitrage price of European downandin
put options: i = 0, ..., N
G
T
(t
N
, i) :=
_
_
_
0 if S(t
0
)u
i
d
Ni
> H
[K S(t
0
)u
i
d
Ni
]
+
if S(t
0
)u
i
d
Ni
H
(7)
and t
n
T t
N
; i = 0, ..., n
G
T
(t
n
, i) :=
_

_
1
1+r
[pG
T
(t
n+1
, i + 1) + (1 p)G
T
(t
n+1
, i)] if S(t
0
)u
i
d
ni
,= H
_
1
1+r
_
Nn Nn

j=0
_
Nn
j
_
p
j
(1 p)
Nnj
_
K S(t
0
)u
i+j
d
N(i+j)

+
if S(t
0
)u
i
d
ni
= H
(8)
For the American case, the algorithm must be changed slightly. The early exercise of an in
option is only admissible if the price path has already satised the incondition. The initial
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 5
condition (7) is the same as before, whereas (8) is now changed to
3
A
T
(t
N
, i) := [K S(t
0
)u
i
d
Ni
]
+
i = 0, , N
and n = N 1, , 0
A
T
(t
n
, i) := max
_
K S(t
0
)u
i
d
ni
,
1
1 +r
[pA
T
(t
n+1
, i + 1) + (1 p)A
T
(t
n+1
, i)]
_
G
T
(t
n
, i) :=
_

_
1
1+r
[p G
T
(t
n+1
, i + 1) + (1 p) G
T
(t
n+1
, i)] if S(t
0
)u
i
d
ni
> H
A
T
(t
n
, i) if S(t
0
)u
i
d
ni
= H
max
_
K S(t
0
)u
i
d
ni
;
1
1+r
[pG
T
(t
n+1
, i + 1) + (1 p)G
T
(t
n+1
, i)]
_
if S(t
0
)u
i
d
ni
< H
(8a)
3. Closedform binomial formulae for European barrier options
As a general pricing principle, the backward induction method can be used to price European and
American type barrier options in a somehow straightforward manner. To study the convergence
behaviour of this method a closedform binomial formula for barrier options can be constructed.
Therefore we redene the notion of ArrowDebreusecurities, such that the barrier is reected.
Denition 1.
i) A downandinArrowDebreusecurity for state S(t
N
) = x is dened by the payo at
time t
N
g
d
(x, H) :=
_
1 i S(t
N
) = x and t
n
T such that S(t
n
) H
0 otherwise
(9)
ii) An upandinArrowDebreusecurity for state S(t
N
) = x is dened by the payo at
time t
N
g
u
(x, H) :=
_
1 i S(t
N
) = x and t
n
T such that S(t
n
) H
0 otherwise
(10)
Given the arbitrage prices of such conditioned ArrowDebreusecurities at time t
0
we can im-
mediately apply the argument which supports the pricing rule (4).
Proposition 1. Let H be a possible terminal realization of the asset price at time t
N
and
J
H
H such that H = S(t
0
)u
J
H
d
NJ
H
3
With A
T
(t
n
, i) we recursively calculate the arbitrage price of the standard American put option. For the
downandin call and the upandin call or put, similar algorithms can be applied. If the underlying asset is
dividend protected, then the early exercise for the American downandin call resp. upandin call option is not
optimal (see section 5.1)
6 MATTHIAS REIMER AND KLAUS SANDMANN
i) The arbitrage price
d
(N, i, H) at t
0
of a downandinArrowDebreusecurity for state
S(t
N
) = S(t
0
)u
i
d
Ni
with barrier H < S(t
0
) is equal to

d
(N, i, J
H
) =
_

_
_
1
1+r
_
N_
N
i
_
p
i
(1 p)
Ni
if i J
H
_
1
1+r
_
N_
N
2J
H
i
_
p
i
(1 p)
Ni
if J
H
i 2J
H
0 if 2J
H
< i
(11)
ii) The arbitrage price
u
(N, i, H) at t
0
of an upandinArrowDebreusecurity for state
S(t
N
) = S(t
0
)u
i
d
Ni
with barrier H > S(t
0
) is equal to

u
(N, i, J
H
) :=
_

_
0 if i <
J
H
2
_
1
1+r
_
N _
N
2J
H
i
_
p
i
(1 p)
Ni
if
J
H
2
i J
H
_
1
1+r
_
N _
N
i
_
p
i
(1 p)
Ni
if i J
H
(12)
Remark.
(1) By arbitrage we have
u
(N, i, H) +
d
(N, i, H)
_
1
1+r
_
N _
N
i
_
p
i
(1 p)
Ni
since the
payo of the left hand side portfolio weakly dominates the unconditional payo for
J
H
2
i 2J
H
and coincides otherwise.
(2) For J
H

N
2
, i.e. H S(t
0
) the downandinArrowDebreusecurity coincides with
the unconditional ArrowDebreusecurity. If J
H

N
2
, i.e. H S(t
0
) the upandin
ArrowDebreusecurity is equal to an unconditional ArrowDebreusecurity.
Proof. For H < S(t
0
), i.e. J
H
<
N
2
, the reection principle (Feller [1968]) yields the number
Z
d
(N, i, J
H
) of price paths with terminal value S(t
0
)u
i
d
Ni
which touch or cross the barrier
H = S(t
0
)u
J
H
d
NJ
H
Z
d
(N, i, J
H
) :=
_

_
_
N
i
_
if i J
H
_
N
2J
H
i
_
if J
H
i 2J
H
i = 0, .., N
0 if i > 2J
H
Since the transition probability p denes the unique equivalent martingal measure P, the arbi-
trage price of the downandinArrowDebreusecurity is given by

d
(N, i, H) =
_
1
1 +r
_
N
E
P
[g
d
(N, i, H)]
which yields (11). With simular arguments we can derive formula (12).

Consequently, these conditioned ArrowDebreusecurities can be used to compute the binomial

formulae of all European type barrier options. The following theorem summarizes this for a
European downandout call. The remaining formulae are given in the appendix (Proposition
2).
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 7
Theorem 1. Suppose the barrier H is a terminal knot of the binomial asset price process at
time t
N
, i.e. J
H
IIN
0
such that H = S(t
0
)u
J
H
d
NJ
H
, then the arbitrage price of an European
downandout call
4
with H < S(t
0
) is equal to
C
d
0
[S, K, T, H] = S(t
0
)
N

i=aJ
H
_
N
i
_
p
i
(1 p)
Ni

_
1
1 +r
_
N
K
N

i=aJ
H
_
N
i
_
p
i
(1 p)
Ni
S(t
0
)
2J
H

i=aJ
H
_
N
2
J
H
i
_
p
i
(1 p)
Ni
+
_
1
1 +r
_
N
K
2J
H

i=aJ
H
_
N
2J
H
i
_
(1 p)
Ni
(13)
where
a = infi IIN[S(t
0
)u
i
d
Ni
K,
a J
H
:= maxa, J
H

p =
pu
1 +r
, p =
1 +r d
u d
Proof. By denition of the ArrowDebreu-securities and the down-and-in ArrowDebreu secu-
rities, we have
C
do
[S, K, T, H] =
N

i=0
(N, i)
_
S(t
0
)u
i
d
Ni
K

i=0

d
(N, i, J
H
)
_
S(t
0
)u
i
d
Ni
K

+
Since H := S(t
0
)u
J
H
d
NJ
H
< S(t
0
) we have J
H
<
N
2
and by assumption J
H
0.

Under the usual assumptions, these binomial formulas converge in distribution to the well known
formulae for European type barrier options
5
in continuous time. As an example consider the
European upandout put and downandout call
6
.
Theorem 2. Let t =
Tt
0
N
be the grid size of the binomial lattice. For u = exp

t,
d = u
1
and r =
1
t
ln(1 + r) (the continuously compounded interest rate) the convergence in
the distribution of the binomial formulae is given by
lim
t0
C
do
[S(t), K, T, H] = S(t)N(x(K H)) Ke
rs
N(x(K H)

s) (14)
S(t)
_
S(t)
H
_
1
N(y(K H)) +Ke
rs

_
S(t)
H
_
1
N(y(K H)

s)
lim
t0
P
uo
[S(t), K, T, H] = Ke
rs
N
_
x(K H) +

s
_
S(t)N(x(K H)) (15)
Ke
rs
_
S(t)
H
_
1
N
_
y(K H) +

s
_
+S(t)
_
S(t)
H
_
1
N (y(K H))
4
A reasonable down barrier H < S(t
0
) should not be too low with respect to the strike level K. If H is too
small, no asset price path that touches or crosses the barrier can reach a terminal knot that yields a positive
option payo. Formally, 2J
H
a, otherwise the down-and-out-call is equal to a standard call, i.e. the last two
sums of equation (13) are by denition equal to zero.
5
see Cox, Rubinstein [1985] and Rubinstein, Reiner [1991]
6
For completeness, the remaining limit formulae are given in the appendix (Proposition 4)
8 MATTHIAS REIMER AND KLAUS SANDMANN
where
s := T t the time to maturity; K H = maxK, H; K H = minK, H
:=
2 r

2
and
x(z) :=
_
ln(
S
ze
rs
) +
1
2

2
s
_

s
; y(z) :=
_
ln(
H
2
S ze
rs
) +
1
2

2
s
_

s
Proof: see appendix.
Remark.
(1) The rst two terms of (14) are just equal to the arbitrage price of a standard European
call option. The remaining part of (14) corrects the price with respect to the barrier
condition. This correction term gives the arbitrage price of a down-and-in call option in
the case of K > H.
(2) For K < H the rst two terms of (15) are equal to the arbitrage price of a standard
European put option. In this situation the correction terms corresponds to the arbitrage
price of a European upandin call option.
4. Binomial approximation
The binomial formulae for barrier options cover only cases where the barrier H is exactly an
endpoint of the binomial tree. But application of the reection principle requires nothing more
than that the barrier H is located within the tree lattice. For barrier levels at tree knots in
between terminal knots, the binomial formula remains valid if we have H = S(t
0
)u
J
H
d
N1J
H
and the binomial coecients in (13) are computed with 2 J
H
+ 1 instead of 2 J
H
.
The arbitrage price computed by the binomial formulae with a xed grid size remains constant
for all barriers H between two knotlevels of the binomial tree. Consequently, for a given pa-
rameter constellation only with a very small number of specic tree renements the valuation
algorithm behaves properly. With deviating renements we cannot expect a monotonic conver-
gence behaviour to the limit especially when there are small grid sizes. Consider a European
downandout option. The endpoint condition on H requires that there exists a J
H
IIN such
that S(t
0
)u
J
H
d
NJ
H
= H. Dene the number k as the minimum number of immediate down
movements such that S(t
0
)d
k
= H. Obviously we have S(t
0
)u
i
d
i+k
= H for all i. Now we can
interprete the time grid or the tree renement as a function of the number k , i.e.
t =
_
ln
S
H
k
_
2
N(k) = N =
(T t
0
)k
2

2
(ln
S
H
)
2
The optimal renement number for a downandout call with rst touch after k down movements
is then dened as
7
N

(k) = max
_
i IIN[ i N(k) =
(T t
0
)k
2

2
(ln
S
H
)
2
, i k is an even number
_
7
This has been observed by Boyle and Sok Hoon Lau [1994] in an independent study. They consider the
recursive algorithms and dene the optimal renement number in a similar way.
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 9
The following gure underlines the important role of these optimal renement numbers.
0
0.5
1
1.5
2
2.5
0 50 100 150 200 250 300 350

O
P
T
I
O
N

P
R
I
C
E

TREE REFINEMENT
Figure 1: Binomial formula for a downandout call with S(t
0
) = 40, K = 40, r = 5%, =
15%, T = 365 days, H = 39 and optimal renement N

(k) = 35, 140, 315 for k = 1, 2, 3.

The appropriate grid size in a binomial model depends in a crucial manner on the barrier H.
This is obviously an unfortunate feature. If the discrete time framework is used to approximate
the continuous time model, in some sense better or quicker approximations are desirable.
Although closedform solutions for European barrier options are known, a better numerical
approximation technique is useful as a test for situations where closedform solutions are not
available or unknown.
In the case of a European downandout call the following technique appears to be very sucessful.
For a xed number of periods N resp. grid size t and a xed barrier H which is not a barrier
level of the binomial tree we can select three barriers H
1
, H
2
, H
3
of the binomial tree lattice such
that
H
1
:= S(t
0
)u
J

H
d
NJ

H
< H
2
= S(t
0
)u
J

H
d
NJ

H
1
< H
3
= S(t
0
)u
J

H
+1
d
NJ

H
1
for J

H
= maxi IIN [ S(t
0
)u
i
d
Ni
H H
1
H < H
3
Using the binomial formula we can compute the arbitrage prices of the downandout call
options with these barriers. The price of a downandout call option with barrier H [H
1
, H
3
]
is now simply approximated by the Lagrange interpolation polynomium of degree 2, i.e.
C
d
0
[S, K, T, H] f(H) =
3

i=1
L
i
(H) C
d
0
[S, K, T, H
i
] (16)
L
i
(H) =
3

j=i
(H H
j
) /
3

j=i
(H
i
H
j
)
Figure 2 gives a typical example of the success of this approximation for a xed grid size and
barriers H between 35 and the initial asset price S. There is basically no dierence between
the continuous time solution and the approximation. Actually, you cannot recognize the result,
10 MATTHIAS REIMER AND KLAUS SANDMANN
because of the precision of the approximation.
0
0.5
1
1.5
2
2.5
3
3.5
35 36 37 38 39 40

O
P
T
I
O
N

P
R
I
C
E

BARRIER
Figure 2: Approximation of the continuous time solution for a down-and-out call with a binomial
model of N = 200 periods with and without Lagrange interpolation. S(t
0
) = 40, K = 40, r =
5%, = 15% and T = 365 days.
5. American barrier options on dividend protected securities
From Merton (1973) we know that a standard American type call option on a dividend protected
asset is always more worth alive than dead, i.e. early exercise does not occur. In the case of an
out barrier option, this is not always true, since when the underlying asset reaches the barrier,
the contract becomes worthless. Thus in general, there is an incentive for early exercise just
before reaching the barrier. The following proposition extends Mertons result to the case of
barrier call options:
Proposition 5. Let the underlying security be a dividend protected security, then
a) an American down-and-in and an American up-and-in call option will never be exercised
before maturity.
b) for an American up-and-out call option with barrier H > S(t
0
) early exercise can become
optimal if and only if H > K .
c) for an American down-and-out call option with barrier H < S(t
0
) and continuous price
paths of the underlying security early exercise can become optimal if and only if H > K.
Proof.
ad a) By denition, the option can only be exercised if it is already in. In this situation, the
barrier option is equivalent to a standard call option for which Mertons result applies.
ad b) If H < K a European up-and-out call is worthless, and furthermore whenever the inner
value MaxS
t
K, 0 at time t < T is greater than zero, the barrier condition implies
that the contract is already out.
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 11
Suppose H > K and that the option is still alive at time t before maturity. The inner
value at time t is then by denition equal to
g(S
t
) =
_
S
t
K K < S
t
< H
0 S
t
H, S
t
< K
Due to H > K the early exercise payo is discontinuous at S
t
= H and bounded by
H K from above. A sucient condition for early exercise at time t is therefore given
by
(S
t
K)e
r(Tt)
H K > g(S
T
) S
T
S
t
He
r(Tt)
K
_
1 e
r(Tt)
_
ad c) i)
8
First consider the situation H K. Let t [0, T[ and assume that the option is
still alive, i.e. S
t
> H t

[0, t]. In the case H < S

t
K, there is no early
exercise, since the inner value [S
t
K]
+
is equal to zero. For H K < S
t
consider
the following portfolio: buy the down-and-out call with the barrier H and time to
maturity T t, sell the underlying asset and place the exercise price into the money
account. At t, the portfolio is worth
C
do
[S
t
, K, T t, H] S
t
+K
Now, in case the barrier is not reached until time T, the down-and-out call yields
the same payo as the standard call, and therefore the nal payo at time T is
given by
[S
T
K]
+
S
T
+Ke
r(Tt)
=
_
K(e
r(Tt)
1) 0 if S
T
K
Ke
r(Tt)
S
T
> 0 if S
T
< K
Now assume the barrier is reached at time t

]t, T[ for the rst time. Since

by assumption S
t
= H, the value of the portfolio at time t

is equal to H +
Ke
r(t

t)
0, which can be placed into the money account until time T. Thus, the
nal payo of this portfolio strategy yields a non-negative payo (even positive if
r > 0) and by means of no arbitrage, this implies a non-negative initial value of the
portfolio: C
do
S
t
K
ii) Second, consider the situation H > K. Suppose that the down-and-out call is still
alive at time t < T, i.e. S
t
> H t

[0, t]. With the same portfolio argument

as in case i), where instead of K the discounted exercise price Ke
r(Tt)
is placed
into the money account, we can conclude that for the European down-and-out call
the following boundary conditions must be satised:
C
do
[S
t
, K, T t, H] S
t
Ke
r(Tt)
if S
t
> H > K
and furthermore that
C
do
[S
t
, K, T t, H] S
t
H if S
t
> H > K
8
The portfolio argument and the proof of Proposition 5 part c was rst given by Daniel Sommer.
12 MATTHIAS REIMER AND KLAUS SANDMANN
where both bounds are tight. Since S
t
K > S
t
H for H > K, there are situations
possible, when early exercise is optimal for the option holder.

We can now apply these distribution free results to the special structure of the binomial model.
Theorem 3.
a) The arbitrage price of an American up-and-out call option with barrier Su
J
H
d
NJ
H
=
H > K and a grid size t =
Tt
0
N
such that dH > K is equal to
C
am
uo
[S, K, T, H] = C
eur
uo
[S, K, T, dH] (17)
+
S
Hd
N(h)

i=1
__
h 2 + 2i
i
_

_
h 2 + 2i
i 1
__
p
h1+i
(1 p)
i
[dH K]
where h = 2J
H
N for H > S, p =
pu
1+r
, and N(h) = supi IIN[i
N+2h
2

b) The arbitrage price of an American down-and-out call option with barrier Su
J
H
d
NJ
H
=
H > K and grid size t =
Tt
0
N
such that uH < K is equal to
C
am
do
[S, K, T, H] = C
eur
do
[S, K, T, uH] (18)
+
S
Hu
N(h)

i=1
__
h 2 + 2i
i
_

_
h 2 + 2i
i 1
__
p
i
(1 p)
h1+i
[uH K]
where h = N 2J
H
for H < S.
Proof: see appendix
Remark.
1) The reason for these binomial closedform solution is the existence of a constant early
exercise boundary. Thus American type barrier options are in some cases equivalent to
European barrier options with a constant rebate.
2) Applying the continuous time closedform solutions for European barrier options with
a constant rebate (Rubinstein,Reiner (1991)) we have the following limit results:
lim
t0
H>K
C
am
uo
[S, K, T, H] = lim
t0
H>K
C
eur
uo
[S, K, T, H] (19)
+ [H K]
_
_
S
H
_

N (y
1
(H)) +
_
S
H
_
N
_
y
2
(H) +

s
_
_
lim
t0
H>K
C
am
do
[S, K, T, H] = lim
t0
H>K
C
eur
do
[S, K, T, H] (20)
+ [H K]
_
_
S
H
_

N (y
1
(H)) +
_
S
H
_
N
_
y
2
(H)

s
_
_
with =
2 r

2
, s = T t
0
and y
1,2
(z) =
_
ln
_
H
2
Sze
rs
_

2
s

s
_
.
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 13
3) The argument for American put options is similar but we cant expect to nd closed
form solutions for all cases. The basic diculty is that it can be optimal to exercise a
standard put option when the value of the underlying is small. Thus for the case of the
up-and-out and up-and-in put option, it is not possible to nd a closedform binomial
expression. In the case of a down-and-out put or down-and-in put, it is possible to nd
closedform solutions for some barriers H. If the barrier H < K is greater than the
critical value S

(t) of the underlying, which indicates the early exercise for standard put
option at time t, then the American down-and-out put will be exercised just before the
barrier. This can be expressed by a binomial formula, which includes again a rebate of
KH. For the American down-and-in put, a binomial formula can be constructed in the
case, where H < S

(t) < K where S

(t) is again the critical value for early exercise at

time t in the standard case. In both cases, the limit result is given by the corresponding
European type down barrier puts plus a rebate of [K H]. In all the other cases, we
have to apply a recursive algorithm.
6. European options with local or partial barrier condition
We consider now situations, where the barrier condition has to be satised only on a subset of
spots, but not on the whole time interval. We restrict the analysis to the following three basic
cases, which we dene in the discrete framework.
Denition 2. Let T = 0 = t
0
< t
1
< ... < t
N
1
< t
N
1
+1
< ... < t
N
= T [0, T] be an
equidistant discretization of the time axis.
a) A barrier option with maturity T, underlying security S, and barrier H is called
i) a front partial barrier option with barrier period T(t
0
, t
N
1
) = t
0
< ... < t
N
1

T, if the path dependency of the payo is restricted to the period T(t
0
, t
N
1
) and
independent of the security realizations at times t t
N
1
+1
< ... < t
N1
.
ii) a back partial barrier option with barrier period T(t
N
1
, t
N
) = t
N
1
< ... < t
N
,
if the path dependency of the payo is restricted to the period T(t
N
1
, t
N
) and
independent of the security realizations at times t
i
t
0
< ... < t
N
1
1
.
b) A barrier option with maturity T is called a local barrier option with barrier times
T
H
= t

0
< t

1
< ... < t

n
T if the path dependency of the payo is restricted to the
set T
H
and independent of the security realizations at times t T T
H
.
The payo of a front partial downandout call option with maturity T > t
N
1
, barrier H, and
barrier period is dened by T(t
0
, t
N
1
) is given by
_
[S
T
K]
+
if S
t
i
> H t
i
T(t
0
, t
N
1
)
0 if t

T(t
0
, t
N
1
) with S
t

H
With reference to the previous discussion we can compute a binomial formula for a partial
barrier option if we can compute the corresponding prices of partial down-and-in, resp. partial
upandin, ArrowDebreusecurities. Given the binomial model for the underlying security,
these prices can be computed by applying the reection principle (see proposition 5 in the
14 MATTHIAS REIMER AND KLAUS SANDMANN
appendix). With these prices, we can compute the arbitrage prices of all partial down barrier
options. The following theorem demonstrates this for the partial downandout call option.
Theorem 4.
i) Let the barrier H be a knot of the binomial security process at time t
N
1
, i.e. J
H
N
0
such that H = S(t
0
)u
J
H
d
N
1
J
H
. The arbitrage price of a European front partial down-
and-out call with barrier period T(t
0
, t
N
1
) = t
0
< ... < t
N
1
is equal to
C
fp
do
[S, K, T, H, T(t
0
, t
N
)]
=
_
1
1 +r
_
N N

i=J
H
+1
_
N
i
_
p
i
(1 p)
Ni
_
Su
i
d
Ni
K

+
(21)

_
1
1 +r
_
N J
H
+NN
1

i=J
H
+1
J
H

k=0(i+N
1
N)
_
N
1
k
__
N N
1
i k
_
p
i
(1 p)
Ni
[Su
i
d
Ni
K]
+

_
1
1 +r
_
N 2J
H
+NN
1

i=J
H
+1
2J
H
i

k=(J
H
+1)(i+N
1
N)
_
N
1
2J
H
k
__
N N
1
i k
_
p
i
(1 p)
Ni
[Su
i
d
Ni
K]
+
ii) Let the barrier H be a knot of the binomial security price process at time t
N
, i.e. J
H

N
0
such that H = S(t
0
)u
J
H
d
NJ
H
. The arbitrage price of a European back partial
downandout call with barrier H, maturity t
N
= T, exercise price K and barrier pe-
riod T(t
N
1
, t
N
) = t
N
1
< ... < t
N
is equal to
C
bp
do
[S, K, T, H, T(t
N
1
, t
N
)]
=
_
1
1 +r
_
N N

i=J
H
_
N
i
_
p
i
(1 p)
Ni
[Su
i
d
Ni
K]
+
(22)

_
1
1 +r
_
N
2J
H
(J
H
+
NN
1
2
)

i=J
H
N
1
(2J
H
i)

k=0(i+N
1
N)
_
N
1
k
__
N N
1
2J
H
i k
_
p
i
(1 p)
Ni
[Su
i
d
Ni
K]
+
As the last extension of the binomial approach, we consider the situation of local barrier options.
Let T
H
= t

1
< ... < t

n
T be a given subset of T. For each local barrier option, there exists a
recursive algorithm to compute the arbitrage price. Consider for example a local down-and-out
call with barrier H. As in section 2 denote by G
T
(t
n
, i) the value of such a local down-and-out
call with xed maturity t
N
= T issued at time t
n
T and state i, i.e. S(t
n
, i) = S(t
0
)u
i
d
ni
.
The initial condition of the algorithm is therefore
(23) G
T
(T, i) =
_

_
[S(t
0
)u
i
d
Ni
K]
+
if T = t
N
, T
H
[S(t
0
)u
i
d
Ni
K]
+
if t
N
T
H
and S(t
0
)u
i
d
Ni
> H
0 if t
N
T
H
and S(t
0
)u
i
d
Ni
H
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 15
By backward induction we have C
local
do
[S, K, T, H, T
H
] = G
T
(t
0
, 0) with k = 0, . . . , N 1 and
i = 0, . . . , k
(24) G
T
(t
k
, i) =
_

_
1
1 +r
[pG
T
(t
k+1
, i + 1) + (1 p)G
T
(t
k+1
, i)] if t
k
, T
H
1
1 +r
[pG
T
(t
k+1
, i + 1) + (1 p)G
T
(t
k+1
, i)] if t
k
T
H
and S(t
k
, i) > H
0 if t
k
T
H
and S(t
k
, i) H
Furthermore if we assume that H is a knot of the binomial security price process at any time
t

k
T
H
it is possible to construct a binomial formula. For simplicity let us assume that both
sets T and T
H
are equidistant sets, i.e. there exists a number N
H
IIN such that t

j
T
H
is
equal to t
jN
H
T. Thus t
j+1
t
j
= t N
H
t
j
T
H
, t

n
= t
nN
H
= t
N
T and t is the grid
size of the set T. Furthermore assume that N
H
is an even number. Let H be a terminal knot,
i.e. J
H
IIN such that H = S(t
0
)u
J
H
d
NJ
H
and since N
H
is even, H is also a knot at time
t

j
T
H
. With these simplications the arbitrage price of a local downandout call is equal to
C
local
do
[S, K, T, H, T
H
]
=
_
1
1 +r
_
N N
H

i
1
=j
H
(1)+1
N
H

i
2
=01+j
H
(2)i
1
. . .
N
H

i
n
=01+j
H
(n)i
n1
_
N
H
i
1
_ _
N
H
i
2
_
. . .
_
N
H
i
n
_
p

n
k=1
i
k
(1 p)
N

n
k=1
i
k
_
S(t
0
)u

n
k=1
i
k
d
N

n
k=1
i
k
K
_
+
(25)
where j
H
(n) = J
H
and j
H
(k) = j
H
(k + 1)
N
H
2
= j
H
(n k)
N
H
2
is the number of up
movements needed at time t
k
such that S(t
k
, j
H
(k) = H. Obviously this formula is only useful
in situations where the number of local checks of the barrier is small.
7. Summary
Cox, Ross, Rubinstein [1979] and Rendleman, Bartter [1979] have developed a binomial model for
the pricing of European and American type standard options. For European type options they
derived closedform binomial formulae which converge to the BlackScholes formulae under the
usual assumptions. Within the binomial framework we have derived recursive algorithms which
can be used for both European and American barrier options. Furthermore the general argument
supporting these algorithms can be used in the case of modications of the contract denition
or/and to dividend paying securities. In analogy to Cox, Ross, Rubinstein and Rendleman,
Bartter we give binomial formulae for European barrier options and prove the convergence
towards the continuous time solutions. In addition the convergence behaviour is analyzed and
a robust approximation with Lagrange interpolation is proposed. This interpolation method
reduces the complexity of the lattice and is therefore of practical use for the implementation of
numerical procedures. Furthermore we solve the case of American barrier options explicitly and
derive closedform solutions within the binomial and continuous time framework. The Merton
(1973) result for American type call options is extended to American barrier call options. As a
consequence the binomial approach choosen can be generalized immediately to European type
16 MATTHIAS REIMER AND KLAUS SANDMANN
barrier options with rebate. Finally barrier options with local or partial barrier condition are
discussed within the binomial framework.
8. Appendix
Proposition 2. Suppose the barrier H is a terminal knot of the binomial asset price process at time t
N
;
i.e. J
H
IIN such that H = S(t
0
)u
J
H
d
NJ
H
. Dene for a, b IIN the following binomial sums:
B(p, a, b) :=
_

_
0 for a > b
b

i=a
_
N
i
_
p
i
(1 p)
Ni
for a b N

B(p, a, b) :=
_

_
0 for a > b
b

i=a
_
N
2J
H
i
_
p
i
(1 p)
Ni
for a b 2J
H
Under these assumptions the arbitrage price of the following barrier options (where we assume H < S(t
0
)
in the down case and H > S(t
0
) in the up case) is equal to
C
d
i
[S, K, T, H] = S(t
0
) B( p, a, J
H
1)

K B(p, a, J
H
1)
+ S(t
0
)

B( p, a J
H
, 2J
H
)

K

B(p, a J
H
, 2J
H
)
C
uo
[S, K, T, H] = S(t
0
) B( p, a, J
H
)

K B( p, a, J
H
)
S(t
0
)

B( p, a
J
H
|
2
, J
H
) +

K B(p, a
J
H
|
2
, J
H
)
C
ui
[S, K, T, H] = S(t
0
) B( p, a (J
H
+ 1), N)

K B( p, a (J
H
+ 1), N)
+ S(t
0
)

B( p, a
J
H
|
2
, J
H
)

K

B(p, a
J
H
|
2
, J
H
)
P
do
[S, K, T, H] =

K B(p, J
H
, b) S(t
0
) B( p, J
H
, b)

K

B(p, J
H
, 2J
H
b) +S(t
0
)

B( p, 2J
H
k)
P
di
[S, K, T, H] =

K B(p, 0, (J
H
1) b) S(t
0
) B( p, 0, (J
H
1) b)
+

K

B(p, J
H
, 2J
H
b) S(t
0
)

B( p, J
H
, 2J
H
b)
P
uo
[S, K, T, H] =

K B(p, 0, J
H
b) S(t
0
) B( p, 0, J
H
b)

K

B(p,
J
H
|
2
, b J
H
) +S(t
0
)

B( p,
J
H
|
2
, b J
H
)
P
ui
[S, K, T, H] =

K B(p, (J
H
+ 1), b) S(t
0
) B( p, (J
H
+ 1), b)
+

K

B(p,
J
H
|
2
, b J
H
) S(t
0
)

B( p,
J
H
|
2
, b J
H
)
where
a := infi IIN[S(t
0
)u
i
d
Ni
K b := supi IIN[S(t
0
)u
i
d
Ni
K
p :=
p u
1 +r
p :=
1 +r d
u d
a J
H
:= maxa, J
H
a J
H
:= mina, J
H

K :=
_
1
1 +r
_
N
K.
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 17
Proof of Theorem 2.
1) Consider the binomial formula (13) for the European downandoutcall. Since for H K
a J
H
the rst two terms coincide with the usual binomial formula for European call options
for which we already know that under the given assumptions the limit in distribution of
S(t
0
)
N

i=aJ
H
_
N
i
_
p
i
(1 p)
Ni

_
1
1 +r
_
N
K
N

i=aJ
H
_
N
i
_
p
i
(1 p)
Ni
for H K
is given by
9
S(t
0
)N(x) Ke
r(Tt
0
)
N(x
_
T t
0
)
with
x(K) =
_
ln
_
S(t)
Ke
r

Tt
0
_
+
1
2

2
(T t
0
)
_
1

T t
0
For H > K it is easy to see that we only have to consider x(H) instead of x(K) as the argument
of the standard normal distribution. It remains to proof that under the assumption of theorem
2 the correction term (for N suciently large)
() S(t
0
)
2J
H

i=aJ
H
_
N
2J
H
i
_
p
i
(1 p)
Ni

_
1
1 +r
_
N
K
2J
H

i=aJ
H
_
N
2J
H
i
_
p
i
(1 p)
Ni
with N 2J
H
converges in distribution to
S(t
0
)
_
S(t)
H
_
(+1)
N(y(K H)) Ke
r(Tt)
_
S(t)
H
_
1
N(y(K H))
_
T t
0
)
For simplicity let us assume K H and
10
therefore a J
H
. By index transformation () can
be rewritten as
S(t
0
)
_
1 p
p
_
N2J
H
2J
H
a

i=0
_
N
i
_
p
(Ni)
(1 p)
i

K
(1 +r)
N
_
1 p
p
_
N2J
H
2J
H
a

i=0
_
N
i
_
p
(Ni)
(1 p)
i
For suciently small t =
Tt
0
N
the martingale transition probability p can be approximated by
p =
e
rt
d
u d

1
2
+
1
2
r

2
2

t
which yields
lim
t0
E
p
_
ln
S(T)
S(t
0
)
_
= (r

2
2
)(T t
0
)
lim
t0
V
p
_
ln
S(T)
S(t
0
)
_
=
2
(T t
0
)
9
See for example Cox, Rubinstein [1985] , for simplicity let r be the continuously compounded interest rate.
10
The case H < K is similar and can be done by a change of variables.
18 MATTHIAS REIMER AND KLAUS SANDMANN
Furthermore we have the following approximation for the ratio
p
1 p
:
p
1 p
=
1 +
r
2
/2

t
1
r
2
/2

t
+o(t)
= 1 + 2
_
r
2
/2

t
1
1
r
2
/2

t
+o(t)
= 1 + 2
_
r
2
/2

i=0
_
n
2
/2

t
_
i
+o(t)
= 1 + 2

i=0
_
r
2
/2

t
_
i+1
+o(t)
= 1 + 2
_
r
2
/2

t + 2
_
r
2
/2

_
2
t +o(t)
Observing that for small t the Taylor-expansion of the exponential function is given by
exp
_
2
_
r
2
/2

t
_
=

i=0
1
i!
_
2
_
r
2
/2

t
_
i
= 1 + 2
_
r
2
/2

t + 2
_
r
2
/2

_
t +o(t)
yields the approximation
p
1p
exp
_
2
_
r
2
/2

t
_
Therefore we obtain the following results
11
:
i) S(t
0
)u
J
H
d
NJ
H
= H N 2J
H
=
ln(
H
S(t
0
)
)
lnd
J
H
=
ln(
H
S(t
0
)
d
N
)
ln(
u
d
)
ii) lim
t0
_
1p
p
_
N2J
H
= lim
t0
exp
_
2
_
r
2
/2

t ln
H
S(t
0
)

1

t
_
=
_
S(t
0
)
H
_
1
2r

2
iii) lim
t0
_
d
u
_
N2J
H
= lim
t0
exp
_
2

t ln
H
S(t
0
)

1

t
_
=
_
H
S(t
0
)
_
2
lim
t0
_
1 p
p
_
N2J
H
= lim
t0
_
1p
p

d
u
_
N2J
H
=
_
S(t
0
)
H
_
1
2r

2
Finally we have to consider the two sums. Let J(N) be the sum of N independent binomially
distributed variables with up and down probabilities 1 p resp. p. Thus we have
E
p
[J(N)] = N(1 p) and V
p
[J(N)] = Np(1 p).

2J
H
a

i=0
_
N
i
_
p
Ni
(1 p)
i
= prob[J(N) 2J
H
a]
11
Now we explicitly use the assumption that H is an endpoint of the binomial tree.
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 19
Obviously, the Central Limit Theorem can be applied. By construction we have
12
a)
J(N)E
p
[J(N)]

V
p
[J(N)]
=
ln
_
S
T
S(t
0
)
_
E
p
[ln
S
T
S(t
0
)
]
_
V
p
[ln
S
T
S(t
0
)
]
b) 2J
H
a =
2
_
ln
H
S(t
0
)d
N
_
ln
K
S(t
0
)d
N
ln
u
d

2J
H
aE
p
[J(N)]

V
p
[J(N)]
=
2lnHlnS(t
0
)lnKln
u
d
+E
p
[ln
S
T
S(t
0
)
]
_
V
p
[ln
S
T
S(t
0
)
]
t0

2lnHlnS(t
0
)lnK+(r
2
/2)(Tt
0
)

(Tt
0
)
=: y
2
By the Central Limit Theorem therefore we have
lim
t0
2J
H
a

i=0
_
N
i
_
p
Ni
(1 p)
i
= N(y
2
)
For the second sum we can use the same argument. The only change concerns the transition
probability p. The Taylor-expansion for p yields
p
1
2
+
1
2
r +
2
/2

t
lim
t0
E
p
[ln
S
T
S(t
0
)
] =
_
r +

2
2
_
(T t
0
)
lim
t0
V
p
[ln
S
T
S(t
0
)
] =
2
(T t
0
)
Again, by the Central Limit Theorem we obtain
lim
t0
2J
H
a

i=0
_
N
i
_
p
Ni
(1 p)
Ni
= N(y
1
)
with y
1
:=
2lnH lnS(t
0
) lnK +
_
r +

2
2
_
(T t
0
)

_
(T t
0
)
For the case H > K the same analysis can be done. The only change concerns the summation.
2) In the case of a European upandout put, again two cases have to be considered: H K and
H < K. For simplicity let us assume H K and thus J
H
b = supi IIN[S(t
0
)u
i
d
Ni
K.
Again we only consider the correction term in (14) since the rst two terms will converge in
distribution to the BlackScholes formula for put options. The correction term can be rewritten
as:
_
1
1 +r
_
N
K
b

i=J
H
/2
_
N
2J
H
i
_
p
i
(1 p)
Ni
S(t
0
)
b

i=J
H
/2
_
N
2J
H
i
_
p
i
(1 p)
Ni
=
_
1 p
p
_
N2J
H
_
_
_
1
1 +r
_
N
K
2J
H
J
H
/2

i=2J
H
b
_
N
i
_
p
Ni
(1 p)
i
S(t
0
)
2J
H
J
H
/2

i=2J
H
b
_
N
2J
H
i
_
p
Ni
(1 p)
i
_
_
Given the results in 1) we only have to consider these two sums. The rst sum is equal to
N

i=2J
H
b
_
N
i
_
p
Ni
(1 p)
i

i=2J
H
+1J
H
/2
_
N
i
_
p
Ni
(1 p)
i
prob[J(N) 2J
H
b] prob[J(N) > 2J
H
J
H
/2|]
12
We use the fact that a = inf{ IIN|S(t
0
)u
i
d
Ni
K}.
20 MATTHIAS REIMER AND KLAUS SANDMANN
where J(N) is the sum of N independent binomially distributed variables with up and down
probabilities 1 p resp. p. From the denition of J
H
and b we have
2J
H
b E
p
[J(N)]
_
V
p
[J(N)]
=
2lnH lnS(t
0
) lnK lnu/d +E
p
[lnS
T
/S(t
0
)]
_
V
p
[lnS
T
/S(t
0
)]
t0

2lnH lnS(t
0
) lnK + (r
2
/2)(T t
0
)

T t
0
=: y
2
2J
H
J
H
/2| E
p
[J(N)]
_
V
p
[J(N)]
=
3/2 (lnH/S(t
0
)) 1/2Nlnd lnu/d +E
p
[lnS
T
/S(t
0
)]
_
V
p
[lnS
T
/S(t
0
)]
t0
+ since Nlnd =
T t
0

t
Therefore the Central Limit Theorem yields
prob[J(N) 2J
H
b] prob
_
J(N) > 2J
H

J
H
|
2
_
t0
N(y
2
)
The same argument applies for the second sum where again the transition probability p has to
be replaced by p =
pu
1+r
.

Proposition 4. Let t =
Tt
0
N
the grid size of the binomial lattice. For u = exp

t, d =
exp

t and r =
1
t
ln(1 + r) the convergence in distribution of the binomial formulae in theo-
rem 1 are given by
13
a) for K > H, S > H
lim
t0
C
di
[S, K, T.H] = S
_
S
H
_
1
N(y
1
(K))

K
_
S
H
_
1
N(y
2
(K))
for K < H, S > H
lim
t0
C
di
[S, K, T, H] = S[N(x
1
(K)) N(x
1
(H))]

K[N(x
2
(K)) N(x
2
(H))]
+ S
_
S
H
_
1
N(y
1
(H))

K
_
S
H
_
1
N(y
2
(H))
b) for K > H, S < H
lim
t0
C
uo
[S, K, T, H] = 0
for K < H, S < H
lim
t0
C
uo
[S, K, T, H] = S[N(x
1
(K)) N(x
1
(H))]

K[N(x
2
(K)) N(x
2
(H))]
S
_
S
H
_
1
[N(y
1
(K)) N(y
1
(H))]
+

K
_
S
H
_
1
[N(y
2
(K)) N(y
2
(H))]
c) for K > H, S < H
13
We assume H < S(t
0
) in all downcases and H > S(t
0
) for all upcases since otherwise the value of an
outoption is equal to zero and the value of an inoption coincides with that of a standard option.
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 21
lim
t0
C
ui
[S, K, T, H] = SN(x
1
(K))

KN(x
2
(K))
for K < H, S < H
lim
t0
C
ui
[S, K, T, H] = SN(x
1
(H))

KN(x
2
(H))
+S
_
S
H
_
1
[N(y
1
(K)) N(y
1
(H))]

K
_
S
H
_
1
[N(y
2
(K)) N(y
2
(H))]
d) for K < H, S > H
lim
t0
P
do
[S, K, T, H] = 0
for K > H, S > H
lim
t0
P
do
[S, K, T, H] =

K[N(x
2
(H)) N(x
2
(K))] S[N(x
1
(H)) N(x
1
(K))]

K
_
S
H
_
1
[N(y
2
(H)) N(y
2
(K))]
+S
_
S
H
_
1
[N(y
1
(H)) N(y
1
(K))]
e) for K < H, S > H
lim
t0
P
di
[S, K, T, H] =

KN(x
2
(K)) SN(x
1
(K))
for K > H, S > H
lim
t0
P
di
[S, K, T, H] =

KN(x
2
(H)) SN(x
1
(H))
+

K
_
S
H
_
1
[N(y
2
(H)) N(y
2
(K))]
S
_
S
H
_
1
[N(y
1
(H)) N(y
1
(K))]
f) for K < H, S < H
lim
t0
P
ui
[S, K, T, H] =

K
_
S
H
_
1
N(y
2
(K)) S
_
S
H
_
1
N(y
1
(K))
for K > H, S < H
lim
t0
P
ui
[S, K, T, H] =

K[N(x
2
(H)) N(x
2
(K))] S[N(x
1
(H)) N(x
1
(K))]
+

K
_
S
H
_
1
N(y
2
(H)) S
_
S
H
_
1
N(y
1
(H))
where

K = Ke
rs
, =
2 r

2
, s = T t
0
and
x
1,2
(z) =
ln
_
S
ze
rs
_

2
s

s
y
1,2
(z) =
ln
_
H
2
Se
rs
_

2
s

s
.
Proof. The proof of the above formulae is an application of the Central Limit Theorem already demon-
strated in Theorem 2.
Proof of Theorem 4.
22 MATTHIAS REIMER AND KLAUS SANDMANN
a) Let C
am
uo
[S, K, H, T] be the arbitrage price of an American up-and-out call option which is still
alive at time t. Let H > K be the barrier. By assumption H is an endpoint of the binomial
tree. Thus at time t
n
T = t
0
< . . . < t
N
the option is still alive if S(t
n
) = S(t
0
)u
j
d
nj
< H.
There are two possible cases of interest. First S(t
n
) d
2
H and second S(t
n
) = dH. Suppose
S(t
n
) d
2
H which implies that at time t
n+1
the option is still alive. Consider now the dierence
between immediate exercise or exercise at time t
n+1
. Since we know that H > K we have
C
am
uo
[S, K, T, H] C
eur
uo
[S, K, H, T] S(t
n
) d
2
H

1
1 +r
E
P
[[S(t
n+1
) K]
+
[ S(t
n
) d
2
H] S(t
n
) d
2
H
= Max
_
0, S(t
n
)
K
1 +r
_
> Max 0, S(t
n
) K
and therefore it is not optimal to exercise the option at time t
n
.
Suppose now S(t
n
) = dH. Since d = exp
_

t
_
and H > K implies that for t <
_
1

ln
H
K
_
2
N > (T t
0
)

2
(ln(H/K))
2
the inner value dH K is positiv. Since dH K is
also the maximum possible payo of the contract at time T = t
N
, early exercise at any time
t
n
< T is optimal in the situation S(t
n
) = dH. This implies that within the binomial setup the
arbitrage price of an American upandout call option with barrier H is equal to the European
upand out call option with the barrier dH plus a rebate of dH K when the barrier dH is
reached, assuming that the grid size is small enough such that dH > K. Dene h IIN such that
Su
h
= H with 0 < h < N since S < H and H is an element of the binomial tree. Furthermore
since H = Su
J
H
d
NJ
H
we have h = 2J
H
N. From the reection principle we know that for
h 2
_
h 2 + 2i
h 2 +i
_

_
h 2 + 2i
(h 1) +i
_
for i = 1, . . . ,
_
N + 2 h
2
_
= : N(h) ,
is equal to the number of paths which at time t
h2+2i
T end in the knot Su
h2+i
d
i
= Su
h
d
2
=
Hd
2
and have not crossed or touched the barrier Su
h1
= Hd. This is then equal to the number
of paths which at time t
h1+2i
reach for the rst time the knot Hd. Summing up, the arbitrage
price of the American upandout call with barrier H is equal to
C
am
uo
[S, K, T, H] = C
eur
uo
[S, K, T, dH]
+
N(h)

i=1
_
h 2 + 2i
i
_
p
h1+i
(1 p)
i
_
1
1 +r
_
h1+2i
[dH K]

N(h)

i=1
_
h 2 + 2i
i 1
_
p
h1+i
(1 p)
i
_
1
1 +r
_
h1+2i
[dH K]
where the grid size t <
_
1

ln
H
K
_
2
.
b) Let C
am
do
[S, K, T, H] with S > H be the arbitrage price of an American downandout call which
is still alive at time t. Suppose H > K, then we know that S(t
i
) Hu
2
immediate exercise
is not optimal. Therefore consider the situation S(t
i
) = uH. Suppose that there are N
1
< N
periods left and dene J
H
such that (uH)u
J
H
d
N
1
J
H
= Hu 2J
H
= N
1
. For the European
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 23
downandout call we can now use the binomial formulae (13) with T

= t
i
< . . . < t
N

C
eur
do
[uH, K, T, H]
=
_
1
1 +r
_
N
1
_
N
1

i=J
H
_
N
1
i
_
p
i
(1 p)
N
1
i
[(uH)u
i
d
N
1
i
K]
+

2J
H

i=J
H
_
N
1
2J
H
i
_
p
i
(1 p)
N
1
i
[(uH)u
i
d
N
1
i
K]
+
_
=
_
1
1 +r
_
N
1
_
N
1

i=J
H
_
N
1
i
_
p
i
(1 p)
N
1
i
((uH)u
i
d
N
1
i
K)

N
1

i=J
H
_
N
1
N
1
i
_
p
i
(1 p)
N
1
i
((uH)u
i
d
N
1
i
K)
_
=
_
1
1 +r
_
N
1
_
N
1

i=0
_
N
1
i
_
p
i
(1 p)
N
1
i
((uH)u
i
d
N
1
i
K)

J
H
1

i=0
_
N
1
i
_
p
i
(1 p)
N
1
i
(uHu
i
d
N
1
i
K)

N
1

i=J
H
_
N
1
i
_
p
i1
(1 p)
N
1
+1i
(Hu
i
d
N
1
+1i
K)
_
Since uH > H > K and for t such that dH > K we have
1 p
p
((dH)u
i
d
N
1
i
K) > Ku
i
d
N
1
i
K i J
H
and the European arbitrage price in this situation is bounded from above by
C
eur
do
[uH, K, H, T] <
_
1
1 +r
_
N
1
_
N
1

i=0
_
N
1
i
_
p
i
(1 p)
N
1
i
(uH K)u
i
d
N
1
i
_
= (uH K)
which implies that early exercise is optimal in the situation S(t
i
) = uH > dH > K. With this
we can now use the same counting algorithm as for the upandout option, where
_
h 2 + 2i
h 2 +i
_

_
h 2 + 2i
h 1 +i
_
for i = 1, . . . ,
_
N + 2 h
2
_
=: N(h)
and Sd
h
= H is equal to the number of paths which at time t
h2+2i
end in a knot uH for the
rst time.

Proposition 5.
i) Let H be a barrier such that there exists a J
H
IIN with S(t
0
)u
J
H
d
N
1
J
H
= H. The arbitrage
price of a front partial downandin ArrowDebreusecurity
fp
d
(T(t
0
, t
N
1
), i, J
H
) with payo
at time t
N
_
1 if S
t
N
= S
t
0
u
i
d
Ni
and S
t
i
H t
i
T(t
0
, t
N
1
)
0 otherwise
is given by:
24 MATTHIAS REIMER AND KLAUS SANDMANN

fp
d
_
T(t
0
, t
N
1
); i, J
H
_
=
_

_
_
1
1 +r
_
N
_
N
i
_
p
i
(1 p)
Ni
for 0 i J
H
_
1
1 +r
_
N
_
_
J
H

k=0(i(NN
1
))
_
N
1
k
__
N N
1
i k
_
+
2J
H
i

k=J
H
+1
_
N
1
2J
H
k
__
N N
1
i k
_
_
_
p
i
(1 p)
Ni
for J
H
< i J
H
+N
2
_
1
1 +r
_
N
_
_
2J
H
i

k=i(NN
1
)
_
N
1
2J
H
k
__
N N
1
i k
_
_
_
p
i
(1 p)
Ni
for J
H
+N
2
i 2J
H
+N
2
0 for i > 2J
H
+N
2
ii) Let J
H
N such that S(t
0
)u
J
H
d
NJ
H
= H. The arbitrage price of a backpartial downandin
ArrowDebreusecurity
bp
d
_
T(t
N
1
, t
N
), i, J
H
_
is given by

bp
d
_
T(t
N
1
, t
N
2
); i, J
H
_
=
_

_
_
1
1 +r
_
N
_
N
i
_
p
i
(1 p)
Ni
for 0 i J
H
_
1
1 +r
_
N
_
_
N
1
(2J
H
i)

k=0(i(NN
1
))
_
N
1
k
__
N N
1
2J
H
i k
_
_
_
p
i
(1 p)
Ni
for J
H
i min
_
2J
H
, J
H
=
NN
1
2
_
0 for i > min
_
2J
H
, J
H
+
NN
1
2
_
Proof of Proposition 5.
ad i) Dene K
bp
d
(0, N
1
, i, J
H
) as the number of paths from the origin to the knot S(t
0
)u
i
d
Ni
which
reach or cross the barrier H = S(t
0
)u
J
H
d
N
1
J
H
at least at one time t t
0
< . . . < t
N
1
. Set
N
2
= N N
1
, then the following picture summarizes the arguments:
A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 25
t
0

P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
S

t
N
1

P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P

Su
N
1
d
0
Su
2J
H
d
N
1
2J
H
H = Su
J
H
d
N
1
J
H
Su
0
d
N
1
t
N
= t
N
1
+N
2
Su
N
1
+N
2
Su
2J
H
+N
2
d
N
1
2J
H
Su
J
H
+N
2
d
N
1
J
H
= Hu
N
2
Su
N
2
d
N
1
Su
J
H
d
NJ
H
= Hd
N
2
Sd
N
1
+N
2
K
fp
(0, N
1
, i, J
H
) =
_

_
_
N
i
_
0 i J
H
J
H

k=0iN
2
_
N
1
k
__
N
2
i k
_
+
2J
H
i

k=J
H
+1
_
N
1
2J
H
k
__
N
2
i k
_
J
H
< i J
H
+N
2
2J
H
i

k=iN
2
_
N
1
2J
H
k
__
N
2
i k
_
J
H
+N
2
< i 2J
H
+N
2
0 i > 2J
H
+N
2
26 MATTHIAS REIMER AND KLAUS SANDMANN
ad ii) The argument is the same in both cases. Consider now the back partial case. The problem is to
compute the number of paths K
bp
d
(N
1
, N
2
, i, J
H
) from the origin to the knot S(t
0
)u
i
d
Ni
which
reach or cross the barrier H = S(t
0
)u
J
H
d
NJ
H
at least at one time t t
N
1
< . . . < t
N
. For
simplicity let N
2
:= N N
1
be an even number.
t
0

P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
S

t
N
1

P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
P

Su
N
1
d
0
Su
0
d
N
1
Su
2J
H
i
d
N
1
2J
H
+i
Su
J
H
d
N
1
J
H
t
N
= t
N
1
+N
2
Su
N
1
+N
2
Su
2J
H
d
N2J
H
Hu
N
2
= Su
J
H
+
N
2
2
d
NJ
H

N
2
2
Su
N
2
d
N
1
Su
i
d
Ni
H = Su
J
H
d
NJ
H
Sd
N
1
+N
2
K
bp
d
(N
1
, N
2
, i, J
H
) =
_

_
_
N
i
_
0 i J
H
N
1
(2J
H
i)

k=0(i(NN
1
))
_
N
1
k
__
N N
1
2J
H
i k
_
J
H
i min
_
2J
H
, J
H
+
N
2
2
_
0 i > min
_
2J
H
, J
H
+
N
2
2
_
where for i N N
1
= N
2
N
1
2J
H
i

k=0
_
N
1
k
__
N N
1
2J
H
i k
_
=
_
N
2J
H
i
_

A DISCRETE TIME APPROACH FOR EUROPEAN AND AMERICAN BARRIER OPTIONS 27

References
Black, F.; J. Cox [1976]: Valuing Corporate Securities: Some Eects of Bond Indenture
Provisions, Journal of Finance, Vol. 31, No.2, May 1976, pp. 351376.
Black, F.; M. Scholes [1973]: The Pricing of Options and Corporate Liabilities, Jour-
nal of Political Economy, Vol. 81, MayJune 1973, pp. 637659.
Boyle, P.P.; Sok Hoon Lau [1994]: Bumping Up Against the Barrier with the Bino-
mial Method, The Journal of Derivatives, Vol. 1, No. 4, 1994, pp.614.
Conze, A.; Viswanathan [1991]: Path Dependent Options: The Case of Lookback Op-
tions, Journal of Finance, Vol. 16, No. 5, 1991, pp. 18931907.
Cox, J.; St.A. Ross; M. Rubinstein [1979]: Option Pricing: A Simplied Approach,
Journal of Financial Economics, Vol. 7, September 1979, pp. 229263.
Cox, J.; M. Rubinstein [1985]: Options Markets, 1985, PrenticeHall, New Jersey.
Feller, W. [1968]: An Introduction to Probability Theory and its Applications, Wiley
Series in Probability and Mathematical Statistics, Vol. I+II, 1968, Wiley & Son, Inc.,
New York, London, Sidney.
Goldman, M.B.; H.B. Sosin; M. Gatto [1979]: Path Dependent Options: Buy at the
Low, Sell at the High, it Journal of Finance, Vol. 34, December 1979, pp. 1111-1127.
Harrison, J.; D.M. Kreps [1979]: Martingales and Arbitrage in Multiperiod Securities
Markets, Journal of Economic Theory,[1979/cb Vol. 20, 1979, pp. 381408.
Ingersoll, J.E.jr. [1987]: Option Pricing Theory, in Eatwell, J; M. Milgate; P. Newman:
The New Palgrave, , Vol 3, 1987, pp. 745751.
Merton, R.C. [1973]: Theory of Rational Option Pricing, it Bell Journal of Economics
and Management, 1973, pp. 141183.
Reimer, M.; Sandmann, K. [1993]: DownandoutCall: Bewertungstheorie, numerische
Verfahren und Simulationsstudie, Discussion Paper B239 , 1993, University of Bonn.
Rendlemann, R.; B. Bartter [1979]: TowState Option Pricing, Journal of Finance,
Vol. 34, December 1979, pp. 10931110.
Rubinstein, M.; E. Reiner [1991]: Breaking Down the Barriers, RISK, September 1991.
Sondermann, D. [1988]: Option Pricing with Bounds on the Underlying Securies, in
Rudolph, B.; J. Wilhelm (Eds.): Bankpolitik, nanzielle Unternehmensf uhrung und die
Theorie der Finanzm arkte, 1988, Berlin