Appendix 2

A2: American Options in the Black-Scholes Model

In lecture 4 we saw that the value of some european (exotic or not) option with payoff
C = C({St }0≤t≤T ) is given by

Vt (St ) = e−r(T −t) EW̃ C | St

 
(2.1)

Furthermore we saw that, if C = C(ST ), the function Vt (S) = V (S, t) is a solution of the
Black-Scholes equation
∂V σ2 ∂ 2V ∂V
+ S 2 2 + rS − rV = 0 (2.2)
∂t 2 ∂S ∂S
with the final condition

V (S, T ) = C(S) (2.3)

The analog of (2.1) for american options is given by the continuous time version of Theorem


8.2 and is quite straight forward. If Ut = U {Su }0≤u≤t is the option value at time t then

Ut = sup EW̃ [e−r(τ −t) C({Su }0≤u≤τ ) | St ] (2.4)

τ ≥t

= EW̃ [e−r(τU =C −t) C({Su }0≤u≤τU =C ) | St ] (2.5)

where τU =C is the stopping time defined by

τU =C = min{s ≥ t | Us = Cs } (2.6)

and Cs = C({Su }0≤u≤s ). Observe that in continuous time one cannot longer define Ut by
backward induction as in (1.10,1.11).

We now want to consider the question whether there is an analog of (2.2) for american
options whose payoff at time t is given by C = C(St ). To this end recall from lecture 8
the relation
Xk  

vk = uk + max cj−1 , ũj−1 − ũj−1 (2.7)
j=1

199
200 A2

Here uk is the discounted Snell envelope for the american claim C(Stk ) which is equal to the
discounted option value at time tk . vk is the discounted portfolio value of a selffinancing
strategy with initial value u0 ,
k
X
vk = u0 + δj−1 (sj − sj−1 ) (2.8)
j=1

where the δ’s are given by (1.13). Finally ũj−1 = EB̃ [uj |sj−1 ]. From (2.7) we get
  
vk − vk−1 = uk − uk−1 + max ck−1 , ũk−1 − ũk−1 (2.9)

We write vk = v(Stk , tk ) = e−rtk V (Stk , tk ), tk = k∆t and consider the limit ∆t → 0. We

have to distinguish two cases: For u(Stk , tk ) > c(Stk , tk ) or U (Stk , tk ) > C(Stk ) the round
brackets in (2.9) vanish and we get

dv = du (2.10)

This is equivalent to the Black-Scholes equation. Namely, because of (2.8),

∂u ∂U
d(e−rt S) = ∂U
−r e−rt Sdt + e−rt dS


dv = δ ds = ∂s
ds = ∂S ∂S
(2.11)

and, using the Ito-Lemma for U (S, t),

du = d(e−rt U ) = −r e−rt U dt + e−rt dU

 
−rt ∂U ∂U 1 ∂2U 2 2
= e −rU dt + ∂t dt + ∂S dS + 2 ∂S 2 σ S dt (2.12)

Substituting this in (2.10),

∂U ∂U ∂U 1 ∂2U 2 2


∂S
−r Sdt + dS = −rU dt + ∂t
dt + ∂S
dS + 2 ∂S 2
σ S dt (2.13)

which is equivalent to

∂U σ2 ∂ 2U ∂U
+ S 2 2 + rS − rU = 0 (2.14)
∂t 2 ∂S ∂S
The second case is U (Stk , tk ) = C(Stk ). In that case the round brackets in (2.9) are positive
such that (2.10) has to be substituted by

dv ≥ du (2.15)

which gives, as above,

∂U σ2 2 ∂ 2U ∂U
+ S + rS − rU ≤ 0 (2.16)
∂t 2 ∂S 2 ∂S
A2 201

Thus we arrive at
∂U σ2 ∂ 2U ∂U
+ S 2 2 + rS − rU ≤ 0 , U (S, t) ≥ C(S) (2.17)
∂t 2 ∂S ∂S
 ∂U σ2 ∂ 2U ∂U  
+ S 2 2 + rS − rU U (S, t) − C(S) = 0 (2.18)
∂t 2 ∂S ∂S

U (S, T ) = C(S) (2.19)

In order to get a feeling for the system (2.17-2.19) we consider two examples which can
be explicitely solved, the perpetual american put (example 9.1 below) and an american
cash or nothing call (excercise 9.1). Perpetual means that the maturity date is T = ∞.
The finite time T < ∞ american put is not explicitely solvable. The value of an american
call coincides with the value of a european call.

Example 9.1 (perpetual american put): We consider an american put with payoff

C(S) = max{K − S, 0} ≡ (K − S)+ (2.20)

which can be excercised at an arbitrary time t ≥ 0, that is, its maturity is T = ∞. It is

clear that its price cannot explicitely depend on t but only on S = St , the stock price at
time t. Thus the system (2.17-2.19) reduces to
σ2 2 ∂ 2 U
2
S ∂S 2 + rS ∂U
∂S
− rU ≤ 0 , U (S) ≥ C(S) (2.21)
 2 2
 
σ 2∂ U ∂U
2
S ∂S 2
+ rS ∂S
− rU U (S) − C(S) =0 (2.22)

As in excercise 8.1 one can show that there is a stock price S∗ such that U (S) = C(S) for
S ≤ S∗ and U (S) > C(S) for S > S∗ . At S = S∗ , U and U 0 has to be continuous. We
first verify that U (S) = C(S) satisfies the first inequality of (2.21). Indeed, since S∗ < K,
C(S) = K − S and
σ2 2 ∂ 2 C
2
S ∂S 2 + rS ∂C
∂S
− rC = 0 + rS(−1) − r(K − S) = −rK < 0 (2.23)

For S > S∗ , U (S) has to be a solution of the Black-Scholes equation

σ2 2 ∂ 2 U
2
S ∂S 2 + rS ∂U
∂S
− rU = 0 (2.24)

The Ansatz U (S) = S α gives

2
α(α − 1) σ2 + αr − r = 0
⇒ α1 = 1, α2 = − σ2r2 =: −κ (2.25)
202 A2

Thus,

U (S) = c1 S + c2 S −κ (2.26)

Since we must have U (S) → 0 as S → ∞ it follows that c1 = 0. Using the continuity of

U and U 0 at S∗ we find (recall S∗ < K)

c2 S∗−κ = K − S∗ (2.27)
−κc2 S∗−κ−1 = −1 (2.28)

Deviding (2.27) by (2.28) gives − κ1 S∗ = S∗ − K or

κ
S∗ = κ+1
K (2.29)

2.27
Furthermore U (S) = c2 S −κ = c2 S∗−κ (S/S∗ )−κ = (K − S∗ )(S/S∗ )−κ . Hence we arrive at
the following solution for the perpetual american put

K −S if S ≤ S∗
U (S) = −κ (2.30)
(K − S∗ )(S/S∗ ) if S > S∗

with κ = 2r/σ 2 and S∗ given by (2.29).

Now we use formula (2.5) to compute U (S) and show that the result coincides with
(2.30). We have to compute

U (S) = EW̃ e−rτ C(Sτ(µ) )

 

= EW e−rτ C(Sτ(r) )
 
(2.31)

where τ is the stopping time

(r)  (r) (r)
τ ({St }) = inf t | U (St ) = C(St )
 (r)
= inf t | St ≤ S∗
σ2
= inf t | S eσxt +(r− 2 )t ≤ S∗


= inf t | xt + ct ≤ σ1 log SS∗


(2.32)

and we abbreviated c = σr − σ2 . For S ≤ S∗ we have τ = 0 and U (S) = C(S) as in (2.30).

(r) (r)
Suppose S > S∗ . Then τ > 0 and Sτ = S∗ , hence C(Sτ ) = C(S∗ ) = K − S∗ . Thus we
have to compute

U (S) = (K − S∗ )EW e−rτ

 
(2.33)
A2 203

The expectation on the right hand side of (2.33) is computed in Lemma 9.2 below. We
get
1 S∗
√ c S∗
2
U (S) = (K − S∗ )e−| σ log S | 2r+c + σ log S
√
 − 2r+c 2
σ
− σc
S
= (K − S∗ ) S∗
 −κ
= (K − S∗ ) SS∗ (2.34)

since
√ q 
2r+c2 c 1 r2 2
σ
+ σ
= σ
2r + − r + σ4 + σr − σ2
σ2
1 r σ
+ σr − σ2 = σ2r2 = κ


= σ σ
+ 2
(2.35)

It remains to verify the value of S∗ . To this end recall from (2.4,2.5) that the stopping
time (2.32) is optimal. That is, a stopping time defined by some S̃∗ 6= S∗ would give a
lower expectation. In other words, we can determine S∗ by differentiating (K −S∗ )(S∗ /S)κ
with respect to S∗ and putting the derivative equal to zero:

−(S∗ /S)κ + κ(K − S∗ )(S∗ /S)κ−1 S1 = 0

⇔ −S∗ + κ(K − S∗ ) = 0
κ
⇔ S∗ = κ+1
K (2.36)

and this coincides with (2.29). This completes our discussion of the american put with
infinite maturity time.

Lemma 9.2: Let b, c be real numbers and λ > 0. Let EW be the expectation with respect
to the Wiener measure dW (0,∞) ({xt }t>0 ) and let xt be a Brownian motion. Define the
stopping time

τbc := inf t ≥ 0 | xt + ct = b

(2.37)

with the convention inf ∅ = ∞. Then

c
√
2
EW e−λτb = e−|b| 2λ+c +cb

(2.38)

σ2
Proof: We first prove (2.38) for c = 0. To this end, consider the martingale St = eσxt − 2
t
2
with σ2 = λ. The stopping theorem states that for every bounded stopping time τ

EW [Sτ ] = EW [S0 ] = S0 = 1 (2.39)

204 A2

Since τb0 is not necessarily finite, one has to consider τb,t

0
:= min{τb0 , t} and apply Lebesgue’s
theorem (assume without loss of generality b > 0 if σ > 0) to obtain
EW [Sτb0 ] = lim EW [Sτb,t
0 ]
t→∞
= lim EW [S0 ] = S0 = 1 (2.40)
t→∞
σ2 0
Since xτb0 = b we have Sτb0 = eσb e− τ
2 b and therefore
σ2 0
√
EW [e− τ
2 b ] = e−σb = e− 2λ |b|
(2.41)
which proves (2.38) for c = 0. The case c 6= 0 is obtained with Girsanov’s theorem.
We make the substitution of variables ys = xs + cs and write τ̃bc ({ys }) = τbc ({xs }) =
τbc ({ys − cs}) = τb0 ({ys }) to obtain
Z
 −λτ c  c
EW e b = e−λτb ({xs }) dW (0,∞) ({xs }s>0 )
Z 2
c cx c + c τ c ({x })
= e−λτ̃b ({ys }) e τb 2 b s dW (0,∞) ({ys }s>0 )
Z 2
0 cy 0 − c τ 0 ({ys })
= e−λτb ({ys }) e τb 2 b dW (0,∞) ({ys }s>0 )
Z
c2 0
= e cb
e−(λ+ 2 )τb ({ys }) dW (0,∞) ({ys }s>0 )
(2.41) √
2λ+c2
= ecb e−|b| (2.42)
This proves the lemma 

Example 9.2 (american cash-or-nothing call): The cash-or-nothing call has a payoff

1 falls St ≥ K
C(St ) = χ(St ≥ K) = (2.43)
0 falls St < K
at time t. For the price process we assume a geometric Brownian motion. We consider
first the european case, then we compute the price of the american claim using the system
(2.17-2.19) and finally we use (2.5) to calculate the price.

(i) The european case: According to (2.1), we get, using Lemma 6.7 in the second line
and abbreviating c = σr − σ2 , θ = T − t (τ is reserved for a stopping time)
Vt (St ) = e−r(T −t) EW̃ C(ST | St
 
Z
σ2
−r(T −t)
χ St eσyT −t +(r− 2 )(T −t) ≥ K pT −t (0, yT −t ) dyT −t


= e
Z R

1 − yθ2
= e−rθ χ yθ + cθ ≥ σ1 log SKt √2πθ e 2θ dyθ
R
n  1 log K −cτ o
−rθ
= e 1 − N σ √Sθt (2.44)
A2 205

(ii) The american case, using the Black-Scholes equation: We have to solve the
system
∂U σ2 2 ∂ 2 U
∂t
+ 2
S ∂S 2 + rS ∂U
∂S
− rU ≤ 0 , U (S, t) ≥ χ(S ≥ K) (2.45)
 2 2
 
∂U
∂t
+ σ2 S 2 ∂∂SU2 + rS ∂U
∂S
− rU U (S, t) − χ(S ≥ K) =0 (2.46)

U (S, T ) = χ(S ≥ K) (2.47)

Since U (S, t) > 0 for all t ∈ [0, T ), we have

U (S, t) = χ(S ≥ K) ⇔ U (S, t) = 1 = χ(S ≥ K)

⇔ S≥K (2.48)

Furthermore we have the boundary condition

U (S, t) → 0 for S → 0 (2.49)

In excercise 9.1 it has been shown that the transformation (κ = 2r/σ 2 )

S = Kex , t = T − 2τ /σ 2 , (x, τ ) ∈ R × [0, σ 2 T /2] (2.50)

and
1 1 1 2
u(x, τ ) := K
e 2 (κ−1)x+ 4 (κ+1) τ U (S, t) (2.51)

leads to the system

∂u ∂2u
∂τ
− ∂x2
≥ 0, u(x, τ ) ≥ g(x, τ ) (2.52)
 2
 
∂u ∂ u
∂τ
− ∂x2
u(x, τ ) − g(x, τ ) = 0 (2.53)
u(x, 0) = g(x, 0) (2.54)

where
1 1 1 2
g(x, τ ) = K
e 2 (κ−1)x+ 4 (κ+1) τ χ(x ≥ 0) (2.55)

Because of (2.48) we have

u(x, τ ) = g(x, τ ) for all x ≥ 0, τ ≥ 0 (2.56)

and u(x, τ ) > g(x, τ ) for x < 0. Hence, for negative x, the function u has to be a solution
of the diffusion equation
∂u ∂2u
∂τ
= ∂x2
, for x < 0 (2.57)
u(x, 0) = 0 , for x < 0 (2.58)
206 A2

with boundary conditions

1 1 2τ
u(0, τ ) = K
e 4 (κ+1) (2.59)
lim u(x, τ ) = 0 (2.60)
x→−∞

Such a system can be solved with a Laplace transformation with respect to τ . Let
Z ∞
v(x, ω) := e−ωτ u(x, τ ) dτ (2.61)
0

Then (2.57) becomes

d2 v
ω v(x, ω) − u(x, 0) = ω v(x, ω) = dx2
(x, ω) (2.62)

with general solution

√ √
v(x, ω) = aω e ωx
+ bω e− ωx
(2.63)

Because of (2.60) the coefficient aω ≡ 0 and the boundary condition (2.59) gives

bω = v(0, ω)
R∞
= 0 e−ωτ u(0, τ ) dτ
=: L[u(0, · )](ω) (2.64)
√
Let f (τ ) = fx (τ ) be the function whose Laplace transform is e−x ω
,
√
L[fx ](ω) = e−x ω
(2.65)

Then, by the properties of the Laplace transform,

√
v(x, ω) = bω e− ωx

= L[u(0, · )](ω) L[fx ](ω)

= L[u(0, · ) ∗ fx ](ω) (2.66)

such that

u(x, τ ) = u(0, · ) ∗ fx (τ )
Rτ
= 0 u(0, τ − s) fx (s) ds (2.67)

It remains to determine the function fx . To this end we use the following integral [7],
3.471.9:
R∞ β β

ν2 √
0
τ ν−1 e−γτ − τ dτ = 2 γ
Kν (2 βγ) (2.68)
A2 207

for Reβ > 0, Reγ > 0. Here Kν = K−ν is a modified Bessel function [7], 8.407. We only
need the values for ν = n + 21 . In that case [7], 8.468:
n
X
−z (n+k)!
(2z)−k
pπ
Kn+ 1 (z) = 2z
e (n−k)! k!
(2.69)
2
k=0

In particular,
R∞ 1 x2 /4
1
x2 /4 − 4
p
0
τ − 2 −1 e−ωτ − τ dτ = 2 ω
K− 1 (2 ωx2 /4)
2
√
2 ω
1
4
√
= 2 √x K 1 ( ωx)
2
√ 1 q √
2 ω √π e− ωx
4
= 2 √x 2 ωx
√ √
4π − ωx
= x
e (2.70)

which gives
x2
fx (τ ) = √ x
4πτ 3
e− 4τ (2.71)

Thus, recalling (2.67), we end up with

Z τ
1 2 (τ −s) x2
u(x, τ ) = 1
K
e 4 (κ+1) √x
4πs3
e− 4s ds (2.72)
0

or, recalling x = log(S/K),

1 1 2
U (S, t) = Ke− 2 (κ−1)x− 4 (κ+1) τ u(x, τ )
Z τ
2
− 21 (κ−1)x 1 2 − x4s
= e e− 4 (κ+1) s √4πs
x
3
e ds (2.73)
0

(iii) The american case, using stopping times: According to (2.5), we have

U (St , t) = EW̃ e−r(τU =C −t) C(SτU =C ) | St

 
(2.74)

where τU =C is the stopping time given by τU =C = inf u ≥ t | U (Su ) = C(Su ) . Ap-
parently, U ≤ 1 and U = 1 ⇔ Su ≥ K. By applying again Lemma 6.7, we have to
compute an integral with respect to the Wiener measure whose integrand is a function of
σ2
{St eσys +(r− 2 )s }0<s≤T −t . Hence, the condition Sτ ≥ K becomes (x = log(S/K))

1
yτ + cτ ≥ σ
log SKt = − σx (2.75)
208 A2

and the stopping time τU =C , because U > 0 for t < T , is given by


τU =C = inf U (Su ) = C(Su )
u∈[t,T ]

= inf U (Su ) = 1 = C(Su )
u∈[t,T ]

= inf Su = K
u∈[t,T ]

= inf yu−t + c(u − t) = − σx


(2.76)
u∈[t,T ]

or, abbreviating b = − σx and θ = T − t,

τbc := τU =C − t

= inf ys + cs = b (2.77)
s∈[0,T −t]

Thus,

EW̃ e−r(τU =C −t) χ(SτU =C ≥ K) | St

 
U (St , t) =
Z
c
e−rτb χ τbc ≤ θ dW ({ys }0<s≤θ )


=

cx 0 − c2 τ 0
Z
Girsanov 0
= e−rτb χ τb0 ≤ θ e τb 2 b dW ({xs }0<s≤θ )
Z
c2 0
cb
e−(r+ 2 )τb χ τb0 ≤ θ dW ({xs }0<s≤θ )


= e
Z θ
− σc x 1 r σ 2
e−(r+ 2 ( σ − 2 ) )u P τb0 ∈ [u, u + du)


= e
0
Z θ
1 1 r2 σ2
− 2 (κ−1)x
e−ru+ 2 ( σ2 −r+ 4 )u du d


= e P τb0 ≤ u du
0
Z θ
(10.21) 1 1 r2 σ2 R∞ y2
= e− 2 (κ−1)x e− 2 ( σ2 +r+ 4 )u du d
2 b/√u √12π e− 2 dy
0
Z T −t
1 1 r σ 2 b2
= e− 2 (κ−1)x e− 2 ( σ + 2 ) u √−b u
3
√1 e− 2u du
2π
0
u= 2s2
Z σ 2 (T −t)/2
− 12 (κ−1)x r 1 2 (x/σ)σ 3 1 x2
= σ
e e−( σ2 + 2 ) s
√ 3 √
2π
e− 4s σ22 ds
2s
0
Z σ 2 (T −t)/2
− 12 (κ−1)x 1 2s x2
= e e− 4 (κ+1) √x
4πs3
e− 4s ds (2.78)
0

and this coincides with the solution (2.73) of the PDE-system (2.45-2.47) since 2τ /σ 2 =
T − t.