Notes on Solving the Black-Scholes Equation

Eola Investments, LLC

[email protected]
October 2009
All Rights Reserved
Quantitative Discovery
Page 1 of 20 Eola Investments LLC All Rights Reserved
ABSTRACT
Solving the Black-Scholes equation is a good exercise for oneself in that it involves several very
useful techniques such as asymptotic analysis, change of variables, Green's function, Fourier transform,
and complex integration that can be applied elsewhere. Presented here is a systematic and self-
contained exercise to derive a closed-form solution of the Black-Scholes equation for a European call
option for the purpose of fun.
Quantitative Discovery
Page 2 of 20 Eola Investments LLC All Rights Reserved
STEP 1: TRANSFORM BLACK-SCHOLES EQUATION TO DIFFUSION EQUATION
It should not be surprised that the Black-Scholes equation can be reduced to a diffusion
equation, given that the Black-Scholes equation comes from a geometrical Brownian motion. However,
the transformation is not obvious. The following derivations can be regarded as side notes to
Silverman's work [1] .
With common notations, the Black-Scholes equation is
V
t
+rS
V
S
+
1
2
c
2
S
2
2
V
S
2
rV =0 (1)
It is easy to spot that the coefficients of the second and third terms can be changed to constants by a
transformation of variables S -lnS . However, strictly speaking, S has an unit and the log operator
can only apply to unitless quantities. We need to normalize S by some constant with the same unit as S.
A natural choice is the strike price K. Let x=ln( S/ K) , we have
V
S
=
V
x
dx
dS
=
V
x
1
S

2
V
S
2
=

S
V
S
=

S
(
1
S
V
x
)=
1
S
2
V
x
+
1
S

x
V
x
dx
dS
=
1
S
2
V
x
+
1
S
2

2
V
x
2
and Eq. (1) becomes
V
t
+(r
1
2
c
2
)
V
x
+
1
2
c
2

2
V
x
2
rV =0 (2)
In solving a partial differential equation we typically specify initial conditions, where the values
of the function at the start are known. That is, we would like to work on time to maturity, Tt , to
solve the equation, since at the time to maturity the value of a call option is exactly known. Using time
to maturity as the time variable, we have
Quantitative Discovery
Page 3 of 20 Eola Investments LLC All Rights Reserved

V
(T t )
+( r
1
2
c
2
)
V
x
+
1
2
c
2
2
V
x
2
rV =0 (3)
Now let us consider a special case that r=c
2
/ 2 , (which may not be so remote if the current
account deficits of US continue skyrocketing and Fed continues printing papers to finance it), and a
region that x is large enough that the curvature can be neglected. In this case the second and third terms
are dropped from Eq. (3) and the equation asymptotically becomes
V
(T t )
rV
, r=c
2
/ 2 and x - (4)
The solution to Eq. (4) is straightforward; it takes the form
V Ce
r ( T t )
(5)
where C is not a function of t. This relationship strongly suggests that if we construct a new function
W=Ve
r (Tt )
(6)
this function can absorb the rV term. Let us check:
V
(T t )
=
We
r ( T t )
(Tt )
=e
r ( Tt )
W
(Tt )
We
r (T t )
V
x
=
We
r ( T t)
x
=e
r ( T t )
W
x

2
V
x
2
=

2
We
r (Tt )
x
2
=e
r (Tt )
2
W
x
2
Inserting above terms into Eq. (3) we have a more simplified Black-Scholes equation

W
(T t )
+( r
1
2
c
2
)
W
x
+
1
2
c
2
2
W
x
2
=0 (7)
Again let us consider another special case where the time to maturity is so long that a few days
change in time will not affect the value of an option, given the same stock price. In this case, the time
Quantitative Discovery
Page 4 of 20 Eola Investments LLC All Rights Reserved
derivative term is neglected, resulting in a second order ordinary differential equation.
(r
1
2
c
2
)
W
x
+
1
2
c
2
2
W
x
2
0 , (Tt )- (8)
Rearranging Eq. (8), we get

2
W
x
2
+
rc
2
/ 2
c
2
/ 2
W
x
0 (9)
This equation is very easy to solve. Its solution takes the form of
WA+Be

rc
2
/ 2
c
2
/ 2
x
where A and B are not functions of x. Note the coefficient of x. If we group it into x to construct a new
variable y, we may be able to eliminate the coefficients in Eq (8), thus in Eq (7). Let us check.
y=
rc
2
/ 2
c
2
/ 2
x (10)
W
x
=
W
y
dy
dx
=
W
y
rc
2
/ 2
c
2
/ 2

2
W
x
2
=

x
W
x
=

x
(
W
y
r c
2
/ 2
c
2
/ 2
)=
rc
2
/ 2
c
2
/ 2

y
(
W
y
)
dy
dx
=(
rc
2
/ 2
c
2
/ 2
)
2

2
W
y
2
Combining the above equations with Eq. (7), we have

W
(T t )
+
(rc
2
/ 2)
2
c
2
/ 2
W
y
+
(rc
2
/ 2)
2
c
2
/ 2

2
W
y
2
=0

|
(rc
2
/ 2)
2
c
2
/ 2
(T t )

+
W
y
+

2
W
y
2
=0
Let
Quantitative Discovery
Page 5 of 20 Eola Investments LLC All Rights Reserved
t=
(rc
2
/ 2)
2
c
2
/ 2
(Tt ) (11)
we get a cleaner version of the equation
W
t
=
W
y
+

2
W
y
2
(12)
Once again let us consider another special case where the curvature effect can be neglected
(deep out of money options with short time to maturity) . In this case Eq. (12) is reduced to
W
t

W
y
, y- and t-0 (13)
Its solution is
WD(t+y) (14)
where D is a constant. This solution prompts us to replace y with a new variable,
z=t+y (15)
Now we regard W as a function of t and z, i.e., W=W(t , z (t, y)) . The consequences of this
change of variable are
W(t , y)
t
=
W(t , z(t , y))
t
+
W(t , z(t , y))
z
z
t
=
W(t , z(t , y))
t
+
W(t , z(t , y))
z
W(t, y)
y
=
W(t , z(t , y))
z
z
y
=
W(t , z(t , y))
z

2
W(t, y)
y
2
=

y
W(t, y)
y
=

y
W(t , z(t, y))
z
=

2
W (t, z(t, y))
z
2
z
y
=

2
W(t, z(t, y))
z
2
Substituting the above relationships into the Eq. 12, we finally have the diffusion equation
W
t
=

2
W
z
2
, 0t and z [16]
Quantitative Discovery
Page 6 of 20 Eola Investments LLC All Rights Reserved
Let us summarize all changes of variables here:
W=Ve
r (T t )
(6)
t=
(rc
2
/ 2)
2
c
2
/ 2
(Tt ) (11)
z=
(rc
2
/ 2)
2
c
2
/ 2
(Tt )+
rc
2
/ 2
c
2
/ 2
ln (
S
K
) (17)
Quantitative Discovery
Page 7 of 20 Eola Investments LLC All Rights Reserved
STEP 2: GREEN'S FUNCTION FOR DIFFUSION EQUATION
Let us use the diffusion equation to review the Green's function method for solving differential
equations. A general explicit one dimensional diffusion equation using our notations here is like this:
W
t

2
W
z
2
=Q(t , z) , 0t and z (17)
which is subject to certain initial condition W(0, z) and boundary conditions. A very lucky
characteristics of the above equation is its linearity. For example, if
W
t

2
W
z
2
=Q
1
(t, z) has a solution
W= f
1
(t , z)
, and
W
t

2
W
z
2
=Q
2
(t, z) has a solution
W= f
2
(t , z)
under the same initial and boundary conditions, we can immediately know that
W
t

2
W
z
2
=Q
1
(t, z)+Q
2
(t , z ) has a solution
W= f
1
(t , z)+f
2
(t , z)
.
Another aspect of linearity is that if
W
t

2
W
z
2
=Q(t , z ) has a solution W= f (t , z)
then we know that
W
t

2
W
z
2
=cQ(t , z) has a solution W=cf (t , z)
where c is a constant.
Having these two extremely handy properties in mind, we can solve Eq. (17) by first dividing
the function Q(t , z) into tiny pieces and solve them individually, then add the solutions together to
obtain the global solution. Actually we do not need to solve numerous equations as one might think so
Quantitative Discovery
Page 8 of 20 Eola Investments LLC All Rights Reserved
since we have divided the function Q(t , z) numerous times. We only need to solve it once with one
single piece of Q(t, z) which can be representative to any other tiny pieces. Such a representative
piece is described by a unit impulse and the representative diffusion equation is
W(t, z)
t

2
W (t , z)
z
2
=6(tt' )6( zz' ) (18)
where 6 is the Dirac delta function, which is an extension of the Kronecker delta function in discrete
space
Kronecker
6
ij
={
1, i =j
0, i j
(19)
to continuous space. The Dirac delta function has the following two properties:
6( xx' )=0, xx ' (20)

6( xx' ) dx' =1 (21)

The above properties leads to several very useful results

f ( x)6( xx' ) dx' = f ( x) (22)

6( xx' )=
d
dx
h( xx' )
, h( xx ' )=

6( xx' ) dx (23)
where h(x-x') is the Heaviside step function
h( xx ' )={
0, xx'
1/ 2, x=x '
1, x>x'
(24)
Note the notations in Eq. (18). t and x are the variables spanning the domain of interest
while t' and x' are the location of the impulse. We call the solution of Eq. (18) the Green's function,
G(t, t' , z , z ' ) , i.e.,
Quantitative Discovery
Page 9 of 20 Eola Investments LLC All Rights Reserved
G(t, t' , z , z' )
t

2
G(t, t' , z , z' )
z
2
=6(tt' ) 6( zz ' ) (25)
By moving the impulse at (t' , z ' ) to cover the whole surface of Q(t, z) , we can use Eq. (18) to
reconstruct Eq. (17). And the solution of Eq. (17) is just the integration of G(t , t' , z , z ' ) with
respect to (t' , z ' ) over (t , z) domain and with amplitude corrections defined by Q(t, z) .
(Remember the impulse we have studied is an unit impulse. And the amplitude of a real tiny piece of
Q(t, z) is, of course, defined by the amplitude of Q at (t' , z ' ) ). Mathematically, the solution of
Eq. (17) is
W(t, z)=

G(t, t' , z , z ' )Q(t' , z' ) dz ' d t' (26)

Let us check this claim:

W(t, z)
t

2
W (t , z)
z
2
=

G(t , t' , z , z ' )Q(t' , z ' ) dz ' d t' +

2
z
2

G(t , t' , z , z ' )Q(t' , z' ) dz ' d t'

=

0

t
G(t, t' , z , z' ) Q(t' , z ' )dz ' d t' +

2
z
2
G(t, t' , z , z' )Q(t' , z ' )dz ' d t'
=
0

Q(t' , z ' )
|

t
G(t , t' , z , z ' )+

2
z
2
G(t , t' , z , z ' )

dz ' d t'
=

0

Q(t' , z ' )6(tt' ) 6( z z ' ) dz ' d t'

= Q(t , z)
Wait! Not so quick. In the above discussion we have intentionally omitted the initial conditions.
In fact, we can consider the initial conditions are related to a special type of Q(t , z) located at
Quantitative Discovery
Page 10 of 20 Eola Investments LLC All Rights Reserved
t=0 . That is, for a homogeneous diffusion equation (Q = 0), we can write
W
t

2
W
z
2
=0=6(t)Q(t , z) , 0t and z (27)
Let us define an effective half width of the Dirac delta function, c , so that

c
c
6(t) d t-1
Integrating Eq. (27) in the domain t|c , c , we have

c
c
W(t, z)
t
d t

c
c

2
W(t , z)
z
2
d t=

c
c
6(t)Q(t , z) d t
lim
c -0
W (t , z)
t=c
t=c
lim
c-0

c
c

2
W(t , z)
z
2
d t=Q(0, z)
Now, W(c , z) is out of bound so we set it to be zero, and we obtain an astonishing result: the
source function at t=0 is the initial condition itself!
lim
c -0
W (c , z)=Q(0, z)
(28)
And the solution to Eq. (27) is then
W(t , z)=

G(t , t' , z , z' )6(t' 0)Q(t' , z ' ) dz ' d t'

=

G(t , z , z ' )Q(0, z' ) dz '

=

G(t , z , z ' )W(0, z ' )dz ' (29)

The Green's function of a homogeneous diffusion equation is reduced from Eq. (25) by setting t' =0
to
Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 11 of 20
G(t, z , z' )
t

2
G(t, z , z' )
z
2
=6(t)6( zz' ) (30)
And using the same argument that we derived Eq. (28) we know that the Green's function at t=0 is
a delta function
G( 0, z , z ' )=6( zz' ) (31)
Now let us check that the Eq. (29) is indeed the solution of (27).

W(t, z)
t

2
W (t , z)
z
2
=

G(t, z , z ' )W(0, z ' )dz ' +

2
z
2

W(t, z , z ' ) Q(0, z ' ) dz '

=

t
G(t , z , z ' )W(0, z ' ) dz ' +

2
z
2
G(t , z , z ' ) W(0, z' ) dz '
=

W(0, z' )
|

t
G(t , z , z ' )+

2
z
2
G(t , z , z ' )

dz'
=

W(0, z' ) 6(t)6( zz ' ) dz'

= 6(t)W (0, z)=6(t)Q(0, z)=6(t)Q(t, z)
And check that the initial condition is fulfilled:
W(0, z)=

G(0, z , z ' )W(0, z ' )dz ' =

6( z , z ' ) W(0, z' ) dz ' =W(0, z)

Combining Eq. (26) and Eq. (29), we obtain the general solution of a diffusion equation:
W(t , z)=

G(t , t' , z , z' )Q(t' , z ' ) dz ' d t' +

G(t, z , z' )W ( 0, z ' ) dz ' (32)

Next, we need to find out what exactly the Green's function is by solving Eq. (24). Denote the
Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 12 of 20
spacial Fourier transform of the Green's function as g(t, t' , k , k ' ) , we have
G(t , t' , z , z ' )=
1
.2n

g(t , t' , k , k ' )e

ikz
dk
(33)
G(t, t' , z , z' )
t
=
1
.2n

t
g (t ,t' , k , k ' ) e
ikz
dk

2
G(t, t' , z , z ' )
z
2
=
1
.2n

2
z
2
g(t , t' , k , k ' )e
ikz
dk=
k
2
.2n

g(t , t' , k , k ' )e

ikz
dk
The Fourier transform pair of the delta function are

6( k)=
1
.2n

6( zz ' )e
ikz
dz=
e
ikz '
.2n
6( zz ' )=
1
. 2n

6(k )e
ikz
dk=
1
2n

e
ikz '
e
ikz
dk
Therefore, the Fourier transform of Eq. (24) is
g (t , t' , k , k ' )
t
+k
2
g(t ,t' , k , k ' )=
1
.2n
6(tt' )e
ikz '
(34)
We just transformed a partial differential equation to a first order ordinary differential equation. When
tt' Eq. (34) becomes
g (t , t' , k , k ' )
t
+k
2
g(t ,t' , k , k ' )=0
Now, we are concerning with the causal Green's function which means when tt' the Green's
function is zero. This is because the impulse at t' can never give a response before t' . The arrow
of time has only one direction. when t>t' the solution to the above equation is
g(t, t' , k , k ' )=Ee
k
2
t
, t>t'
where E is a constant to be determined by the continuity of g(t, t' , k , k ' ) at t' below. The value
Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 13 of 20
of g(t, t' , k , k ' ) at t' is determined by integrating Eq. (34) with respect to t| t' c ,t' +c
and take limit c-0 :

t' c
t' +c
g(t ,t' , k , k ' )
t
d t+

t' c
t' +c
k
2
g (t, t' , k , k ' ) d t=
1
.2n

t' c
t' +c
6(tt' )e
ikz '
d t
lim
c -0
g(t ,t' , k , k ' )
t=t' c
t=t' +c
=lim
c-0
g (t' +c , t' , k , k ' )=
1
.2n
e
ikz '
Therefore
lim
c -0
Ee
k
2
( t' +c)
=
1
.2n
e
ikz '
E=
1
.2n
e
ikz '
e
k
2
t'
and the Green's function in the k domain is
g(t , t' , k , k ' )={
0, tt'
1
.2 n
e
k
2
( tt' )ikz '
, t>t'
(35a)
Or we use the Heaviside step function
g(t , t' , k , k ' )=
h(tt' )
.2n
e
k
2
(tt' )ikz '
(35b)
Substituting Eq. (35b) into Eq. (33)
G(t, t' , z , z' )=
1
2n

h(tt' ) e
k
2
(tt' )+ik ( zz ' )
dk
Now the integration as-is is not readily solvable. Here is a trick. It looks like a error function
erf ( x)=
2
.n

0
x
e
u
2
du (36)
and we are going to transform it to a error function. Let
Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 14 of 20
a(k+b)
2
+c=k
2
(tt' )+ik ( zz ' )
it is easy to find out that
a=(tt' ) , b=
i ( zz ' )
2(tt' )
, c=
( zz ' )
2
4(tt' )
Therefore we can transform the integral to
G(t, t' , z , z' )=
h(tt' )
2n
e

( zz ' )
2
4(tt' )

e
(tt' )
(
k
i( zz ' )
2(tt' )
)
2
dk
Let
u=k
i ( zz ' )
2(tt' )
then
G(t, t' , z , z' )=
h(tt' )
2n
e

( zz ' )
2
4(tt' )

i ( zz ' )/ 2( tt' )
i( z z ' )/ 2(tt' )
e
(tt' ) u
2
du
Now the integration becomes along a complex line and we need to construct a closed integration path
to evaluate this integral, as shown below.
Since there is no singularity in the function, the integration along the closed line
I
1
+I
2
+I
3
I
4
is
zero. Therefore
I
1
+I
2
+I
3
=I
4
. Furthermore, at infinite the function become zero so
I
1
and
I
3
are all zero and
I
2
=I
4
. That means the integral along the complex line is the same as the integral
along the real line, i.e.,

( k)=i ( zz ' )/ 2(tt' )
( k)=0

I
1
I
2
I
3
I
4
Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 15 of 20
G(t, t' , z , z ' )=
h(tt' )
2n
e

( zz ' )
2
4(tt' )

e
(tt' ) u
2
du
=
h(tt' )
2n.tt'
e

( zz ' )
2
4( tt' )

e
(tt' ) u
2
d (.tt' u)
=
h(tt' )
n.tt'
e

( z z ' )
2
4( tt' )

e
( tt' ) u
2
d (.tt' u)
=
h(tt' )
2.n(tt' )
e

( zz ' )
2
4( tt' )
|
2
.n

e
(tt' ) u
2
d (.tt' u)

=
h(tt' )
2.n(tt' )
e

( zz ' )
2
4( tt' )
erf ()
Therefore
G(t, t' , z , z' )=
h(tt' )
.4n(tt' )
e

( zz ' )
2
4(tt' )
(37)
Quantitative Discovery
Page 16 of 20 Eola Investments LLC All Rights Reserved
STEP 3: SOLVE BLACK-SCHOLES EQUATION USING GREEN'S FUNCTION
The procedure we derived in Step 2 is more than enough to solve the Black-Scholes equation
for vanilla European options. In fact, the procedure is more reserved for solving other exotic options
That we will demonstrate in other places.
To solve the Black-Scholes equation for a European call option, the general Green's function for
the diffusion equation is reduced to
G(t , z , z' )=
1
.4nt
e

( zz ' )
2
4t
(38)
The initial condition in (t,S) domain is
V ( 0, S)=h(SK )( SK) (39)
Again here we use a Heaviside step function. In the (t, z) domain, the initial condition is
W(0, z)=h( z0) K
(
e
c
2
/ 2
rc
2
/2
z
+1
)
(40)
The solution is, according to Eq. (29)
W(t , z)=

1
.4nt
e

( z z ' )
2
4t
h( z ' 0) K
(
e
c
2
/ 2
rc
2
/2
z '
+1
)
dz'
=
K
.4nt

( z z ' )
2
4t
+
c
2
/2
rc
2
/2
z '
dz ' +
K
.4nt

( zz ' )
2
4t
dz '
The second integral is easier to solve
K
.4nt

( z z ' )
2
4t
dz' =
K
.n

z
.4t

( z ' z)
2
4t
d
(
z ' z
.4t
)
=
K
.n
|

z
.4t
0
e
u
2
d u+

e
u
2
d u

=
K
2
|
2
.n

0
z
.4t
e
u
2
d u+
2
.n

e
u
2
d u

Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 17 of 20
=
K
2 |
erf
(
z
.4t
)
+1

Noticing that the error function relates to the normal cumulative distribution function by
N ( x)=
1
2 |
erf
(
x
.2
)
+1

(41)
we obtain
K
.4nt

( z z ' )
2
4t
dz' =K N
(
z
.2t
)
We use the same trick (besides using Matlab, Mathematica, Mathcad, or any software stating with M)
presented in Step 2 to carry out the first integration
K
.4nt

( z z ' )
2
4t
+
c
2
/2
rc
2
/2
z '
dz '
= K
.4nt

(
z ' z 2t
c
2
/ 2
r c
2
/ 2
)
2
4t
+
c
2
/2
r c
2
/ 2
(
z+t
c
2
/ 2
rc
2
/2
)
dz '
= K
.4nt
e
c
2
/ 2
rc
2
/2
(
z +t
c
2
/2
r c
2
/ 2
)

(
z ' z 2t
c
2
/2
r c
2
/ 2
)
2
4t
dz '
=
K
2
e
c
2
/ 2
rc
2
/2
(
z+t
c
2
/2
r c
2
/2
)
|
2
.n

z 2t
c
2
/ 2
r c
2
/ 2
.4t

(
z ' z2t
c
2
/ 2
rc
2
/2
)
2
4t
d (
z ' z2t
c
2
/ 2
rc
2
/ 2
.4t
)

=
K
2
e
c
2
/ 2
rc
2
/2
(
z+t
c
2
/2
r c
2
/2
)
|
erf
(
z+2t
c
2
/ 2
rc
2
/ 2
.4t
)
+1

=
K e
c
2
/ 2
r c
2
/ 2
(
z+t
c
2
/ 2
rc
2
/2
)
N
(
z
.2t
+.2t
c
2
/ 2
rc
2
/ 2
)
Therefore, the solution to Eq. (16) for a European call option is
Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 18 of 20
W(t , z)=K e
c
2
/2
r c
2
/ 2
(
z +t
c
2
/ 2
r c
2
/ 2
)
N
(
z
.2t
+.2t
c
2
/ 2
rc
2
/ 2
)
+K N
(
z
.2t
)
(42)
Recall that
W=Ve
r (T t )
(6)
t=
(rc
2
/ 2)
2
c
2
/ 2
(Tt ) (11)
z=
(rc
2
/ 2)
2
c
2
/ 2
(Tt )+
rc
2
/ 2
c
2
/ 2
ln (
S
K
) (17)
we can easily show
z
.2t
=d
2
=
ln
(
S
K
)
+
(
r
c
2
2
)
(T t )
c.Tt
(43)
z
.2t
+.2t
c
2
/ 2
rc
2
/ 2
=d
1
=
z
. 2t
+c .T t =d
2
+c.T t (44)
and
V ( t , S)=S N ( d
1
)+K e
r (T t)
N ( d
2
) (45)
which is the well known Black-Scholes formula. And some people prefer to call it Bachelier-Thorp
formula. [2]
Happy typing with OpenOffice Writer!

Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 19 of 20
REFERENCES
[1] Dennis Silverman, Solution of the Black Scholes Equation Using the Green's Function of the
Diffusion Equation (1999) http://www.physics.uci.edu/~silverma/bseqn/bs/bs.html. (Last access
October 16 2009). Note: there is a typo in Eq. (21) of the paper by Silverman.
[2] Espen Gaarder Haug & Nassim Nicholas Taleb, Why we have never used the Black-Scholes-
Merton Option Pricing Formula Fifth Version (2009), Social Science Research Network,
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1012075&rec=1&srcabs=8156 (Last access
October 22, 2009)
Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 20 of 20