Derivation of Local Volatility

by Fabrice Douglas Rouah

www.FRouah.com
www.Volopta.com
The derivation of local volatility is outlined in many papers and textbooks
(such as the one by Jim Gatheral [1]), but in the derivations many steps are left
out. In this Note we provide two derivations of local volatility.
1. The derivation by Dupire [2] that uses the Fokker-Planck equation.
2. The derivation by Derman et al. [3] of local volatility as a conditional
expectation.
We also present the derivation of local volatility from Black-Scholes implied
volatility, outlined in [1]. We will derive the following three equations that
involve local volatility o = o(o
t
, t) or local variance
L
= o
2
.
1. The Dupire equation in its most general form (appears in [1] on page 9)
0C
0T
=
1
2
o
2
1
2
0
2
C
01
2
+ (r
T

T
)
_
C 1
0C
01
_
. (1)
2. The equation by Derman et al. [3] for local volatility as a conditional
expected value (appears with
T
= 0 in [3])
0C
0T
= 1(r
T

T
)
0C
01

T
C +
1
2
1
2
1
_
o
2
T
[o
T
= 1

0
2
C
01
2
. (2)
3. Local volatility as a function of Black-Scholes implied volatility, =
(1, T) (appears in [1]) expressed here as the local variance
L

L
=
@w
@T
_
1
y
w
@w
@y
+
1
2
@
2
w
@y
2
+
1
4
_

1
4

1
w
+
y
2
w
__
@w
@y
_
2
_. (3)
where n = (1, T)
2
T is the Black-Scholes total implied variance and j =
ln
K
F
T
where 1
T
= exp
_
_
T
0
j
t
dt
_
is the forward price with j
t
= r
t

t
(risk free
rate minus dividend yield). Alternatively, local volatility can also be expressed
in terms of as

2
+ 2T
_
@
@T
+ (r
T

T
) 1
@
@K

_
1 +
Ky

@
@K
_
2
+ 1T
_
@
@K

1
4
1T
_
@
@K
_
2
+ 1
@
2

@K
2
_
.
Solving for the local variance in Equation (1), we obtain
o
2
= o (1, T)
2
=
@C
@T
(r
T

T
)
_
C 1
@C
@K
_
1
2
1
2
@
2
C
@K
2
. (4)
1
If we set the risk-free rate r
T
and the dividend yield
T
each equal to zero,
Equations (1) and (2) can each be solved to yield the same equation involving
local volatility, namely
o
2
= o (1, T)
2
=
@C
@T
1
2
1
2
@
2
C
@K
2
. (5)
The local volatility is then
L
=
_
o
2
(1, T). In this Note the derivation of
these equations are all explained in detail.
1 Local Volatility Model for the Underlying
The underlying o
t
follows the process
do
t
= j
t
o
t
dt + o(o
t
, t)o
t
d\
t
(6)
= (r
t

t
) o
t
dt + o(o
t
, t)o
t
d\
t
.
We sometimes drop the subscript and write do = jodt + ood\ where o =
o(o
t
, t). We need the following preliminaries:
Discount factor 1(t, T) = exp
_

_
T
t
r
s
d:
_
.
Fokker-Planck equation. Denote by )(o
t
, t) the probability density func-
tion of the underlying price o
t
at time t. Then ) satises the equation
0)
0t
=
0
0o
[jo)(o, t)] +
1
2
0
2
0o
2
_
o
2
o
2
)(o, t)

. (7)
Time-t price of European call with strike 1, denoted C = C(o
t
, 1)
C = 1(t, T)1
_
(o
T
1)
+

(8)
= 1(t, T)1
_
(o
T
1)1
(S
T
>K)

= 1(t, T)
_
1
K
(o
T
1))(o, T)do.
where 1
(S
T
>K)
is the Heaviside function and where 1 [] = 1 [[T
t
]. In the
all the integrals in this Note, since the expectations are taken for the underlying
price at t = T it is understood that o = o
T
, )(o, T) = )(o
T
, T) and do = do
T
.
We sometimes omit the subscript for notational convenience.
2 Derivation of the General Dupire Equation (1)
2.1 Required Derivatives
We need the following derivatives of the call C(o
t
, t).
2
First derivative with respect to strike
0C
01
= 1(t, T)
_
1
K
0
01
(o
T
1))(o, T)do (9)
= 1(t, T)
_
1
K
)(o, T)do.
Second derivative with respect to strike
0
2
C
01
2
= 1(t, T) [)(o, T)]
S=1
S=K
(10)
= 1(t, T))(1, T).
We have assumed that lim
S!1
)(o, T) = 0.
First derivative with respect to maturityuse the chain rule
0C
0T
=
0C
0T
1(t, T)
_
1
K
(o
T
1))(o, T)do + (11)
1(t, T)
_
1
K
(o
T
1)
0
0T
[)(o, T)] do.
Note that
@P
@T
= r
T
1(t, T) so we can write (11)
0C
0T
= r
T
C + 1(t, T)
_
1
K
(o
T
1)
0
0T
[)(o, T)] do. (12)
2.2 Main Equation
In Equation (12) substitute the Fokker-Planck equation (7) for
@f
@t
at t = T
0C
0T
+ r
T
C = 1(t, T)
_
1
K
(o
T
1) (13)
_

0
0o
[j
T
o)(o, T)] +
1
2
0
2
0o
2
_
o
2
o
2
)(o, T)

_
do.
This is the main equation we need because it is from this equation that the
Dupire local volatility is derived. In Equation (13) have two integrals to evaluate
1
1
= j
T
_
1
K
(o
T
1)
0
0o
[o)(o, T)] do, (14)
1
2
=
_
1
K
(o
T
1)
0
2
0o
2
_
o
2
o
2
)(o, T)

do.
Before evaluating these two integrals we need the following two identities.
3
2.3 Two Useful Identities
2.3.1 First Identity
From the call price Equation (8), we obtain
C
1(t, T)
=
_
1
K
(o
T
1))(o, T)do (15)
=
_
1
K
o
T
)(o, T)do 1
_
1
K
)(o, T)do.
From the expression for
@C
@K
in Equation (9) we obtain
_
1
K
)(o, T)do =
1
1(t, T)
0C
01
.
Substitute back into Equation (15) and re-arrange terms to obtain the rst
identity
_
1
K
o
T
)(o, T)do =
C
1(t, T)

1
1(t, T)
0C
01
. (16)
2.3.2 Second Identity
We use the expression for
@
2
C
@K
2
in Equation (10) to obtain the second identity
)(1, T) =
1
1(t, T)
0
2
C
01
2
. (17)
2.4 Evaluating the Integrals
We can now evaluate the integrals 1
1
and 1
2
dened in Equation (14).
2.4.1 First integral
Use integration by parts with n = o
T
1, n
0
= 1,
0
=
@
@S
[o)(o, T)] , =
o)(o, T)
1
1
= [j
T
(o
T
1)o
T
)(o, T)]
S=1
S=K
j
T
_
1
K
o)(o, T)do
= [0 0] j
T
_
1
K
o)(o, T)do.
We have assumed lim
S!1
(o 1)o)(o, T) = 0. Substitute the rst identity (16)
to obtain the rst integral 1
1
1
1
=
j
T
C
1(t, T)
+
j
T
1
1(t, T)
0C
01
. (18)
4
2.4.2 Second integral
Use integration by parts with n = o
T
1, n
0
= 1,
0
=
@
2
@S
2
_
o
2
o
2
)(o, T)

, =
@
@S
_
o
2
o
2
)(o, T)

1
2
=
_
(o
T
1)
0
0o
_
o
2
o
2
)(o, T)
_
_
S=1
S=K

_
1
K
0
0o
_
o
2
o
2
)(o, T)

do
= [0 0]
_
o
2
o
2
)(o, T)

S=1
S=K
= o
2
1
2
)(1, T)
where o
2
= o(1, T)
2
. We have assumed that lim
S!1
@
@S
_
o
2
o
2
)(o, T)
_
= 0.
Substitute the second identity (17) for )(1, T) to obtain the second integral 1
2
1
2
=
o
2
1
2
1(t, T)
0
2
C
01
2
. (19)
2.5 Obtaining the Dupire Equation
We can now evaluate the main Equation (13) which we write as
0C
0T
+ r
T
C = 1(t, T)
_
1
1
+
1
2
1
2
_
.
Substitute for 1
1
from Equation (18) and for 1
2
from Equation (19)
0C
0T
+ r
T
C = j
T
C j
T
1
0C
01
+
1
2
o
2
1
2
0
2
C
01
2
Substitute for j
T
= r
T

T
(risk free rate minus dividend yield) to obtain the
Dupire equation (1)
0C
0T
=
1
2
o
2
1
2
0
2
C
01
2
+ (r
T

T
)
_
C 1
0C
01
_
.
Solve for o
2
= o(1, T)
2
to obtain the Dupire local variance in its general form
o(1, T)
2
=
@C
@T
+
T
C + (r
T

K
)1
@C
@K
1
2
1
2
@
2
C
@K
2
Dupire [2] assumes zero interest rates and zero dividend yield. Hence r
T
=

T
= 0 so that the underlying process is do
t
= o(o
t
, t)o
t
d\
t
. We obtain
o(1, T)
2
=
@C
@T
1
2
1
2
@
2
C
@K
2
.
which is Equation (5).
5
3 Derivation of Local Volatility as an Expected
Value, Equation (2)
We need the following preliminaries, all of which are easy to show
@
@S
(o 1)
+
= 1
(S>K)
@
@S
1
(S>K)
= c(o 1)
@
@K
(o 1)
+
= 1
(S>K)
@
@K
1
(S>K)
= c(o 1)
@C
@K
= 1(t, T)1
_
1
(S>K)

@
2
C
@K
2
= 1(t, T)1 [c(o 1)]
In the table, c() denotes the Dirac delta function. Now dene the function
)(o
T
, T) as
)(o
T
, T) = 1(t, T)(o
T
1)
+
.
Recall the process for o
t
is given by Equation (6). By Itos Lemma, ) follows
the process
d) =
_
0)
0T
+ j
T
o
T
0)
0o
T
+
1
2
o
2
T
o
T
0
2
)
0o
2
T
_
dT +
_
o
T
o
T
0)
0o
T
_
d\
T
. (20)
Now the partial derivatives are
0)
0T
= r
T
1(t, T)(o
T
1)
+
,
0)
0o
T
= 1(t, T)1
(S
T
>K)
,
0
2
)
0o
2
T
= 1(t, T)c (o
T
1) .
Substitute them into Equation (20)
d) = 1(t, T) (21)
_
r
T
(o
T
1)
+
+ j
T
o
T
1
(S
T
>K)
+
1
2
o
2
T
o
2
T
c(o
T
1)
_
dT
+1(t, T)
_
o
T
o
T
1
(S
T
>K)

d\
T
Consider the rst two terms of (21), which can be written as
r
T
(o
T
1)
+
+ j
T
o
T
1
(S
T
>K)
= r
T
(o
T
1)1
(S
T
>K)
+ j
T
o
T
1
(S
T
>K)
= r
T
11
(S
T
>K)

T
o
T
1
(S
T
>K)
.
When we take the expected value of Equation (21), the stochastic term drops
out since 1 [d\
T
] = 0. Hence we can write the expected value of (21) as
dC = 1 [d)] (22)
= 1(t, T)1
_
r
T
11
(S
T
>K)

T
o
T
1
(S
T
>K)
+
1
2
o
2
T
o
2
T
c(o
T
1)
_
dT
6
so that
dC
dT
= 1(t, T)1
_
r
T
11
(S
T
>K)

T
o
T
1
(S
T
>K)
+
1
2
o
2
T
o
2
T
c(o
T
1)
_
. (23)
Using the second line in Equation (8), we can write
1(t, T)1
_
o
T
1
(S
T
>K)

= C + 11(t, T)1
_
1
(S
T
>K)

so Equation (23) becomes

dC
dT
= 11(t, T)r
T
1[1
(S
T
>K)
]
T
_
C + 11(t, T)1
_
1
(S
T
>K)
_
(24)
+
1
2
1(t, T)1
_
o
2
T
o
2
T
c(o
T
1)

= 1(r
T

T
)
0C
01

T
C +
1
2
1(t, T)1
_
o
2
T
o
2
T
c(o
T
1)

where we have substituted

@C
@K
for 1(t, T)1[1
(S
T
>K)
]. The last term in the
last line of Equation (24) can be written
1
2
1(t, T)1
_
o
2
T
o
2
T
c(o
T
1)

=
1
2
1(t, T)1
_
o
2
T
o
2
T
[o
T
= 1

1[c(o
T
1)]
=
1
2
1(t, T)1
_
o
2
T
[o
T
= 1

1
2
1[c(o
T
1)]
=
1
2
1
_
o
2
T
[o
T
= 1

1
2
0
2
C
01
2
where we have substituted
@
2
C
@K
2
for 1(t, T)1[c(o
T
1)]. We obtain the nal
result, Equation (2)
0C
0T
= 1(r
T

T
)
0C
01

T
C +
1
2
1
2
1
_
o
2
T
[o
T
= 1

0
2
C
01
2
.
When r
T
=
T
= 0 we can re-arrange the result to obtain
1
_
o
2
T
[o
T
= 1

=
@C
@T
1
2
1
2
@
2
C
@K
2
which, again, is Equation (5). Hence when the dividend and interest rate are
both zero, the derivation of local volatility using Dupires approach and the
derivation using conditional expectation produce the same result.
4 Derivation of Local Volatility From Implied
Volatility, Equation (3)
To express local volatility in terms of implied volatility, we need the three deriv-
atives
@C
@T
,
@C
@K
, and
@
2
C
@K
2
that appear in Equation (1), but expressed in terms of
7
implied volatility. Following Gatheral [1] we dene the log-moneyness
j = ln
1
1
T
where 1
T
= o
0
exp
_
_
T
0
j
t
dt
_
is the forward price (j
t
= r
t

t
, risk free rate
minus dividend yield) and 1 is the strike price, and the "total" Black-Scholes
implied variance
n = (1, T)
2
T
where (1, T) is the implied volatility. The Black-Scholes call price can then
be written as
C
BS
(o
0
, 1, (1, T) , T) = C
BS
(o
0
, 1
T
c
y
, n, T) (25)
= 1
T
(d
1
) c
y
(d
2
)
where
d
1
=
ln
S0
K
+
_
T
0
(r
t

t
) dt +
w
2
_
n
= jn

1
2
+
1
2
n
1
2
(26)
and d
2
= d
1

_
n = jn

1
2

1
2
n
1
2
.
4.1 The Reparameterized Local Volatility Function
To express the local volatility Equation (1) in terms of j, we note that the
market call price is
C(o
0
, 1, T) = C(o
0
, 1
T
c
y
, T)
and we take derivatives. The rst derivative we need is, by the chain rule
0C
0j
=
0C
01
01
0j
=
0C
01
1. (27)
The second derivative we need is
0
2
C
0j
2
=
0
0j
_
0C
01
_
1 +
0C
01
01
0j
(28)
=
0
2
C
01
2
1
2
+
0C
0j
,
since by the chain rule
@A
@y
=
@A
@K
@K
@y
, so that
@
@y
_
@C
@K
_
=
@
2
C
@K
2
@K
@y
=
@
2
C
@K
2
1. The
third derivative we need is
0C
0T
=
0C
0T
+
0C
01
01
0T
(29)
=
0C
0T
+
0C
01
1j
T
=
0C
0T
+
0C
0j
j
T
8
since 1 = o
0
exp
_
_
T
0
j
t
dt + j
_
so that
@K
@T
= 1j
T
. Equation (28) implies
that
0
2
C
01
2
1
2
=
0
2
C
0j
2

0C
0j
.
Now we substitute into Equation (1), reproduced here for convenience
0C
0T
=
1
2
o
2
1
2
0
2
C
01
2
+ j
T
_
C 1
0C
01
_
0C
0T

0C
0j
j
T
=
1
2
o
2
_
0
2
C
0j
2

0C
0j
_
+ j
T
_
C
0C
0j
_
which simplies to
0C
0T
=

L
2
_
0
2
C
0j
2

0C
0j
_
+ j
T
C (30)
where
L
= o
2
(1, T) is the local variance. This is Equation (1.8) of Gatheral
[1].
4.2 Three Useful Identities
Before expression the local volatility Equation (1) in terms of implied volatility,
we rst derive three identities used by Gatheral [1] that help in this regard. We
use the fact that the derivatives of the standard normal cdf and pdf are, using
the chain rule,
0
(r) = :(r)r
0
and :
0
(r) = r:(r)r
0
. We also use the relation
:(d
1
) =
1
_
2
c

1
2
(d2+
p
w)
2
=
1
_
2
c

1
2
(d
2
2
+2d2
p
w+w)
= :(d
2
)c
d2
p
w
1
2
w
= :(d
2
)c
y
.
From Equation (25) the rst derivative with respect to n is
0C
BS
0n
= 1
T
[:(d
1
)d
1w
c
y
:(d
2
)d
2w
]
= 1
T
_
:(d
2
)c
y
_
d
2w
+
1
2
n

1
2
_
c
y
:(d
2
)d
2w
_
=
1
2
1
T
c
y
_
:(d
2
)n

1
2
_
9
where d
1w
is the rst derivative of d
1
with respect to n and similarly for d
2
.
The second derivative is
0
2
C
BS
0n
2
=
1
2
1
T
c
y
_
:(d
2
)d
2
d
2w
n

1
2

1
2
:(d
2
)n

3
2
_
(31)
=
1
2
1
T
c
y
:(d
2
)n

1
2
_
d
2
d
2w

1
2
n
1
_
=
0C
BS
0n
__
jn

1
2
+
1
2
n
1
2
__
1
2
jn

3
2

1
4
n

1
2
_

1
2
n
1
_
=
0C
BS
0n
_

1
8

1
2n
+
j
2
2n
2
_
.
This is the rst identity we need. The second identity we need is
0
2
C
BS
0n0j
=
1
2
1
T
n

1
2
0
0j
[c
y
:(d
2
)] (32)
=
1
2
1
T
n

1
2
[c
y
:(d
2
) c
y
:(d
2
)d
2
d
2y
]
=
0C
BS
0n
[1 d
2
d
2y
]
=
0C
BS
0n
_
1
2

j
n
_
where d
2y
= n

1
2
is the rst derivative of d
2
with respect to j. To obtain the
third identity, consider the derivative
0C
BS
0j
= 1
T
[:(d
1
)d
1y
c
y
(d
2
) c
y
:(d
2
)d
2y
]
= 1
T
c
y
[:(d
2
)d
1y
(d
2
) :(d
2
)d
2y
]
= 1
T
c
y
(d
2
).
The third identity we need is
0
2
C
BS
0j
2
= 1
T
[c
y
(d
2
) + c
y
:(d
2
)d
2y
] (33)
= 1
T
c
y
(d
2
) + 1
T
c
y
:(d
2
)n

1
2
=
0C
BS
0j
+ 2
0C
BS
0n
.
We are now ready for the main derivation of this section.
4.3 Local Volatility in Terms of Implied Volatility
We note that when the market price C(o
0
, 1, T) is equal to the Black-Scholes
price with the implied volatility (1, T) as the input to volatility
C(o
0
, 1, T) = C
BS
(o
0
, 1, (1, T), T). (34)
10
We can also reparameterize the Black-Scholes price in terms of the total implied
volatility n = (1, T)
2
T and 1 = 1
T
c
y
. Since n depends on 1 and 1 depends
on j, we have that n = n(j) and we can write
C(o
0
, 1, T) = C
BS
(o
0
, 1
T
c
y
, n(j), T). (35)
We need derivatives of the market call price C(o
0
, 1, T) in terms of the Black-
Scholes call price C
BS
(o
0
, 1
T
c
y
, n(j), T). From Equation (35), the rst deriv-
ative we need is
0C
0j
=
0C
BS
0j
+
0C
BS
0n
0n
0j
(36)
= a(n, j) + /(n, j)c(j).
It is easier to visualize the second derivative we need,
@
2
C
@y
2
, when we express the
partials in
@C
@y
as a, /, and c.
0
2
C
0j
2
=
0a
0j
+
0a
0n
0n
0j
+ /(n, j)
0c
0j
+
_
0/
0j
+
0/
0n
0n
0j
_
c(j) (37)
=
0
2
C
BS
0j
2
+
0
2
C
BS
0j0n
0n
0j
+
0C
BS
0n
0
2
n
0j
2
+
_
0
2
C
BS
0n0j
+
0
2
C
BS
0n
2
0n
0j
_
0n
0j
=
0
2
C
BS
0j
2
+ 2
0
2
C
BS
0j0n
0n
0j
+
0C
BS
0n
0
2
n
0j
2
+
0
2
C
BS
0n
2
_
0n
0j
_
2
.
The third derivative we need is
0C
0T
=
0C
BS
0T
+
0C
BS
0n
0n
0T
(38)
= j
T
C +
0C
BS
0n
0n
0T
.
Gatheral explains that the second equality follows because the only explicit
dependence of C
BS
on T is through the forward price 1
T
, even though C
BS
depends implicitly on T through j and n. The reparameterized Dupire equation
(30) is reproduced here for convenience
0C
0T
=

L
2
_
0
2
C
0j
2

0C
0j
_
+ j
T
C.
We substitute for
@C
@T
,
@
2
C
@y
2
, and
@C
@y
from Equations (38), (37), and (36) respec-
tively and cancel j
T
C from both sides to obtain
0C
BS
0n
0n
0T
=

L
2
_
0
2
C
BS
0j
2
+ 2
0
2
C
BS
0j0n
0n
0j
+
0C
BS
0n
0
2
n
0j
2
+
0
2
C
BS
0n
2
_
0n
0j
_
2

0C
BS
0j
+
0C
BS
0n
0n
0j
_
. (39)
11
Now substitute for
@
2
C
BS
@w
2
,
@
2
C
BS
@w@y
, and
@
2
C
BS
@y
2
from the identities in Equations
(31), (32), and (33) respectively, the idea being to end up with terms involving
@C
BS
@w
on the right hand side of Equation (39) that can be factored out.
0C
BS
0n
0n
0T
=

L
2
0C
BS
0n
_
2 + 2
_
1
2

j
n
_
0n
0j
+
_

1
8

1
2n
+
j
2
2n
__
0n
0j
_
2
+
0
2
n
0j
2

0n
0j
_
.
Remove the factor
@C
BS
@w
from both sides and simplify to obtain
0n
0T
=
L
_
1
j
n
0n
0j
+
1
2
0
2
n
0j
2
+
1
4
_

1
4

1
n
+
j
2
n
__
0n
0j
_
2
_
.
Solve for
L
to obtain the nal expression for the local volatility expressed in
terms of implied volatility n = (1, T)
2
T and the log-moneyness j = ln
K
F
T

L
=
@w
@T
_
1
y
w
@w
@y
+
1
2
@
2
w
@y
2
+
1
4
_

1
4

1
w
+
y
2
w
__
@w
@y
_
2
_.
4.4 Alternate Derivation
In this derivation we express the derivatives
@C
@K
,
@
2
C
@K
2
, and
@C
@T
in the Dupire
equation (1) in terms of j and n = n(j), but we substitute these derivatives
directly in Equation (1) rather than in (30). This means that we take derivatives
with respect to 1 and T, rather than with j and T. Recall that from Equation
(35), the market call price is equal to the Black-Scholes call price with implied
volatility as input
C(o
0
, 1, T) = C
BS
(o
0
, 1
T
c
y
, n(j), T).
Recall also that from Equation (25) the Black-Scholes call price reparameterized
in terms of j and n is
C
BS
(o
0
, 1
T
c
y
, n(j), T) = 1
T
(d
1
) c
y
(d
2
)
where d
1
is given in Equation (26), and where d
2
= d
1

_
n. The rst derivative
we need is
0C
01
=
0C
BS
0j
0j
01
+
0C
BS
0n
0n
01
(40)
=
1
1
0C
BS
0j
+
0C
BS
0n
0n
01
.
12
The second derivative is
0
2
C
01
2
=
1
1
2
0C
BS
0j
+
1
1
0
01
_
0C
BS
0j
_
. (41)
+
0
01
_
0C
BS
0n
_
0n
01
+
0C
BS
0n
0
2
n
01
2
Let =
@C
@y
for notational convenience. Then
@
@K
_
@C
@y
_
=
@A
@K
and
0
01
_
0C
BS
0j
_
=
0
01
(42)
=
0
0j
0j
01
+
0
0n
0n
01
=
0
2
C
BS
0j
2
1
1
+
0
2
C
BS
0j0n
0n
01
.
Similarly
0
01
_
0C
BS
0n
_
=
0
2
C
BS
0j0n
1
1
+
0
2
C
BS
0n
2
0n
01
. (43)
Substituting Equations (42) and (43) into Equation (41) produces
0
2
C
01
2
=
1
1
2
0C
BS
0j
+
1
1
_
0
2
C
BS
0j
2
1
1
+
0
2
C
BS
0j0n
0n
01
_
(44)
+
_
0
2
C
BS
0j0n
1
1
+
0
2
C
BS
0n
2
0n
01
_
0n
01
+
0C
BS
0n
0
2
n
01
2
=
1
1
2
_
0
2
C
BS
0j
2

0C
BS
0j
_
+
2
1
0
2
C
BS
0j0n
0n
01
+
0
2
C
BS
0n
2
_
0n
01
_
2
+
0C
BS
0n
0
2
n
01
2
.
The third derivative we need is
0C
0T
=
0C
BS
0T
+
0C
BS
0j
0j
0T
+
0C
BS
0n
0n
0T
(45)
= j
T
C
BS
+
0C
BS
0j
j
T
+
0C
BS
0n
0n
0T
,
again using the fact that
@C
BS
@T
depends explicitly on T only through 1
T
. Now
substitute for
@C
@K
,
@
2
C
@K
2
, and
@C
@T
from Equations (40), (44), and (45) respectively
into Equation (4) for Dupire local variance, reproduced here for convenience.
o
2
=
@C
@T
j
T
_
C
BS
1
@C
@K

1
2
1
2
@
2
C
@K
2
.
13
We obtain, after applying the three useful identities in Section 4.2,
o
2
=
j
T
C
BS
+
@C
BS
@y
j
T
+
@C
BS
@w
@w
@T
j
T
_
C
BS
1
_
1
K
@C
BS
@y
+
@C
BS
@w
@w
@K
__
1
2
1
2
_
1
K
2
_
@
2
C
BS
@y
2

@C
BS
@y
_
+
2
K
@
2
C
BS
@y@w
@w
@K
+
@
2
C
BS
@w
2
_
@w
@K
_
2
+
@C
BS
@w
@
2
w
@K
2
_.
Applying the three useful identities in Section 4.2 allows the term
@C
BS
@w
to be
factored out of the numerator and denominator. The last equation becomes
o
2
=
_
@w
@T
+ j
T
1
@w
@K

1
2
1
2
_
2
K
2
+
2
K
_
1
2

y
w
_
@w
@K
+
_

1
8

1
2w
+
y
2
2w
2
_
_
@w
@K
_
2
+
@
2
w
@K
2
_. (46)
Equation (46) can be simplied by considering deriving the partial derivatives
of the Black-Scholes total implied variance, n = (1, T)
2
T. We have
@w
@T
=
2T
@
@T
+
2
,
@w
@K
= 2T
@
@K
, and
@
2
w
@K
2
= 2T
_
_
@
@K
_
2
+
@
2

@K
2
_
. Substitute into
Equation (46). The numerator in Equation (46) becomes

2
+ 2T
_
0
0T
+ j
T
1
0
01
_
(47)
and the denominator becomes
1 + 21T
_
1
2

j
n
_
0
01
+ 21
2

2
T
2
_

1
8

1
2n
+
j
2
2n
2
__
0
01
_
2
+1
2
T
_
_
0
01
_
2
+
0
2

01
2
_
.
Replacing n with
2
T everywhere in the denominator produces
1 + 21T
_
1
2

j

2
T
_
0
01
+ 21
2

2
T
2
_

1
8

1
2
2
T
+
j
2
2
4
T
2
__
0
01
_
2
+1
2
T
_
_
0
01
_
2
+
0
2

01
2
_
= 1 + 1T
0
01

21j

0
01

1
2

2
T
2
4
_
0
01
_
2
+
1
2
j
2

2
_
0
01
_
2
+1
2
T
0
2

01
2
=
_
1
1j

0
01
_
2
+
_
1 2
1j

0
01
+
_
1j

0
01
_
2
_
. (48)
Substituting the numerator in (47) and the denominator in (48) back to Equa-
tion (46), we obtain

2
+ 2T
_
@
@T
+ j
T
1
@
@K
_
_
1 +
Ky

@
@K
_
2
+ 1T
_
@
@K

1
4
1T
_
@
@K
_
2
+ 1
@
2

@K
2
_
14
See also the dissertation by van der Kamp [4] for additional details of this
alternate derivation.
References
[1] Gatheral, J. (2006). The Volatility Surface: A Practitioners Guide. New
York, NY: John Wiley & Sons.
[2] Dupire, B. (1994). "Pricing With a Smile." Risk 7, pp. 18-20.
[3] Derman, E., Kani, I., and M. Kamal (1996). "Trading and Hedging Local
Volatility." Goldman Sachs Quantitative Strategies Research Notes.
[4] van der Kamp, Roel (2009). "Local Volatility Modelling." M.Sc. disserta-
tion, University of Twente, The Netherlands.
15