January 3, 2020 20:32 Fitting Local Volatility – 9in x 6in b3761-main page 3

Chapter 1

Local Volatility and Dupire’s

by 182.70.15.86 on 05/21/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

Equation
Fitting Local Volatility Downloaded from www.worldscientific.com

Local volatility model was invented around 1994 in [Dupire (1994)] for the
continuous case and [Derman and Kani (1994a)] for the discrete case in
response to the following problem.
In the celebrated Black-Scholes model, see e.g. [Hull (1997)], the dynam-
ics of the stock price is modeled as a Geometric Brownian motion process
with constant volatility parameter σ
dSt = µSt + σSt dWt , St = S0 . (1.1)
t=0
Here, St is the stock price at the time t, µ is the drift, Wt is the standard
Brownian motion. It can be shown that under the risk-neutral measure Q
the drift becomes µ = r − q, where r, q are the constant interest rate and
continuous dividends functions.
Next, we introduce a notion of the Black-Scholes implied volatility. In
financial mathematics, the implied volatility σBS of an option contract is
that value of the volatility of the underlying instrument which, when input
in an Black-Scholes option pricing model will return a theoretical value
equal to the current market price of the option. The implied volatility shows
what the market implies about the underlying stock volatility in the future.
For instance, the implied volatility is one of six inputs used in a simple
option pricing (Black-Scholes) model, but is the only one that is not directly
observable in the market. The standard way to determine it by knowing the
market price of the contract and the other five parameters, is solving for the
implied volatility by equating the model and market prices of the option
contract. There exist various reasons why traders prefer considering option
positions in term of the implied volatility, rather than the option price itself,
see e.g., [Natenberg (1994)].
The Black-Scholes implied volatility is a useful measure, as it is a market
practice instead of quoting the option premium in the relevant currency,

3
 January 3, 2020 20:32 Fitting Local Volatility – 9in x 6in b3761-main page 4

4 Fitting Local Volatility

the options are quoted in terms of the Black-Scholes implied volatility. Over
the years, option traders have developed an intuition in this quantity. How-
ever, it can be further generalized by using a similar concept, but replacing
the Black-Scholes framework with another one. For instance, in [Corcuera
et al. (2009)] this is done under a Lévy framework, and, therefore, based on
distributions that match more closely historical returns. Here we don’t con-
by 182.70.15.86 on 05/21/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

sider these generalizations, and are concentrated only on the Black-Scholes

implied volatility.
Assume we are given an underlying and the continuous dividends func-
tion for this underlying. Also assume that the market constant interest rate
is somehow known. Finally, assume that we are given a snapshot of market
Fitting Local Volatility Downloaded from www.worldscientific.com

prices of, say European Call and Put options written on this underlying with
the same expiration date T . Then one can compute the Black-Scholes im-
plied volatilities for these options which will be a function of strikes K, and
plot them against the strike price. Thus obtained line is called a volatility
smile if it slopes upward on either end, or volatility skew if it slopes upward
only on the left. The former behavior is typical for the stock options, while
the latter — for the index options.
The important observation is that the volatility smiles should never oc-
cur based on standard Black-Scholes option theory, which normally requires
a completely flat volatility curve. However, the first notable volatility smile
was apparently seen back to 1987 following the stock market crash. Since
that time the topic attracted a lot of attention in the financial industry. As
mentioned in [Derman et al. (2016)], “After the crash, and ever since, equity
index option markets have displayed a volatility smile, an anomaly in bla-
tant disagreement with the Black-Scholes-Merton model. Since then, quants
around the world have labored to extend the model to accommodate this
anomaly”. There exist various books on local and implied volatility with
main focus on modeling, description, understanding, etc. We mentioned
just few of them based on our own preferences, [Derman et al. (2016);
Gatheral (2006); Natenberg (1994); Bossu (2014)], but the reader can also
find numerous references therein.
The existence of the smile forced the quants to move from the simple
Black-Scholes model to more sophisticated ones that would be able to de-
scribe this pattern. The idea of the local volatility model as proposed in
[Dupire (1994); Derman and Kani (1994a)] was as follows. Assuming that
only minimal changes should be applied to the Black-Scholes model, they
proposed to replace the constant volatility σ with that which is a determin-
istic function of St and t. In other words, in this model instead of Eq.(1.31)
 January 3, 2020 20:32 Fitting Local Volatility – 9in x 6in b3761-main page 5

Local Volatility and Dupire’s Equation 5

we have

dSt = µSt + σ(St , t)St dWt , St = S0 . (1.2)

t=0

Thus, in this model the local volatility is a function of the stock level St
and time t (rather than the constant value) which might be sufficient to
by 182.70.15.86 on 05/21/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

build a smile. With that, several natural questions become subject of an

immediately concern [Derman et al. (2016)]:

(1) Is there exist a unique local volatility function or surface σ(S, t) to

match the observed implied volatility surface σBS (S, t, K, T, r, q)?
Fitting Local Volatility Downloaded from www.worldscientific.com

(2) If yes, that means that we can explain the observed smile by means of
a local volatility process for the stock. Is the explanation meaningful?
Does the stock actually evolve according to an observable local volatility
function? There are many different models that can match the implied
volatility surface, but achieving a match doesn’t mean that model is
“correct.”
(3) What does the local volatility model tell us about the hedge ratios of
vanilla options and the values of exotic options? How do the results
differ from those of the classic Black-Scholes model?

The first question was positively answered by [Derman and Kani

(1994a)] who constructed a binomial-tree model with volatilities at every
model being calibrated to some stock and options market data. This mod-
ification of the binomial tree is called now the implied tree, [Hull (1997);
Derman et al. (2016)]. In turn, [Dupire (1994)] considered the stock’s local
volatility function as σ(K, T ), i.e. this is the local volatility σ(S, t) when
the future stock price is K at time T . He derived a forward PDE (known
now as the Dupire equation) which describes the dynamics of σ(K, T ) in
the continuous case, which means the derivatives of σ(K, T ) represent the
market prices of infinitesimal strike spreads, calendar spreads, and butterfly
spreads. We discuss this derivation in the next Section.

1.1 Dupire’s equation

The derivation of the Dupire equation is provided in many textbooks and

papers, e.g., [Dupire (1994); Gatheral (2004); Derman et al. (2016); Rouah
(2001)]. Here, we consider two ways to obtain it.
 January 3, 2020 20:32 Fitting Local Volatility – 9in x 6in b3761-main page 6

6 Fitting Local Volatility

1.1.1 Derivation using the Fokker-Planck equation

Below suppose that the interest rate is deterministic, r = r(t) and denote
the discount factor D(t, T ) as
!
Z T
D(t, T ) = exp − r(k)dk . (1.3)
by 182.70.15.86 on 05/21/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

By definition, a European option Call C(S, t, T ) and Put P (S, t, T ) prices

can be defined as, [Hull (1997)]

C(S, K, T − t) = D(t, T )EQ [(ST − K)+ ], (1.4)

Fitting Local Volatility Downloaded from www.worldscientific.com

+
P (S, K, T − t) = D(t, T )EQ [(K − ST ) ],

where for simplicity we dropped the dependence of the option prices on

r, q, σ, x+ = max(x, 0), and the expectation EQ is taken under the risk-
neutral measure Q. Therefore, it reads
Z
EQ [x] = xp(x, xT , T − t)dxT ,

where p(x, xT , T − t) is the transition probability density from the state

(x, t) into the state (xT , T ).
The function p(S, ST , T − t) satisfies the forward Kolmogorov (Fokker-
Planck) equation (see, e.g. [Risken and Haken (1989); Soize (1994); Van
Kampen (2007)] and references therein). With the simplified notation
p(S, t) instead of p(S, ST , T − t), it reads
∂ ∂ 1 ∂2 2 2
p(S, t) = − [µSp(S, t)] + [σ S p(S, t)]. (1.5)
∂t ∂S 2 ∂σ 2
This equation should be solved subject to the initial condition p(S, 0) =
δ(S − S0 ), where δ(x) is the Dirac delta-function.
The next step is to find an explicit expression for the option Theta:
Θ = ∂C
∂T by using the definition of C in Eq.(1.4). Using the chain rule we
get
∂D(t, T ) ∞
Z
∂C
= (ST − K)p(S, ST , T − t)dST (1.6)
∂T ∂T
Z ∞K
∂p(S, ST , T − t)
+ D(t, T ) (ST − K) dST
K ∂T
Z ∞
∂p(S, ST , T − t)
= −r(T )C + D(t, T ) (ST − K) dST .
K ∂T
 January 3, 2020 20:32 Fitting Local Volatility – 9in x 6in b3761-main page 7

Local Volatility and Dupire’s Equation 7

Now the derivative under the integral in Eq.(1.7) can be substituted with
the right hands side of Eq.(1.5) taken at t = T which yields
Z ∞
∂C
+ r(T )C = D(t, T ) (ST − K) (1.7)
∂T K
1 ∂2 2 2
 
∂
· − [µ(T )ST p(S, ST , T − t)] + [σ ST p(S, ST , T − t)] dST
by 182.70.15.86 on 05/21/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

∂ST 2 ∂σ 2
 
1
= D(t, T ) −µ(T )I1 + I2 ,
2
where the short notation I1 , I2 is introduced for the integrals in the second
line of Eq.(1.7). To evaluate these integrals we need two identities.
Fitting Local Volatility Downloaded from www.worldscientific.com

1.1.1.1 First identity

From the definition of the Call option price in Eq.(1.4) we have
Z ∞ Z ∞
C
= ST p(S, ST , T − t)dST − K p(S, ST , T − t)dST . (1.8)
D(t, T ) K K

On the other hand, from Eq.(1.4)

Z ∞
∂C
= −D(t, T ) p(S, ST , T − t)dST . (1.9)
∂K K
Therefore,
Z ∞  
1 ∂C
ST p(S, ST , T − t)dST = C −K . (1.10)
K D(t, T ) ∂K

1.1.1.2 The Breeden-Litzenberger identity

Differentiating both sides of Eq.(1.9) on K and using the fundamental the-
orem of calculus, we obtain
∂2C
= D(t, T )p(S, K, T − t). (1.11)
∂K 2
This identity is known as the Breeden-Litzenberger formula, [Breeden and
Litzenberger (1978)], which states that the risk-neutral probability of mak-
ing a transition from S at time t to K at time T is proportional to the
second partial derivative of the call price with respect to strike. 
With these identities we can proceed with evaluating the integrals I1 , I2 .
For I1 we get
Z ∞
∂
I1 = (ST − K) [ST p(S, ST , T − t)]dST (1.12)
K ∂S T
 January 3, 2020 20:32 Fitting Local Volatility – 9in x 6in b3761-main page 8

8 Fitting Local Volatility

h i∞ Z ∞
= (ST − K) ST p(S, ST , T − t) − ST p(S, ST , T − t)dST
K K
Z ∞  
1 ∂C
=− ST p(S, ST , T − t)dST = K −C , (1.13)
K D(t, T ) ∂K
where in the last line the result in Eq.(1.10) is used, and also it is assumed
by 182.70.15.86 on 05/21/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

that
lim (ST − K)ST p(S, ST , T − t) = 0.
ST →∞

In other words, the first and the second moment of the density p(S, ST , T −t)
are finite.
Fitting Local Volatility Downloaded from www.worldscientific.com

For I2 we obtain
Z ∞
∂2
I2 = (ST − K) 2 [σ 2 ST2 p(S, ST , T − t)]dST (1.14)
K ∂ST
∂ h 2 2 i∞
= (ST − K) σ ST p(S, ST , T − t)
∂ST K
Z ∞
∂
− [ST p(S, ST , T − t)]dST
K ∂S T
h i∞
= − σ 2 ST2 p(S, ST , T − t) = σ 2 K 2 p(S, K, T − t).
K
2 2
Here σ = σ(K, T ) , and it is assumed that
∂
lim (ST − K) [ST p(S, ST , T − t)] = 0.
ST →∞ ∂ST
Using the Breeden-Litzenberger we finally obtain
σ2 ∂2C
I2 = K2 . (1.15)
D(t, T ) ∂K 2

1.1.1.3 The final step

Substituting Eq.(1.12) and Eq.(1.15) into Eq.(1.7) we obtain
∂C ∂C 1 ∂2C
+ r(T )C = µ(T )C − µ(T )K + σ2 K 2 . (1.16)
∂T ∂K 2 ∂K 2
Taking into account that under the risk-neutral measure, the drift reads
µ(T ) = r(T ) − q(T ), [Brigo and Mercurio (2006)], we finally get the Dupire
equation
∂C 1 ∂C
= σ 2 K 2 − [(r(T ) − q(T )]K − q(T )C. (1.17)
∂T 2 ∂K
 January 3, 2020 20:32 Fitting Local Volatility – 9in x 6in b3761-main page 9

Local Volatility and Dupire’s Equation 9

This equation can also be solved for the local variance σ 2 (K, T ) to obtain
∂C ∂C
+ [(r(T ) − q(T )]K ∂K + q(T )C
σ 2 (K, T ) = ∂T
1 2 ∂2C
. (1.18)
2 K ∂K 2

1.1.2 A probabilistic approach

by 182.70.15.86 on 05/21/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

Here we derive the Dupire equation by using a probabilistic argument pro-

posed in [Derman and Kani (1998)]. Let us define the stochastic variable

f (ST , t, T ) = D(t, T )(ST − K)+ , (1.19)

where St is the Geometric Brownian motion introduced in Eq.(1.31).

Fitting Local Volatility Downloaded from www.worldscientific.com

Obviously, based on Eq.(1.4),

EQ [f (ST , t, T )|St = S] = C(S, K, T − t). (1.20)

Using Itö’s lemma at time t = T , one can find that f (ST , t, T ) follows
the process
2
 
∂f ∂f 1 2 2 ∂ f ∂f
df = + µ(T )ST + σ(T ) ST dT + σ(T )ST dWT .
∂T ∂ST 2 ∂St2 ∂ST
(1.21)
The partial derivatives in this expression could be easily found using the
definition of f in Eq.(1.19):
∂f
= −r(T )D(t, T )(ST − K)+ , (1.22)
∂T (
∂f 1, x ≥ 0
= D(t, T )1ST >K , 1x =
∂ST 0, x < 0
∂2f
= D(t, T )δ(ST − K).
∂ST2
Substituting this into Eq.(1.21) yields
h
df = D(t, T ) − r(T )(ST − K)+ + µ(T )ST 1ST >K (1.23)
1  i
+ σ(T )2 ST2 δ(ST − K) dT + σ(T )ST 1ST >K dWT .
2
The first two terms in parentheses can be re-written as follows

−r(T )(ST − K)+ + µ(T )ST 1ST >K = 1ST >K [−r(T )(ST − K) + µ(T )ST ]
= r(T )K1ST >K − q(T )ST 1ST >K . (1.24)
 January 3, 2020 20:32 Fitting Local Volatility – 9in x 6in b3761-main page 10

10 Fitting Local Volatility

Since Wt is a martingale, taking the risk-neutral expectation of Eq.(1.23)

and using Eq.(1.20) yields
h
dC = D(t, T )EQ r(T )K1ST >K − q(T )ST 1ST >K (1.25)
1 i
+ σ(T )2 ST2 δ(ST − K) dT.
2
by 182.70.15.86 on 05/21/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

Since
D(t, T )EQ [ST 1ST >K ] = C + KD(t, T )EQ [1ST >K ],
and
∂C
Fitting Local Volatility Downloaded from www.worldscientific.com

= −D(t, T )EQ [1ST >K ]. (1.26)

∂K
Eq.(1.25) can also be represented as
∂C
= D(t, T )K[r(T ) − q(T )]EQ [1ST >K ] − q(T )C (1.27)
∂T
1
+ D(t, T )EQ [σ 2 (T )ST2 δ(ST − K)]
2
∂C 1
= −K[r(T ) − q(T )] − q(T )C + D(t, T )EQ [σ 2 (T )ST2 δ(ST − K)].
∂K 2
The last term in this equation can be simplified by using the sifting property
of the Dirac delta function
1 1
D(t, T )EQ [σ 2 (T )ST2 δ(ST − K)] = D(t, T )EQ [σ 2 (T )ST2 |ST = K] (1.28)
2 2
1
· EQ [δ(ST − K)] = D(t, T )EQ [σ 2 (T )|ST = K]K 2 EQ [δ(ST − K)]
2
1 ∂2C
= EQ [σ 2 (T )|ST = K]K 2 ,
2 ∂K 2
since it follows from Eq.(1.26) that
∂2C
= D(t, T )EQ [δ(St − K)].
∂K 2
Thus, from Eq.(1.27) and Eq.(1.28) we finally obtain
∂C ∂C 1 ∂2C
= −K[r(T ) − q(T )] − q(T )C + EQ [σ 2 (T )|ST = K]K 2 .
∂T ∂K 2 ∂K 2
(1.29)
Comparing this with the Dupire equation Eq.(1.17) we see that
σ 2 (K, T ) = EQ [σ 2 (ST , T )|ST = K].
 January 3, 2020 20:32 Fitting Local Volatility – 9in x 6in b3761-main page 11

Local Volatility and Dupire’s Equation 11

This means that the local variance is the risk-neutral expectation of the
instantaneous variance conditional on the final stock price ST being equal
to the strike price K.

1.2 Local volatility via the implied volatility

by 182.70.15.86 on 05/21/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

In this Section we derive the identity that connects the local and Black-
Scholes implied variances. The identity was introduced in [Lipton (2002);
Gatheral (2006)] and reads
∂T w
σ 2 (T, K) =  2 , (1.30)
(∂y w)2 1 ∂y2 w
 
y∂y w 1
Fitting Local Volatility Downloaded from www.worldscientific.com

1− − + +
2w 4 w 4 2
where y = log K/F , F = Se(r−q)T is the stock forward price, and w = σBS
2
T
is the total implied variance. In terms of these variables the Black-Scholes
formula for the future value of the Call option price becomes, [Gatheral
(2006)]
CBS (FT , y, w) = D(t, T )FT [N (d1 ) − ey N (d2 )] (1.31)
  √   √ 
y w y w
= D(t, T )FT N − √ + − ey N − √ − ,
w 2 w 2
where N (x) is the normal CDF, and
σ2 √
ln FKT + BS T 1
d1 = √2 , d2 = d1 − σBS T . (1.32)
σBS T 2
The Dupire equation Eq.(1.17) can also be re-written in terms of the
new variables. It is easy to check that this yields
 2 
∂C 1 ∂ C ∂C
= σ2 − − qC. (1.33)
∂T 2 ∂y 2 ∂y
Computing derivatives of the Black-Scholes formula in Eq.(1.31), we
obtain
∂ 2 CBS y2
 
1 1 ∂CBS
= − − + − , (1.34)
∂w2 8 2w 2w2 ∂w
∂ 2 CBS
 
1 y ∂CBS
= − ,
∂w∂y 2 w ∂w
∂ 2 CBS ∂CBS ∂CBS
− =2 .
∂y 2 ∂y ∂w
 January 3, 2020 20:32 Fitting Local Volatility – 9in x 6in b3761-main page 12

12 Fitting Local Volatility

Taking into account that w = w(y, T ), the Dupire equation in Eq.(1.33)

can be also transformed by using the identities
∂C ∂CBS ∂CBS ∂w
= + , (1.35)
∂y ∂y ∂w ∂y
2
∂2C ∂ 2 CBS ∂ 2 CBS ∂w ∂ 2 CBS ∂w ∂CBS ∂ 2 w

= + 2 + + ,
by 182.70.15.86 on 05/21/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

∂y 2 ∂y 2 ∂w∂y ∂y ∂w2 ∂y ∂w ∂y 2
∂C ∂CBS ∂CBS ∂w ∂CBS ∂w
= + = − qCBS .
∂T ∂T ∂w ∂T ∂w ∂T
Substituting these expressions into Eq.(1.33) yields
∂CBS ∂w 1 ∂CBS
= σ2 A, (1.36)
Fitting Local Volatility Downloaded from www.worldscientific.com

∂w ∂T 2 ∂w
2
y2 ∂2w
   
∂w 1 1 ∂w
A=2− + − − + 2
+
∂y 8 2w 2w ∂y ∂y 2
 
1 ∂y ∂w
+2 − .
2 ∂w ∂y
Taking out a factor ∂C
∂w and simplifying, we finally obtain Eq.(1.30).
BS

An interesting particular case is when ∂w ∂y = 0. This implies

∂σBS (S,K,T )
∂K = 0, i.e., there is no skew and the implied volatility is flat.
Then it follows from Eq.(1.30) that σ 2 = ∂w ∂T . In other words, in this case
the local variance reduces to the forward Black-Scholes implied variance,
and
Z T
w(T ) = σ 2 (k)dk. (1.37)
0

Also, it is possible to show that the implied volatility σBS (S, K, T ) of

an option is approximately the average of the local volatilities σ(S, t) en-
countered over the life of the option between the current underlying price
and the strike. A detailed discussion of this rule of thumb is provided in
[Derman et al. (2016)].