Mathematical Methods for Financial Markets

by M. Jeanblanc, M. Yor and M. Chesney

Springer Verlag 2009, Chapter 6.

1 Stochastic Volatility

1.1 Option Pricing in Presence of Non-normality of Returns: The Mar-

tingale Approach

The property of non-normality of stock or currency returns which

has been observed and studied in many articles is usually taken into
account by relying either on a stochastic volatility or on a mixed
jump diffusion process for the price dynamics.
Strong underlying assumptions concerning the information arrival
dynamics determine the choice of the model. Indeed, a stochastic
volatility model will be adapted to a continuous information flow
and mixed jump-diffusion processes will correspond to possible dis-
continuities in this flow of information. In this context, option valu-
ation is difficult. Not only is standard risk-neutral valuation usually
no longer possible, but these models rest on the valuation of more
parameters. In spite of their complexity, some semi-closed-form so-
lutions have been obtained.
Let us consider the following dynamics for the underlying (a cur-
rency): ³ ´
e
dSt = St µdt + σt dBt , (1.1)
and for its volatility:
³ ´
ft .
dσt = σt f (σt )dt + γdW (1.2)

Here, (Bet , t ≥ 0) and (W

ft , t ≥ 0) are two correlated Brownian mo-
tions under the historical probability, and the parameter γ is a con-
stant.
In the Hull and White model, the underlying is a stock, the function
f is constant, hence σ follows a geometrical Brownian motion. In
2
the Scott model, it has the form: f (σ) = β(a − ln σ) + γ2 , where the

1
parameters a, β and γ are constant. By relying on Ito’s lemma, the
logarithm of the volatility follows a Vasicek process:
ft
d ln σt = β(a − ln(σt ))dt + γdW

The standard Black and Scholes approach of riskless arbitrage is not

enough to produce a unique option pricing function CE . Indeed, the
volatility is not a traded asset and there is no asset perfectly corre-
lated with it. The three assets required in order to eliminate the two
sources of uncertainty and to create a riskless portfolio will be the
foreign bond and, for example, two options of different maturities.
Therefore, it will be impossible to determine the price of an option
without knowing the price of another option on the same underly-
ing (see Scott (1987)). Under any risk-adjusted probability Q, the
dynamics of the underlying spot price and of the volatility are given
by ½
dSt = St ((r − δ)dt + σt dBt ) ,
(1.3)
dσt = σt (f (σt ) − Φσt )dt + γσt dWt
where (Bt , t ≥ 0) and (Wt , t ≥ 0) are two correlated Q-Brownian mo-
tions and where Φσt , the risk premium associated with the volatility,
is unknown since the volatility is not traded.
The expression of the underlying price at time T
µ Z Z T ¶
1 T 2
ST = S0 exp (r − δ)T − σ du + σu dBu , (1.4)
2 0 u 0

which follows from (1.3), will be quite useful in pricing European

options as is now detailed.
We first start with the case of 0 correlation between B and W .

Proposition 1.1.1 Assume that the dynamics of the underlying spot

price (for instance a currency) and of the volatility are given, under
the risk-adjusted probability, by equations (1.3), with a zero correla-
tion coefficient, where the risk premium Φσ associated with the volatil-
ity is assumed constant. The European Call price can be written as

2
follows:
Z +∞
CE (S0 , σ0 , T ) = BS(S0 e−δT , a, T )dF (a) (1.5)
0

where BS is the Black and Scholes price:

BS(x, a, T ) = xN (d1 (x, a)) − Ke−rT N (d2 (x, a)) .
Here,
ln(x/K) + rT + a/2 √
d1 (x, a) = √ , d2 (x, a) = d1 (x, a) − a ,
a
δ is the foreign interest rate and F is the distribution function (under
the risk-adjusted probability) of the cumulative squared volatility ΣT
defined as Z T
ΣT = σu2 du . (1.6)
0

Proof: The stochastic integral which appears on the right-hand side

of (1.4) is a stochastic time-changed Brownian motion:
Z t
σu dBu = BΣ∗ t
0

where (Bs∗ , s ≥ 0) is a Q-Brownian motion. Therefore,

µ ¶
Σ T
ST = S0 exp (r − δ)T + BΣ∗ T −
2
and the conditional law of ln(ST /S0 ) given ΣT is
ΣT
N((r − δ)T − , ΣT ) .
2
By conditioning with respect to ΣT , returns are normally distributed,
and the Black and Scholes formula can be used for the computation
of the conditional expectation
bT : = EQ (e−rT (ST − K)+ |ΣT ) .
C
Formula (1.5) is therefore obtained from
bT ) = EQ (BS(S0 , ΣT , K)).
EQ (e−rT (ST − K)+ ) = EQ (C

3
In order to use this result, the risk-adjusted distribution function F
of ΣT is needed. One method of approximating the option price is
the Monte Carlo method. By simulating the instantaneous volatility
process σ over discrete intervals from 0 to T , the random variable
ΣT can also be simulated. This requires the solution of (1.3) for the
volatility process (see the Appendix 1 and 2). Each value of ΣT is
substituted into the Black and Scholes formula with ΣT in place of
σ 2 T . The sample mean converges in probability to the option price as
the number of simulations increases to infinity. The estimates for the
parameters of the volatility process could be computed as described
in Scott (1987) and in Chesney and Scott (1989) for currencies: by
relying on the method of moments and on ARMA processes. The risk
premium associated with the volatility, i.e., Φσ , which is assumed to
be a constant, should also be estimated.

1.2 Hull and White Model

Hull and White (1987) consider a stock option (δ = 0); they assume
that the volatility follows a geometric Brownian motion and that
it is uncorrelated with the stock price and has zero systematic risk.
Hence, the drift f (σt ) of the volatility is a constant k and Φσ is taken
to be null:
dσt = σt (kdt + γdWt ) . (1.7)
They also introduce the following random variable: VT = ΣT /T . In
this framework, they obtain another version of equation (1.5):
Z +∞
CE (S0 , σ0 , T ) = BS(S0 , vT, T )dG(v) (1.8)
0

where G is the distribution function of VT .

4
1.2.1 Approximation

By relying on a Taylor expansion, the left-hand side of (1.8) may be

approximated as follows: introduce C(y) = BS(S0 , yT, T ), then
1 d2 C
CE (S0 , σ0 , T ) ≈ C(∇T ) + (∇T )Var(VT ) + ...
2 dy 2
where
E(ΣT )
∇T : = = E(VT ) .
T
We recall the following results:
( κT
E(VT ) = he κT−1 σ02 i
(2κ+ϑ2 )T
eκT (1.9)
E(VT2 ) = (κ+ϑ2e2 )(2κ+ϑ2 )T 2 + κT2 2 ( 2κ+ϑ
1
2 − κ+ϑ2 ) σ04

where κ and ϑ are respectively the drift and the volatility of the
squared volatility:
κ = 2k + γ 2 , ϑ = 2γ .
When κ is zero, i.e., when the squared volatility is a martingale, the
following approximation is obtained:
1 d2 C 1 d3 C
CE (S0 , σ0 , T ) ≈ C(∇T )+ (∇T )Var(VT )+ (∇T )Skew(VT )+...
2 dy 2 6 dy 3
where Skew(VT ) is the third central moment of VT . The first three
moments are obtained by relying on the following formulae (note that
the first two moments are the limits obtained from (1.9) as κ goes to
0) :
E(VT ) = σ02 ,
ϑ2 T
2 2(e − ϑ2 T − 1) 4
E(VT ) = σ0 ,
ϑ4 T 2
2 2
3 e3ϑ T − 9eϑ T + 6ϑ2 T + 8 6
E(VT ) = σ0 .
3ϑ6 T 3

1.2.2 Closed-form Solutions in the Case of Uncorrelated Processes

As we have explained above in Proposition 1.1.1, in the case of un-

correlated Brownian motions, the knowledge of the law of ΣT yields

5
the price of a European option, at least in a theoretical way. In fact,
this law is quite complicated, and a closed-form result is given as a
double integral.

Proposition 1.2.1 Let (σt , t ≥ 0) be the GBM solution of (1.7) and

RT
ΣT = 0 σs2 ds. The density of ΣT is Q(ΣT ∈ dx)/dx = g(x) where
µ ¶ν µ 2 ¶
1 xγ 2 1 π σ02 ν 2γ 2T
g(x) = exp − 2 −
x σ02 (2π 3 γ 2 T )1/2 2γ 2 T 2γ x 2
Z ∞
1
× dy y ν exp(− 2 γ 2 xy 2 )Υy (γ 2 T )
0 2σ0
where the function Υy is defined by
Z ∞ ³ πy ´
2
Υy (t) = dy exp(−y /2t) exp[−r(cosh y)] sinh(y) sin
0 t
k
and where ν = γ2 − 12 . the integral found in (??)

1.3 Closed-form Solutions in some Correlated Cases

We now present the formula for a European call in a closed form

(up to the computation of some integrals). We consider the general
case where the correlation between the two Brownian motions W f, Be
given in (1.1, 1.2) under the historical probability (hence between the
risk-neutral Brownian motions W, B defined in (1.3)) is equal to ρ.
Thus, we write the risk-neutral dynamics of the stock price (δ = 0)
and the volatility process as
( ³ p ´
2 (1)
dSt = St rdt + σt ( 1 − ρ dWt + ρ dWt ) ,
(1.10)
dσt = σt (kdt + γ dWt ),
where the two Brownian motions (W (1) , W ) are independent. In an
explicit form
Z Z t p Z t
1 t 2
ln(ST /S0 ) = rt − σs ds + ρ σs dWs + 1 − ρ2 σs dWs(1)
2 0 0 0
(1.11)
We first present the case where the two Brownian motions W, B b are
independent, i.e., when ρ = 0.

6
Proposition 1.3.1 Case ρ = 0: Assume that, under the risk-
neutral probability
( ³ ´
(1)
dSt = St r dt + σt dWt ,
(1.12)
dσt = σt (k dt + γ dWt ),
where the two Brownian motions (W (1) , W ) are independent. The
price of a European call, with strike K is
C = S0 f1 (T ) − Ke−rT f2 (T )
where the functions fj , defined as
Z ∞
f1 (T ) := E(N (d1 (ΣT ))) = N (d1 (x))g(x)dx
0
and Z ∞
f2 (T ) := E(N (d2 (ΣT ))) = N (d2 (x))g(x)dx
0
where the function g(.) is defined in Proposition 1.2.1.

Proof: The solution of the system (1.12) is:

( ³R R ´
rt t (1) 1 t 2
St = S0 e exp 0 σs dWs − 2 0 σs ds ,
σt = σ0 exp(γWt + (k − γ 2 /2)t) = σ0 exp(γWt + γ 2 νt) ,
(1.13)
2 W
where ν = k/γ − 1/2. Conditionally on F , the process ln S is
Gaussian, and ln(ST /S0 ) is a Gaussian variable with mean rT −ΣT /2
RT
and variance ΣT where ΣT = 0 σs2 ds. It follows that
C = S0 E (N(d1 (ΣT ))) − Ke−rt E (N(d2 (ΣT )))
By relying on Proposition 1.2.1, the result is obtained. ¤
We now consider the case ρ 6= 0, but k = 0, i.e., the volatility is a
martingale.
In this case the following result is required:
Proposition 1.3.2 Let {Wt : t ≥ 0} be a Brownian motion and let
us define:
Rt
At = 0 e2Wu du then
Z ∞ Z ∞
Wt 1 1 −v(1+y2 )/2
E(f (e , At )) = dy dv f (y, )e Υyv (t)
(2π 3 t)1/2 0 0 v
7
where the function Υ is defined in Proposition 1.2.1 and where the
function f is continuous and positive.
Proposition 1.3.3 (Case k = 0) Assume that
( ³ p ´
2 (1)
dSt = St rdt + σt ( 1 − ρ dWt + ρ dWt ) ,
dσt = γσt dWt .
with ρ 6= 1. The price of a European call with strike K is given by

C = S0 f1∗ (γ 2 T ) − Ke−rT f2∗ (γ 2 T )

where the functions fj∗ , defined as fj∗ (t) := E(fj (eWt , At )), j = 1, 2
are given in Proposition 1.3.2. Here, the functions fi (x, y) are:
1 2 2 2 2
f1 (x, y) = e−γ T /8
√ eρσ0 (x−1)/γ e−σ0 ρ y/(2γ ) N(d∗1 (x, y)),
x
2 1
f2 (x, y) = e−γ T /8 √ N(d∗2 (x, y)),
x
where
µ ¶
γ S0 ρσ0 (x − 1) σ02 y(1 − ρ2 )
d∗1 (x, y) = p ln + rT + + ,
σ0 (1 − ρ2 )y K γ 2γ 2
σ0 p
d∗2 (x, y) = d1 (x, y) − (1 − ρ2 )y ,
γ
Rt
and At = 0 e2Ws ds.
Rt
Proof: In that case, one notices that 0 σs dWs = γ1 (σt − σ0 ) where

σt = σ0 exp(γWt − γ 2 t/2) .

Hence, from equation (1.4),

µ Z p Z t ¶
1 t 2 ρ
St = S0 exp rt − σs ds + (σt − σ0 ) + 1 − ρ2 σs dWs(1) ,
2 0 γ 0

and conditionally on FW the law of ln(St /S0 ) is Gaussian with mean

Z
1 t 2 ρ
rt − σs ds + (σt − σ0 )
2 0 γ

8
and variance
Z t Z t
2
2 2 2 2 2
(1 − ρ ) σs ds = (1 − ρ )Σt = (1 − ρ )σ0 e2(γWs −γ s/2) ds .
0 0

The price of a call option is

EQ (e−rT (ST − K)+ ) = EQ (e−rT ST 11{ST ≥K} ) − Ke−rT Q(ST ≥ K) .

We now recall that, if Z is a Gaussian random variable with mean

m and variance β, then
1
P(eZ ≥ k) = N( √ (m − ln k))
β
µ ¶
β + m − ln k
E(eZ 11{Z≥k} ) = em+β/2 N √ .
β
It follows that Q(ST ≥ K) = EQ (N(d2 (σT , ΣT ))) where
µ ¶ µ ¶
1 S0 1 ρ ∗ u γ2
d2 (u, v) = p ln + rT − v + (u − σ0 ) = d2 , v .
(1 − ρ2 )v K 2 γ σ0 σ02

By scalingR and by relying on Girsanov’s theorem, setting τ = T γ 2

t ∗
and A∗t = 0 e2Ws ds, we obtain
µ µ 2
¶ ¶
Wτ∗ σ0 ∗ − 12 Wτ∗ −τ /8
Q(ST ≥ K) = EQ∗ N d2 (σ0 e , 2 Aτ ) e
γ
∗
= EQ∗ (f2 (eWτ , A∗τ )) .

where
dQ∗ ν2t
|Ft = e− 2 −νWt
dQ
and Wt∗ = Wt + νt is a Q∗ Brownian motion. Indeed the scaling
property of BM implies that
2 Z γ2T Z 2
law σ0 σ02 γ T ∗ σ02 ∗
ΣT = exp (2(W s + νs)) ds = exp (2W s ) ds = A .
γ2 0 γ2 0 γ2 τ
where ν = −1/2, because k = 0.

9
For the term E(e−rT ST 11{ST ≥K} ), we obtain easily
E(e−rT ST 11{ST ≥K} )
µ µ 2
¶
Wτ∗ σ0 ∗
= S0 EQ∗ N d1 (σ0 e , 2 Aτ )
γ
µ ¶¶
Wτ∗ τ ρσ0 Wτ∗ ρ2 σ02 ∗
exp − − + (e − 1) − A
2 8 γ 2γ 2 τ
∗
= S0 EQ∗ (f1 (eWτ , A∗τ )) .

The functions f1 and f2 are defined in Proposition 1.3.3. and the

result is obtained. ¤

1.4 PDE Approach

We come back to the simple Hull and White model 1.12, where
( ³ ´
(1)
dSt = St r dt + σt dWt ,
(1.14)
dσt = σt (k dt + γ dWt ),
for two independent Brownian motions. Itô’s lemma allows us to
obtain the equation:
· ¸
∂CE ∂CE ∂CE 1 2 2 ∂ 2 CE 1 2 2 ∂ 2 CE
dCE = dSt + dσt + + σt St + γ σt dt .
∂x ∂σ ∂t 2 ∂x2 2 ∂σ 2
Then, the martingale property of the discounted price leads to the
following equation:
1 2 2 ∂ 2 CE 1 2 2 ∂ 2 CE ∂CE ∂CE ∂CE
σ x + γ σ + rx + σk + − rCE = 0 .
2 ∂x2 2 ∂σ 2 ∂x ∂σ ∂t
In the Scott model (see 1.3), the following PDE is obtained:
1 2 2 ∂ 2 CE 1 2 2 ∂ 2 CE ∂CE σ ∂CE ∂CE
σ x + γ σ +(r−δ)x +σ(f (σ)−Φ t) + −rCE = 0 .
2 ∂x2 2 ∂σ 2 ∂x ∂σ ∂t
Indeed
dCE St ∂CE σ ∂CE
= σ dB̃t + γ dW̃t (1.15)
CE CE ∂x CE ∂σ
µ ¶
1 ∂CE ∂CE ∂CE 1 2 2 ∂ 2 CE 1 2 2 ∂ 2 CE
+ µSt + f (σt )σt + + γ σt + σ x dt
CE ∂x ∂σ ∂t 2 ∂σ 2 2 ∂x2
10
Relying on a continuous time version of a two factor APT (Arbitrage
Pricing Theory) model:
µ ¶
dCE St ∂CE S σ ∂CE σ
E = (r + Φt + Φ )dt (1.16)
CE CE ∂x CE ∂σ t
Since the foreign currency is a traded asset and investors can earn
the foreign interest rate on funds invested in the foreign currency,
the risk premium associated to the currency ΦSt , can be written as
follows:
ΦSt = µ + δ − r
However, since the volatility is not a traded asset, the risk premium
associated to the volatility Φσt remains unknown. By equating the
expected return of the option price generated by equations (1.15)
and (1.16) the PDE is obtained.

1.5 Heston’s Model

In the Heston (1993) model, the underlying process follows a geo-

metric Brownian motion under the risk-neutral probability Q (with
δ = 0): ³ ´
b
dSt = St rdt + σt dBt ,
and the squared volatility follows a square-root process. The model
allows arbitrary correlation between volatility and spot asset returns.
The dynamics of the volatility are given by:
dσt2 = κ(θ − σt2 )dt + γσt dWt .
The parameters κ, θ and γ are constant, and κθ > 0, so that the
square-root process remains positive. The Brownian motions (Wt , t ≥
0) and (Bbt , t ≥ 0) have correlation coefficient equal to ρ. Setting
Xt = ln St and Yt = σt2 these dynamics can be written under the
risk-neutral probability Q as

 σt2 bt ,
dXt = (r − )dt + σt dB (1.17)
2
 dY = κ(θ − Y )dt + γ √Y dW .
t t t t

11
As usual, the computation of the value of a call reduces to the com-
putation of
b T ≥ K)−Ke−rT Q(ST ≥ K)
EQ (ST 11ST ≥K )−Ke−rT Q(ST ≥ K) = S0 Q(S
b X follows dXt = (r + σ 2 /2)dt + σt dBt , where B is a
where, under Q, t
b
Q-Brownian motion. In this setting, the price of the European call
option is:
b − Ke−rT Γ ,
CE (S0 , σ0 , T ) = S0 Γ

with
b T ≥ K), Γ = Q(ST ≥ K) .
b = Q(S
Γ
We present the computation for Γ, then the computation for Γb follows
from a simple change of parameters. The law of X is not easy to
compute; however, the characteristic function of XT , i.e.,

f (x, σ, u) = EQ (eiuXT |X0 = x, σ0 = σ)

can be computed using the results on affine models. From Fourier

transform inversion, Γ is given by
Z µ −iu ln(K) ¶
1 1 +∞ e f (x, σ, u)
Γ= + Re du .
2 π 0 iu
As in Heston (1993) one can check that

EQ (eiuXT |Xt = x, σt = σ) = exp(C(T − t, u) + D(T − t, u)σ + iux) .

The coefficients C and D are given by

κθ 1 − geds
C(s, u) = i (rus) + {(κ − i(ργu) + d)s − 2 ln[ ]}
γ2 1−g
κ − i(ργu) + d 1 − eds
D(s, u) =
γ2 1 − geds
where
κ − ργu + d p
g= , d = (κ − i(ργu))2 + γ 2 (iu + u2 ) .
κ − i(ργu) − d

12
Appendix 1

Consider the Ornstein-Uhlenbeck process X,

dXt = κ(θ − Xt )dt + σdBt

where {Bt ; t ≥ 0} is a Brownian motion. Apply Ito’s formula to the

function Yt = eκt Xt . This gives
¡ ¢
d eκt Xt = κeκt Xt dt + eκt dXt
= κeκt Xt dt + eκt (κ(θ − Xt )dt + σdBt )
= eκt κθdt + eκt σdBt

since the second partial derivative equals zero. Now integrate to

obtain Z t Z t
κt κu
e Xt − X0 = κθ e du + σ eκu dBu
0 0
which then yields
µ κt Z t ¶
(e − 1)
Xt = e−κt θκ + X0 + σ eκu dBu
κ 0
µ Z t ¶
= θ − e−κt θ − X0 − σ eκu dBu
0

The stochastic integral is a time-changed Brownian motion. Hence

³ ´
−κt
Xt = θ − e θ − X0 − σ B̃ t e2κu du
R
³ 0
´
−κt
= θ−e θ − X0 − σ B̃ e2κt −1
µ 2κ
¶
−κt σ
= θ−e θ − X0 − √ B̃e2κt −1
2κ

Appendix 2

By applying Ito’s lemma, the Q dynamics of ln(σ) is obtained.

d ln(σt ) = (β(a − ln(σt )) − φσ ) dt + γdBt

= β(a∗ − ln(σt ))dt + γdBt

13
where a∗ = a − φβσ . By relying on Appendix 1 the solution of the
equation is known.
µ ¶
γ
ln(σt ) = a∗ − e−βt a∗ − ln(σ0 ) − √ Ze2βt −1
2β
Therefore,
µ ¶
∗ −β(t−1) ∗ γ
ln(σt−1 ) = a − e a − ln(σ0 ) − √ Ze2β(t−1) −1
2β
where {Zt ; t ≥ 0} is a Brownian motion. By combining the above
two equations, we obtain
γ
ln(σt ) − e−β ln(σt−1 ) = a∗ (1 − e−β ) + √ e−βt (Ze2βt −1 − Ze2β(t−1) −1 )
2β
The random variable:
Ze2βt −1 − Ze2β(t−1) −1
is a Gaussian variable N (0, e2βt − e2β(t−1) ) = N (0, e2βt (1 − e−2β )).
Hence:
ln(σt ) = c∗1 + c2 ln(σt−1 ) + εt
2
γ
where c∗1 = a∗ (1 − e−β ), c2 = e−β and var(εt ) = (1 − e−2β ) 2β .

Appendix 3

Method of moments:
By relying on equation (1.1):
∆ ln(St ) = m + σt−1 ut
where var(ut ) = 1. Let us define xt = ∆ ln(St ) − m where E(x2t ) =
E(u2t )E(σt−1
2
). Because the two Brownian motions B̃t and W̃t are
independent
E(x2t ) = E(e2 ln σt−1 ).
Now ln σt−1 is a Gaussian variable:
µ ¶
c1 var(εt )
N ,
1 − c2 1 − c22
14
Therefore:
22
E(x2t ) = e 2 var(ln σt−1 )+2E(ln(σt−1 ))
2
= e2(c1 /(1−c2 ))+2var(εt )/(1−c2 )

Along the same lines:

2
E(x4t ) = e4(c1 /(1−c2 ))+8var(εt )/(1−c2 ) .

ln |xt | follows an ARMA(1,1) process:

ln |xt | = ln σt−1 + ln |ut |

hence:

ln |xt | − c2 ln |xt−1 | = ln σt−1 + ln |ut | − c2 (ln σt−2 − ln |ut−1 |)

= c1 + εt−1 + ln |ut | − c2 ln |ut−1 | = ... = cst + et − θet−1 .

References

Chesney, M. and Scott, L. (1989), “Pricing European currency op-

tion: a comparison of the modified Black-Scholes model and a ran-
dom variance model”, Journal of Financial and Quantitative Anal-
ysis , Vol. 24, pp. 267–285.
Heston, S. (1993), “A closed-form solution for options with stochastic
volatility with applications to bond and currency options”, The
Review of Financial Studies , Vol. 6, pp. 327–343.
Hull, J. and White, A. (1987), “The pricing of options on assets with
stochastic volatilities”, J. of Finance , Vol. 42, pp. 281–300.
Scott, L. (1987), “Option pricing when the variance changes ran-
domly: theory, estimation, and an application”, Journal of Finan-
cial and Quantitative Analysis , Vol. 22, pp. 419–438.

15