Chapter 3

General Models of Stochastic Volatilities

3.1 The Governing PDE in a Complete Market
Consider the following price dynamics of an asset under P, the physical measure,
dS
t
=
t
S
t
dt +
t
S
t
d

Z
1
t
,
d
t
= (S
t
,
t
, t)dt +(S
t
,
t
, t)d

Z
2
t
,
with
d

Z
1
t
d

Z
2
t
=
t
dt.
Here, both and are deterministic functions. We call the volatility of volatility. When
0, the model reduces to the Black-Scholes model. For simplicity we assume deter-
ministic interest rates and zero dividend yield.
In this model, there are two sources of risks: asset price and its volatility. For arbitrage
pricing, we need another security, in addition to the underlying, to complete the market.
Suppose there is another asset, with price F
1
, whose value depends on both the underlying
asset price and its volatility. Let F denote the price of a generic derivative to be priced.
For pricing purpose, we form a portfolio consisting the generic derivative, the underlying,
and the 2
nd
asset:
II = F S
1
F
1
,
where and
1
are yet to be determined. The change in this portfolio over [t, t + dt] is
given by
dII =
_
F
t
+
1
2

2
S
2

2
F
S
2
+S

2
F
S
+
1
2

2
F

2
_
dt

1
_
F
1
t
+
1
2

2
S
2

2
F
1
S
2
+S

2
F
1
S
+
1
2

2
F
1

2
_
dt
+
_
F
S

1
F
1
S

_
dS +
_
F

1
F
1

_
d
1
2 CHAPTER 3. GENERAL MODELS OF STOCHASTIC VOLATILITIES
To make the portfolio riskless, we must eliminate the stochastic terms by choosing and

1
according to the following equations:
_

_
F
S

1
F
1
S
= 0,
F

1
F
1

= 0,
which yield

1
=
F

_
F
1

,
=
F
S

1
F
1
S
.
Note that the resulted can be quite different from the Black-Scholes delta. With those
choice of deltas, we eliminate risks and create a portfolio which, by arbitrage arguments,
should earn a return equal to the risk-free rate:
dII =
_
F
t
+
1
2

2
S
2

2
F
S
2
+S

2
F
S
+
1
2

2
F

2
_
dt

1
_
F
1
t
+
1
2

2
S
2

2
F
1
S
2
+S

2
F
1
S
+
1
2

2
F
1

2
_
dt
=r
t
IIdt
=r
t
(F S
1
F
1
)dt
=r
t
_
F
F
S
S
1
_
F
1

F
1
S
S
__
dt.
Collecting all F terms on one side of the equation and all F
1
terms on the other side, we
obtain
F
t
+

LF +r
t
S
F
S
r
t
F
F /
=
F
1
t
+

LF
1
+r
t
S
F
1
S
r
t
F
1
F
1
/
where

L =
1
2

2
S
2

2
S
2
+S

2
S
+
1
2

2

2

2
.
The LHS of the equation depends on F only, while the RHS depends on F
1
only, this is
possible only if both sides equal to a function, (S, , t) thus leading us to
F
t
+

LF +r
t
S
F
S
r
t
F = (S, , t)
F

. (3.1.1)
Function is actually the market price of the volatility risk, for its specication we need
additional information. For later use we denote
L =

L +r
t
S

S
+

r
t
I.
3.2. THE MARKET PRICE OF VOLATILITY RISK 3
3.2 The Market Price of Volatility Risk
We now rewrite the asset price process under the objective measure as
dS
t
= S
t
(
t
dt +
t
d

Z
1
t
),
d
t
= (S
t
,
t
, t)dt +(S
t
,
t
, t)(d

Z
1
t
+
_
1
2
d

Z
2
t
),
where
d

Z
1
t
d

Z
2
t
= 0.
Let
S
and

be the market prices of risks for the Brownian motion

Z
1
t
and

Z
2
t
, respectively,
with

S
=

t
r
t

,
and

yet to be determined, then the Radon-Nikodym derivative of the risk-neutral mea-

sure w.r.t. the objective measure is
dQ
dP

Ft
= exp
__
t
0

1
2
(
2
S
+
2

)du
S
d

Z
1
u

Z
2
u
_
.
The P-Brownian motion (

Z
1
t
,

Z
2
t
) relates to Q-Brownian motion (Z
1
t
,

Z
2
t
) through
dZ
1
t
=d

Z
1
t
+
S
dt,
d

Z
2
t
=d

Z
2
t
+

dt.
In terms of Z
1
t
and

Z
2
t
, we write the asset price dynamics under the risk-neutral measure
Q as
dS
t
=S
t
(
t
dt +
t
(dZ
1
t

S
dt))
=S
t
[(r
t
q
t
)dt +
t
dZ
1
t
],
d
t
=(S
t
,
t
, t)dt +(S
t
,
t
, t)[(dZ
1
t

S
dt) +
_
1
2
(d

Z
2
t

dt)]
=(

)dt +[dZ
1
t
+
_
1
2
d

Z
2
t
],
where

=
S
+
_
1
2

.
We call
S
and

the market prices of asset price risk and market price of volatility risk,
respectively. Next, we will relate to these market prices of risks.
Consider the partially delta-hedged portfolio:
II
1
= F
F
S
S
4 CHAPTER 3. GENERAL MODELS OF STOCHASTIC VOLATILITIES
By Itos lemma
dII
1
=
_
F
t
+

LF
_
dt +
F

d
The excess of return of the hedged portfolio is
dII
1
rII
1
dt
=
_
F
t
+

LF +rS
F
S
rF
_
dt +
F

d
=(S
t
,
t
, t)
F

dt +
F

[(

)dt +(dZ
1
t
+
_
1
2
d

Z
2
t
)].
Under the pricing measure, the excess of return must be a martingale, which dictates that
(S
t
,
t
, t) =

,
or

= ( )/.
The interpretation are
1.

is the market price per unit of volatility risk,

2.

is the risk premium for volatility risk, and

3. is the drift term of the volatility process under the risk-neutral pricing measure.
3.3 Interpreting the PDE using Sharpe Ratios
Under the objective measure P, the price dynamics of the derivative is
dF = F(
F
dt +
F
S
d

Z
1
t
+
F
d

Z
2
t
)
where

F
S
=
S
F
F
S

F
=
1
F
F

and, from (3.1.1),

F
=
1
F
_
F
t
+

LF +S
F
S
+
F

_
=(S
t
,
t
, t)

+ ( r
t
)

Fs

+r
t
,
3.4. UTILITY-BASED EQUILIBRIUM MODEL 5
or

F
r
t
=
F
S
r
t

+
F

.
The volatility of F is

F
=
_

2
F
S
+
2
F
+ 2
F
S

F
,
so the Sharpe ratio of the derivative is

F
r
t

F
=

F
S

F
r
t

+

F

=

F
S

S
+
F

For the derivative, if we call

Fs
percentage volatility w.r.t. equity,

F
percentage volatility w.r.t. volatility,

F
total percentage volatility,
then we can say that the Sharpe ratio of the option is the weight average of the Sharpe
ratios of the asset and the volatility.
One can verify that
(
F
r
t
)dt =
S
_
dF
F
, d

W
t
_

Z=0
+

_
dF
F
, d

Z
t
_

d

Wt=0
,
saying that the excess of return equals to the risk exposures time to the market prices of
risks. In reality,

is not known, and is determined by calibration.

3.4 Utility-based Equilibrium Model
Our investment universe consists of only these types of securities
(i) market index shares (the stock) with price S
t
and instantaneous variance V
t
,
(ii) general index option with price F(S
t
, V
t
, t),
(iii) money market investment (MMA), with price B
t
= e
rt
, under constant interest rate
r
t
= r.
The equity price process is given by
_

_
dS
t
=
t
S
t
dt +
t
S
t
dZ
1
t
,
t
=
_
V
t
,
dV
t
=(V
t
)dt +(V
t
)dZ
2
t
,
dB
t
=rB
t
dt,
6 CHAPTER 3. GENERAL MODELS OF STOCHASTIC VOLATILITIES
and
dZ
1
t
dZ
2
t
= (V
t
)dt.
By Itos lemma, the option price F(S
t
, V
t
, t) evolves according to
dF
t
=
F
t
F
t
dt +
t
F
t
dZ
1
t
+
t
F
t
dZ
2
t
.
The representative agent decides at each point in time (i) how to invest societys wealth W
t
and (ii) at what rate to consume the wealth, simultaneously. The investment allocation can
be represented by a 3-Dvector, x
t
= (x
t
, y
t
, z
t
) for stock, option and MMAwith x
t
+y
t
+z
t
=
1. With the allocation x
t
, wealth evolves according to
dW
t
= x
t
W
t
dS
t
/S
t
+y
t
W
t
dF
t
/F
t
+ (z
t
W
t
r c)dt
=x
t
W
t
(
t
dt +
t
dZ
1
t
) +y
t
W
t
(
F
t
dt +
t
dZ
1
t
+
t
dZ
2
t
) + [(1 x
t
y
t
)W
t
r c]dt
=[x
t
W
t
(
t
r
t
) +y
t
W
t
(
F
t
r
t
) + (rW
t
c)]dt +x
t
W
t

t
dZ
1
t
+y
t
W
t
(
t
dZ
1
t
+
t
dZ
2
t
).
The representative agent determine the optimal value {x

t
, c

t
} at each point in time by
maximizing an expected utility function
J(W
t
, V
t
, t) max
{xt,ct}
E
t
__
T
t
U(c
s
, s)ds +B(W
T
, T)
_
. (3.4.2)
Here,
T planning horizon,
U(c
t
, t) consumption utility,
B(W
T
, T) utility function for all consumption occurs after T, also called the
bequest function.
The solution of (3.1) needs the following
Assumption (Market cleaning): The market clears so that in equilibrium, the optimal
wealth allocation is (1,0,0).
Two special cases are simple but interesting
(I) The innite horizon consumption-investment problem
J(W
t
, V
t
, t) = max
{xt,ct}
E
t
__

t
U(c
s
, s)ds
_
.
(II) The pure investment problem (consumption at last):
J(W
t
, V
t
, t) = max
{xt,ct}
E
t
[B(W
T
, T)] .
3.5. HAMILTON-JACOBI-BELLMAN EQUATION 7
Power utility: it sufces to consider the power utility, corresponding to constant pro-
portional risk aversion (CPRA) with
U(c
t
, t) = e
Rt
c

, B(W
T
, T) = e
RT
W

,
where is the CPRA parameter and R 0 is the impatience parameter, both constants.
For any > 0, U(c, t) is an increasing function of c, here, more is preferred to less. With <
1, U has a negative second derivative, U
cc
< 0. This means that the representative is risk-
averse, and prefer a certain outcome to an uncertain outcome with the same expectation.
Clearly, B(W
T
, T) has the same properties.
If we replace c

by c

1 in U and then take the limit 0, we will have U(c

t
, t) =
e
Rt
ln c
t
, the log utility function.
When = 1, we have the case of risk neutrality. As we shall see, the corresponding
volatility risk premium is zero.
We assume that the representative agent is either risk averse or risk neutral, i.e., 0
1
3.5 Hamilton-Jacobi-Bellman Equation
The problem of utility maximization can be reduced to solving a dynamical programming
problem of PDE (Merton, 1990). The solution is known to satisfy the following HJB equa-
tion:
0 = max
{xt,ct}
_
U(c
t
, t) +
J
t
+L(J)
_
(3.5.3)
where L(J) is the differential generator of J, dened by
L(J)dt =E
t
_
J
W
dW
t
+
1
2
J
WW
(dW
t
)
2
+J
V
dV
t
+
1
2
J
V V
(dV
t
)
2
+J
WV
dW
t
dV
t
_
=
__
(
t
r)x
t
W
t
+ (
F
t
r)y
t
W
t
+ (rW
t
c
t
)

J
W
+
1
2
[x
2
t
V
t
+y
2
t
(
2
t
+
2
t
+ 2
t

t
) + 2xy(
t
+
t
)]W
2
J
WW
+ J
V
+
1
2

2
J
V V
+ [x
t

t
+y
t

t
(
t
+
t
)]WJ
WV
_
dt.
Differentiate (3.2) w.r.t. c
t
, x
t
and y
t
, set x = 1 and y = 0, and nally make use of J
c
= J
x
=
J
y
= 0, we obtain
0 = U
c
(c

t
, t) J
W
,
0 = J
W
(
t
r)W
t
+J
WW
x
t
W
2
t
V
t
+J
WV
W
t

t
,
0 = J
W
(
F
t
r)W
t
+J
WW
W
2
t
(
2
t
+
2
t
+ 2
t

t
) +J
WV
W
t
(
t
+
2
t
).
8 CHAPTER 3. GENERAL MODELS OF STOCHASTIC VOLATILITIES
Rearrange using
t
= SF
S
/F,
t
= F
V
/F, we obtain
U
c
(c

t
, t) = J
W
, the envelope condition,

t
r =
WJ
WW
J
W
V
t

J
WV
J
W

t
,

F
t
r = (
t
r)
S
F
F
S
+ ( )
1
F
F
V
,
where
=
WJ
WW
J
W

J
WV
J
W

2
t
,
U
c
(c, t) =e
Rt
c
1
.
Substituting c

and {x
t
, y
t
} = {1, 0} back to (3.2) and making use of the necessary condi-
tions, we obtain
0 = U(c

t
, t) +
J
t
+ (rW
t
c

t
)J
W

1
2
V
t
W
2
t
J
WW
+J
V
+
1
2

2
J
V V
. (3.5.4)
Because of the choice of the power utility, there is a solution of the form
J(W, V, t) = e
Rt
g(V, t)
W

.
We call g(V, t) the risk premium coefcient. Then
(V, t)
J
WV
J
W
=
g
V
(V, t)
g(V, t)
.
In terms of the new notation, the expected excess of returns on the market are

t
r = (1 )V
t
(V
t
, t),

F
t
r = ( r)
S
F
F
s
+ ( )
1
F
F
V
,
where
= (1 )
_
V
t

2
(V
t
, t)
Substituting
c

t
= g
1
1
W
t
into (3.3), we obtain an equation for g:
_
_
_
g
t
= (1 )g
/
1
+ [(R r)
1
2
(1 )V ]g
g
V

1
2

2

2
g
V
2
,
g(V, T) = 1.
3.6. THE EQUILIBRIUM OPTION VALUATION 9
3.6 The Equilibrium Option Valuation
According to Itos lemma,

F
t
F =
F
t
+

AF,
where

AF =
1
2

2
S
2
F
SS
+SF
SV
+
1
2

2
F
V V
+SF
S
+F
V
.
Combining with the equation for Sharpe ratios, we have
F
t
+

AF rF =(
F
t
r)F
=(
t
r)SF
S
+ ( )F
V
.
After cancelation, we end up with
F
t
+AF rF = 0, (3.6.5)
where
AF =
1
2

2
S
2
F
SS
+SF
SV
+
1
2

2
F
V V
+ (r q)SF
S
+F
V
.
Equation (3.5) implies that under the pricing measure, the asset price process is
_
dS
t
= S
t
[(r q)dt +
t
dZ
1
t
],
dV
t
= (V
t
, t)dt +dZ
2
t
.
To specify (V, t) we must
solve for g(V, t),
compute
t
= g
V
/g,
obtain (V, t) = [(1 )

V
t

2
(V, t)].
Once we have obtain (V, t) will also obtain the market prices of risks:

S
=
t
r/
t
= (1 )
t
(V
t
, t),

=
(V
t
, t)

= (1 ) (V
t
, t).
Next, we demonstrate the calculation of (V, t) in two special cases.
Example 1: Log-utility.
If = 0, (3.4) becomes
_
_
_
g
t
= 1 +Rg
g
V

1
2

2

2
g
V
2
,
g(V, T) = 1.
(3.6.6)
10 CHAPTER 3. GENERAL MODELS OF STOCHASTIC VOLATILITIES
Equation (3.6.6) has a solution independent of V :
g(V, t) = g(t) =
1
R
+ (1
1
R
) exp[R(T t)].
It follows that
= 0
and

S
=
_
V
t
,

=
_
V
t
.
Example 2: Pure investment problem with the square root volatility process.
The risk premium process satises
_
_
_
g
t
= [(R r)
1
2
(1 )V ]g
g
V

1
2

2

2
g
V
2
,
g(V, T) = 1.
Consider, in particular, the square-root process:
dV
t
=( V )dt +
_
V
t
dZ
2
t
,
dZ
1
t
dZ
2
t
=dt.
The risk premium coefcient then satises
_
_
_
g
t
= [(R r)
1
2
(1 )V ]g ( V )
g
V

1
2

2
V

2
g
V
2
,
g(V, T) = 1.
Let = T t. The solution is of the form
g(V, ) = exp{a
1
() +a
2
()V },
with
a
1
() =
_
1
2
(

+d) R +r
_

2

2
ln
_
1 K exp
_
1
2

2
d
_
1 K
_
,
a
2
() =
1
2
(

+d)
_
1 exp (
1
2

2
d)
1 K exp
_
1
2

2
d
_
_
,
where

=
2

2
, c =
1

2
( 1), d =
_

2
+ 4 c, K =

+d

d
.
3.7. PRICES VIA EXPECTATION 11
It follows that
(V, t) =
g
V
(V, t)
g(V, t)
= a
2
(T t),
which is independent of V again. It follows that

S
= [1 a
2
(T t)]
_
V
t
,

= [(1 ) a
2
(T t)]
_
V
t
.
The drift for the risk-neutral volatility process is
= V +

= V +

V [(1 ) a
2
(T t)]
_
V
t
= [ + ((1 ) a
2
(T t))]V
t
=

V
t
.
For the consumption-investment problem with general CPRA parameter, (0, 1), it is
not clear whether there is still

_
V
t
.
3.7 Prices via Expectation
Consider the pricing of a call option under the risk-neutral measure
C = e
rT
E
Q
_
(S
T
K)
+

.
The risk-neutral dynamics is
dS
t
=S
t
_
(r
t
q
t
)dt +
t
dZ
1
t
_
,
d
t
=
t
dt +
t
dZ
2
t
,
with dZ
1
t
dZ
2
t
= dt. Let
Z
1
t
=
_
1
2
Z
1
t
+Z
2
t
,
_

Z
1
t
, Z
2
t
_
= 0.

2
T
=
1
T
_
T
0

2
t
dt =

V .
Suppose = 0, then
C = e
rT
E
Q
_
E
Q
[ (S
T
k)
+

T
]

= E
Q
[C
BS
(
T
)].
12 CHAPTER 3. GENERAL MODELS OF STOCHASTIC VOLATILITIES
Williad (1996) extends this formula to the care of = 0 by writing
S
T
= S
0
exp
__
T
0
(r
t
q
t

1
2

2
t
)dt +
t
(
_
1
2
d

Z
1
t
) +dZ
2
t
_
= S
0

T
exp
__
T
0
_
r
t
q
t

1
2
(
_
1
2

t
)
2
_
dt +
t
_
1
2
d

Z
1
t
_
,
with

T
:= exp
_

1
2
_
T
0

2
t
dt +
t
dZ
2
t
_
.
Conditional on
T
, there is
log S
T
N
_
log (S
0

T
)
1
2
(1
2
)
2
T, (1
2
)
2
T
_
.
It follows that
C = E
Q
_
C
BS
(S
0

T
,
_
1
2
)
_
.
If we have the density function of or

V , the option can be calculated via numerical
integration.