Sei sulla pagina 1di 9

Option Pricing Under Stochastic

Volatility with Incomplete Information

Mondher Bellalah
Professeur of finance, THEMA, University of Cergy, 33 boulevard du port,
95011 Cergy-Pontoise, (France). Mail: Mondher.Bellalah@eco.u-cergy.fr and
ESG Group, Paris
Sana Mahfoudh Besbes
Assistant professor in the University of Maine, UFR Exact Sciences, Le Mans
(France), and Phd Student, GREFI, University of Maine, UFR Management
sciences, Avenue OLivier Messiean, 72085 Le Mans (France), and Member
of THEMA. Mail: Sana.Mahfoudh@eco.u-cergy.fr
Abstract
Options are analyzed and valued in the context of Merton’s (1987) “Simple Model of Capital Market Equilibrium with Incomplete Information”. We show now the derivation of
the partial differential equation for options in the presence of shadow qcosts of incomplete information and stochastic volatility. We illustrate our approach by specific appli-
cations and show the dependancy of the option price on information and stochastic volatility. Then, we introduce information costs in a general diffusion model for asset prices
which allows the description of stochastic volatility in an incomplete market. As in Norbert, Platen and Schweizer (1992), we show that the investor’s choice of the minimal
equivalent martingale measure is not changing, but the process of the price of the asset depends on incomplete information.

1 Introduction (1993) proposed a dynamic approach for the volatility which is repre-
sented by an Ornstein-Ulhenbeck. It is difficult to find an analytic solu-
Volatility is an important parameter in option pricing theory. Black and tion for the stochastic volatility option pricing problem.
Scholes (1973)proposed an option valuation equation under the assump- Merton (1987) proposed a capital asset pricing model in the presence
tion of a constant volatility in a complete market without frictions. of the shadow costs of incomplete information. Bellalah (1990) applied
Engle (1982) developed a discrete-time model, to show that the volatility the Merton (1987) model to the valuation of options under incomplete
depends on its previous values. information. Bellalah and Jacquillat (1995) and Bellalah (1999) re derived
The stochastic volatility problem has been examined by several the Black and Scholes (1973) equation in the context of Merton (1987)
authors. For example, Hull and White (1987), Wiggins (1987), Johnson model, they obtained another version of the Black and Scholes equation
and Shanno (1987) studied the general case in which the instantaneous within information uncertainty.
variance of the stock price follows some geometric process. Scott (1989) In this paper, we propose a general context for the pricing of options
and Stein and Stein (1991) used an arithmetic volatility in the study of under stochastic volatility and information costs.
option pricing. All these models describe (with precision) the effects of The first section provides a general concept for the valuation of
the volatility on the options prices. Stein and Stein (1991) and Heston options with shadow costs when the volatility is random. The second

50 Wilmott magazine
TECHNICAL ARTICLE 2

Section examines some applications to several known models. The third All the sources of randomness in the portfolio value resulting from dS
section investigates a general process of a compatible asset with incom- can be eliminated by setting the quantity before dS equal to zero, or
pleteness in the market under information uncertainty and stochastic
∂V ∂V1
volatility. −  − 1 =0
∂S ∂S
and also by setting the quantity before dσ equal to zero, or
2 Valuation of Options In the Presence
∂V ∂V1
of a Stochastic Volatility and Shadow ∂σ
− 1
∂σ
=0

Costs of Incomplete Information After eliminating the stochastic terms, the terms in dt must yield the
deterministic return as in a Black-Scholes “hedge” portfolio. Hence, the
2.1 The valuation model instantaneous return on the portfolio must be the risk-free rate plus infor-
mation costs on each asset in the portfolio as in Bellalah (1999). This gives
The pricing of derivative securities in the presence of a random volatil-
ity needs the use of two processes : one for the underlying asset and ∂V 1 ∂2V ∂2V 1 ∂2V
d = dt + σ 2 S2 2 dt + ρσ Sq dt + q2 2 dt
one for the volatility. Consider the following dynamics for the under- ∂t 2 ∂S ∂S∂σ 2 ∂σ
 
lying asset ∂V1 1 2 2 ∂ 2 V1 ∂ 2 V1 1 2 ∂ 2 V1
− 1 dt + σ S dt + ρσ Sq dt + q dt
∂t 2 ∂S2 ∂S∂σ 2 ∂σ 2
dS = µSdt + σ SdW 1
= [(r + λV )V − (r + λS )S − (r + λV1 )1 V1 ]dt,
and the following process for the volatility
Isolating the terms in V and V1 gives
dσ = p(S, σ, t)dt + q(S, σ, t)dW 2 2 2 2
∂V + 1 σ 2 S2 ∂ V + ρσ Sq ∂ V + 1 q2 ∂ V
∂t 2 ∂S2 ∂S∂σ 2 ∂σ 2
The two processes dW 1 dW 2 are Brownian-motions with a correlation
∂V
coefficient ρ . The functions p(S, σ, t) and q(S, σ, t) are specified in a way +(r + λS )S − (r + λV )V
∂S
that fits the dynamics of the volatility over time. Hence, the derivative ∂V
asset price V(S, σ, t) can be expressed as a function of the dynamics of the ∂σ
underlying asset price S, the volatility σ and time t. Since the volatility is 2 2 2
not a traded asset, a problem arises because this new source of random- ∂V1 + 1 σ 2 S2 ∂ V1 + ρσ Sq ∂ V1 + 1 q2 ∂ V1
∂t 2 ∂S 2 ∂S∂σ 2 ∂σ 2
ness can not be easily hedged away. The pricing of options in this context ∂V1
needs the search for two hedging contracts. The first is the underlying +(r + λS )S − (r + λV1 )V1
= ∂S
asset. The second can be an option that allows a hedge against volatility ∂V1
risk. Following the same logic as in the original Black-Scholes model ∂σ
(1973), consider a portfolio comprising a long position in the option V , a Since the two options differ by their strikes, payoffs and maturities, this
short position of  units of the underlying asset and a short position of implies that both sides of the equation are independent of the contract
−1 units of an other option with value V1 (S, σ, t) : type. Since both sides are functions of the independent variables S, σ and
t, we have
= V − S − 1 V1 (1) ∂V 1 ∂2V ∂2V 1 ∂2V ∂V
+ σ 2 S2 2 + ρσ Sq + q2 2 + (r + λS )S − (r + λV )V
∂t 2 ∂S ∂S∂σ 2 ∂σ ∂S
Over a short interval of time dt , applying Ito’s lemma for the functions
S, σ and t gives the change in the value of this portfolio as: ∂V
= −(p − δq) ,
∂σ
  for a function δ(S, σ, t) referred to as the market price for risk or volatility
∂V 1 ∂2V ∂2V 1 ∂2V
d = + σ 2 S2 2 + ρσ qS + q2 2 dt risk. This equation can also be written as
∂t 2 ∂S ∂S∂σ 2 ∂σ
  ∂V 1 ∂2V ∂2V 1 ∂2V ∂V
∂V1 1 2 2 ∂ 2 V1 ∂ 2 V1 1 2 ∂ 2 V1 + σ 2 S2 2 + ρσ Sq + q2 2 + (r + λS )S
− 1 + σ S + ρσ qS + q dt ∂t 2 ∂S ∂S∂σ 2 ∂σ ∂S
∂t 2 ∂S2 ∂S∂σ 2 ∂σ 2 (2)
   
∂V ∂V1 ∂V ∂V1 ∂V
+ − 1  dS + − 1 dσ + (p − δq) − (r + λV )V = 0
^
∂S ∂S ∂σ ∂σ ∂σ

Wilmott magazine 51
This equation shows two hedge ratios ∂V
∂S
and ∂V
∂σ
. The term (p − δq) is The market price of risk for a traded asset in the presence of information
known as the risk-neutral drift rate. costs
µ − (r + λS )
δS =
σ
2.2 Market price of volatility risk
Substituting δS in (3) gives the following equation
Suppose the investor holds only the option V which is hedged only by the
underlying asset S in the following portfolio ∂V 1 ∂2V ∂V
+ σ 2 S2 2 + (r + λS )S − (r + λV )V = 0
∂t 2 ∂S ∂S
= V − S
This is the Black-Scholes equation in the presence of information costs.
Over a short interval of time dt , the change in the value of this portfolio
can be written as
 
3 Generalization of Certain Model
∂V 1 ∂2V ∂2V 1 ∂2V
d =
∂t
+ σ 2 S2 2 + ρσ qS
2 ∂S ∂S∂σ
+ q2 2 dt
2 ∂σ
with Stochastic Volatility
 
+
∂V
−  dS +
∂V

and Information Costs
∂S ∂σ

In the standard delta-hedging, the coefficient of dS is zero and we have 3.1 Generalization of the Hull and White 1987 model

We consider the following model
∂V 1 ∂2V ∂2V
d − [(r + λV )V − (r + λS )S]dt = + σ 2 S2 2 + ρσ qS dBt = (r + λB )Bt dt
∂t 2 ∂S ∂S∂σ
1 2∂ V 2
∂V dSt = µ(St , σt , t)St dt + σt St dW t1 (4)
+ q + (r + λS )S
2 ∂σ 2 ∂S dνt = γ (σt , t)νt dt + δ(σt , t)νt dW t2

∂V
− (r + λV )V dt + dσ where St denotes the stock price at time t, νt = σt2 its instantaneous vari-
∂σ
∂V ance, and r the riskless interest rate, which is assumed to be constant. W 1
=q (δdt + dW 2 ) and W 2 are Brownian motions under P , they are independent. νt has no
∂σ
systematic risk. This yields a unique option price which can be computed
This results from equations (1) and (2). The term dW 2 represents a unit of as the (conditional) expectation of the discounted terminal payoff under
volatility risk. There are δ units of extra-return, given by dt for each unit a risk-neutral probability measure P̃ . Put differently, P̃ is obtained from P
of volatility risk. by means of a Girsanov transformation such that

2.3 The market price of risk for traded assets dBt = (r + λB )Bt dt

In the Black-Scholes analysis, the hedging portfolio is constructed using dSt = (r + λS )St dt + σt St dW̃t1 (5)
the option and its underlying tradable asset. Consider the construction dνt = γ (σt , t)νt dt + δ(σt , t)νt dW̃t2
of a portfolio as before using two options V and V1 with different charac-
teristics, the initial portfolio value would be under P̃ , where W̃ 1 , W̃ 2 are independent Brownian motions under P̃ . The
risk-neutral dynamics of the bond and the underlying asset are used in
= V −  1 V1 Bellalah (1999). The portfolio value would be

Note that there are none of the underlying asset in this portfolio. Using = V − S −  Bt
the same methodology as before gives the following equation with  units of the bond. When we apply the methodology of the previ-
ous section to this model, equation (2) gives:
∂V 1 ∂2V ∂V
+ σ 2 S2 2 + (µ − δS σ )S − (r + λV )V = 0 (3)
∂t 2 ∂S ∂S ∂V 1 ∂2V ∂2V 1 ∂2V
+ νt S2t 2 + ρσt3 St ξ + ξ 2 νt2
∂t 2 ∂St ∂S∂νt 2 ∂νt2
The variable asset S is the value of a traded asset. Then V = S must be a (6)
∂V ∂V ∂V
solution to this last equation. Substituting V = S in the last equation + (r + λS )St + (γ − δξ )νt + (r + λB )Bt − (r + λV )V = 0
∂St ∂νt ∂Bt
gives
with W̃ 1 , W̃ 2 independent Brownian motion under the probability
(µ − δS σ )S − (r + λS )S = 0 P̃ (ρ = 0)and λS is the information cost of the security St . The investor paid

52 Wilmott magazine
TECHNICAL ARTICLE 2

the shadow cost λS if he does not know the asset. Also λB is the informa- with dW St , dW σt are processes of Wiener, the correlation coefficient
tion cost of the bond Bt and it is equal to zero if the asset is reskless. We between stock returns and volatility movements is ρdt = dW St dW σt and
suppose that δ = 0, The option price is then given by (dP/P)(dSt /St ) = 0 . The instantaneous rate of return on the hedge port-
  folio P is
Bt
V(t, St ) = Ẽ (ST − K)+ |Ft = e−(r+λB )(T−t) Ẽ[(ST − K)+ |Ft ] (7)
BT dP/P = wdV /V + (1 − w)dSt /St
To obtain a more specific form for V , we use the additional assumption with w the fraction invested in the contingent claim V and (1 − w) the
contained in (5) and the independence of W 1 , W 2 that the instanta- fraction invested in the stock S. Equation (2) is equivalent in this case to:
neous variance ν is not influenced by the stock price S. Setting
 T ∂V 1 ∂2V ∂2V 1 ∂2V
1 + σt2 S2t 2 + ρσt2 θ St + θ 2 σt2
ν t,T = νs ds (8) ∂t 2 ∂St ∂St ∂σt 2 ∂σt2
T−t t (13)
∂V ∂V
They show that the conditional distribution of SSTt under P̃, given ν t,T , is log- + (r + λS )St + (f (σt ) − δθ σt ) − (r + λV )V = 0
∂St ∂σt
normal with parameters (r + λB )(T − t) and ν T−t . This allows to write V as
 ∞ As in Wiggins (1987), we can write the following equation
V(t, St , σt2 ) = uBS (t, St , ν t,T )dF(ν t,T |St , σt2 ) (9)
0
∂V 1 ∂2V ∂2V 1 ∂2V ∂V
where VBS denotes the usual Black-Scholes (1973) price corresponding to + σt2 S2t 2 + ρσt2 θ St + θ 2 σt2 2
+ (r + λS )St
∂t 2 ∂St ∂S∂σt 2 ∂σt ∂St
 ∂V (14)
the variance ν t,T and F is the conditional distribution under 
P̃ of ν t,T given St and σt2 . This is equivalent to write : + f (σt ) − (µ − r − λS )ρθ + (.)θ σt (1 − ρ 2 ) − (r + λV )V = 0
∂σt
 ∞
V(t, St , σt2 ) = VBS (t, St , ν t,T )h(ν t,T |St , σt2 )dν t,T (10) We conclude that the market price of risk affects the term given by
0
Wiggins (1987)(.) = (µP − r − λP )/σ P . This term is the expected excess
with return per unit risk, or the market price of risk, for the hedge portfolio. It
VBS (ν) = St N(d1 ) − Xe−(r+λS )(T−t) N(d2 ) represents the return-to-risk tradeoff required by investors for bearing
the volatility risk of the stock.
log(St /K) + 
(r + λS + ν/2)(T − t) √
and d1 = , d2 = d1 − ν(T − t) .
ν(T − t) δσt − (µ − r − λS )ρ
(.) =  (15)
When µ = 0 and as in H and White (1987) we have: σt (1 − ρ 2 )

1 S T − tN  (d1 )(d1 d2 − 1) The market price of risk depends on the information cost of the stock and
V(S, σt2 ) = VBS (ν) +
2 4σ 3 the stochastic volatility.
 4 k 
2σ (e − k − 1)
× − σ 4
k2 3.3 Generalization of Stein and Stein’s model
√ (11)
1 S T − tN  (d1 )[(d1 d2 − 1)(d1 d2 − 3) − (d21 + d22 )] In this model, the stock price dynamics are given by the following
+
6 8σ 5 process:
 3k 
6 e − (9 + 18k)e + (8 + 24k + 18k2 + 6k3 )
k
×σ + ..., dSt = µ(St , σt , t)St dt + σt St dW 1
3k3

avec k = ξ 2 (T − t), N  (x) = √1 e− 2


x2
. The volatility follows an Ornstein-Uhlenbeck process:

dσt =  (σt − θ )dt + kdW 2


3.2 Generalization of Wiggins’s model
Under the assumption of the continuous trading, without frictions, in a The Weiner processes dW 1 , dW 2 are uncorrelated. When equation (2) is
complete market, Wiggins (1987) use the following dynamics for the applied in this context, we have:
asset and the volatility: ∂V 1 ∂2V 1 ∂2V
+ σt2 S2t 2 + k2
∂t 2 ∂St 2 ∂σt2
dSt = µ(St , σt , t)St dt + σt St dW St (16)
(12) ∂V ∂V
+ (r + λSt )St + [− (σt − θ ) − δk] − (r + λV )V = 0
^
dσt = f (σt )dt + θ σt dW σt ∂St ∂σt

Wilmott magazine 53
When δ = 0 or to be a constant, equation (16) has a solution with the equation (19), (20)P1 and P2 must satisfy the following equation:
same form as in Stein and Stein (1991). The solution depends on informa-
tion costs of V and the underlying asset S. The option price has the fol- ∂Pj 1 ∂2V ∂2V 1 ∂ 2 Pj
lowing form: + νt + ρσt νt St + σt2 νt
∂t 2 ∂x ∂St ∂νt 2 ∂νt2
 ∞ (21)
∂Pj √ ∂Pj
V = e−(r+λV ) [St − K]H(St , t | , r + λSt , k, θ )dSt (17) + (r + λS + uj νt ) + (a − bj νt ) =0
S t =K
∂x ∂νt

with H(St , t) is the price distribution of the underlying asset at the time t √
for j = 1, 2 where u1 = 1/2, u2 = −1/2 , a = κθ , b1 = (κ − ρσt ) νt + δσt ,
with a non-zero drift of St . √
b2 = κ νt + δσt
Following the same resolution method in Heston (1993) for the equa-
3.4 Generalization of Heston’s model tion (21), we obtain the solution of the characteristic function:
The underlying asset and the volatility follow the diffusion process:
fj (x, νt , t; φ) = exp[C(T − t; φ) + D(T − t; φ)νt + ixφ] (22)

dSt = µ(St , σt , t)St dt + νt St dW t1 when
√ (18)
dνt = κ(θ − νt )dt + σt νt dW t2
 
a 1 − gedτ
C(τ ; φ) = i(r + λSt )φτ + (bj − iρσ t φ + d)τ − 2 ln
with ρ the correlation coefficient between dW t1 , dW t2 . σt2 1−g
 
In this case, the value of any option V(St , νt , t) must satisfy the follow- bj − iρσt φ + d 1 − edτ
D(τ ; φ) =
ing partial differential equation σt2 1 − gedτ

∂V ∂2V 1 ∂2V 1 ∂2V


+ ρσt νt St + νt S2t 2 + σt2 νt bj − iρσt φ + d
and g = , d = (iρσt φ − bj )2 − σt2 (2iuj φ − φ 2 )
∂t ∂St ∂νt 2 ∂St 2 ∂νt2 bj − iρσt φ − d
(19) By inverting the characteristic functions fj , we obtain the desired
∂V √ ∂V
+ (r + λS )St + [κ(θ − νt ) − δσt νt ] − (r + λV )V = 0 probabilities:
∂St ∂νt
  −iφ ln K ∗ 
1 1 ∞ (e ) fj (x, νt , T; φ)
Under the same assumption as in Heston (1993), it is possible to obtain Pj (x, νt , t; ln K) = + Re dφ (23)
2 π 0 iφ
solution to equation (19). This solution depends on information costs λS .
In fact, an European call with a strike price K and maturing at time T, sat- with fj (x, νt , T; φ) = eiφx .
isfies the equation (19) subject to the following boundary conditions
3.5 Generalization of Johnson and Shanno’s model
V(S, νt , t) = Max(o, S − K)
We consider the following model:
V(0, νt , t) = 0
∂V
(∞, νt , t) = 1 dSt = µSt St dt + σt Sαt dW t1
∂St (24)
dσt = µσt σt dt + σt σσt dW t2
∂V ∂V ∂V
(r + λS )St + κ(θ ) − (r + λV )V + =0
∂St ∂νt ∂t
with dW t1 dW t2 = ρdt .
V(S, ∞, t) = S
When equation (2) is applied to the model, we obtain:
By analogy with the Black et Scholes (1973) formula, Heston (1993) gives a
solution of the form ∂V 1 ∂2V ∂2V 1 ∂2V
+ σt2 S2α
t 2
+ ρσt3 Sαt σσt + σt2 σσ2t
∂t 2 ∂St ∂S∂σt 2 ∂σt2
(25)
V(S, νt , t) = SP1 − KP(t, T)P2 (20) ∂V ∂V
+ (r + λS )St + (µσt − δσσt )σt − (r + λV )V = 0
∂St ∂σt
with P(t, t + τ ) = e−(r+λS )τ the price at time t of a unit discount bond that
matures at time t + τ . The first term of the right side of the solution
Johnson and Shanno (1987), suppose that the risk premium of the volatil-
V(S, νt , t) is the present value of the underlying asset upon optimal exer-
ity is zero. Consequently, we have:
cise. The second term is the present value of the strike-price. Both of
these terms must satisfy the equation. It is convenient to write them in µσ t
δ=
terms of the logarithm, (19)x = ln(S) . By substitution of the solution in σσ t

54 Wilmott magazine
TECHNICAL ARTICLE 2

3.6 A general Markovian model with shadow costs of where (Xst,x )t≤s≤T denotes the solution of (26) starting from x at time t, i.e.,
incomplete information with Xtt,x = x ∈ R m+1 . To illustrate the previous analysis, we give the fol-
lowing example:
In this subsection we present a general model in which we include infor-
mation cost and volatilities stochastic. We consider the following multi-
dBt = (r(t, Xt ) + λB )Bt dt
dimensional diffusion process:
dSt = (r(t, Xt ) + λS )St dt + σt St dW t1

n
dXti = a (t, Xt )dt +
i
b ij j
(t, Xt )Wt (26) dσt = −q(σt − ςt )dt + pσt dW t2 (30)
j=1
1
for i = 1, . . . , m , where dςt = (σt − ςt )dt
α
ai , bij : [0, T] × R m+1 → R with p > 0, q > 0, α > 0 and dW t1 , dW t2 are independent Brownian
motion under the probability Q . The processes of σ, ς are respectively the
are measurable functions. The process W = (W 1 , . . . , W n ) is an n-dimen-
instantaneous and weighted average volatility of the stock. The equation
sional Brownian motion on a probability space (, F , Q ) , and
for σ shows that the instantaneous volatility σt is distributed by some
F = (Ft )0≤t≤T is the Q-augmentation of the filtration generated by W. We
external noise (with an intensity p) and at the same time continuously
assume that the coefficients ai , bij satisfy appropriate growth and Lip-
pulled back toward the average volatility ςt . The parameter q measures
schitz conditions so that the solution of (26) is a Markov process. We also
the strength of this restoring force or speed of adjustment. The equation
remark that under suitable continuity and nondegeneracy conditions on
(2) becomes in this case as follows
the coefficients, F coincides with the natural filtration F X of X. This
model will be interpreted in the following way. The component X0
describes the risk less asset; setting B := X0 , we shall take b0j ≡ 0 for ∂V 1 ∂2V ∂2V 1 ∂2V ∂V
j = 1, . . . , n and a0 (t, x) = (r(t, Xt ) + λB )x0 , so + σt S2t 2 + ρσ 2 pSt + p2 σt2 + (r + λS )St
∂t 2 ∂St ∂S∂σt 2 ∂σt2 ∂St
∂V ∂V 1 ∂V (31)
dBt = (r(t, Xt ) + λB )Bt dt (27) + (−q(σt − ςt ) − δpσt ) + (r + λB )Bt + (σt − ςt )
∂σt ∂Bt α ∂ςt
We assume that − (r + λV )V = 0
 T
|r(s, Xs ) + λX |ds ≤ L < ∞ Q − a.s. (28) Or
0
∂V ∂V ∂σt
for some L > 0 and X = (B, X1 , . . . Xm ) . According to the analysis of =
∂ςt ∂σt ∂ςt
Merton (1987),we assume that λX = λ(s, Xs ) is measured in units of
expected return. We shall work with only one stock. The component X1 The equation (31) becomes
describes its price process and is denoted by S. The other components of X
can then be used to model the additional structure of the market in
which S is embedded. In this general framework, an option or contingent ∂V 1 ∂2V ∂2V 1 ∂2V ∂V
+ σt S2t 2 + ρσ 2 pSt + p2 σt2 + (r + λS )St
claim will be a random variable of the form g(XT ). The classical example ∂t 2 ∂St ∂S∂σt 2 ∂σt2 ∂St
is provided by European call option with strike price K which corre- 1 ∂σt ∂V ∂V (32)
+ (−q(σt − ςt ) − δpσt + (σt − ςt ) ) + (r + λB )Bt
sponds to the claim (ST − K)+ . Since the process X will usually contain α ∂ςt ∂σt ∂Bt
more components than just the bond B = X0 and the stock price S = X1 , − (r + λV )V = 0
claim can depend on many things other than just the terminal stock
price ST . In fact, the only serious restriction is that the underling process This equation can be written as follows:
X (but not necessarily S) should be Markovian. This implies that (subject
to some integrability conditions) we can associate to any contingent ∂V 1 ∂2V ∂2V 1 ∂2V ∂V
+ σt S2t 2 + ρσ 2 pSt + p2 σt2 + (r + λS )St
claim g(XT ) an option pricing function ∂t 2 ∂St ∂S∂σt 2 ∂σt2 ∂St
  
1 ∂σt ∂V ∂V (33)
V : [0, T] × R m+1 → R + − q (σt − ςt ) − δpσt + (r + λB )Bt
α ∂ςt ∂σt ∂Bt
defined by − (r + λV )V = 0
    
T
V(t, x) = EQ exp − r(s, Xs ) + λX ds g(XTt,x ) (29) with λB , λS and λV indicate the information costs respectively for the
^
t bond, the stock and the option.

Wilmott magazine 55
4 The Incomplete Market and the processes in R m+1 the first is (ηt )0≤t≤T the quantity of the riskless asset or
the bond and the second one is ξt = (ξt1 , . . . , ξtm )0≤t≤T the quantity of the
Minimal Equivalent Martingale Measure risky assets hold in such portfolio. the wealth process V must satisfy the
following equation:
with Information Costs
m

We shall work with a model considerably more general than. It contains dVt = ηt dBt + ξti dSit (35)
one risk less asset B and m risky assets (26)Si , i = 1, . . . , m . The bond i=1

price B and the stock prices Si are given by the stochastic differential Substituting equation (34) in equation (35) and setting πti = ξti Sit we have:
equation

m 
n 
i,j j
dBt = (r + λB )Bt dt dVt = πti µit dt + σt dW t

n
i,j j (34)
i=1 j=1
(36)
dSit = µit Sit dt + Sit σt dW t
m

j=1 + (Vt − πti )(rt + λB )dt


i=1
Here, W = (W 1 , . . . , W n )∗ is an n-dimensional Brownian motion on a ⇐⇒
probability space (, F , P) and F = (Ft )0≤t≤T denotes the P-augmentation
m 
n 
i,j j
of the filtration generated by W. We take n ≥ m so that there are at least dVt = πti µit dt + σt dW t
i=1 j=1
as many sources of uncertainty as there are stocks available for trading.
All processes will be defined on [0, T], where the constant T > 0 denotes
m
+ Vt (rt + λV ) − πti (rt + λi )dt
the terminal time for our problem. We assume that the interest rate
i=1
r = (rt )0≤t≤T , the vector µ = (µt )0≤t≤T = (µ1t , . . . , µmt )0≤t≤T of stock appre-
ij
ciation rates, the volatility matrix σ = (σ )
t 0≤t≤T = (σt )0≤t≤T,i=1,...,n,j=1,...,m With no arbitrage, we have:
and the vector (λX )0≤t≤T = (λB , λS1 .., λSm )0≤t≤T of the assets’information
costs , are progressively measurable with respect to F . The interest rate
m
m
n
i,j j
dVt = πti (µit − rt − λi )dt + πti σt dW t + Vt (rt + λV )dt
r and λX satisfies i=1 i=1 j=1
 T
With λV is the vector of information costs (λB , λ1 , . . . , λm ) because the
|ru + λX |du ≤ L < ∞ P − a.s. for some L > 0
0 value of the option is equal to the value of the portfolio in this strategy.
Equation (36) is equal to:
This implies that the bond price process B is bounded above and away
from 0, uniformly in t and ω . We also assume that the matrix σt has full  
rank m for every t so that the matrix (σt σt∗ )−1 is well defined. This means dVt = πt∗ [µt − rt 1 − λS ] + (rt + λV )Vt dt
(37)
that the basic assets, namely the stock prices, have been chosen in such a
way that they are all nonredundant. Consider a “small investor”, i.e. , an + πt∗ σt dW t , 0≤t≤T
economic agent whose actions do not influence prices, who trades in the
The discounted wealth process V  = V/B is then given by
stocks and the bond. His trading strategy can be described at any time t
by his total wealth Vt and by the amounts πti invested in the ith stock for
dVt = πt∗ [µt − rt 1 − λS ]dt + Vt (λV − λB 1) + πt∗ σt dW t (38)
i = 1, . . . , m . The amount invested in the bond is then given by
 m ∗ ⇐⇒
Vt − m i=1 πt . We shall call π = (π t )0≤t≤T = (π t , . . . , πt )0≤t≤T a portfolio
i 1
dVt = πt∗ [µt − rt 1 − λS ]dt + Vt (λS − λB 1) + πt∗ σt dW t (39)
process if is progressively measurable with respect to F and satisfies
π
 T with πt∗ = πt∗ /Bt . Thus, any portfolio, process π uniquely determines a
||σu∗ πu ||2 du < ∞ P − a.s. wealth process V such that π and V together constitute a self-financing
0
strategy. We remark that the process give by the equation (37) depend
and
 on the vector of the shadow costs λS [because that the value of the
T
|πu∗ (µu − ru 1 − λS |du < ∞ P − a.s market price depend on λS ] and on λB . If we interpret the process V as
0 the price of some assets, we can remind the definition of any “general”
where 1 = (1, . . . , 1)∗ ∈ R m and λS = (λS1 , . . . , λSm ) the vector of shadow asset:
costs of the risky assets. The trading strategy is called self-financing if all
changes in the wealth process are entirely due to gains or losses from trad- Definition: A general asset is any asset whose value A is a semi martin-
ing in the stocks and bond. For such a strategy, we denote two predictable gale with respect to P and F .

56 Wilmott magazine
TECHNICAL ARTICLE 2

Remark: For the following study, the general asset A has the following Then the equation (43) becomes
form
γ̃ = ϑ + σ ∗ (σ σ ∗ )−1 [µt − rt 1 − λS ]
dAt = υt∗ dW t + dFt 0≤t≤T (40)
This allows us to prove the following result.
with F un F -adapted process with paths of finite variation and the
process υ = (υ 1 , . . . , υ n )∗ is progressively measurable with respect to F Lemma: Every compatible asset has a value process A of the form
and satisfies:  
 T
dAt = πt∗ [µt − rt 1 − λπ ∗ ] + (rt + λA )At dt
||υu ||2 du < ∞ P − a.s. (45)
0
+ πt∗ σt dW t + νt∗ dW t + νt∗ ϑt dt
We recall also the concept of an equivalent martingale for S:
for some portfolio process π and some process ν, ϑ ∈ K(σ ).
Definition 1: A probability measure P̃ on (, F ) is called equivalent
proof: Let A and his equivalent A a continuous processes. P̃ an
martingale measure for S if
equivalent martingale measure for S such that A is a local P̃ -martingale.
i)- P̃ and P have the same null sets, P̃ ≈ P . In particular, this implies Then A P̃ is a local P-martingale and therefore continuous, since F is a
P̃ = P on F0 . Brownian filtration. We denote by γ̃ the process corresponding to P̃ by
ii)- The discounted price process S = S/B is a vector martingale under P̃ . the relation (42). Under P, A = A/B has the form:
Definition 2: An equivalent martingale measure P̂ for S is called mini- 1 υ∗
dAt = dFt − At [rt + λB ]dt + t dW t
mal if any local P -martingale, orthogonal to Si , i = 1, . . . , m 0remains a Bt Bt
local martingale under P̂ .
where we have used equation. If we decompose (39)υ ∈ L2a [0, T] as:
We begin by describing more precisely the equivalent martingale
measures for S. If P̃ is any equivalent martingale measure for S and υ = ν + σt∗ π with ν ∈ K(σ )
   
dP̃  dP̃ 
Z̃t = EP Ft = , 0≤t≤T (4.5) then applying Girsanov’s theorem to W shows that A can be written
dP  dP F t
under P̃ as
denotes a continuous version of the density process of P̃ with respect to P ,  
1 υ∗ υ∗
then Z̃ can be written as dAt = dFt − At [rt + λB ] + t γ̃t dt + t dW̃t
Bt Bt Bt
  t  
1 t for some P̃ -Brownian motion W̃t .
Z̃t = exp − γ̃u∗ dW u − ||γ̃u ||2 du , 0≤t≤T (41)
0 2 0 Since A is a continuous local P̃ -martingale, by substitution of
γ̃ = γ̂ + ϑ and using equation (44) in the previous equation we obtain:
where γ̃ = (γ̃ 1 , . . . , γ̃ n )∗ is adapted to F and satisfies the following  
condition dFt = At [rt + λA ] + υt∗ γ̃t dt
 T
||γ̃u ||2 du < ∞ P − a.s. (42)  
0
= At [rt + λAt ] + (νt∗ + πt∗ σt )(γ̂t + ϑt ) dt
and
 
σt γ̃t = [µt − rt 1 − λS ], 0≤t≤T (43) = At [rt + λAt ] + νt∗ γ̂t + νt∗ ϑt + πt∗ [µt − rt 1 − λπ ∗ ] dt
if we suppose the existence and the uniqueness of the minimal equiva-
lent martingale measure1 P̂ , we have with F un F -adapted process with paths of finite variation.
This equation confirm that the process of a general asset defined in
γ̂t = σt∗ (σt σt∗ )−1 [µt − rt 1 − λS ], 0≤t≤T (44) equation is equivalent to the process defined in equation (45).

If γ̃ ∈ L2a [0, T] satisfies (44), then γ̃ can be written as2


5 Conclusion
γ̃ = γ̂ + ϑ for some ϑ ∈ K(σ )
This paper developes a general context for the valuation of options with
Indeed, decomposing γ̃ as γ̃ = ϑ + σ ∗ π with ϑ ∈ K(σ )yields by (43) stochastic volatility and information costs. The shadow costs are integrat-
ed in the investor’s portfolio wealth process in the same vein as in
^
[µt − (rt + λB )1] = σ γ̃ = σ σ ∗ π Merton (1987), Bellalah and Jacquillat (1995) and Bellalah (1999). The

Wilmott magazine 57
TECHNICAL ARTICLE 2

information costs appear naturally in the derivation proposed in this


analysis. There is also another reformulation of the compatible asset’s
process, that gives more information for the drift term, which depends
on information costs. In the same way, several extensions of existing
models can be used for the development of the option valuation with sto-
chastic volatility and information costs.

FOOTNOTES & REFERENCES


1
see Norbert, Platen and Schweizer[1992] (page 162, 163)
2
see Norbert, Platen and Schweizer[1992] (page 165)
■ Bellalah M : “Quatre Essais Sur Lévaluation des Options: Dividendes, Volatilités des
Taux d’intérêt et Information Incompléte .”, Doctorat de l’université de Paris-Dauphine,
(June 1990).
■ Bellalah M, Jacquillat B : ”Option Valuation with Information Costs :Theory and Tests.”,
The Financial Review , Vol 30, N 3, (August 1995), 617–635.
■ Bellalah M : ”The Valuation of Futures and Commodity Options with Information
Costs.“, Journal of Futures Markets (September 1999).
■ Black F, Scholes M : ”The Pricing of Options and Corporate Liabilities.”, Journal of
Political Economy, 81 (May-June 1973), 637–659.
■ Cox J C, Ross S A : ”A survey of Some New Results In financial Option Pricing Theory.”,
Journal of Finance, Vol 31, N 2 (May 1976), 383–402.
■ Engle R : ”Autoregressive Conditional Hetero-elasticity with Estimates of the Variance
of U.K.Inflation .”, Econometrica 50, (1982), 987–1008.
■ Heston S : ”A closed-Form Solution for Options with Stochastic Volatility with
Applications to Bond and Currency Options.”, The Review of Financial Studies, Vol 6
(1993), 327–343.
■ Hofmann N, Platen E, Schweizer M : ”Option Pricing Under Incompleteness”,
Mathematical Finance, Vol 2, N 3, (July 1992), 153–187.
■ Hull J, White A : ”The Pricing of Options on Assets with Stochastic Volatilities.”,
Journal of Finance,(June 1987), 281–320.
■ Hull J, White A : ”An Analysis of The Bias in Option Pricing Caused by a Stochastic
Volatility.”, Advances in Futures and Options Research, Vol 3, (1988), 29–61.
■ Johnson H, Shanno D : ”Option Pricing When The Variance is Changing.”, Journal of
Financial Quant. Anal, Vol 22, (1987), 143–151.
■ Karatzas I : ”Optimization Problems In The Theory of Continuous Trading.”, SIAM.
Journal. Control Optima; Vol 27, (November 1989), 1221–1259.
■ Merton R C : ”A Simple Model of Capital Market Equilibrium with Incomplete
Information.”, Journal of Finance, N 3 (July 1987), 483–509.
■ Scott L : ”Option Pricing when The Variance Changes Randomly: Theory, Estimation
and an Application.”, Journal of Financial and Quantitative Analysis, Vol 22 (December
1987), 419–437.
■ Stein E M, Stein J C : ”Stock Price Distributions with Stochastic Volatility: An Analytic
Approach.”, The Review of Financial Studies, Vol 4 (1991), 727–752.
■ Wiggins J B :”Option Values Under Stochastic Volatility.” The Journal of Financial
Economics, Vol 19 (1987), 351–372.

58 Wilmott magazine