MAFS 5030 - Quantitative Modeling of Derivative Securities

Topic 4 – Extended option models

4.1 Continuous dividend yield models: Time dependent parameters

4.2 Exchange options: Change of numeraire

4.3 Quanto options – equity options with exchange rate risk expo-
sure

4.4 Implied volatility and volatility smiles

4.5 Volatility exposure generated by delta hedging options

4.6 VIX and timer options

4.7 Guaranteed minimum withdrawal benefit

4.8 Transaction costs models

1
4.1 Continuous dividend yield models

Let q denote the constant continuous dividend yield, that is, the
holder receives dividend of amount equal to qS dt within the in-
finitesimal interval (t, t + dt). The asset price dynamics is assumed
to follow the Geometric Brownian motion
dS
= ρ dt + σ dZ P .
S

We form a riskless hedging portfolio by short selling one unit of the

European call and long holding △ units of the underlying asset. The
differential change of the portfolio value Π is given by

−dc + △ dS + q△S dt )
dΠ = (
( )
∂c σ 2 2 ∂ 2c ∂c
= − − S + q△S dt + △ − dS.
∂t 2 ∂S 2 ∂S

2
The last term q△S dt is the wealth added to the portfolio due to
∂c
the dividend payment received. By choosing △ = , we obtain
∂S
a riskless hedge for the portfolio. By no-arbitrage argument, the
hedged portfolio should earn the riskless interest rate.

We then have
( ) ( )
∂c σ2 ∂ 2c
∂c ∂c
dΠ = − − S 2 2 + qS dt = r −c + S dt,
∂t 2 ∂S ∂S ∂S
which leads to
∂c σ 2 2 ∂ 2c ∂c
= S 2
+ (r − q)S − rc, τ = T − t, 0 < S < ∞, τ > 0.
∂τ 2 ∂S ∂S

3
Martingale pricing approach

Suppose all the dividend yields received are continuously used to

purchase additional units of asset, then the wealth process of holding
one unit of the underlying asset initially is given by

Sbt = eqtSt,
where eqt represents the growth factor in the number of units. The
wealth process Sbt follows
dSbt P.
= (ρ + q) dt + σ dZ t
Sbt
We would like to find the equivalent risk neutral measure Q under
which the discounted wealth process Sbt∗ is Q-martingale. We choose
γ(t) in the Radon-Nikodym derivative to be
ρ+q−r
γ(t) = .
σ

4
 Q
Now Zt is a Q-Brownian motion and

Q P ρ+q−r
dZt = dZt + dt.
σ
Also, Sbt∗ becomes Q-martingale since

dSbt∗ Q
b ∗
= σ dZt .
St
The asset price St under the equivalent risk neutral measure Q be-
comes
dSt Q
= (r − q) dt + σ dZt .
St
Hence, the risk neutral drift rate of St is r − q.

Analogy with the foreign currency options

The continuous yield model is also applicable to options on foreign

currencies where the continuous dividend yield can be considered as
the yield due to the interest earned by the foreign currency at the
foreign interest rate rf .
5
Call and put price formulas

The price of a European call option on a continuous dividend paying

asset can be obtained by changing S to Se−qτ in the price formula
since the asset price S will be depleted at the rate q due to payment
of dividend yield. Suppose we let Se = Se−qτ , then the option pricing
equation becomes
∂c σ 2 e2 ∂ 2c ∂c
= S e + rSe e − rc.
∂τ 2 ∂S2 ∂S

The European call price formula with continuous dividend yield q is

c(S, τ ) = Se−qτ N (dc1) − Xe−rτ N (dc2),

where
S σ 2
c
ln X + (r − q + 2 )τ √
d1 = √ , dc2 = dc1 − σ τ.
σ τ

6
Alternatively, knowing that the expected rate of return of St under
Q is δS = r − q, we can deduce that

c(S, τ ) = e−rτ [Se(r−q)τ N (dc1) − XN (dc2)].

More explicitly, we have
{ [ ] [ ]}
c(S, τ ) = e−rτ t
EQ ST 1
{ST >X} − XEQ
t 1{ST >X}
so that
[ ]
t
E Q ST 1{ST >X} = SeδS τ N (dc1)
[ ]
EQt 1{ST >X} = N (dc2).
Similarly, the European put formula with continuous dividend yield
q can be deduced from the Black-Scholes put price formula to be

p = Xe−rτ N (−dc2) − Se−qτ N (−dc1).

In a similar manner, we deduce that
[ ]
t
EQ ST 1{ST <X} = SeδS τ N (−dc1)
[ ]
EQt 1{ST <X} = N (−dc2).
7
Put-call parity and put-call symmetry

The new put and call prices satisfy the put-call parity relation

p = c − Se−qτ + Xe−rτ .

Furthermore, the following put-call symmetry relation can also be

deduced from the above call and put price formulas

c(S, τ ; X, r, q) = p(X, τ ; S, q, r).

That is, the put price formula can be obtained from the correspond-
ing call price formula by interchanging S with X and r with q in the
formula.

8
 • Recall that a call option entitles its holder the right to exchange
the riskless asset for the risky asset, and vice versa for a put
option. The dividend yield earned from the risky asset is q while
that from the riskless asset is r.
• If we interchange the roles of the riskless asset and risky asset
in a call option, the call becomes a put option, thus giving the
justification for the put-call symmetry relation.

As a verification, consider
 ( ) 
ln X + q−r− σ2 τ
 S 2
p(X, τ ; S, q, r) = Se−qτ N 
 − √ 

σ τ
 ( ) 
ln X + q−r+ σ2 τ
−rτ  S 2
− Xe 
N − √ 

σ τ

= Se−qτ N (db1) − Xe−rτ N (db2)

= c(S, τ ; X, r, q).

9
Time dependent parameters

Suppose the model parameters become time dependent functions,

the Black-Scholes equation has to be modified as follows
∂V σ 2(τ ) 2 ∂ 2V ∂V
= S 2
+[r(τ )−q(τ )] S −r(τ )V, 0 < S < ∞, τ > 0,
∂τ 2 ∂S ∂S
where V is the price of the derivative security.

When
∫τ we apply the following transformations: y = ln S and w =
e 0 r(u) duV , then
[ ]
∂w σ 2(τ ) ∂ 2w σ 2(τ ) ∂w
= + r(τ ) − q(τ ) − .
∂τ 2 ∂y 2 2 ∂y
Consider the following form of the fundamental solution
( )
1 [y + e(τ )]2
f (y, τ ) = √ exp − ,
2πs(τ ) 2s(τ )

which satisfies the initial condition: f (y, 0+) = δ(y).

10
By direct differentiation, it can be shown that f (y, τ ) satisfies the
parabolic equation
∂f 1 ′ ∂ 2f ∂f
= s (τ ) 2 + e′(τ ) .
∂τ 2 ∂y ∂y
Suppose we let
∫ τ
s(τ ) = σ 2(u) du
∫0τ
s(τ )
e(τ ) = [r(u) − q(u)] du −
,
0 2
then f satisfies the same differential equation as that for w(y, τ ).
One can deduce that the fundamental solution is given by
 
∫τ 2
σ (u)
1  {y + 0 [r(u) − q(u) − ] du}2 
ϕ(y, τ ) = √ ∫ exp − ∫τ
2
2
.
τ
2π 0 σ 2(u) du 2 0 σ (u) du

Given the initial condition w(y, 0), the solution can be expressed as
∫ ∞
w(y, τ ) = w(ξ, 0) ϕ(y − ξ, τ ) dξ.
−∞

11
Note that the time dependency of the coefficients r(τ ), q(τ ) and
σ 2(τ ) will not affect the spatial integration with respect to ξ. We
may simply make the following substitutions in the option price
formulas
∫
1 τ
r is replaced by r(u) du
τ ∫0
1 τ
q is replaced by q(u) du
τ 0∫
2 1 τ 2
σ is replaced by σ (u) du.
τ 0

For example, the European call price formula is modified as follows:

∫τ ∫τ
c = Se− 0 q(u) du
N (df1) − Xe− 0 r(u) du
N (df2)
where
S ∫τ σ 2(u) √∫
ln + 0 [r(u) − q(u) + 2 ] du τ
df1 = X √∫ , df2 = df1 − σ 2(u) du.
τ 2 0
0 σ (u) du

12
4.2 Exchange options

• An exchange option is an option that gives the holder the right

but not the obligation to exchange one risky asset for another.

• Let Xt and Yt be the price processes of the two risky assets.

• The terminal payoff of a European exchange option at maturity

T of exchanging YT for XT is given by max(XT − YT , 0).

Under the risk neutral measure Q, let Xt and Yt be governed by

dXt Q dYt Q
= (r − qX ) dt + σX dZX,t and = (r − qY ) dt + σY dZY,t,
Xt Yt
where r is the constant riskless interest rate, σX and σY are the
constant volatility of Xt and Yt, respectively, qX and qY are the
dividend yield of Xt and Yt, respectively. Also, the two standard
Q Q
Brownian motions are correlated with dZX,t dZY,t = ρ dt, where ρ is
correlation coefficient.

13
Numeraire Invariance Theorem
(∫ t )
With the use of the money market account M (t) = exp rudu
0
as the numeraire (accounting unit), the effect of the time value of
money with respect to normalized asset values becomes immaterial.
Discounted security price St/Mt is a martingale under a risk neutral
measure Q.

The choice of the money market account M (t) as the numeraire

is not unique in order that no-arbitrage pricing principle holds. We
may choose the price of a tradable asset N (t) as the numeraire.

A numeraire is any strictly positive (FT )t∈R+ -adapted stochastic

process (Nt)t∈R+ that can be used as a unit of reference.

In order to effect the no-arbitrage pricing approach, one has to

determine a probability measure under which Ŝt = St/Nt is a mar-
tingale. With the martingale property remains intact ( ৏ሱнࣘ ),
there will be absence of arbitrage.
14
How to construct the new measure QN from the risk neutral measure
Q such that the deflated prices are QN -martingale?

N ∫T
It is observed that Nt∗ = = e− 0 ruduNt is an Ft-martingale under
t
Mt
Q since Nt is a traded asset.

Given (Nt)t∈[0,T ], we define the associated measure QN via

dQN NT∗ ∫
− 0T ru du NT
= ∗ =e .
dQ N0 N0
This is equivalent to stating that
∫ ∫ ∫
− 0T ru du NT
X(ω) dQN (ω) = e X(ω) dQ(ω).
Ω Ω N0
For any integrable FT -measurable random variable X, we have
[ ∫T ]
ru du NT
EQN [X] = EQ e− 0 X .
N0

15
From the martingale property of Nt∗, we deduce that
[  ] [  ] ∫T
dQN  ∫  Nte− 0 ru du Nt∗
− 0T ru du NT 
EQ  Ft = EQ e  Ft = = ∗.
dQ  N0  N0 N0
By the tower rule, for any Ft-measurable random variable G and
integrable random variable X, we obtain
[ ∫T ]
− 0 ru du NT
EQN [GX] = EQ GXe
N0
[ [  ]]
∫
Nt − t ru du ∫ T NT 
− r du
= EQ G e 0 EQ Xe t u  Ft
N0 Nt 
[ [  ] [  ]]
dQN  ∫ 
− t ru du NT 
T
= EQ GEQ  Ft EQ Xe  Ft
dQ  Nt 
[ [  ]]
dQN ∫ 
− t ru du NT 
T
= EQ G EQ Xe  Ft
dQ Nt 
[ [  ]]
∫T 
− t ru du NT 
= EQN GEQ Xe  Ft .
Nt 

16
By comparing with

EQN [GX] = EQN [GEQN [X|Ft]],

we deduce that
[  ]
∫T 
du NT 
EQN [X|Ft] = EQ Xe− t ru
 Ft .
Nt 
Consider an option with integrable claim payoff C ∈ L1(Q, FT ), by
C
taking X = , we obtain
NT
[  ] [  ]
C  ∫T 
− t ru du
NtEQN  Ft = EQ e C  Ft ,
NT 

C
provided that ∈ L1[QN , FT ].
NT
Let Vt denote the time-t price of the contingent claim, where VT =
C. We then deduce that
[  ]
Vt VT 
= EQN  Ft
Nt NT 
so that the deflated price Vt/Nt is QN -martingale.
17
Remark One motivation of choosing another tradeable security
(discount bond) as the numeraire arises from the pricing of an equity
Mt ∫T
option under stochastic interest rate. Recall = e− t ru du when
MT
the interest rate is stochastic so that
[ ∫T ]
t e−
Vt = EQ t ru du
V (ST ) .

With the choice of the bond price Bt as the numeraire, we have

[ ]
t VT (ST ) t [V (S )]
Vt = BtEQ T
= BtEQ T T T since BT = 1.
BT
Since Vt/Bt is the time-t forward price of forward delivery of VT at
time T , so QT is termed the T -forward measure.

18
Use of the underlying asset as the numeraire (share measure) and
the associated change of measure

Recall that the risk neutral measure uses the money market account
as the numeraire. Let the starting time be time zero for notation-
al convenience. Consider the Radon-Nikodym derivative Lt as a
stochastic process that is defined by taking the ratio of the asset
numeraire and the money market account
 /
S
dQ   St Mt
Lt =  = eqt , t ∈ (0, T ],
dQ F S0 M0
0

where Mt = ert is the money market account and q is the dividend

yield of the underlying asset.

19
The inclusion of the factor eqt means one unit of the risky asset
initially grows to eqt units after time t if all dividends are invested
into the purchase of new units of the risky asset.

Let Sbt = eqtSt, then Sbt∗ = Sbt/ert is a martingale under Q. Note that
Sbt is chosen as the numeraire asset instead of St since we require
the discounted numeraire asset is a Q-martingale.

We examine the change of measure from Q to QS as effected by Lt.

20
Symbolically, we write the Radon-Nikodym derivative as

dQ 
S
Lt =  , t ∈ (0, T ].
dQ  F0
Under the risk neutral measure Q, the dynamics of St is governed
by
dSt Q Q
= (r − q) dt + σ dZt , Zt is Q-Brownian.
St
The solution to St is given by
( 2
)
σ Q
r−q− 2 t+σZt
St = S0e
so that
/
St 2 Q
− σ2 t+σZt
Lt = eqt rt
e =e , t ∈ (0, T ].
S0

21
Recall from p.41, Topic 3 that under the Radon-Nikodym derivative:
dPe 2
− γ2 t−γZtP
=e ,
dP
the Brownian motion with drift defined by ZeP (t) = ZP (t) + γt be-
comes Pe -Brownian. Here, ZP (t) is P -Brownian.

In the current context, this corresponds to the choice of γ = −σ in

the Radon-Nikodym derivative. We then deduce that
QS Q
Zt = Zt − σt is a QS -Brownian.
We commonly call QS to be the share measure with respect to St.

22
As a check, we recall
[ / ]
S
dQ 
−rT −rT
V0 = e EQ[VT (ST )|F0] = e EQS VT (ST )  F0
dQ 
[ /  ]
−rT rT qT 
= e EQS VT (ST )e S0 e ST  F0
[  ]

VT (ST ) 
= S0EQS  F0 ,
ST e 
qT

so that
[ ]
V0 VT (ST )
b
= EQS , where SbT = eqT ST and Sb0 = S0.
S0 SbT
This verifies that Vt/Sbt is QS -martingale.

23
Write QX as the share measure with respect to the asset price
process Xt. We have shown that
QX Q
ZX,t = ZX,t − σX t

is QX -Brownian.

Since Xt and Yt have correlation coefficient ρ, we expect that sub-

Q Q
tracting ρσX t from ZY,t would make ZY,t −ρσX t to be QX −Brownian.

QX Q
How to verify that ZY,t = ZY,t − ρσX t is QX -Brownian?

24
By considering the moment generating function, a random variable
U is normal with mean m and variance σ 2 under QX if and only if
( )
α2
EQX [exp(αU )] = exp αm + σ2 .
2
It suffices to show
( )
QX Q α2
EQX [exp(αZY (T ))] = EQX [exp(αZY (T ) − αρσX T )] = exp T .
2
Recall the Radon-Nikodym derivative, where
( )
dQ X σX2
Q
LT = = exp − T + σX ZX (T ) , so
dQ 2

QX
EQX [exp(αZY (T ))]
[ ( )]
σX2
Q Q
= EQ exp(αZY (T ) − αρσX T ) exp − T + σX ZX (T )
2
( )
2
σX ( )
Q Q
= exp − T − αρσX T EQ[exp αZY (T ) + σX ZX (T ) ]
2
( ) ( ) ( )
2
σX 2
α + 2ρασX + σX 2 α 2
= exp − T − αρσX T exp T = exp T .
2 2 2
25
Derivation of the price formula of an exchange option

Suppose we choose eqX tXt as the numeraire, and MT /M0 = erT , the
corresponding Radon-Nikodym derivative that effects the change
from Q to QX is given by

(qX −r)T XT
LT = e .
X0
The price function of the exchange option with maturity T and initial
asset values X0 and Y0 is given by

V (X0, Y0; T ) = e−rT EQ[max(XT − YT , 0)]

 ( ) 
X e(r−qX )T YT
= e−rT EQX  0
XT 1− 1{YT /XT <1} .
XT XT
Note that XT is canceled. Setting WT = YT /XT , then
V (X0, Y0; T )
= e−qX T EQX [(1 − WT )1{WT <1}].
X0
This nice feature of dimension reduction of the option model does
not work if the terminal payoff becomes max(XT − YT − K, 0).
26
From Ito’s lemma, the dynamics of Wt under Q is given by
dWt 2 ] dt + σ dZ Q − σ dZ Q .
= [(r − qY ) − (r − qX ) − ρσX σY + σX Y Y,t X X,t
Wt
QX QX
We observe that ZX,t and ZY,t as defined by

QX Q QX Q
dZX,t = dZX,t − σX dt and dZY,t = dZY,t − ρσX dt

are QX -Brownian motions. The dynamics of Wt under QX becomes

dWt QX QX
= (qX − qY ) dt + σY dZY,t − σX dZX,t .
Wt
We deduce that Wt remains to be a Geometric Brownian motion,
2 = σ 2 − 2ρσ σ + σ 2 and µ
and σW Y X Y X W = qX − qY = (r − qY ) − (r − qX )
under QX .

27
We may write
dWt QX
= (qX − qY )dt + σW dZW,t ,
Wt
QX
where ZW,t is QX -Brownian.

We retain the nice analytical tractability for Yt/Xt since the ratio of
the two lognormal distributions Xt and Yt remains to be lognormal.
However, the difference Xt − Yt has no nice analytic form of joint
distribution function. This explains why we choose to normalize
the payoff function by XT instead of choosing the apparently more
obvious choice of Wf =X −Y .
T T T

28
The payoff (1 − WT )1{WT <1} resembles a put payoff with unit strike
and underlying Wt. Using the put price formula, we deduce
Y0
EQX [(1 − WT )1{WT <1}] = N (dX ) − W0e(qX −qY )T N (dY ), W0 = ,
X0
where
2
σW 2
σW
X0 X0
ln Y + (qY − qX )T + 2 T ln Y + (qY − qX )T − 2 T
dX = 0 √ , dY = 0 √ .
σW T σW T
Finally, the price function of the exchange option is given by
V (X0 , Y0 ; T ) = e−qX T X0 N (dX ) − e−qY T Y0 N (dY ).
  [ 2
] 
σW
ln Y0 + (r − qX ) − (r − qY ) + 2 T
X0
−rT  (r−qX )T
= e e X0 N  √ 
σW T
 [ 2
] 
σW
ln X0 + (r − qY ) − (r − qX ) + 2 T
Y0

− e (r−qY )T 
Y0 N − √  .
σW T

Suppose we take Yt to be the fixed strike price K. By setting

Y0 = K, qY = r (so the drift rate r − qY of Yt becomes zero) and
2 = σ 2 (since σ = 0), we recover the usual call price formula.
σW X Y

29
4.3 Quanto option – equity options with exchange rate risk
exposure

• A quanto option is an option on a foreign currency donominated

asset but the payoff is in domestic currency.
• The holder of a quanto option is exposed to both exchange rate
risk and equity risk. Essentially, quanto option pricing models
are two-dimensional with exchange rate and asset price as the
pair of state variables.

Some examples of quanto call options are listed below:

1. Foreign equity call struck in foreign currency

c1(ST , FT , T ) = FT max(ST − Xf , 0).

Here, FT is the terminal exchange rate, ST is the terminal price
of the underlying foreign currency denominated asset and Xf is
the strike price in foreign currency.

30
2. Foreign equity call struck in domestic currency

c2(ST , T ) = max(FT ST − Xd, 0)

Here, Xd is the strike price in domestic currency.

3. Fixed exchange rate foreign equity call

c3(ST , T ) = F0 max(ST − Xf , 0)
Here, F0 is some predetermined fixed exchange rate.

4. Equity-linked foreign exchange call

c4(ST , T ) = ST max(FT − XF , 0).

Here, XF is the strike price on the exchange rate. The holder
plans to purchase the foreign asset any way but wishes to place
a floor value XF on the exchange rate. If it happens that the
terminal exchange rate FT shoots beyond XF , she receives com-
pensation from the positive payoff received through holding the
foreign exchange call.

31
Quanto prewashing techniques

• Let St and Ft denote the stochastic process of the foreign asset

price and exchange rate, respectively.

• Define St∗ = FtSt, which is the foreign asset price in domestic

currency.

• Let rd and rf denote the constant domestic and foreign interest

rate, respectively, and let q denote the dividend yield of the
foreign asset. Note that the dividend yield is the same in both
currency worlds.

• We assume that both St and Ft follow the Geometric Brownian

motion.

32
• Under the domestic risk neutral measure Qd, the drift rate of S ∗
and F are
d =r −q
δS and d =r −r .
δF
∗ d d f

• The reciprocal of F can be considered as the foreign currency

price of one unit of domestic currency.

• The drift rate of S and 1/F under the foreign risk neutral mea-
sure Qf are given by
f f
δ S = rf − q and δ1/F = rf − rd,
respectively. Note that the dividend yield is the same for the
foreign asset in the two-currency world.

• “Quanto prewashing” means finding δS d , that is, the drift rate in

the stochastic price process of the foreign currency denominated

asset S under the domestic risk neutral measure Qd.

33
Quanto-prewashing formula

Let the dynamics of St and Ft under Qd be governed by

dSt d dt + σ dZ d
= δS S S
St
dFt d dt + σ dZ d ,
= δF F F
Ft
where dZSd dZFd = ρ dt, σS and σF are the volatility of St and Ft,
respectively. Since St∗ = FtSt, we obtain from Ito’s lemma (see
Problem 3 in HW3):
d = δ d = δ d + δ d + ρσ σ .
δS ∗ FS F S F S

The extra drift rate ρσF σS arises from the correlated diffusion move-
ments of Ft and St, where dZSd dZFd = ρ dt. We then obtain
d = δ d − δ d − ρσ σ = (r − q) − (r − r ) − ρσ σ = r − q − ρσ σ .
δS S∗ F F S d d f F S f F S
f
d = δ − ρσ σ . It is necessary to add the quanto pre-
We obtain δS S F S
washing term −ρσF σS when we specify the dynamics of St changing
from Qf to Qd.
34
 f
d
Siegel’s paradox δ1/F = rf − rd + σF
2 =δ
1/F
+ σ 2
F

Given that the dynamics of Ft under Qd is

dFt
= (rd − rf ) dt + σF dZd,
Ft
then the dynamics of 1/Ft under Qd is (see Problem 3 in HW3)
d(1/Ft) 2 ) dt − σ dZ .
= (rf − rd + σF F d
1/Ft
To show an alternative proof of the dynamic equation for 1/Ft, we
observe that Ft admits the solution as exponential Brownian motion:
(( ) )
2
σF
Ft = F0 exp rd − rf − t + σF Zd
2
Taking the reciprocal, we obtain
(( ) )
1 1 2
σF
= exp 2
rf − rd + σF − t − σF Zd .
Ft F0 2
Working in the reverse manner, from the knowledge of the solution,
we can deduce the governing stochastic differential equation for
1/Ft as given in the above.
35
This is seen as a puzzle to many people since the risk neutral drift
rate for 1/F is expected to be rf − rd instead of rf − rd + σF
2.

We observe directly from the above SDE’s for 1/Ft that

σF = σ1/F and ρF,1/F = −1.

f
Note that δ1/F = rf − rd. This is also consistent with the quanto
prewashing technique when it is applied to 1/F , where the added
prewashing term −ρσF σ1/F becomes −(−1)σF 2 = σ2 .
F

36
An interesting application of Siegel’s paradox

Suppose the terminal payoff of an exchange rate option is FT 1{FT >K}.

Let V d(F, t) denote the value of the option in the domestic currency
world. Define
V f (Ft, t) = V d(Ft, t)/Ft,
so that the terminal payoff of the exchange rate option in foreign
currency world is 1{FT >K}. Now

V f (F, t) = e−rf (T −t)EQ

t [1
f {FT >K} |Ft = F ].

37
 d f 2 and observing σ = σ
From δ1/F = δ1/F + σF F 1/F , we deduce that

f d + σ2 .
δF = δF F
This result is consistent with the Siegel formula if we interchange
the foreign and domestic currency worlds. We obtain
d f −rf (T −t) −rd τ dτ
δF
V (F, t) = F V (F, t) = e F N (d) = e e F N (d)
where τ = T − t and
( 2 ) ( 2 )
F + f σF F + σF
ln K δF − 2 τ ln K rd − rf + 2 τ
d = √ = √ .
σF τ σF τ

Note that the discount factor is e−rf τ in the foreign currency world
and it becomes e−rdτ in the domestic currency world. Also, observe
that
[ ]
t
EQf 1{F >K} = N (d)
[ ] d
δF
t
EQd F 1{F >K} = e F N (d).
38
Price formulas of various quanto options

1. Foreign equity call struck in foreign currency

f
Let c1(S, τ ) denote the usual vanilla call option on the foreign
currency asset in the foreign currency world. The terminal payoff
is
f
c1(S, 0) = max(S − Xf , 0).
We treat this call as if it is structured in the foreign currency
world. Its value can always be converted into domestic currency
using the prevailing exchange rate.

39
 [ ]
c1(S, F, τ ) = F c1(S, τ ) = F Se−qτ N (d1 ) − Xf e−rf τ N (d2 ) ,
f (1) (1)

where
( 2 )
f σS
ln XS + δS + 2 τ √
(1) f (1) (1)
d1 = √ , d2 = d1 − σS τ ,
σS τ

f
δS = rf − q.

Note that both the correlation risk ρ and exchange rate risk σF do
not appear in the price formula! This is reasonable since we allow
the exchange rate to float and do not set the exchange rate to some
fixed value F0.

40
2. Foreign equity call struck in domestic currency
The terminal payoff in domestic currency is

c2(S, F, 0) = max(S ∗ − Xd, 0),

where S ∗ = F S is the price of a domestic currency denominated
asset. Note that
d =r −q
δS and 2 = σ 2 + 2ρσ σ + σ 2 .
σS
∗ d ∗ S S F F
The price formula of the foreign equity call is then given by

c2(S, F, τ ) = S ∗e−qτ N (d1 ) − Xde−rdτ N (d2 ),

(2) (2)

where
( )
∗ 2
σS
S d + ∗
ln X + δS ∗ τ
(2) d 2
(2) (2) √
d1 = √ , d2 = d1 − σS ∗ τ .
σS ∗ τ
Note that rf does not appear since we perform valuation in the
domestic currency world and no foreign-denominated asset is
involved. However, ρ and σF are involved since the volatility of
S ∗ comes into play.
41
3. Fixed exchange rate foreign equity call
The terminal payoff is denominated in the domestic currency
d of the foreign asset in Q should
world, so the drift rate δS d
be used. The price function of the fixed exchange rate foreign
equity call is given by
[ ]
c3(S, τ ) = F0e−rdτ
dτ
δS (3) (3)
Se N (d ) 1 − Xf N (d2 ) ,

where
( 2 )
σS
ln XS + d
δS + 2 τ √
(3) f (3) (3)
d1 = √ , d2 = d1 − σS τ .
σS τ
• The price formula does not depend on the exchange rate F since
the exchange rate has been chosen to be the fixed value F0.
• The currency exposure of the call is embedded in the quanto-
prewashing term −ρσS σF in δSd . This call option has exposure to

both correlation risk and exchange rate risk (in terms of ρ and
σF , respectively).

42
4. Equity-linked foreign exchange call
Write the terminal payoff in the form of an exchange option

c4(S, F, 0) = max(S ∗ − XS, 0).

Taking the two assets to be an exchange XS for S ∗, the ratio of
S∗ F
the two assets is = and the difference of the drift rates
XS X
under Qd is δS ∗ − δS = rd − q − (rf − q − ρσS σF ) = rd − rf +
d d

ρσF σS . By recalling the exchange option price formula, where

X F
ln 0 becomes ln , r − qX becomes δS d , r − q becomes δ d , σ
∗ Y S W
Y0 X
becomes σF , etc., we obtain
[ ]
d τ
c4(S, τ ) = e−rdτ ∗ δS (4) d (4)
S e ∗ N (d )
1 − XSeδS τ N (d )
2
[ ]
= Se−qτ F N (d1 ) − Xe(rf −rd−ρσF σS )τ N (d2 ) ,
(4) (4)

where
( 2 )
F + σF
ln X rd − rf + ρσF σS + 2 τ √
(4) (4) (4)
d1 = √ , d2 = d1 − σF τ .
σF τ
43
Digital quanto option relating 3 currency worlds

FS\U = SGD currency price of one unit of USD currency

FH\S = HKD currency price of one unit of SGD currency

We visualize FS\U as Singaporean currency denominated asset.

Example 1

Digital quanto option payoff: pay one HKD if FS\U is above some
strike level K.

The dynamics of FS\U under QS is governed by

dFS\U
= (rSGD − rU SD ) dt + σFS\U dZFS .
FS\U S\U

44
 S
Given δF = rSGD − rU SD , how to find δF
H , the risk neutral drift
S\U S\U
rate of the SGD asset denominated in Hong Kong dollar?

Taking Hong Kong as the domestic world. Treating FS\U as the

foreign asset denominated in Singaporean currency and FH\S as the
exchange rate, by the quanto-prewashing technique
H
δF S
= δF − ρσFS\U σFH\S ,
S\U S\U

where ρ dt = dZFH dZFH . Here, FS\U is visualized as the foreign

S\U H\S
asset S and FH\S as the exchange rate F in the quanto prewashing
formula.
[ ]
Digital option value = e−rHKD τ EQ
t
H 1{FS\U >K} = e−rHKD τ N (d)

where
 2

FS\U σF
ln K + δF
H − 2
S\U 
τ
S\U
d= √ .
σFS\U τ

45
Example 2

The quanto option pays FH\S Hong Kong dollars when FS\U > K.
This is equivalent to pay one Singaporean dollar. Value of the
quanto option in Singaporean dollar is
[ ]
e−rSGD τ EQ
t
S 1{FS\U >K} = e−rSGD τ N (d)
b

where
 2

FS\U σF
ln K + δF
S − 2
S\U τ
S\U
db = √ , S
δF = rSGD − rU SD .
σFS\U τ S\U

This option model is similar to c1(S, F, τ ), where the option payoff

in foreign currency is converted into domestic currency using the
prevailing exchange rate at maturity. The most efficient approach
is to perform valuation of the option under the foreign currency
world. The present value of the quanto option in Hong Kong dollar
b
is FH\S e−rSGD τ N (d).
46
Example 3

The quanto option pays FH\U Hong Kong dollars when FS\U > K.
This is equivalent to pay one US dollars.

Method One – Valuation in the Singaporean currency world

Observe that FH\U = FH\S FS\U so that it is like paying FS\U Singa-
porean dollars (equivalent to one US dollar) when FS\U > K. Note
that S
δF = rSGD − rU SD .
S\U
The present value of the quanto option in Hong Kong dollars is
[ ]
−rSGD τ
FH\S e t
EQ S
FS\U 1{F S\U >K} = FH\S e−rSGD τ e(rSGD τ −rU SD )τ FS\U N (d1 )
= FH\U e−rU SD τ N (d1 )

where
 2

FS\U σF
ln K + rSGD − rU SD + S\U 
2 τ
d1 = √ .
σFS\U τ

The discount factor is e−rSGD τ and multiplication of FH\S converts

the Singaporean dollar payoff into Hong Kong dollars payoff.
47
Method Two – Valuation in the US currency world

The quanto option pays one US dollars when FS\U > K ⇔ K 1 >
1 =F
FS\U U \S . Later, we multiply the option value in US currency by
the exchange rate FH\U to convert into Hong Kong dollars.

The present value of the quanto option in Hong Kong dollars is

 

FH\U e−rU SD τ EQ
t 1{
U
} = F
H\U e−rU DS τ N (−d ),
2
<1
FU \S K

where
 2

F σF
U \S U \S 
ln 1/K + rU SD − rSGD − 2 τ
d2 = √ = −d1.
σFU \S τ

By noting σFU \S = σFS\U , we can check easily that the quanto option
value in Hong Kong dollars using the two approaches agree with each
other.
48
4.4 Implied volatilities and volatility smiles

The difficulties of setting volatility value in the option price for-

mulas lie in the fact that the input value should be the forecast
volatility value over the remaining life of the option rather than an
estimated volatility value from the past market data of the asset
price (historical volatility).

The Black-Scholes model assumes a lognormal probability distri-

bution of the asset price at all future times. Since volatility is the
only unobservable parameter in the Black-Scholes model, the model
gives the option price as a function of volatility. The Black-Scholes
implied volatility σimp(X, T ) is the unique solution to

Vmarket(X, T ) = V BS (S, t; X, T, σimp(X, T )).

The above equation is an answer to: What volatility is implied in

the observed option prices, if the Black-Scholes model is a valid
description of the market conditions?

49
Remark

Implied volatility derives from the Black-Scholes model, and it trans-

lates into an option price through the Black-Scholes equation. It is
similar to the yield to maturity of a bond, where the implied yield is
implied from the bond price through the bond price formula.

There is one-to-one correspondence between the implied volatility

and option price, same as the implied yield and bond price.

50
Volatility smiles and volatility term structures

• In financial markets, it becomes a common practice for traders

to quote an option’s market price in terms of implied volatility
σimp.

• In particular, several implied volatility values obtained simulta-

neously from different options with varying maturities and strike
prices on the same underlying asset provide an extensive market
view about the volatility at varying strikes and maturities.

• The Black-Scholes (BS) implied volatility computed from the

market option price by inverting the BS price formula varies
with the strike price and time to expiration – volatility smile
(skew) and volatility term structure, respectively. The plot of
the implied volatilities against moneyness (X/S) and time to
expiration T − t generates the implied volatility surface.

51
Implied volatility surface

DAX option implied volatilities (as black dots) on 2000/05/02. The

lower left axis is moneyness X/S and the right axis is time to expi-
ration measured in years.

• σimp(X, T ) is non-linear in strikes and time to expiration; and if

observed over in calendar time, it is also time-dependent.

52
Volatility skew

The plots of the implied volatility σimp against moneyness K/S for
traded call and put options with the same maturity date typically
show the skew shape as shown in the following figure.

ı LPS

SXW

FDOO

.6
RXWRIWKH PRQH\ RXWRIWKHPRQH\
IRUFDOOV IRUSXWV

53
• The volatility skew across moneyness occurs since the option
prices for deep out-of-the-money options are bid up higher than
around-the-money counterparts.

• Since the stock markets tend to crash downward faster than

they move upward, option buyers are willing to bid at higher
prices for puts than calls, so σimp of puts are typically higher
than σimp of calls.

• Investors have stronger motive to use deep out-of-the-money

puts to hedge against drastic market decline, so out-of-the-
money puts are trading more expensively than out-of-the-money
calls for the same level of out-of-the-moneyness (say, K/S = 0.8
for a call and K/S = 1.25 for a put).

54
Comparison of the risk neutral probability density of asset price
(solid curve) implied from market data and the theoretical lognormal
distribution (dotted curve). The risk neutral probability density is
thicker at the left tail and thinner at the right tail, indicating that
there is a higher change of more acute drop when S is low and a
lower chance of further increase when S is high.

55
Negative correlation between stock price process and volatility pro-
cess

In real market situation, it is a common occurrence that when the

asset price is high, volatility tends to decrease, making it less prob-
able for a higher asset price to be realized. When the asset price
is low, volatility tends to increase, that is, it is more probable that
the asset price plummets further down. In other words, stock price
process and volatility process are in general negatively correlated.

56
Extreme events in stock price movements

Probability distributions of stock market returns have typically been

estimated from historical time series. Unfortunately, common hy-
potheses may not capture the probability of extreme events. The
crash events are rare and may not be present in the historical record.

Examples
1. On October 19, 1987, the two-month S&P 500 futures price
fell 29%. Under the lognormal hypothesis of annualized volatility
of 20%, this is a −27 standard deviation event with probability
10−160 (virtually impossible).

2. On October 13, 1989, the S&P 500 index fell about 6%, a −5
standard deviation event. Under the maintained hypothesis, this
should occur only once in 14, 756 years.

57
Implied volatilities across time

Supply and demand

When markets are very quiet, the implied volatilities of the near
month options are generally lower than those of the far month.
When markets are very volatile, the reverse is generally true.

• Recall that gamma is the second order derivative of the option

price function with respect to the stock price, which is highly
dependent on volatility. In very volatile markets, everyone wants
or needs to load with gamma so that the portfolio increases
in value when volatility increases. Near-dated options provide
the most gamma and the resulting buying pressure will have the
effect of pushing prices up.

• In quiet markets, no one wants a portfolio long of near dated

options since the loss of over time over the passage of time is
higher for short-lived options.

58
Implied volatilities for different strike prices

1. Stock options – higher implied volatilities at lower strike and

lower implied volatilities at higher strikes

• In a falling market, everyone needs out-of-the-money puts

for insurance and pushes a higher price for the lower strike
options due to good demand.

• Equity fund managers are long billions of dollars worth of

stock and writing out-of-the-money call options against their
holdings as a way of generating extra income. This pushes
the value of out-of-the-money call options down due to good
supply.

59
2. Commodity options – higher implied volatilities at higher strike
and lower implied volatilities at lower strikes

• Government intervention – no worry about a large price fall.

Speculators are tempted to sell puts aggressively.

• Risk of shortages – no upper limit on the price. Demand for

higher strike price options.

60
Term structure of volatility

The Black-Scholes formulas

√ remain valid under time dependent
∫ T
1
volatility except that σ(τ )2 dτ is used to replace σ.
T −t t

How to obtain the term structure of volatility σ(t) given the implied
volatility measured at time t∗ of a European option expiring at time
t? For an option with time to expiry t − t∗, the substitution of the
implied volatility σimp(t∗, t) into the standard Black-Scholes formula
under constant volatility gives the option price.

Remark
The implied volatility σimp(t∗, t) may be obtained by averaging the
implied volatility values for options at different strikes but with same
maturity t. This treatment is acceptable since we focus on the
consideration of the term structure of volatility.

61
The equivalence of giving the same observed option price by adopt-
ing the two different forms of volatility in the two separate option
price formulas leads to
∫ t
2 (t∗, t)(t − t∗ ).
σ(u)2 du = σimp
t∗
The left hand side is the assumption of time dependent volatility σ(t)
and the right hand side is the application of the implied volatility
formula based on constant volatility in the Black-Scholes pricing
formula.

Differentiating with respect to t, we obtain the term structure of

volatility in terms of the term structure of implied volatility
√
∗ 2 ∗ ∗ ∂σimp(t∗, t)
σ(t) = σimp(t , t) + 2(t − t )σimp(t , t) .
∂t

62
Approximation of σ(t) as a piecewise constant function

Practically, we do not have a continuous differentiable implied volatil-

ity function σimp(t∗, t), but rather implied volatilities are available at
discrete instants ti, i = 1, 2, . . . , n. Suppose we assume σ(t) to be
piecewise constant over (ti−1, ti), where σ(t) = σi, ti−1 < t < ti,
i = 1, 2, . . . , n. We then have
∫ t ∫ t
i i−1
σ 2(τ ) dτ − σ 2(τ ) dτ
t∗ t∗
= (ti − t∗)σimp
2 (t∗ , t ) − (t
i
∗ 2 ∗
i−1 − t )σimp (t , ti−1 )
∫ t
i
= σ 2(τ ) dτ = σi2(ti − ti−1), ti−1 < t < ti,
ti−1
giving
v
u
u (ti − t∗)σ 2 (t∗, ti) − (ti−1 − t∗)σ 2 (t∗, ti−1)
t imp imp
σi = , ti−1 < t < ti.
ti − ti−1

63
Risk neutral density function

• Let ψ(ST , T ; St, t) denote the transition density function of the

asset price. The time-t price of a European call with maturity
date T and strike price X is given by
∫ ∞
c(St, t; X, T ) = e−r(T −t) (ST − X)ψ(ST , T ; St, t) dST .
X

• If we differentiate c with respect to X, we obtain

∞ ∫
∂c −r(T −t)
= −e ψ(ST , T ; St, t) dST ;
∂X X
and differentiate once more, we have

r(T −t) ∂ 2c
ψ(X, T ; St, t) = e .
∂X 2
• Suppose that market European option prices at all strikes are
available, the risk neutral density function can be inferred com-
pletely from the market prices of options with the same maturity
and different strikes, without knowing the volatility function.

64
Dupire equation and local volatility function

Assuming that the asset price dynamics under the risk neutral mea-
sure is governed by
dSt
= (r − q)dt + σ(St, t)dZt,
St
where the local volatility function σ(St, t) is assumed to have both
state and time dependence.

Suppose we visualize the call price function as a function of X and

T , where c = c(X, T ), the Dupire equation takes the form
∂c ∂c σ 2(X, T ) 2 ∂ 2c
= −qc − (r − q)X + X 2
.
∂T ∂X 2 ∂X

The forward asset price ST is substituted by X, treating as an inde-

pendent variable in the price function c(X, T ).

65
Proof

We differentiate ψ(X, T ; St, t) with respect to T to obtain

( )
∂ψ ∂ 2c ∂ 2 ∂c
= er(T −t) r 2
+ ,
∂T ∂X ∂X 2 ∂T
and ψ(X, T ; S, t) satisfies the forward Fokker-Planck equation, where
[ ]
∂ψ ∂2 σ 2(X, T ) ∂
= X 2ψ − [(r − q)Xψ]
∂T ∂X 2 2 ∂X
{ [ ] [ ]}
∂ 2 2 2
σ (X, T ) 2 ∂ c ∂ 2
∂ c
=e r(T −t) X − (r − q)X .
∂X 2 2 ∂X 2 ∂X ∂X 2

See Problem 3.8 on P.166 in Kwok’s text for a proof of the forward
Fokker-Planck equation.

66
Combining the above equations and eliminating the common factor
er(T −t), we have
∂ 2c ∂ 2 ∂c
r +
∂X [ ∂X 2 ∂T
2
] [ ]
∂ 2 2 2
σ (X, T ) 2 ∂ c ∂ 2
∂ c
= X − (r − q)X .
∂X 2 2 ∂X 2 ∂X ∂X 2
Integrating the above equation with respect to X twice, we obtain
( )
∂c ∂c
+ rc + (r − q) X −c
∂T ∂X
2
σ (X, T ) 2 ∂ c2
= X 2
+ α(T )X + β(T ),
2 ∂X
where α(T ) and β(T ) are arbitrary functions of T .

Since all functions involving c in the above equation vanish as X

tends to infinity, hence α(T ) and β(T ) must be zero (one can check
∂c 2 ∂ 2c
lim X = 0 and lim X 2
= 0). Grouping the remaining
X→∞ ∂X X→∞ ∂X
terms in the equation, we obtain the Dupire equation.

67
From the Dupire equation, we may express the local volatility σ(X, T )
explicitly in terms of the call price function and its derivatives, where
[ ]
∂c ∂c
2 ∂T + qc + (r − q)X ∂X
2
σ (X, T ) = .
∂ 2c
X 2 ∂X 2

• Suppose a sufficiently large number of market option prices are

available at many maturities and strikes, we can estimate the
local volatility from the above equation by approximating the
derivatives of c with respect to X and T using the market data.

• In real market conditions, market prices of options are available

only at limited number of maturities and strikes.

The local volatility σ(S, T ) of an asset price process at some future

market level S and time T is the future volatility the asset must have
at that market level and time in order to make the current option
prices fair.

68
Relationship between local volatility and implied volatility

Dupire’s equation shows how to compute σloc(X, T ) from market

prices of European options. On the other hand, the market quote
prices for European options are in terms of their implied volatilities.
One may want to relate σloc(X, T ) with σimp(X, T ). We have (see
Problem 11 in HW4)
∂σ ∂σ
2 (X, T ) =
σ 2 + 2T σ
imp imp
imp
∂T + 2(r − q)XT σ imp ∂X
imp
σloc ( ) [ ( )2] ,
√ ∂σimp 2 2
∂ σ √ ∂σimp
1 + Xd1 T ∂X + X 2T σimp ∂Ximp 2 − d 1 T ∂X

where
[ ]
2 (X,T )
σimp
S + r−q+
ln X T
2
d1 = √ .
σimp(X, T ) T

69
4.5 Volatility exposure generated by delta hedging options

Delta hedging an option based on some chosen time dependent

hedge volatility generates a profit and loss (P&L) that is related
to the realized variance and the cash gamma position (defined as
product of option gamma and square of asset price).

Consider the underlying asset price which follows the Geometric

Brownian motion under a risk neutral measure Q as specified by
dSt
= (r − q)dt + σt dWt, (1)
St
where r and q are constant riskfree rate and dividend yield, respec-
tively, σt is the instantaneous volatility process, and Wt is a standard
Brownian motion under Q. The assumed dynamics of St allows for
stochastic volatility σt but no jump.

70
Let σti be the time dependent implied volatility derived from traded
option prices at varying times. We write the time-t option price
as Vti = V (St, t; σti) with reference to implied volatility σti. Sup-
pose an option trader sells an option at time zero priced at the
current market implied volatility σ0i , the option price is given by
V0i = V (S0, 0; σ0i ).

The seller’s short position in the option is delta hedged at some

chosen time dependent hedge volatility σth for the remaining life
of the option, that is, by holding ∆h = ∂ V (S , t; σ h ) units of the
t ∂S t t
underlying asset at time t ∈ [0, T ], plus a money market account Mt
worth Vti − ∆ht St . Recall that the hedger replicates the option by
adopting Vti = ∆h t St + Mt .

In summary, there are 3 volatilities:

(i) σt - Mother Nature’s choice

(ii) σti - market’s choice
(iii) σth - hedger’s choice

71
After selling one unit of the option, the hedger replicates the option
by holding ∆h t units of stock and money market account. These
three instruments constitute the delta hedged portfolio that hedges
against the stock price risk. However, variance risk remains since
only stock is used as the hedging instrument.

The P&L of the delta hedged portfolio over the infinitesimal time
interval [t, t + dt] consists of the following three components:

• change of the option value from shorting one unit of option:

−dVti;
• P&L resulted from the dynamic position of the underlying asset
and dividend income: ∆ht (dSt + qSt dt);
• riskfree interest income earned from the money market account:
r(Vti − ∆ht St )dt.

72
At time t, the hedger adopts his own hedging volatility σth. We
allow the flexibility that the hedger may not choose the implied
volatility σti from the market option price. The hedger computes
Vth = V (St, t; σth) based on the choice of the volatility parameter
value σth in the Black-Scholes option price formula and use Vth to
h ∂Vth
compute the hedge ratio ∆t = ∂S . The various Greek parameters
h ∂Vth h ∂Vth h ∂ 2 Vth
Θt = ∂t , ∆t = ∂S and Γt = ∂S 2 are related by (based on the
Black-Scholes equation)

h Γh St2 h 2
Θt = − t (σt ) + rVth − (r − q)∆h
t St . (2a)
2
By applying Itô’s lemma to the price function Vth = V (St, t; σth), we
have
h h h 1 ∂ 2Vth 2
dVt = Θt dt + ∆t dSt + 2
(dS t )
( 2 ∂S )
h 2
= ∆h dS + Θh + Γt St σ 2 dt, (2b)
t t t t
2
where dSt2 = σt2St2dt. The diffusion term in (2b) involves σt2 instead
of (σth)2 since St follows the dynamics as specified in (1).
73
Substituting Θh
t in (2a) into (2b), we obtain

h h Γh St2 2
dVt = ∆t dSt + t [σt − (σth)2]dt + r(Vth − ∆h h
t St )dt + q∆t St dt.
2
(2c)

Let Πt denote the time-t value of the P&L, then dΠt over [t, t + dt]
is given by the sum of the three components:

dΠt = ∆h i h i
t (dSt + qSt dt) + r(Vt − ∆t St )dt − dVt .
We eliminate the hedge ratio term by substituting (2c) into the
above equation to obtain

h i h i Γh St2 2
dΠt = (dVt − dVt ) − r(Vt − Vt )dt − t [σt − (σth)2]dt
2
−r(T −t) d [ r(T −t) h ] Γh S 2
=e e (Vt − Vt ) dt − t t [σt2 − (σth)2]dt.
i
dt 2

74
The total P&L at maturity T is given by the accumulated sum of
the forward value of the differential P&L, where
∫ T
ΠT = er(T −t) dΠt
0
∫ T
−t) Γh St2 2
= (VTh − VTi ) − erT (V0h − V0i) − er(T t [σt − (σth)2] dt.
0 2
At maturity T , the option price is independent of the volatility pa-
rameter yielding
VTh = VTi = V (ST ).
The total P&L at maturity T is
∫ T
−t) Γh S 2
ΠT = erT [V (S0, 0; σ0i ) − V (S0, 0; σ0h)] + e r(T t t [(σ h )2 − σ 2 ] dt.
t t
0 2
Since the factor Γh 2
t St appears as cash term in ΠT , it is commonly
called the cash gamma or dollar gamma since it is expressible in
dollar value.

75
Remarks

1. The total P&L generated by delta hedging an option at hedge

volatility σth can be decomposed into two parts.

• The first component is the future value of the time-0 price

difference of the two options priced at the implied volatility
and the hedge volatility, respectively.

• The second component arises since the option is hedged at

the hedge volatility instead of the realized volatility. It is
equal to the future value of the weighted difference of the
expected variance (with reference to the hedge volatility) and
integrated realized variance, and the weight is half of the cash
Γh
i St2
gamma, 2 .

76
2. Suppose the trader is delta hedging the option at the implied
volatility, where σth = σti, then the total P&L becomes
∫ T iS2
Γ
ΠT = er(T −t) t t [(σti)2 − σt2] dt.
0 2
In this case, the total P&L is equal to the future value of the
weighted sum of the difference between the implied variance and
realized variance, where the weight factor is half of the cash
Γit St2
gamma, 2 . Usually, options are over priced, which means
σti > σt. As most options observe the property Γit > 0, we then
deduce that delta hedging strategy generates positive P&L.
Note that the variance exposure associated with the delta hedged
option is also dependent on the realized path of St. It is still
possible to obtain
∫ T
[(σti)2 − σt2] dt > 0,
0
while the P&L can be negative due to the path dependent factor
Γit St2
2 .

77
4.6 VIX and timer options

Characteristics of volatility (hidden stochastic process)

• Volatility is likely to grow when uncertainty and risk increase.
May serve as a proxy for market confidence – fear gauge.

• Volatilities appear to revert to the mean (non-linear drift).

– After a large volatility spike, the volatility can potentially
decrease rapidly.
– After a low volatility period, it may start to increase slowly.

• Volatility is often negatively correlated with stock or index level,

and tends to stay high after large downward moves.

• Stock options are impure in terms of volatility exposure. They

provide exposure to both direction of the stock price and its
volatility. If one hedges an option according to the Black-Scholes
prescription, then she can remove the exposure to the stock
price.

78
Mathematical derivation of VIX

VIX expresses volatility in percentage points. It is calculated as 100

times the square root of the expected 30-day variance of the rate
of return of the forward price of the S&P 500 index.
√
VIX = 100 forward price of realized cumulative variance
Suppose the forward price Ft of the S&P index under Q follows
dFt σt2
= σt dWt so that d ln Ft = − dt + σt dWt.
Ft 2
Here, the volatility function σt is assumed to be stochastic. Note
that Ft has zero drift rate under Q since Ft = e(r−q)(T −t)St and St
has the drift rate equals r − q under Q.

σt2
The drift term − dt in d ln Ft arises from the Ito lemma, where
2
1 2 2 2 ∂2 σt2
σt Ft (dWt) 2
ln Ft = − dt.
2 ∂Ft 2

79
Subtracting the two equations, we obtain the cumulative variance
over [0, T ] under continuous time model as follows:
∫ T [∫ ]
dFt σ 2 T dFt F
− d ln Ft = t dt, so σt2 dt = 2 − ln T .
Ft 2 0 0 Ft F0
[∫ ]
T
Our goal is to find the mathematical formula for EQ 0 σt2 dt , visu-
alized as the forward price of the realized cumulative variance over
[0, T ]. Recall that the T -maturity forward price of a risky asset St
is given by EQ[ST ]. In the current context, the underlying is the
∫
random cumulative variance 0T σt2 dt.

Note that
[∫ ] [∫ ]
T dFt T
E = EQ σt dWt = 0
0 Ft 0
so that
[∫ ] [ ]
T F
EQ σt2 dt = −2EQ ln T .
0 F0
The log contract with terminal payoff ln F T
F0 appears naturally. How
to relate the log contract with the usual call and put options?
80
Technical result For any twice-differentiable function f : R → R,
and any non-negative S∗, we have
∫ S
∗
f (ST ) = f (S∗) + f ′(S∗)(ST − S∗) + f ′′(K)(K − ST )+ dK
∫ ∞ 0
+ f ′′(K)(ST − K)+ dK.
S∗
The sum of the two integrals is the integral representation of the
remainder term in the Taylor expansion of f (ST ) up to the first
power term.

Proof

Assuming x0 > 0, we have

∫ ξ {
0
if ξ < x0
δ(x − x0) dx =
0 1
if ξ > x0
= 1{x0<ξ};
∫ ∞ {
1
if ξ < x0
δ(x − x0) dx =
ξ 0
if ξ > x0
= 1{x0>ξ}.
81
To establish the technical results, we perform repeated integration
by parts in order to generate the option payoff terms: (ST − K)+
and (K − ST )+. For any choice of S∗, we have
∫ S ∫ ∞
∗
f (ST ) = f (K)δ(ST − K) dK + f (K)δ(ST − K) dK
0 S∗
]S∗ ∫ S∗
= f (K)1{ST <K} − f ′(K)1{ST <K} dK
0 0∫
]∞ ∞
− f (K)1{ST ≥K} + f ′(K)1{ST ≥K} dK
S∗ S∗
[ ]S∗ ∫ S
∗
= f (S∗)1{ST <S∗} − f ′(K)(K − ST )+ + f ′′(K)(K − ST )+ dK
0 0∫
[ ]∞ ∞
′
+ f (S∗)1{ST ≥S∗} − f (K)(ST − K) + + f ′′(K)(ST − K)+ dK
S∗ S∗
= f (S∗) + f ′(S∗)[(ST − S∗)+ − (S∗ − ST )+]
∫ S∗ ∫ ∞
+ f ′′(K)(K − ST )+ dK + f ′′(K)(ST − K)+ dK
0 S∗

82
Rearranging the terms and rewriting FT and F0 for ST and S ∗, re-
spectively, we have
∫ F
f (FT ) − f (F0) = f ′(F0)(FT − F0) + f ′′(K)(K − FT )+dK
0

∫ ∞ 0
+ f ′′(K)(FT − K)+dK.
F0
Here, F0 is the time-0 forward price of the S&P index, an observable
known quantity. Taking f (FT ) = ln FT , we have
∫ ∫
∞ (FT − K)+
FT FT − F0 F0 (K − FT )+
ln = − 2
dK − 2
dK.
F0 F0 0 K F0 K
[ ]
FT
Recall EQ[FT ] = F0 so that EQ − 1 = 0. We then have
F0
[∫ ] [∫ ∫ ∞ ]
T F0 (K − FT )+ (FT − K)+
EQ σt2 dt = 2EQ 2
dK + 2
dK .
0 0 K F0 K

83
Note that
EQ[(K − FT )+] = erT putK ,
where putK is the time-0 price of a put on the S&P index with strike
K. At T , forward price and the underlying index coincide in value.

We then obtain
[∫ ] [∫ ∫ ∞ ]
T F0 putK callK
EQ 2
σt dt = 2erT dK + dK .
0 0 K 2 F0 K 2

The two terms represent continuum of puts whose strikes are below
F0 and calls whose strikes are above F0, respectively. They represent
out-of-the-money options with respect to the current forward price
F0. The CBOE’s choice is sensible since out-of-the-money options
tend to be more liquid contracts.

84
Actual implementation of VIX formula

In the actual implementation of replication formula, one has to face

with availability of options with discrete number of strikes. Also, on-
ly a finite range of strikes of traded options are available. In addition,
the option with strike that exactly equals F0 is not available in gen-
eral. In the calculation formula for VIX, the CBOE procedure takes
the out-of-the-money options within the bounded interval [KL, KU ],
and choose K0 to be the closest listed strike below F0. The out-of-
the-money options include all listed put options with strikes at or
below K0, and all listed call options with strike at or above K0.

Recall
∫ ∫
∞ (FT − K)+
FT FT − K0 K0 (K − FT )+
ln = − 2
dK − 2
dK
K0 K0 0 K K0 K
so that
[ ] [ ]
FT FT F0
EQ ln = EQ ln + ln
K0 F0 K0
∫ K rT ∫ ∞ rT
F0 0 e putK e callK
= −1− 2
dK − 2
dK
K0 0 K K0 K
85
We use the approximation
( ) ( )2
F0 F0 1 F0
ln ≈ −1 − −1 ,
K0 K0 2 K0
so that
{[∫ ∫ ∞ ] [∫ ∫ ∞ ]}
F0 putK callK K 0 putK callK
2erT 2
dK + 2
dK − 2
dK + 2
dK
0 K F0 K 0 K K0 K
[ ( )] ( )2
rT F0 F0 F0
= −2e ln − −1 ≈− −1 .
K0 K0 K0

Finally, we obtain
[∫ ]
T
EQ σt2 dt
0
[∫ ∫ ∞ ] ( )2
K0 putK callK F0
= 2erT 2
dK + 2
dK − −1 .
0 K K0 K K0

86
Finally, we multiply the above expected realized cumulative vari-
365
ance by the product of the annualization conversion factor and
30
percentage point factor 100 to obtain

V IXt2
 ( )2 
 2 ∑ ∆Ki 1 F0 
= 1002 e r(30/365) Q(K ) −
i −1 .
 30/365 2 30/365 K0 
i Ki

Here, K0 is the first strike below the forward index level F0, Q(Ki)
is the time-t out-of-the-money option price with strike Ki.

87
Summary: Sources of errors

1. Truncation error
We choose a bounded truncation interval [KL, KU ] of strikes in-
stead of the theoretical interval [0, ∞]. Note that the CBOE may
add new strikes as the underlying S&P 500 index moves. The
added strikes expand the truncation interval. During the period
of frequent spikes, the expansion of interval can be frequent and
significant.

2. Discretization error
The continuous integration of option prices with respect to con-
tinuum of strikes is approximated by the sum of weighted out-
of-the-money option prices.

88
3. Approximation error arising from replacing F0 by K0
F0
The logarithm term ln K is approximated by the Taylor expan-
0
F0
sion in powers of K − 1 up to the quadratic term.
0

4. Linear maturity interpolation

The VIX is calculated based on options with a fixed 30-day
maturity. The CBOE calculation procedure finds two maturities
T1 and T2 that bracket the required 30-day maturity. The VIX2
based on 30-day maturity is computed by applying the linear
maturity interpolation. Errors are introduced since the model
free VIX2 is a nonlinear function of maturity.

89
Plot of VIX for the past two decades

Volatility is bursty in nature: tendency of high volatility to come in

bursts. VIX spiked in 2008: period of financial tsunami.

90
Plot of VIX in 2018

Volatility is negatively correlated to the S&P index return. Fear of

inflation that might prompt the Federal Reserve to raise interest
rate in February. Deepening of the trade war between China and
US pushes up the VIX since September. Reversion of the long-term
and short-term yield curves shocked the market on December 4,
2018.
91
VIX index and SPX

S&P 500 index and VIX are negatively correlated. Growth of SPX
over the last 10 years since 2008 financial tsunami with low level of
VIX.

92
VIX derivatives

Investors directly invest in volatility as an asset class by mean of

VIX derivatives. CBOE began trading in futures on VIX on March
26, 2004 and European options on February 24, 2006. Both are
cash settled. The contract multiplier for each VIX futures contract
is $1000, while that of VIX option is $100.

VIX derivatives can be used to hedge the risks of investments in

the S&P 500 index and/or achieve exposure to S&P 500 volatility
without having to delta hedge their S&P 500 option positions with
the stock index.

93
Trading volumes of VIX futures

On June 24, 2016, in reaction to the Brexit referendum, over

721, 000 VIX futures contracts changed hands and on November
9, 2016, the volume was 644, 892 in reaction to the surprise out-
come of the US election.
94
Increasing popularity of VIX futures and options

Year-to-date through the end of July, 2017, average daily volume in

VIX futures was 283,342 contracts, 20 percent ahead of the same
period a year ago.

In VIX options at CBOE, a reported 2,562,477 contracts traded

on August 10, 2017 , surpassing the previous single-day record of
2,382,752 contracts on February 3, 2014.

Year-to-date through the end of July, 2017, average daily volume in

VIX options was 687,181 contracts, 11 percent ahead of the same
period a year ago.

95
VVIX and modeling of dynamics of VIX

Every asset class deserves its own volatility index, including VIX
itself. The VVIX Index is an indicator of the expected volatility of
the 30-day forward price of the VIX. This volatility drives nearby
VIX option prices.

Approximate fair values of VIX futures prices and their standard

deviations can be inferred from the VVIX term structure.

Direct modeling of VIX dynamics (Bates, 2012)

√
Q
dVt = κV [θ(t) − V (t)]dt + σ V (t) dWV (t)
√
Q
dθt = κθ [θ − θ(t)]dt + σ θ(t) dWθ (t)

96
Timer options

A standard timer call option can be viewed as a call option with ran-
dom maturity which depends on the time needed for a pre-specified
variance budget to be fully consumed. A variance budget is calculat-
ed as the target volatility squared, multiplied by the target maturity.

Suppose the asset price is monitored at tj = j∆t, j = 0, 1, . . . , n.

The annualized realized variance for the period [0, T ] is defined as
n−1 ( )2
1 ∑ Si+1
bT
σ 2 = ln , T = n∆t.
(n − 1)∆t i=0 Si
Here, ∆t is in units of year. The realized log returns over successive
Sti+1
time intervals ln , i = 0, 1, . . . , n − 1, are considered as sampling
Sti
of the independent Brownian increments with variance rate σT2 and
time interval ∆t. The factor n − 1 is related to the Bessel correction
due to the loss of one degree of freedom.

97
The realized variance over time period [0, T ] is
n−1 ( )2
∑ Si+1
bT
VT = n∆tσ 2 ≈ ln .
i=0 Si
The investor specifies a variance budget

B = σ02T0,
where T0 is the estimated investment horizon and σ0 is the forecast
volatility during the investing period.

The discretely monitored finite-maturity timer call with random ma-

turity pays max(Stj − X, 0) at the first time tj when the discrete
realized variance exceeds B. That is,
 ( )2 
 k−1
∑ Si+1 
tj = min k > 0, ln ≥B .
 Si 
i=0

Suppose the variance budget is never met within [0, T ], the timer
call option expires at the mandated maximum expiration date T
with terminal payoff max(ST − X, 0).
98
Option strategy without paying the volatility risk premium

The price of a vanilla call option is determined by the level of im-

plied volatility quoted in the market (as well as maturity and strike
price). The level of impled volatility is often higher than the real-
ized volatility (risk premium due to the uncertainty of future market
direction).

“High implied volatility means call options are often overpriced. In

the timer option, the investor only pays the real cost of the call and
does not suffer from high implied volatility.”

99
The first trade was in April 2007 on HSBC with a June expiry.

• The implied volatility on the plain vanilla call was slightly above
15%, but the client sets a target volatility level of 12%, a little
higher than the prevailing realized volatility level of around 10%.

• The premium of the timer call has 20% discount compared to

the vanilla call counterpart. Here, the vanilla call price is calcu-
lated based on T0 and implied volatility of 15%.

• The realized volatility has been around 9.5% since the inception
of the trade. The maturity of the timer call is 60% longer than
the original vanilla call.

100
Hedge against stock price drop at a lower cost

Exploit the negative correlation that generally exists between stock

return and realized volatility.

An investor wants to hedge against stock price drop by buying a

timer put.

• Suppose the stock price goes down, the timer put expires earlier
since the realized volatility increases. The compensation of the
stock price loss arrives sooner and more timely.

• Suppose the stock price goes up, the realized volatility decreases
giving longer life of the timer put. The insurance protection
provided by the timer put lasts longer. Hence, the hedging cost
is lower.

101
How to gain from the bullish view of the market?

Long a timer call and short a vanilla call, both with the same strike
K. Usually, stock price and volatility are negatively correlated. The
implied volatility in the market is too high currently, and subsequent
realized volatility will be less than that implied in the market.

The strategy is to set the volatility target to be below the current

implied volatility level (price of timer call would be less than that of
the comparative vanilla call).

If the stock shifts higher over the period, the tenor of the timer
option would be longer. The timer call will expire later than the
plain vanilla call. The value of the timer call at expiry of the vanilla
call is the sum of the intrinsic value ST − K and time value.

102
Pricing of discretely monitored finite-maturity timer options

Define the
∫
continuous integrated variance (quadratic variation) to
t
be It = Vs ds.
We use It as a proxy of the discrete realized
0
variance for the monitoring of the first hitting time and let ∆t be
the uniform time interval between successive monitoring dates. We
define τB to be
  
   
τB = min j Itj ≥ B ∆t.
  

This approximation does not introduce a noticeable error for daily

monitored timer options. Note that

C0(X0, I0, V0) = E0[e−r(T ∧τB ) max(ST ∧τB − K, 0)]

= E0[e−rT max(ST − K, 0)1{τB >T }
+ e−rτB max(SτB − K, 0)1{τB ≤T }],
where K is the strike price and r is the constant interest rate.

103
Decomposition into a portfolio of timerlets

The event {τB > t} is equivalent to {It < B}. Note that τB = tj+1
if and only if Itj < B and Itj+1 ≥ B. Therefore, we have

−1
N∪
{τB ≤ T } = {Itj < B, Itj+1 ≥ B}.
j=0

The price of a finite-maturity discrete timer call option can be con-

veniently computed by decomposing it into a portfolio of timerlets
as follows

C0 = E0[ e−rT max(ST − K, 0)1{IT <B}]


−1
N∑ (
+ E0  e−rtj+1 max(Stj+1 − K, 0)1{It <B}
j
j=0
)]
− max(Stj+1 − K, 0)1{It <B} .
j+1

The challenge is the modeling of the joint processes of {Stj+1 , Itj }

and {Stj+1 , Itj+1 } under a stochastic volatility model.
104
4.7 Guaranteed minimum withdrawal benefit

Product Nature

• Variable annuities — deferred annuities that are fund-linked.

• The single lump sum paid by the policyholder at initiation is

invested in a portfolio of funds chosen by the policyholder —
equity participation.

• The GMWB allows the policyholder to withdraw funds on an

annual or semi-annual basis until the entire principal is returned.

• In 2004, 69% of all variable annuity contracts sold in the US

included the GMWB option.

105
Numerical example
• Let the initial fund value be $100, 000 and the withdrawal rate
be fixed at 7% per annum. Suppose the investment account
earns 10% in the first two years but earns returns of −60% in
each of the next three years.

Year Rate of Fund value Amount Fund value Guaranteed

return before withdrawn after withdrawals
during withdrawals withdrawals remaining
the year balance
1 10% 110, 000 7, 000 103, 000 93, 000
2 10% 113, 300 7, 000 106, 300 86, 000
3 −60% 42, 520 7, 000 35, 520 79, 000
4 −60% 14, 208 7, 000 7, 208 72, 000
5 −60% 2, 883 7, 000 0 65, 000

• At the end of year five before any withdrawal, the fund value
$2, 883 is not enough to cover the annual withdrawal payment
of $7, 000.

106
The guarantee kicks in when the fund is non-performing

The value of the fund is set to be zero and the policyholder’s 10

remaining withdrawal payments are financed under the writer’s guar-
antee. The policyholder’s income stream of annual withdrawals is
protected irrespective of the market performance. The investmen-
t account balance will have shrunk to zero before the principal is
repaid and will remain there.

Good performance of the fund

If the market does well, then there will be funds left at policy’s
maturity.

107
Numerical example revisited

Suppose the initial lump sum investment of $100, 000 is used to

purchase 100 units of the mutual fund, so each unit worths $1, 000.

• After the first year, the rate of return is 10% so each unit is
$1, 100. The annual guaranteed withdrawal of $7, 000 represents
$7, 000/$1, 100 = 6.364 units. The remaining number of units
of the mutual fund is 100 − 6.364 = 93.636 units.

• After the second year, there is another rate of return of 10%,

so each unit of the mutual fund worths $1, 210. The withdraw-
al of $7, 000 represents $7, 000/$1, 210 = 5.785 units, so the
remaining number of units = 87.851.

108
• There is a negative rate of return of 60% in the third year, so
each unit of the mutual fund worths $484. The withdrawal of
$7, 000 represents $7, 000/$484 = 14.463 units, so the remain-
ing number of units = 73.388.

• Depending on the performance of the mutual fund, there may

be certain number of units remaining if the fund is performing
or perhaps no unit is left if it comes to the worst senario.

– In the former case, the holder receives the guaranteed total

withdrawal amount of $100, 000 (neglecting time value) plus
the remaining units of mutual funds held at maturity.

– If the mutual fund is non-performing, then the total with-

drawal amount of $100, 000 over the whole policy life is guar-
anteed.

109
How is the benefit funded?

• Proportional fee on the investment account value

— for a contract with a 7% withdrawal allowance, a typical

charge is around 40 to 50 basis points of proportional fee on
the investment account value.

• GMWB can also be seen as a guaranteed stream of 7% per

annum plus a call option on the terminal investment account
value WT , WT ≥ 0. The strike price of the call is zero.

110
Static withdrawal model – continuous version

• The withdrawal rate G (dollar per annum) is fixed throughout

the life of the policy.

• When the investment account value Wt ever reaches 0, it stays

at this value thereafter (absorbing barrier).

τ = inf{t : Wt = 0}, τ is the first passage time of hitting 0.

Under the risk neutral measure Q, the dynamics of Wt is gov-
erned by

dWt = (r − α)Wt dt + σWt dBt − G dt, t<τ

Wt = 0, t≥τ
W0 = w0,
where α is the proportional annual fee charge on the investment
account as the withdrawal allowance.
[∫ ]
T
policy value = EQ Ge−ru du + EQ[e−rT WT ].
0

111
Surrogate unrestricted process

To enhance analytic tractability, the restricted account value process

Wt is replaced by a surrogate unrestricted process W f at the expense
t
of introducing optionality in the terminal payoff (zero strike call
payoff). Consider the modified unrestricted stochastic process:
f = (r − α)W
dW f dt − G dt + σ W
f dB , t > 0,
t t t t
f
W0 = w0 .

f , we obtain
Solving for Wt
( ∫ t )
f =X w −G 1
W t t 0 du
0 Xu
where
( 2
)
σ
r−α+ 2 t+σBt
Xt = e .
The solution is the unit exponential Brownian motion Xt multiplied
by the number of units remaining after depletion by withdrawals.

112
Financial interpretation

Take the initial value of one unit of the fund to be unity for con-
venience. Here, Xt represents the corresponding fund value process
with X0 = 1.

• The number of units acquired at initiation is w0. The total

∫
number of units withdrawn over (0, t] is given by G 0t X1u du.

• Under the unrestricted process assumption, Wf may become neg-

t
ative when the number of units withdrawn exceeds w0. However,
in the actual case, Wt stays at the absorbing state of zero value
once the number of unit withdrawn hits w0.

113
 f > 0 for t ≤ T or W
Either W f remains negative once W reaches
t T t
the negative region at some earlier time prior to T .

114
 f > 0.
Lemma τ0 > T if and only if WT

=⇒ part. Suppose τ0 > T , then by the definition of the first passage

f > 0.
time, we have WT

⇐= part. Recall that

( ∫ t )
f =X w − G
Wt t 0 du
0 Xu
so that
∫ t
f >0 G
Wt if and only if du < w0.
0 Xu

Suppose Wf > 0, this implies that the number of units withdrawn

T
∫
by time T = 0T XGu du < w0. Since Xu ≥ 0, for any t < T , we have
∫ t ∫ T
G G
number of units withdrawn by time t = du ≤ du < w0.
0 Xu 0 Xu
f > 0, then W
Hence, if W f > 0 for any t < T .
T t

115
Intuition of the dynamics

Once the process W f becomes negative, it will never return to the

t
f increases from below back
positive region. This is because when W t
to the zero level, only the drift term −G dt survives. This always
f back into the negative region.
pulls W t

f
Relation between WT and WT

Note that WT = 0 if and only if τ ≤ T . We then have

f 1 f f
WT = WT {τ >T } = WT 1{W
fT >0} = max(WT , 0).

Optionality in the terminal payoff

The terminal payoff from the investment account becomes

( ∫ T )+
f , 0) = GX w0 1
max(WT T − dx , x+ = max(x, 0).
G 0 Xu

116
 ∫
G t 1
Defining Ut = du, we obtain
w0 0 Xu

EQ[e−rT W
f +] = w E [e−rT X (1 − U )+].
T 0 Q T T

Here, Ut represents the fraction of units withdrawn up to time t,

which captures the path dependence of the depletion process of
the investment account due to the continuous withdrawal process.
Lastly, we have
[∫ ]
T [ ]
policy value = EQ Ge −ru du + w0EQ e −rT +
XT (1 − UT ) .
0

The pricing issue is to find the fair value for the participating fee
rate α such that the initial policy value equals the lump sum paid
upfront by the policyholder so that the policy contract is fair to
both counterparties.

117
4.8 Transaction costs models

How to construct the hedging strategy that best replicates the pay-
off of a derivative security in the presence of transaction costs?

Recall that one can create a portfolio containing ∆ units of the

underlying asset and money market account which replicates the
payoff of the option. By the portfolio replication argument, the
value of an option is equal to the initial cost of setting up the
replicating portfolio which mimics the payoff of the option.

Leland proposes a modification to the Black-Scholes model where

the portfolio is adjusted at regular time intervals. His model assumes
proportional transaction costs where the costs in buying and selling
the asset are proportional to the monetary value of the transaction.

118
Let k denote the round trip transaction cost per unit dollar of trans-
action. Suppose α units of assets are bought (α > 0) or sold (α < 0)
k
at the price S, then the transaction cost is given by |α|S in either
2
buying or selling.

We consider a hedged portfolio of the writer of the option, where

he is shorting one unit of option and long holding ∆ units of the
underlying asset. The value of this hedged portfolio at time t is
given by
Π(t) = −V (S, t) + ∆S,
where V (S, t) is the value of the option and S is the asset price at
time t. Let δt denote the fixed and small finite time interval between
successive rebalancing of the portfolio.

119
After the small time interval δt, the change in value of the portfolio
is
k
δΠ = −δV + ∆ δS − |δ∆|S,
2
where δS is the change in asset price and δ∆ is the change in the
number of units of asset held in the portfolio.

A cautious reader may doubt why the proportional transaction cost

k
term − |δ∆|S appears in δΠ while the term S δ∆ is missing.
2

• The transaction cost term represents the single trip transaction

cost paid due to rebalancing of the position in the underlying
asset.

• By following the “pragmatic” approach used by Black and Sc-

holes (1973), the number of units ∆ is taken to be instanta-
neously constant.

120
By Ito’s lemma, the change in option value in time δt to leading
orders is given by
( )
∂V ∂V σ2 ∂ 2V
δV ≈ δS + + S2 2 δt.
∂S ∂t 2 ∂S
∂V
In order to cancel the stochastic terms, one chooses ∆ = . The
∂S
change in the number of units of asset in time δt is given by
∂V ∂V
δ∆ = (S + δS, t + δt) − (S, t).
∂S ∂S
From the dynamics of St, we observe δS ≈ ρSδt + σSδZ. Note that
√
δZ ≈ O( δt) is dominant over ρSδt, so the leading order of |δ∆| is
found to be
   
 ∂ 2V   2 V 
   ∂
|δ∆| ≈  2  |δS| ≈ σS  2  |δZ|.
 ∂S   ∂S 

121
 √
Formally, we may treat δZ as xe δt, where x
e is the standard normal
variable. The expectation of the reflected Brownian motion |δZ| is
given by
(∫ )
∞1 −t 2 /2 √
E(|δZ|) = 2 t√ e dt δt
0 2π
(∫ )
∞ 1
−u
√
= 2 √ e du δt, u = t2/2,
0 2π
√
2√
= δt.
π
The risk of loss associated with transaction costs is investor-specific,
so it should not be compensated. By the Capital Asset Pricing
Model, the hedged portfolio should earn an expected rate of return
same as that of a riskless asset. This gives
( ) √  
∂V σ2 ∂ 2V k 2  ∂ V  √
 2
E[δΠ] = −
− S2 2 δt − σS 2   δt
∂t 2 ∂S 2 π ∂S 
 2
( )
∂V
= r −V + S δt.
∂S

122
By putting all the above results together, the above equation can
be rewritten as
 √   ( )
−
∂V σ 2
2 ∂ 2V σ 2
2 2 k  ∂ 2V  ∂V
− S − S √   δt = r −V + S δt.
∂t 2 ∂S 2 2 
π σ δt ∂S 2  ∂S
√ ( )
2 k
If we define the Leland number to be Le = √ , we obtain
π σ δt
 
 2 V 
∂V σ 2 2 ∂ 2V σ2 
2 ∂  + rS
∂V
+ S 2
+ Le S  2  − rV = 0.
∂t 2 ∂S 2  ∂S  ∂S

The Leland number is related to the ratio of k and standard devi-

ation of the asset price process over the rebalancing time interval
δt.

123
  
σ2  ∂ 2V 
 
In the proportional transaction costs model, the term Le S 2  2 
2  ∂S 
is in general non-linear, except when the comparative static Γ =
∂ 2V
2
does not change sign for all S. The transaction cost term
∂S
∂∆
is dependent on Γ, where Γ = measures the sensitivity of the
∂S
hedge ratio ∆ to the underlying asset price S.

One may rewrite the equation into the form that resembles the
Black-Scholes equation
∂V e 2 2 ∂ 2V
σ ∂V
+ S 2
+ rS − rV = 0,
∂t 2 ∂S ∂S
where the modified volatility under transaction costs is given by

e 2 = σ 2[1 + Le sign(Γ)].
σ
e2
The governing equation becomes mathematically ill-posed when σ
becomes negative. This occurs when Γ < 0 and Le > 1.

124
Modified volatility

It is known that Γ is always positive for the vanilla European call

and put options in the absence of transaction costs. If we postulate
the same sign behavior for Γ in the presence of transaction costs,
then σe 2 = σ 2(1 + Le) > σ 2.

The governing equation then becomes linear under the above as-
sumption so that the Black-Scholes formulas become applicable
e is now used as the volatility
except that the modified volatility σ
parameter.

We can deduce V (S, t) to be an increasing function of Le since we

expect a higher option value for a high value of modified volatility.
Financially speaking, the more frequent the rebalancing (smaller δt)
the higher the transaction costs and so the writer of an option
should charge higher for the price of the option.

125
 e ) and V (S, t; σ) denote the option values obtained from
Let V (S, t; σ
the Black-Scholes formula with volatility values σ e and σ, respectively.
The total transaction costs associated with the replicating strategy
is then given by
T = V (S, t; σ
e ) − V (S, t; σ).

When Le is small, T can be approximated by

∂V
T ≈ e − σ).
(σ
∂σ
[ ]
1/2 Le k
e = σ[1+Le sign(Γ)]
Since σ ≈σ 1+ e −σ ≈ √
sign(Γ) , so σ .
2 2πδt
2
√ d
− 21
∂V S T − te
Note that = √ is the same for both call and put op-
∂σ 2π
tions. For Le ≪ 1, the total transaction costs for either a call or a
put is approximately given by
d2 √
kSe− 2
1
T −t
T ≈ .
2π δt

126