Stochastic Volatility Modelling

Bruno Dupire
Nice 14/02/03

Structure of the talk

1.Model review
2.Forward equations
3.Forward volatility

4.Arbitrageable & admissible smile dynamics

Bruno Dupire

Model Requirements
Has to fit static/current data:
Spot Price
Interest Rate Structure
Implied Volatility Surface

Should fit dynamics of:

Spot Price (Realistic Dynamics)

Volatility surface when prices move
Interest Rates (possibly)

Has to be

Understandable
In line with the actual hedge
Easy to implement
Bruno Dupire

A Brief History of Volatility

(1)

dSt dWt Q

dSt
r dt dWt Q
St

: Bachelier 1900

: Black-Scholes 1973

dSt
r (t ) dt (t ) dWt Q
St

: Merton 1973

dS t
(r k ) dt dWt Q dq
St

: Merton 1976

dSt
Q

r
dt

dW
t
t
S
t

: Hull&White 1987

d 2 a (V V )dt dZ
t
L
t

Bruno Dupire

A Brief History of Volatility

(2)

dS t
t dWt Q
St

2 LT t
d 2
dt dZ tQ
2
T
2
t

Dupire 1992, arbitrage mod

which fits term structure of
volatility given by log contra

dSt
r (t ) dt ( S , t ) dWt Q Dupire 1993, minimal mode

St

C
C
rK
2 K , T 2 T 2 K
2 C K ,T
K
K 2

to fit current volatility surfac

Bruno Dupire

A Brief History of Volatility

(3)
dSt
S
t

r dt t dWt

formulae.
2
d

2
2
t

b
(

)dt dZ t

t
t2

Heston 1993,
semi-analytical

dVK ,T K ,T dt bK ,T dZ tQ

VK ,T

: instantaneous forward variance

conditional to

ST K

Dupire 1996 (UTV),

Derman 1997,
stochastic volatility

model
which fits current
volatility
surface HJM
treatment.

Bruno Dupire

A Brief History of Volatility

(4)
Bates 1996, Heston + Jumps:
dSt
S r dt t dZ t dq
t

2
d

2
2
t

b
(

t ) dt dWt
t2
Local volatility + stochastic volatility:
Markov specification of UTV
Reech Capital Model: f is quadratic
SABR: f is a power function
dSt
r dt t f S , t dZ tQ
St
Bruno Dupire

A Brief History of Volatility

(5)
Lvy Processes
Stochastic clock:

VG (Variance Gamma) Model:

BM taken at random time (t)
CGMY model:
same, with integrated square root process

Jumps in volatility
Path dependent volatility
Implied volatility modelling
Incorporate stochastic interest rates
n dimensional dynamics of
n assets stochastic correlation

Bruno Dupire

Forward Equations (1)

BWD Equation:
price of one option CK0,T0 for different (S,t)
FWD Equation:
price of all options CK,T for current (S0,t0)
Advantage of FWD equation:
If local volatilities known, fast computation
of implied volatility surface,
If current implied volatility surface known,
extraction of local volatilities,
Understanding of forward volatilities
and how to lock them.

Bruno Dupire

Forward Equations (2)

Several ways to obtain them:

Fokker-Planck equation:
Integrate twice Kolmogorov Forward Equat
Tanaka formula:
Expectation of local time
Replication
Replication portfolio gives a much
more financial insight

Bruno Dupire

FWD Equation: dS/S = (S,t) dW

Define CS

T
K ,T

C K ,T T C K ,T

C K ,T T

C K ,T

ST

CS KT,T at T

dT
dT/2

2 K ,T 2
K K ,T
2

T 0

ST

ST
K

C 2 K , T 2 2C
Equating prices at t0:

K
T
2
K 2
Bruno Dupire

FWD Equation: dS/S = r dt + (S,t)

dW
TV

ST Ke rT

IV

ST K

Value + Intrinsic Value

CS KT,T at TTime

C K ,T T

(Strike Convexity)

C K ,T

Ke rT K

ST

2 K ,T 2
K K ,T
2

T 0
TV

(Interest on Strike)

IV

rK
rKDig K ,T

Ke rT K

ST

C 2 K , T 2 2C
C
Equating prices at t0:

rK
T
2
K 2
K
Bruno Dupire

ST

FWD Equation: dS/S = (r-d) dt + (S,t)

dW
ST Ke rT
TV IV

CS KT,T at T TV +

ST e dT Ke rT

C K ,T T
Ke

d r T

C K ,T

Interests on K
Dividends on S

ST

2 K ,T 2
K K ,T
2

T 0

r dK
Ke ( d r )T K

(r d ) K Dig K ,T

ST

d C K ,T

Equating prices at t0:

d C K ,T

C 2 K , T 2 2C
C

d
K
d C
2
T
2
K
K
Bruno Dupire

ST

Stripping Formula
C 2 K , T K 2 2C
C

d
K
d C
2
T
2
K
K

If K , T

S0 ,t0
known, quick computation C
ofK ,Tall

If all C K ,T S0 ,t0

known:

C
C
2
r d K
dC
T
K
K ,T
2
C
K2
K 2

Local volatilities extracted from vanilla prices

and used to price exotics.
Bruno Dupire

FWD Equation for Americans

Local Volatility model in the continuation region

if
if

2S 2
Pt
PSS r SPS P 0
2
,
Pt PT

S,t S
, SP KP
P

S
K
S,t t

2
2
S
P

K
PKK
SS

In Black Scholes
S, t

2 2
PT
K PKK rKPK
2
in the continuation region (Carr 2002)
Same FWD Eq. than in the European case (+Boundary
Condition)
Bruno Dupire

FWD Equation for Americans in

LVM

If

BS

LVM

BWD Eq.

Same for
Eur/Am

Same for
Eur/Am

FWD Eq.

Same for
Eur/Am

dS
rdt S , t dW
S

K ,T 2
PT
K PKK rKPK
2
2

do we have
for Americans?

Bruno Dupire

Counterexample
Assume S , t f t ot t1 ,t 2

for T t1 ,t 2
,
FWD Eq. for Europeans:
PT rKPK 0

0
0

Am
When indeedPT 0

Bruno Dupire

t1

t2

Local Volatility Model

dS
r d dt S , t :dW
Simplest model which fits the smil
S
Models of
interest
Risk
Neutral
Dynamics

Smile
Calibrated

1D
Diffusion

Local Volatility
Model
Bruno Dupire

FWD Equation when dS/S = t dW

1
2

CS
, K ,T

When T 0

2
2
gives at T:
T K K

2
C 1

C
Equating Prices at t0: E T2 | ST K K 2 2
T 2
K
C
E T2 | ST K 2 T2 2 K , T from local vol mode
C
K2
K 2

Any stochastic volatility model which matches

the initial smiles has to satisfy:

E T2 | ST K 2 K , T

Bruno Dupire

Convexity Bias
dS t dW

2
d

t dZ
W , Z 0

E t2 | St K 02 ?

NO! only E t2 02

E t2 | S t K

02
S0

likely
to be high ifSt S 0 or St S 0
t
Bruno Dupire

Impact on Models

Risk Neutral drift for instantaneous forward varianc

Markov Model:
dS
S, t
f S , t t dW
fits initial smile with local vols
S

2 S,t
f S,t
E[ t2 | S t S ]

Bruno Dupire

Locking FWD Variance | ST=K

2

CS KT,T at T

ST

Define C

K ,T

and C K ,,TT

,T
K ,T

at T

Constant
curvature

C K ',T dK '

2
C K ,T T C K ,T

K K K
22 K 2

T 0
1
K

ST

4
12 K 2
4
K

Bruno Dupire

ST

1
K

ST

Conditional Variance Swap

K ,T

PortfolioPF

Dig K ,T Dig K ,T
2

2
2 CS K ,T K , T
K
2
Local Vol stripped
from initial smile

Initial Cost = 0

K ,T

PF

2
2
K ,T

at T: T
2

ST if K , K
if ST K , K

Replicates the pay-off of a conditional variance swa

E T2 ST K 2 K , T

It is possible to lock this value

Bruno Dupire

Deterministic future smiles

It is not possible to prescribe just any future smile
If deterministic, one must have

C K ,T2 S 0 , t0 S 0 , t0 , S , T1 C K ,T2 S , T1 dS

Not satisfied in general

K
S0

t0

Bruno Dupire

T1

T2

Det. Fut. smiles & no jumps => = FWD

smile
2
K , T , K K , T T
imp
If S , t , K , T / VK ,T S , t 2 K , T lim
K 0

T 0

stripped from SmileS.t

Then, there exists a 2 step arbitrage:

DefinePL 2 K , T V S , t 2C S , t , K , T S0
t

At t0 : Sell
At t:

K ,T

PLt Dig S ,t Dig S ,t

if S t S , S

K 2

buy

t0
2
CSK,T , sell
2
K

t
2

K , T K ,T

gives a premium = PLt at t, no loss at

T

VK ,T S , t

S , t VK ,T S 0 , t0 2 K , T
Conclusion:
independent of
Brunofrom
Dupire

initial smile

Consequence of det. future smiles

Sticky Strike assumption: Each (K,T) has a fixed
impl(K,T) independent of (S,t)
Sticky Delta assumption: impl(K,T) depends
only on moneyness and residual maturity
In the absence of jumps,
- Sticky Strike is arbitrageable
- Sticky is (even more) arbitrageable

Bruno Dupire

Example of arbitrage with Sticky

Strike
Each CK,T lives in its Black-Scholes (impl(K,T))world

C1 C K1 ,T1

C2 C K 2 ,T2

assume 1 2

P&L of Delta hedge position over t:

PL C1 12 S 2 1S 2t 1
PL C2

1
2

2 S 2t 2

12 2 2
PL 1C2 2C1
S 1 22 t 0
2
no , free

1C2

If no jump

Bruno Dupire

2C1

S t1

S t t

From Local Vols to implied

Vols
: model with local vols , 0 with 0

1 T 2
C C 0 S , t 02 S , t 0 S , t S , t dS dt
2 0 0

Where 0= of C under 0, and =RN density under

2

Implied vol solution of0

2 2

S 0dSdt
2
S
0dSdt

2
implied
average of 2 weighted by S20

(equivalent to density of Brownian Bridge from (S,t0) to

(K,T))

Bruno Dupire

Market Model of Implied

Volatility

Implied volatilities are directly observable

Can we model directly their dynamics?
r 0

dS
S dW1

d dt u dW u dW
1
1
2
2

C K ,T
where is the implied volatility of a given
Condition on
dynamics?

Bruno Dupire

Drift Condition
Apply Itos lemma to
C
Cancel the drift term

S , , t
C S , , t

Rewrite derivatives of

gives the condition that the drift

satisfy.

dof

must

For short T,

2
2
2 u1 X u2 X

where

X ln K ln S

C S , ,t
t T

(Short Skew Condition :SSC)

Bruno Dupire

Local Volatility Model Case

det. function of S , t
S S
dt
dW1

and d
SSC:

det. function of
S, t

S S
u1

u,2 0

u1 X 1
SX
S

1
SX

solved by

X
K
du
S u u,0
Bruno Dupire

Dual Equation
2 2
The stripping formula

CT
K 2C KK

:
can be expressed in terms
of
When T 0

K
1
KX

X
solved by K
du
S u u,0

Bruno Dupire

Large Deviation
Interpretation
du
K

The important quantityis

S

u u,0
x du
y x
then
x0 a u

dx a x dW

If

satisfies:

dy dt dW

du

xt K Wt y K
and

du
ln K ln S
K
du
u u ,0
S u u,0

Bruno Dupire

Smile dynamics: Local Vol Model 1

Consider, for one maturity, the smiles associated
to 3 initial spot values
Skew case
Local vols

Smile S
Smile S 0
Smile S

S S0 S

-ATM short term implied follows the local vols

-Similar skews
Bruno Dupire

Smile dynamics: Local Vol Model 2

Pure Smile case
Local vols

Smile S
Smile S
Smile S 0

S0

-ATM short term implied follows the local vols

-Skew can change sign

Bruno Dupire

Smile dynamics: Stoch Vol Model 1

Skew case (<0)

Local vols

Smile S
Smile S 0
Smile S

S S0

- ATM short term implied still follows the local vols

ST K 2 K , T

- Similar skews as local vol model for short horizons

- Common mistake when computing the smile for an
spot: just change S0 forgetting the conditioning on
if S : S0 S+ where is the new ?
Bruno Dupire

Smile dynamics: Stoch Vol Model 2

Pure smile case

Local vols

Smile S

Smile S

Smile S0

S0

- ATM short term implied follows the local vols

- Future skews quite flat, different from local vol mo
-Again, do not forget conditioning of vol by S
Bruno Dupire

Smile dynamics: Jump Model

Skew case
Local vols

Smile S

Smile S 0

Smile S

S0 S

- ATM short term implied constant (does not follows

the local vols)
- Constant skew
- Sticky Delta model
Bruno Dupire

Smile dynamics: Jump Model

Pure smile case
Local vols

Smile S

Smile S 0 Smile S

S0

- ATM short term implied constant (does not follows

the local vols)
- Constant skew
- Sticky Delta model
Bruno Dupire

Spot dependency
2 ways to generate skew in a stochastic vol
1) t xt f S , t , W , Z 0
model
2) W , Z 0

S0

ST

S0

ST

-Mostly equivalent: similar (St,t) patterns,

similar future
evolutions
-1) more flexible (and arbitrary!) than 2)
-For short horizons: stoch vol model local vol
model + independent noise on vol.
Bruno Dupire

Conclusion
Local vols are conditional forward
values
that can be locked
All stochastic volatility models have to
respect
the local vols in expectation
Without jumps, the only non
arbitrageable
deterministic future smiles are the
forward smile from the local vol model
In particular, without jump, sticky
strikes and
sticky delta are arbitrageable
Bruno Dupire