Advanced Derivatives Modelling

Stochastic Volatiity

Naoufel El-Bachir

ICMA centre
The University of Reading
Email: [email protected]

Spring 2015
Lecture 4

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 1 / 27

Outline

1 A few simple answers to the question of Why?

Stylised facts on of log return time-series
Stylised facts of traded volatilities

2 Pricing: How do stochastic volatility determine vanilla prices?

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 2 / 27

 A few simple answers to the question of Why? Stylised facts on of log return time-series

Skewed and fat-tailed pdf of log-returns

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 5 / 27

 A few simple answers to the question of Why? Stylised facts on of log return time-series

Skewed and fat-tailed pdf of log-returns

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 6 / 27

 A few simple answers to the question of Why? Stylised facts on of log return time-series

Skewed and fat-tailed pdf of log-returns

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 7 / 27

 A few simple answers to the question of Why? Stylised facts on of log return time-series

Volatility is random

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 8 / 27

 A few simple answers to the question of Why? Stylised facts on of log return time-series

Volatility is mean-reverting

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 9 / 27

 A few simple answers to the question of Why? Stylised facts on of log return time-series

Volatility can be correlated with underlying

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 10 / 27

 A few simple answers to the question of Why? Stylised facts on of log return time-series

Stylised facts on of log return time-series

Empirical evidence:
Skewed and fat-tailed pdf of log-returns
Volatility is random
Volatility is mean-reverting
Volatility can be correlated with underlying

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 11 / 27

 A few simple answers to the question of Why? Stylised facts of traded volatilities

Volatility indices are volatile

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 13 / 27

 A few simple answers to the question of Why? Stylised facts of traded volatilities

Implied volatility surfaces

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 14 / 27

 A few simple answers to the question of Why? Stylised facts of traded volatilities

Implied volatility surfaces

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 15 / 27

 A few simple answers to the question of Why? Stylised facts of traded volatilities

Implied volatility cubes

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 16 / 27

 A few simple answers to the question of Why? Stylised facts of traded volatilities

Persistent implied volatility surfaces

maturity 16 days 44 days 79 days

0.3 0.3 0.3

0.25 0.25 0.25

0.2 0.2 0.2

0.15 0.15 0.15

0.1 0.1 0.1

0.05 0.05 0.05

0 0 0
−0.2 −0.1 0 0.1 −0.2 −0.1 0 0.1 −0.2 −0.1 0 0.1

135 days 233 days 324 days

0.3 0.3 0.3

0.25 0.25 0.25

0.2 0.2 0.2

0.15 0.15 0.15

0.1 0.1 0.1

0.05 0.05 0.05

0 0 0
−0.2 −0.1 0 0.1 −0.2 −0.1 0 0.1 −0.2 −0.1 0 0.1

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 17 / 27

 A few simple answers to the question of Why? Stylised facts of traded volatilities

"Traded" volatilities

Empirical evidence:
Implied volatility surfaces (cubes) vary across strikes /
moneyness, maturities, (and tenors).
Implied volatility surfaces (cubes) move randomly
Volatility indices and volatility derivatives are actively traded

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 18 / 27

 A few simple answers to the question of Why? Stylised facts of traded volatilities

2-D diffusions

h p i
dSt = St µ(t, St , Vt )dt + Vt dWt , (1)
dVt = µv (t, St , Vt )dt + Σ(t, St , Vt )dWtv , (2)

and d [W T , W v ]t = ρdt. Typically we have

Σ(t, St , Vt ) = ηVtα

Captures reasonably well the empirical behaviour, but

Implementation is more involved

Theoretically more complex: incomplete markets,
hedging/replication cannot be done with underlying alone

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 19 / 27

 A few simple answers to the question of Why? Stylised facts of traded volatilities

Fundamental Theorem and EMM

Fundamental Theorem: No-arbitrage essentially equivalent to

existence of risk-neutral measure Q.
Under Q, all assets discounted by the numeraire are martingales.
Valuation is calculating expectations
 
Q VT
Mt E
MT

Perfect replication or hedging not required.

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 20 / 27

 A few simple answers to the question of Why? Stylised facts of traded volatilities

Risk-Neutral measures and incomplete markets

Under Q, for diffusions, martingale requirement means risk-free

drift h p i
dSt = St rt dt + Vt dWt

No-arbitrage and Risk-neutral measure (2-D Girsanov) leaves the

drift of the variance unspecified

dVt = µ˜v (t, St , Vt )dt + Σ(t, St , Vt )dWtv

Two sources of risk but only one tradable asset, thus market is
incomplete.

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 21 / 27

 A few simple answers to the question of Why? Stylised facts of traded volatilities

Risk-Neutral measures and incomplete markets

Pricing is not preference-free as in Black-Scholes.

We are left to choose
µ˜v (t, St , Vt )
Because volatility is not a tradable asset.
Unlimited choices means infinite EMMs candidates.

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 22 / 27

 A few simple answers to the question of Why? Stylised facts of traded volatilities

Or is it?

In practice, many options are actively traded. Also volatility

markets.
Only very few forms of µ˜v (t, St , Vt ) are used.
In a way, the market chooses the measure (the drift function).
Heston is very popular. Another very popular model is SABR.
Both allow for very fast pricing of vanillas, i.e. fast calibration.
You pick one model, calibration to vanillas fixes your parameters.

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 23 / 27

Pricing: How do stochastic volatility determine vanilla prices?

By Cholesky:
p  p 
dFtT = FtT Vt ρdWtv + 1 − ρ2 dWt′ , (3)
dVt = µ˜v (t, St , Vt )dt + Σ(t, St , Vt )dWtv , (4)

Hence
!
Z T Z T Z T
1 p p p
ST = F0T exp − Vt dt + ρ v
Vt dWt + 1 − ρ 2 Vt dWt′
2 0 0 0
(5)

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 25 / 27

Pricing: How do stochastic volatility determine vanilla prices?

Thus given the variance path, i.e. conditionally on the sigma-algebra

σ {Wtv , t ≤ T },
Z T Z Tp
Vt dt, Vt dWtv
0 0
are known, and
!
Z T p Z T
Vt dWt′ ≈ N 0, Vt dt
0 0

Therefore,  
1
ST = F0T exp Σ0,T WT′ − Σ20,T T
2
with
 Z T
Σ20,T T = 1−ρ 2
Vt dt
0

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 26 / 27

Pricing: How do stochastic volatility determine vanilla prices?

Hence given the variance path, we can apply Black-Scholes formula

with volatility given by Σ0,T . Unconditionally, we still have to integrate
with respect of the distribution density of the mean-square volatility
φ(x):
Z
1 1 h T i
BS(Σ0,T )φ(Σ0,T )d Σ0,T = E F0 N(d+ ) − KN(d− ) ,
MT MT
with
log F0T /K √


1
d± = √ ± Σ0,T T
Σ0,T T 2
In summary, vanilla prices in a stochastic volatility model are uniquely
determined by the distribution of the pair
Z T Z Tp !
v
Vt dt, Vt dWt
0 0

Naoufel El-Bachir (ICMA) Stochastic Volatility Spring 2015 27 / 27