Vol. models Vol.

is rough Regularity of time series of data

Rough volatility models

Part I: Introduction

Sergio Pulido (ENSIIE - Évry, France)

Sergio Pulido Rough volatility models Part I: Introduction 1/46

 Vol. models Vol. is rough Regularity of time series of data

Outline

1 Classical volatility models

2 Volatility is rough

3 Regularity of time series of data

Sergio Pulido Rough volatility models Part I: Introduction 2/46

 Vol. models Vol. is rough Regularity of time series of data

Asset price models

We denote by
S = (St )0≤t≤T
the price process of a financial asset, such as a
stock (e.g. Apple)
stock index (e.g. S&P 500)
over the time horizon [0,T]

For each time t, St is a random variable. The collection of these

random variables S constitutes a stochastic process

A stochastic asset price model specifies the dynamics of the

stochastic process S. It can be used for
pricing options
hedging and risk management of financial positions

Sergio Pulido Rough volatility models Part I: Introduction 4/46

 Vol. models Vol. is rough Regularity of time series of data

Derivatives pricing

In general, these are the tasks we need to carry out to compute the price
of financial derivatives:
1 Develop a model for the underlying asset S that fits the empirical
features that are observed in the financial markets. Possibly, the
model should be tractable and possess properties that can be used
to implement it
2 With the chosen model at hand, we can compute the price of
financial derivatives exploiting the model properties. Here,
Theoretical tools (Fundamental theorems of asset pricing )
Computational methods

3 Other computational considerations (efficiency, accuracy, etc...)

The most famous and widely used model is the Black-Scholes model...

Sergio Pulido Rough volatility models Part I: Introduction 5/46

 Vol. models Vol. is rough Regularity of time series of data

Black-Scholes model
Dynamics: In this model one assumes that S is a Geometric
Brownian Motion. Specifically the dynamics of S are of the form

dSt = rSt dt + σSt dWt

Here r denotes the risk-free interest rate earned by the money

market account, σ is the (constant) volatility and W is a Brownian
motion
Explicit representation: In this case there is an explicit formula for
S at each time, namely,
   
1
St = S0 exp σWt + r − σ 2 t
2

Since Wt ∼ N (0, t), St has a log-normal distribution: log St is

normal with mean µ = log S0 + r − 12 σ 2 t and variance σ 2 t

Sergio Pulido Rough volatility models Part I: Introduction 6/46

 Vol. models Vol. is rough Regularity of time series of data

Risk-neutral vs. real-world measure

These dynamics are described with respect to a probability measure

known as the risk-neutral measure, which is used for option pricing

To describe the dynamics with respect to the real-world measure

the drift, rSt dt, would have to be changed to

(r + σλ)St dt

where λ is the market price of risk (per units of volatility σ). This
measure is used for risk management

Sergio Pulido Rough volatility models Part I: Introduction 7/46

 Vol. models Vol. is rough Regularity of time series of data

Black-Scholes model simulation

0
0 1 2 3 4 5 6 7 8 9 10

Figure: Simulated paths of BS model on [0, 10] with r = 0, σ = 0.5

Sergio Pulido Rough volatility models Part I: Introduction 8/46

 Vol. models Vol. is rough Regularity of time series of data

European options

Payoff at time T :
Ψ(ST )

Examples:
Call option, Ψ(ST ) = (ST − K)+ = max(ST − K, 0)
Put option Ψ(ST ) = (K − ST )+ = max(K − ST , 0)
K is called the strike price

Value at time t:

v(t, S) = e−r(T −t) E[Ψ(ST )|St = S]

Sergio Pulido Rough volatility models Part I: Introduction 9/46

 Vol. models Vol. is rough Regularity of time series of data

Black-Scholes formula

The explicit formula for a European call option with strike K and
maturity T in the BS model is

v(t, S) = SΦ(d1 ) − Ke−r(T −t) Φ(d2 )

where Φ is the cumulative distribution function of the standard normal

distribution Z z
1 2
Φ(z) = √ e−u /2 du
2π −∞
and
σ2
    
1 S
d1 = √ ln + r+ (T − t)
σ T −t 2K

σ2
     
1 S
d2 = √ ln + r− (T − t)
σ T −t K 2
√
Notice that d2 = d1 − σ T − t

Sergio Pulido Rough volatility models Part I: Introduction 10/46

 Vol. models Vol. is rough Regularity of time series of data

Log-moneyness
Suppose r = 0. In the case of call / put options it is customary to
write "  #
ST
v(t, S) = S E − ek St = S (call)
S +
"  #
k ST
v(t, S) = S E e − St = S (put)
S +
where
K
ek = (Moneyness)
S 
K
k = log (Log-moneyness)
S

We say that an option is at the money (ATM) if the

log-moneyness k = 0 (i.e. K = St )
Sergio Pulido Rough volatility models Part I: Introduction 11/46
 Vol. models Vol. is rough Regularity of time series of data

Implied volatility and volatility surface

Suppose that St = S and that

v BS (t, S; k, T, σ) = price of a call (put) option with maturity T

and log-moneyness k in the BS model with vol. σ

Let
v obs (t, S; k, T ) = observed price for the option with
maturity T and log-moneyness k

We define the implied volatility σBS (t, S; k, T ) through the formula

v BS (t, S; k, T, σBS (t, S; k, T )) = v obs (t, S; k, T )

The set {σBS (t, S; k, T ) : k, T } forms what is known as the implied

volatility surface (IVS)

Remark: σBS is unique since v BS in monotonically increasing in σ

Sergio Pulido Rough volatility models Part I: Introduction 12/46
 Vol. models Vol. is rough Regularity of time series of data

Implied volatility and volatility surface (IVS)

Limitation BS model: σBS (·, ·; , k, T ) not constant, i.e. the IVS is not flat
Implied Volatility Surface

0.7

0.6

0.5

0.4
BS

0.3

0.2

0.1

0
2.5
2
1.5 -0.6
1 -0.4
-0.2
0.5 0
Expiration T 0.2
0 0.4 Log-moneyness k

Figure: Implied volatility surface for call options on S&P 500 on 2015-12-04
Sergio Pulido Rough volatility models Part I: Introduction 13/46
 Vol. models Vol. is rough Regularity of time series of data

Post Black-Scholes models

Deficiencies of the Black Scholes model yield the development of
many models. Two directions are:
Observed that the volatility is not constant =⇒ local and stochastic
volatility models:

dSt = µSt dt + σt St dWt

with a non-constant σ or with a stochastic process σ

Distributional consideration: Replace the Brownian motion in the
B&S-model by another stochastic process, that better fits the
distribution of the log-returns:

Xt
St = S0 e

e.g. X is a Lévy process

Moreover: More complex models are constantly developed to better fit
market observations
Sergio Pulido Rough volatility models Part I: Introduction 14/46
 Vol. models Vol. is rough Regularity of time series of data

Local volatility models

One possible way to correct the limitations of the BS modes to fit

the IVS at a given date is through a local volatility model

Dynamics: The volatility of the process is no longer constant, but

depends on the time t and current asset price St . More precisely S
has dynamics of the form

dSt = rSt dt + σ(t, St )St dWt

References: [Dupire, 1994], [Derman and Kani, 1994]

Sergio Pulido Rough volatility models Part I: Introduction 15/46

 Vol. models Vol. is rough Regularity of time series of data

Local volatility function and Dupire’s formula

Suppose that we fix St = S and that
C(T, K) = model prices of options
Dupire (1994) shows the following “forward PDE”
1
−CT (T, K) + σ 2 (T, K)K 2 CKK (T, K) − rKCK (T, K) = 0
2
Therefore
  12
CT (T, K) + rKCK (T, K)
σ(T, K) = 2
K 2 CKK (T, K)
Hence if C obs (T, K) are the observed prices for enough maturities and
strikes, the partial derivatives could be approximated in order to
approximate the local volatility by
 obs obs
1
CT (T, K) + rKCK (T, K) 2
2 obs (T, K)
K 2 CKK

Sergio Pulido Rough volatility models Part I: Introduction 16/46

 Vol. models Vol. is rough Regularity of time series of data

Local volatility graphs

Figure: Local volatility σ(T, K): from a given model (left), from S&P 500 data
(right). Taken from Tankov 2005

Sergio Pulido Rough volatility models Part I: Introduction 17/46

 Vol. models Vol. is rough Regularity of time series of data

Stochastic volatility models

Local volatility models have limitations to reproduce the evolution of
the implied volatility surface over time

Stochastic volatility models (SVM) aim to fix this problem (at the
cost of a more complicated calibration of the IVS)

In an SVM the volatility is in itselft a stochastic process. More

precisely S has dynamics of the form

dSt = rSt dt + σt St dWt

where σt is a stochastic process

Popular examples: [Hull and White, 1987],

[Stein and Stein, 1991], [Heston, 1993], [Hagan et al., 2002]
– SABR
Sergio Pulido Rough volatility models Part I: Introduction 18/46
 Vol. models Vol. is rough Regularity of time series of data

Stochastic volatility models - Heston Model

Dynamics: In the Heston Model the volatility is assumed to be

stochastic and it is modeled with an accompanying mean reverting
process. More precisely, S has dynamics
p  p 
dSt = rSt dt + Vt St ρ dWt + 1 − ρ2 dWt⊥

where the spot variance (squared volatility) follows

p
dVt = λ(θ − Vt ) dt + ν Vt dWt

with W, W ⊥ are independent Brownian motions

Key features: θ is the mean-reversion level of the spot variance, λ
is the mean reversion speed and there is empirical evidence that
ρ < 0 for equity markets
An important condition: λθ ≥ 0 to assure that equation for V has
a unique nonnegative solution

Sergio Pulido Rough volatility models Part I: Introduction 19/46

 Vol. models Vol. is rough Regularity of time series of data

Heston Model: Additional comments

(log S, V ) is an affine process. ⇒ explicit characteristic function ⇒
fast transformation methods for pricing
Square root process: The spot variance process V is a square root
process or CIR process - also used to model spot interest rates
Simulation of spot variance V : λ = 2, θ = V0 = 0.04, ν = 0.3

Sergio Pulido Rough volatility models Part I: Introduction 20/46

 Vol. models Vol. is rough Regularity of time series of data

Heston Model: Implied volatility surface

0.35

0.3

0.25

0.2
IV

0.15

0.1

0.05
2.5

1.5 -0.5
-0.4
1 -0.3
-0.2
-0.1
0.5 0
Maturity 0.1
0 0.2
0.3 Log-moneyness

Figure: Implied vol. surface in Heston model. Parameters: S0 = 1, r = 0.03,

λ = 2, ν = 1, V0 = 0.02, θ = 0.01, ρ = −0.6

Sergio Pulido Rough volatility models Part I: Introduction 21/46

 Vol. models Vol. is rough Regularity of time series of data

Calibration problem
For a given stochastic model the price of a derivative v depends on a
set of parameters Θ
For instance in the Heston Model Θ = (V0 , λ, θ, ν, ρ)
There are also parameters coming from the particular payoff
structure: For instance, for call and put options, we have the
maturity date T and log-moneyness k
Denote the corresponding model price by

v(Θ; T, k)

Suppose that we observe m call/put option prices for various

(Ti , ki ). Denote these prices by

v obs (T1 , k1 ), . . . , v obs (Tm , km )

Sergio Pulido Rough volatility models Part I: Introduction 22/46

 Vol. models Vol. is rough Regularity of time series of data

Calibration problem (cont.)

The calibration problem consists in finding the set of parameters Θ

that best fits these observed prices
More precisely, one considers the following optimization problem
m
X 2
min ωi v(Θ; Ti , ki ) − v obs (Ti , ki )
Θ
i=1

for some specified weights, ω1 , . . . , ωm

For stochastic volatility, models prices v(Θ; T, k) are not explicit and
this optimization problem is highly non-linear and difficult
This is in contrast with local volatility models where by definition
the local volatility function is defined to approximately match the
observed IVS

Sergio Pulido Rough volatility models Part I: Introduction 23/46

 Vol. models Vol. is rough Regularity of time series of data

Realized volatility S&P 500

10 -3 S&P realized variance from 2000-01-03 to 2017-12-04

3

2.5

1.5

0.5

0
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018

Figure: S&P 500 (estimated) realized variance from 2000-01-03 to 2017-12-04.

Source: http://realized.oxford-man.ox.ac.uk

Sergio Pulido Rough volatility models Part I: Introduction 25/46

 Vol. models Vol. is rough Regularity of time series of data

Motivations for rough volatility models I

1 Match roughness of time series data: Empirical studies indicate

volatility is rougher than Brownian Motion: e.g.
[Gatheral et al., 2018]; [Bennedsen et al., 2021]

2 Fit implied volatility smiles: Without the introduction of jumps

good fit of implied volatilities for short and long maturities. In
particular the ATM skew defined as

∂
ψ(τ ) = σBS (k, τ )
∂k k=0

where σBS denotes the implied volatility; see e.g.

[Fukasawa, 2017]

Sergio Pulido Rough volatility models Part I: Introduction 26/46

 Vol. models Vol. is rough Regularity of time series of data

Motivations for rough volatility models II

3 Microstructural justification: Taking appropriate limits of

microscopic models for the price of an asset based on Hawkes
processes (that encode stylized facts of high frequency markets)
one obtains models with rough volatility; see e.g.
[El Euch et al., 2018] and [El Euch and Rosenbaum, 2019]

Rough Volatility Literature (sites.google.com/site/roughvol/home/risks-1)

Sergio Pulido Rough volatility models Part I: Introduction 27/46

 Vol. models Vol. is rough Regularity of time series of data

Fractional Brownian motion

Mathematically, some rough volatility models rest on fractional

Brownian motion (fBm) in the tradition of [Kolmogorov, 1940],
[Mandelbrot and Van Ness, 1968]

Fractional Brownian motion (fBm): A continuous Gaussian

process W H with mean 0 and covariance function
1  2H
E[WtH WsH ] = |t| + |s|2H − |t − s|2H
2
where H ∈ (0, 1) is known as the Hurst index or parameter
1
Challenge: Non-Markovian / Not a semimartigale for H 6= 2

Sergio Pulido Rough volatility models Part I: Introduction 28/46

 Vol. models Vol. is rough Regularity of time series of data

Mandelbot and Van Ness representation

Mandelbot and Van Ness representation:

Z t 1
Z 0 h 1 1
i 
WtH = CH (t − s)H− 2 dWs + (t − s)H− 2 − (−s)H− 2 dWs
0 −∞

for t > 0, where W is a (classical) Brownian motion on R and

s
2HΓ 23 − H


CH =
Γ H + 12 Γ (2 − 2H)

Sergio Pulido Rough volatility models Part I: Introduction 29/46

 Vol. models Vol. is rough Regularity of time series of data

Basic properties of fBm

H
Self-similarity: Wat has the same law as aH WtH

H
Stationary increments: Wt+∆ − WtH has the same law as W∆
H

Correlation of increments:
H = 0.5 independent increments (classical BM)
H > 0.5 positively correlated increments
H < 0.5 negatively correlated increments

Regularity of trajectories: For every ε > 0 the paths of W H are

(H − ε)-Hölder continuous a.s.

Sergio Pulido Rough volatility models Part I: Introduction 30/46

 Vol. models Vol. is rough Regularity of time series of data

Simulation of fBm - Cholesky factorization

Since the process is Gaussian with known covariance structure the

simplest way to simulate is using the Cholesky method:

Consider a partition 0, t1 , . . . , tN of the interval [0, T ]

Compute the variance-covariance matrix Σ of the Gaussian vector

(WtHi )i=1,...,N

Calculate the Cholesky factorization of Σ = C > C

Simulate N independent N (0, 1) random variables

Z = (Z1 , . . . , ZN )> ∈ RN (column vector)

Define X = C > Z and Yti = Xi for i = 1, . . . , N

The simulated path of fBm on [0, T ] is 0, Yt1 , . . . , YtN

Sergio Pulido Rough volatility models Part I: Introduction 31/46

 Vol. models Vol. is rough Regularity of time series of data

Simulation of fBm

H = 0.8
1

WH
t 0.5

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

t
H = 0.5
1
WH

0
t

-1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t
H = 0.2

2
WH

1
t

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

t

Figure: Simulation of fBm on [0, 1] for H = 0.8, 0.5, 0.2

Sergio Pulido Rough volatility models Part I: Introduction 32/46
 Vol. models Vol. is rough Regularity of time series of data

Riemann–Liouville fractional Brownian motion

Other rough volatility models rest on the Riemann–Liouville fractional

Brownian motion introduced by [Lévy, 1953]

Riemann–Liouville fractional Brownian motion: This is a process

on R+ defined by
Z t
ftH = 1 1
W (t − s)H− 2 dWs
Γ H + 21


0

Notice that this is just one piece in the Mandelbot and Van
Ness representation of fBm

Sergio Pulido Rough volatility models Part I: Introduction 33/46

 Vol. models Vol. is rough Regularity of time series of data

Measure of the regularity of the log-volatility

Moments of the increments of the log-volatility:

q
E [|log σt+∆ − log σt | ]

or its empirical value:

M∆
1 X q
m(q, ∆) = |log(σti +∆ ) − log(σti )|
M∆ i=1

where tM∆ + ∆ is the index of the last point in the data series
Moments of the increments of the fBM:
h q
i
H
E Wt+∆ − WtH = Kq ∆qH

with Kq the moment of order q of the absolute value of a standard

Gaussian variable

Sergio Pulido Rough volatility models Part I: Introduction 35/46

 Vol. models Vol. is rough Regularity of time series of data

An example with SPX volatility

Consider the variance Vt = σt2 estimates between 2000-01-03 and

2016-05-03
We estimate m(q, ∆) empirically for moment orders
q = 0.5, 1, 1.5, 2, 3 and lags ∆ = 1, 2, . . . , 100
For each value of q find the best linear fit for log(m(q, ∆)) as a
function of log(∆)

log(m(q, ∆)) ∼ ζq log(∆) + Cq

Find the best linear fit of ζq as a function of q

Sergio Pulido Rough volatility models Part I: Introduction 36/46

 Vol. models Vol. is rough Regularity of time series of data

Results: Scaling of the moments m(q, ∆)

Scaling of m(q, )
0

-0.5

-1
Log(m(q, ))

-1.5

q=0.5
-2
q=1

q=1.5
-2.5
q=2

q=3

-3
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Log( )

Figure: log(m(q, ∆)) as a function of log(∆), SPX, 2000-01-03 to 2016-05-03

Sergio Pulido Rough volatility models Part I: Introduction 37/46
 Vol. models Vol. is rough Regularity of time series of data

Resutls: Monofractality of the log-volatility

Monofractality of the log-volatility
0.4

0.35

0.3

0.25
q

0.2

0.15

0.1
q
Linear fit
0.05
0.5 1 1.5 2 2.5 3
q

Figure: ζq empirical estimate and linear fit with slope H ≈ 0.124

Sergio Pulido Rough volatility models Part I: Introduction 38/46
 Vol. models Vol. is rough Regularity of time series of data

An example with fBm

We simulate fBm WtH on [0, 1] with H = 0.124 and 4268 time steps
H
We estimate E[|Wt+∆ − WtH |q ] ≈ m(q, ∆) empirically for moment
orders q = 0.5, 1, 1.5, 2, 3 and lags ∆ = 1, 2, . . . , 100

For each value of q find the best linear fit for log(m(q, ∆)) as a
function of log(∆)

log(m(q, ∆)) ∼ ζq log(∆) + Cq

Find the best linear fit of ζq as a function of q

The slope H gives us an estimate of the Hurst parameter

Sergio Pulido Rough volatility models Part I: Introduction 39/46

 Vol. models Vol. is rough Regularity of time series of data

Results: Scaling of the moments m(q, ∆)

Scaling of m(q, )
0

-0.5

-1
Log(m(q, ))

-1.5

q=0.5
-2
q=1

q=1.5
-2.5
q=2

q=3

-3
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Log( )

Figure: log(m(q, ∆)) as a function of log(∆), fBm

Sergio Pulido Rough volatility models Part I: Introduction 40/46
 Vol. models Vol. is rough Regularity of time series of data

Results: Monofractality of fBm

Monofractality of the fBm
0.35

0.3

0.25
q

0.2

0.15

0.1

q
Linear fit
0.05
0.5 1 1.5 2 2.5 3
q

Figure: ζq empirical estimate and linear fit

Sergio Pulido Rough volatility models Part I: Introduction 41/46
 Vol. models Vol. is rough Regularity of time series of data

Rough Fractional Stochastic Volatility (RFSV)

These observations motivated the introduction of the following model
Rough Fractional Stochastic Volatility (RFSV) model: A
stochastic volatility model with σt satisfying

dlog(σt ) = ν dWtH + α(m − log(σt )) dt

1
with H < 2

Source: [Gatheral et al., 2018]

[Comte and Renault, 1998] had proposed Fractional Stochastic

Volatility (FSV) model with H > 21 ⇒ Long memory
H
Cov(Wt+1 − WtH , W1H ) ∼t→∞ Ct2H−2

Sergio Pulido Rough volatility models Part I: Introduction 42/46

 Vol. models Vol. is rough Regularity of time series of data

References I

Bennedsen, M., Lunde, A., and Pakkanen, M. S. (2021).

Decoupling the Short- and Long-Term Behavior of Stochastic
Volatility.
Journal of Financial Econometrics.
Comte, F. and Renault, E. (1998).
Long memory in continuous-time stochastic volatility models.
Mathematical Finance, 8(4):291–323.
Derman, E. and Kani, I. (1994).
Riding on a smile.
Risk Magazine, pages 32–39.
Dupire, B. (1994).
Pricing with a smile.
Risk Magazine, pages 18–20.

Sergio Pulido Rough volatility models Part I: Introduction 43/46

 Vol. models Vol. is rough Regularity of time series of data

References II

El Euch, O., Fukasawa, M., and Rosenbaum, M. (2018).

The microstructural foundations of leverage effect and rough
volatility.
Finance and Stochastics, 22(2):241–280.
El Euch, O. and Rosenbaum, M. (2019).
The characteristic function of rough Heston models.
Mathematical Finance, 29:3–38.
Fukasawa, M. (2017).
Short-time at-the-money skew and rough fractional volatility.
Quantitative Finance, 17(2):189–198.
Gatheral, J., Jaisson, T., and Rosenbaum, M. (2018).
Volatility is rough.
Quantitative Finance, 18(6):933–949.

Sergio Pulido Rough volatility models Part I: Introduction 44/46

 Vol. models Vol. is rough Regularity of time series of data

References III

Hagan, P. S., Kumar, D., Lesniewski, A. S., and Woodward, D. E.

(2002).
Managing smile risk.
Wilmott Magazine, pages 84–108.
Heston, S. L. (1993).
A closed-form solution for options with stochastic volatility with
applications to bond and currency options.
The review of financial studies, 6(2):327–343.
Hull, J. and White, A. (1987).
The pricing of options on assets with stochastic volatilities.
The Journal of Finance, 42(2):281–300.
Kolmogorov, A. N. (1940).
Wienersche Spiralen und einige andere interessante Kurven im
Hilbertschen Raum.
C. R. (Doklady) Acad. Sci. URSS (N.S.), 26:115–118.

Sergio Pulido Rough volatility models Part I: Introduction 45/46

 Vol. models Vol. is rough Regularity of time series of data

References IV

Lévy, P. (1953).
Random functions: general theory with special reference to Laplacian
random functions.
University of California publications in statistics. University of
California Press.
Mandelbrot, B. and Van Ness, J. (1968).
Fractional Brownian motions, fractional noises and applications.
SIAM Review, 10(4):422–437.
Stein, E. M. and Stein, J. C. (1991).
Stock price distributions with stochastic volatility: an analytic
approach.
The Review of Financial Studies, 4(4):727–752.

Sergio Pulido Rough volatility models Part I: Introduction 46/46