Lognormal Stochastic Volatility

with Applications to Cryptocurrency Options

Artur Sepp
Head Quant, LGT Bank

Minnesota Center for Financial and Actuarial Mathematics Seminar

18 October 2024

1
Content
1. Applications of stochastic volatility (SV) models for options modeling
2. Cryptocurrency inverse options and inverse measures
3. Log-normal stochastic volatility model with quadratic drift
4. Closed-form accurate solution for the moment generating function for
the log-normal SV model
5. Roughness of volatility and its impact on auto-correlation
6. Fitting options data for Bitcoin traded on Deribit exchange
References:
• Sepp A and Rakhmonov P (2023) Log-normal Stochastic Volatility
Model with Quadratic Drift, International Journal of Theoretical and Ap-
plied Finance, 2023, 26(8)
https://www.worldscientific.com/doi/reader/10.1142/S0219024924500031

• Lucic V and Sepp A (2024) Valuation and Hedging of Cryptocurrency

Inverse Options, Quantitative Finance, 2024, 24(7), 851-869
https://ssrn.com/abstract=4606748
2
Applications of stochastic volatility models

1. Inference of volatility from discrete time series

• Open-High-Low-Close estimators, GARCH type models
• Focus on prediction ability of future volatility

2. Vanilla options valuation using dynamic model for returns volatility:

modelling vanilla options and estimation of risk-premia for systematic
strategies
• Heston model
• Exp-OU, Bergomi type models
• Log-normal model
• Focus on prediction ability of implied volatility dynamics

3. Structured products valuation and risk management: applications of

models for managing derivatives books
• Dupire local volatility models
• Local stochastic volatility models
• Focus on close fit to traded vanilla options and risk decomposition
of structured products

3
Heston model
• Heston model dynamics for price St and variance Vt:
p
dSt = µtStdt + VtStdWt, S0 = S,
p (1)
dVt = κ (θ − Vt) dt + ε VtdZt, V0 = V,
where µt is price drift under P and µt = rt under risk-neutral measure Q
Wt and Zt are Brownian motions with return-volatility correlation ρ
κ is mean-reversion speed
θ is long-run variance
ε is volatility of volatility

• Drawbacks:

1) Heston model provides semi-analytic solution for vanilla options but it

is hard to implement numerically for Monte-Carlo and PDE solvers

2) The variance can reach zero is reachable if Feller condition is not

satisfied

4
Exp-OU, Bergomi type models

• Dynamics expressed using the forward variance ξt(u)

q
dSt = µtStdt + ξt(t)StdWt, S0 = S,
1
 
ξt(u) = ξ0(u) exp εe−κ(u−t)Yt − ε2e−2κ(u−t)E[Yt2] (2)
2
dYt = −κYtdt + dZt
where Wt and Zt are Brownians with return-volatility correlation ρ
Yt is Ornstein–Uhlenbeck (OU) driver

• Exp-OU, Bergomi type models provide no analytic solution for vanilla

but they are easier to implement numerically

• Drawbacks:

1) These models cannot be applied for assets with positive volatility-

return correlations because the asset price loses martingale property

2) The model dynamics are not closed (the functional form of the volatil-
ity process is different) under the change of numeraire so that they cannot
be applied for valuation of inverse options using spot price numeraire
5
Log-normal stochastic volatility model
• The dynamics for price St and volatility σt:
(0)
dSt = µtStdt + σtStdWt ,
(3)
(0) (1)
dσt = (κ1 + κ2σt) (θ − σt)dt + βσtdWt + εσtdWt ,
(0) (1)
Wt and Wt are uncorrelated Brownian motions
β ∈ R is the volatility beta to price changes
ε ∈ R+ is residual volatility-of-volatility
θ ∈ R+ is the mean level of the volatility
κ1 + κ2σt is the mean-reversion speed is linear function of the volatility
• The model combines the best of the two worlds: semi-analytic solution
for vanilla options and easy to implement numerically
1) The model can be applied for assets with positive return-volatility
correlation (if κ2 > max{0, β})
2) The model is closed under change of numeraire so that can be applied
for valuation of inverse options
• Classic results do not apply due to super-linear drift in SDE for σt

Theorem 1. i) Boundaries {0, +∞} are unattainable for volatility σt

ii) SDE for σt has a unique strong solution.
6
Model Dynamics consistent with Forward Variance I

• Given a set of put and call option prices, we compute fair strikes of
variance swaps
 
2 2 X δK X δK
Kvarswap (T ) =  P ut(T, K) + Call(T, K)

2 2

T K<F (T )
K K≥F (T )
K

• Exp-OU Bergomi model type are formulated as forward variance models

under which: E[dσt2] = δKvarswap
2 (t)dt

• The model is fitted by construction to the term structure of market-

implied var swaps for any set of model parameters

• Formulate log-normal SV model using volatility backbone function η(t)

(0)
dSt = µtStdt + σtη(t)StdWt , S0 = S,
(0) (1)
dσt = (κ1 + κ2σt) (θ − σt)dt + βσtdWt + εσtdWt , σ0 = σ,
(4)
dItmodel = η 2(t)σt2dt, I0model = 0,
dIt = σt2dt, I0 = 0,

7
Model Dynamics consistent with Forward Variance II

• Consider η(t) to be a deterministic piecewise function

ηn1{t∈(Tn−1,Tn]}
X
η(t) =
n

• Model-implied value of expected quadratic variance

"Z #
T
η 2(t′)σt2′ dt′ | Ft =
 h i h i
I¯tmodel (T ) = EQ 2
X
Q Q
ηn E ITn − E ITn−1 (5)
t n

• We apply the analytic formula for computing expected value of quadratic

variance I(T ) = EQ [IT ] (initial value σ0 and mean θ are fixed externally)

• The term structure of ηn is fitted as follows:

2
TnKvarswap 2
(Tn) − Tn−1Kvarswap (Tn−1)
2
ηn =
I(Tn) − I(Tn−1)

8
Assets with positive volatility return correlations
Many assets have positive implied Black-Scholes volatility skews
Volatilities of out-of-the-money calls are higher than volatilities of out-
of-the-money puts

9
Vanilla and inverse options

• Pay-offs of vanilla call and put are settled in cash at expiry time T :
ucall(ST ) = max {ST − K, 0} , uput(ST ) = max {K − ST , 0} .

• Pay-offs of inverse call and put are converted to units of asset at T :

1 1
ũcall(ST ) = max {ST − K, 0} , ũput(ST ) = max {K − ST , 0} .
ST ST

• In markets for cryptocurrency options, inverse payoffs are popular for

coin-margined accounts and on-chain transfers (Deribit exchange)

• Vanilla options on cryptocurrencies are also traded (Binance, CME ex-

change)

• Are there arbitrage opportunities between inverse and vanilla options?

10
Valuation under MMA and inverse measures
nR o
T
• By choosing money market account (MMA) M (T ), M (T ) = exp t r(s)ds ,
as a numéraire, we consider an equivalent martingale measure Q, induced
by M , where r(t) is risk-free rate.

• Accordingly, the time-t value of an option, denoted by U (t, S), with

payoff function u(ST ) at time T , equals to
" #
1
U (t, S) = M (t)E u(ST )| Ft , (6)
M (T )
where expectation E is taken under the MMA martingale measure Q.

• By choosing S as a numéraire, we consider an inverse martingale mea-

sure Q̃, induced by S.

• The value function of inverse option, denoted by Ũ (t, S), with the payoff
paid in units of S, equals to
" #
1
Ũ (t, S) = StẼ u(ST )| Ft , (7)
ST
where expectation Ẽ is taken under the inverse martingale measure Q̃.
11
Equivalence of MMA and inverse measures

• Are values of the vanilla option under the MMA measure and of inverse
option under the inverse measure equal?
Theorem 2. Assume a complete market and that both the MMA measure
Q and the inverse measure Q̃ are equivalent martingale measures. Then
the values of options under the MMA measure in Eq (6) and under the
inverse measure are in Eq (7) satisfy
" # " #
1 1
M (t)E u(ST )| Ft = StẼ u(ST ) Ft . (8)
M (T ) ST

Proof. Using the change of numéraire technique in theorem 1 in Geman

et al. (1995).

• PV of inverse option = PV of vanilla option / Price

• Dynamic SV models must produce martingale dynamics for both MMA

and inverse measures

12
Importance of quadratic drift for Log-normal SV model

Theorem 3. [Log-normal SV model with quadratic drift] Under model

dynamics (3) for the price processes St and Rt = St−1.
1) The process St is a martingale under the MMA measure Q iff κ2 ≥ β.

2) The process Rt is a martingale under the inverse measure Q̃ iff κ2 ≥ 2β

• Linear Log-normal SV model with κ2 = 0 is not arbitrage-free for

β > 0like classic Exp-OU SV models

• Figure: Admissible regions of model parameters (β, ϑ) for the martingale

property of ZT under the MMA measure Q and of RT under the inverse
measure Q̃ with κ2 = 0 and κ2 = 1.
• Both conditions are satisfied in the area with overlapping hatches.

13
Conventional SV models are ill-defined when return-volatility cor-
relation is positive
• Heston model may not have a stationary distribution of the variance
when κ − ρϑ < 0
• Exp-OU SV model is not arbitrage-free for positive return-vol correlation
• Subplot (A) shows the admissible region for parameters (ρ, ϑ) of Heston
model where Feller condition is satisfied and the stationary distribution
exists under the inverse measure using κ = 1, θ = 1
• Subplot (B) shows the admissible region parameters (ρ, ϑ) of Exp-OU SV
model for the martingale property under the MMA and inverse measures
using κ = 1, θ = 1
• Both conditions are satisfied in the area with overlapping hatches

14
Properties of the log-normal volatility process
• We consider the SDE of volatility process in model dynamics in Eq (3)
represented as follows
(∗)
dσt = (κ1 + κ2σt) (θ − σt)dt + ϑσtdWt , σ0 = σ (9)
(∗)
where Wt is a standard Brownian motion
ϑ2 = β 2 + ε2 is the total variance of volatility.
• Notice that the process σt is not a polynomial diffusion because the
drift coefficient is a second-order polynomial in σt (using Lemma 2.2 in
Filipovic and Larsson (2016))
Lemma 1. Dynamics of volatility in Eq (3) under the inverse measure Q̃
f (0) + εσ dW
f (1).
 
2
dσt = κ1θ − (κ1 − κ2θ) σt − (κ2 − β ) σt dt + βσtdW t t t
Rt
− 0 r(t′ )dt′
Using zero-drift stochastic driver Zt, Zt = e St, the MMA measure
Q and the inverse measure Q̃ are related by the density process by
h i
Λt = Et dQ̃/dQ = Zt/Z0.
Theorem 4. Measures Q and Q̃ are equivalent if and only if κ2 ≥ β.
15
Steady state distribution of the volatility
• The steady state distribution is given by the Generalized Inverse Gaus-
sian distribution (Jorgensen (1982))
q
  
G(σ) = cσ η−1 exp − + bσ , σ>0 (10)
σ
• Subplot (A) shows the steady state PDF of the volatility computed
using Eq (10) with fixed κ1 = 4 and κ2 = {0, 4, 8}.
• Subplot (B) and (C) show the skeweness of the volatility and the excess
kurtosis of the unconditional returns distribution , respectively, both as
functions of κ2, for the three choices of κ1 = 1, 4, 8.
Other model parameters are fixed to θ = 1 and ϑ = 1.5.

16
Moments of Volatility Process, I

• We consider the mean-adjusted process Yt = σt − θ and define its power

(n) (n) (n)
function mt , mt = Ytn, and the m-th moment, mτ , n = 0, 1, 2, .., as
follows
 
(n) (n)
mt (τ ) = Et mτ

(n)
• By applying Itô’s lemma for mt , we obtain
(n) (∗
 
dmt = −κYt − κ2Yt nY n−1dt+c(n) (Yt + θ)2 Y n−2dt+ϑ (Yt + θ) nY n−1dWt
2
(n) 1 ϑ2 n(n − 1)
with m0 = Y0n and c(n) = 2

• We notice a pattern of the powers of n which allows for a recursive

solution as follows.

17
Moments of Volatility Process, II
Proposition 1. [Moments of Volatility Process] The solution to moments,
can be presented as a matrix equation for an infinite-dimensional vector:
∂τ M (0,∞)(τ ) = Λ(0,∞)M (0,∞)(τ ),
where ∂τ is the derivative w.r.t. τ and
 T 
M (0,∞) (0) (1) (2) (3) (4)
(τ ) = m , m , m , m , m , ... , M (0,∞) (0) = 1, Y0, Y02, Y03, Y0
 
0 0 0 0 0 0 ...
 0

−κ −κ2 0 0 0 ...
c(2)θ 2 2c(2)θ (c(2) − 2κ) −2κ2
 
(0,∞) 0 0 ...
Λ = .
 0
 c(3)θ2 2c(3)θ (c(3) − 3κ) −3κ2 0 ...
 0 0 c(4)θ2 2c(4)θ (c(4) − 4κ) −4κ2 ...
 

...

18
Moments of Volatility Process, III
• An approximate solution to ODE system is obtained using a truncation
by fixing the number of terms to k∗ and using a finite dimensional vector
of m-th moments, m = 1, ..., k∗ with analytic solution
(1,k ∗) n
(1,k ∗) o (1,k ∗) 
(1,k ∗ ) −1  n
(1,k ∗) o (k ∗)
M (τ ) = expm Λ t ·M (0)+ Λ · expm Λ t −I
where expm() is the matrix exponent, · and −1 are the matrix product
∗
and inverse, respectively, and I (k ) is k∗ × k∗ identity matrix.

• Subplot (A) and (B): The first four moments of mean-adjusted volatility
computed with the truncation order k∗ = 4 and k∗ = 8 as functions of τ .
Dots and error bars denote the estimate and 95% confidence interval,
respectively, computed using MC simulations of model dynamics.

19
Model Dynamics under MMA and Inverse Measures

We introduce the mean-adjusted volatility process Yt = σt − θ

Corollary 1 (Dynamics under the MMA measure Q). The joint dynamics
of log-price Xt = log St, the mean-adjusted volatility process Yt = σt − θ,
and the QV It:
1 2 (0)
dXt = − η (t) (Yt + θ)2 dt + η(t) (Yt + θ) dWt , X0 = X,
 2
(0) (1)

2
dYt = −κYt − κ2Yt dt + β (Yt + θ) dWt + ε (Yt + θ) dWt , Y0 = σ0 − θ,
dIt = η 2(t) (Yt + θ)2 dt, I0 = I.

Corollary 2 (Dynamics under the inverse measure Q̃). The joint dynamics
of log-price Xt, the mean-adjusted volatility process Yt, and the QV It:
1 2 f (0), X = X,
dXt = η (t) (Yt + θ)2 dt + (Yt + θ) η(t)dW t 0
2
f (0) + ε (Y + θ ) dW
f (1), Y = σ − θ,
 
2
dYt = λ̃ − κ̃Yt − κ̃2Yt dt + β (Yt + θ) dW t t t 0 0
dIt = η 2(t) (Yt + θ)2 dt, I0 = I.
where λ̃ = βθ2η(t), κ̃ = κ1 − κ2θ + 2(κ2 − βη(t))θ, κ̃2 = κ2 − βη(t)
20
Joint Valuation Equation under MMA and Inverse measures
• State variables: Xt = ln St, Yt = σt − θ, It = 0t σt2dt
R

The joint valuation PDE under MMA p = 1 and inverse p = −1 measures:


EQ[u(X ) | F ], p=1
T t
U (τ, X, I, Y ; p) = (11)
ẼQ̃[u(X ) | F ], p = −1.
T t

• The classic Feynman-Kac formula cannot be applied directly to Eq (11)

because the model dynamics do not satisfy linear growth condition

Theorem 5. The value function U(τ, X, Y ; p) solves the PDE:

−Uτ + L(Y ; p) + L(X; p) + L(I; p) U = 0, U (0, X, I, Y ) = u(X, I), (12)
with operators:
1
 
(p)
L(Y ; p)U = ϑ2(Y + θ)2UY Y + λ(p) − κ(p)Y − κ2 Y 2 UY ,
2
1
 
L(X; p)U = (Y + θ)2 η 2(t) (UXX − pUX ) + βη(t)UXY , L(I; p)U = (Y + θ)2UI
2

κ , p = 1,
(p) (p) (p) (p) 2 (p) 2
κ = κ1 − κ2θ + 2κ2 θ, λ = (κ2 − κ2 )θ , κ2 =
κ − βη(t), p = −1
2
21
Moment Generating Function (MGF) for state variables

• Introduce MGF for state variables (Xt, It, Yt) with complex-valued trans-
form parameters Φ, Ψ, Θ ∈ C:

EQ[e−ΦXτ −ΨIτ −ΘYτ | F ], p=1
t
G(τ, Φ, Ψ, Θ; p) = (13)
ẼQ̃[e−ΦXτ −ΨIτ −ΘYτ | F ], p = −1
t

• G solves the PDE:

 
− Gτ + L (Y ; p) +L (X; p) +L (I; p) G = 0,
(14)
G(0, Φ, Ψ, Θ; p) = e−ΦX−ΨI−ΘY

• MGF in (14) has no closed-form solution because volatility generator

L(Y ; p) is not affine in both variance and drift terms:
1
 
(p)
L(Y ; p)U = ϑ2(Y + θ)2UY Y + λ(p) − κ(p)Y − κ2 Y 2 UY
2
• Analytical intractability has limited applications of log-normal SV models
despite their empirical support
22
Existence of MGF for state variables
Theorem 6. Given the transform variable Φ = ΦR + iΦI ∈ C, the MGF
of the log-price Xτ

G(τ, X; Φ; p = 1) = E[e−ΦXτ X0 = X]
exists for ΦR ∈ (−1, 0), if Zτ is a martingale under the MMA measure.
By Theorem 3 1), the necessary condition is κ2 ≥ β.

Similarly, the MGF of the log-price Xτ

G(τ, X; Φ; p = −1) = Ẽ[e−ΦXτ X0 = X]

exists for ΦR ∈ (0, 1) if Rτ is a martingale under inverse measure. By
Theorem 3 2), the necessary condition is κ2 ≥ 2β.

23
Affine expansion for closed-form solution

• Make an anzats forn exponential solution [∞] using an infinite dimen-

o E
sional vector A(τ ) = A(k)(τ ; Φ, Ψ, Θ; p) , k = 0, 1, ..., ∞:
 
 ∞ 
E [∞](τ, Φ, Ψ, Θ; p) = exp −ΦX − ΨI + A(k)(τ ; Φ, Ψ, Θ; p)Y k ,
X
 
k=0
where A(τ ) solves an infinite-dimensional system of quadratic ODEs:
⊤
(k) ⊤

Aτ = (k)
A M A+ L (k) A + H (k), k = 0, 1, ..., ∞, (15)

with M (k), L(k) and H (k) infinite dimensional (very) sparse matrices

• Similarity with the problem of computing the moments of the volatility

process: we need to introduce a finite dimensional basis
Theorem 7. [Second-order affine expansion] The MGF in Eq (13) can
be decomposed into leading term E [4] and remainder term R[4]:
G(τ, Φ, Ψ, Θ; p) = E [4](τ, Φ, Ψ, Θ; p) + R[4](τ, Φ, Ψ, Θ; p), (16)

Theorem 8. Continuous solution A for E [4] exists on [0, τ0)

24
Approximation for MGF using second-order affine expansion
Corollary 3. [Second-order affine approximation for the MGF (13)] is
obtained using the leading term E [4]:
G(τ, Φ, Ψ, Θ; p) = E [4](τ, Φ, Ψ, Θ; p) (17)
The approximation error is estimated using remainder R[4]
Theorem 9. E [4] is consistent with first and second moments of (X, Y, I).
• Illustration of PDFs computed by inversion of the first E [2] and second
order E [4] compared to Monte-Carlo histogram (blue) for τ = 1

25
Extension to rough dynamics (Sepp-Rakhmonov-Motuzenko (2024)
in progress)
• We consider dynamics driven by log-normal Volterra SV
(0)
dSt = r(t)Stdt + η(t)σtStdWt ,
Z t  
(0) (1)
σt = σ0 + K(t − s) (κ1 + κ2σs)(θ − σs)ds + βσs dWs + εσs dWs
0
(18)
where W (0), W (1) are uncorrelated Brownian motions
η(t) is volatility backbone to fit to the term structure of variance swaps
Kernel K : R+ → R+ is a fractional kernel
1 α−1 1
 
K(t) = t , α∈ ,1 (19)
Γ(α) 2

• Boundary points {0, +∞} are unattainable for process σt

• σt has unique strong solution

• Dynamics (18) are closed (have the same functional form) under nu-
meraire changes and under the inverse measure
26
Extension to rough dynamics II
• Analytic/numerical solution for expected quadratic variance for
κ2 = 0 / κ2 > 0 for fitting volatility backbone η(t)

• Affine first/second order expansion can be applied to derive an expo-

nential approximation to MGF of rough log-normal SV model

• However, the expansion results in a multi-variate system of integral

equations which is numerically tedious

• Implement Monte-Carlo based valuation

• Develop Deep Learning (DL) for model calibration

27
Rough dynamics and volatility auto-correlation
• Auto-correlation of the volatility measures the “memory” or the longevity
of periods with high volatility

• Auto-correlation function (ACF) of the volatility path σt observed at

regular sampling times {tn}
ACF (h) = Corr(σt+h, σt), (20)
where h is the lag

• One-factor Markovian SV models produce exponentially decaying ACF:

ACF (h) = ce−qh, |h| → 0 (21)
where c > 0 and q > 0 are constants

• Bennedsen et al (2022) suggest that rough SV models produce the

following ACF:
′
ACF (h) = 1 − c |h|2α +1 , |h| → 0, (22)
where c > 0 is a constant and α′ = α − 1, α′ ∈ (−1/2, 1/2), is called the
roughness index of σt

• Negative values of α′ imply auto-correlation rougher than that of a

Brownian motion
28
Estimation of rough auto-correlation
• We estimate the values of c and α′ by minimizing squired differences
empirical ACF and ACF given in Eq (22)
• To fit ACF implied by log-normal SV model, we fit model ACF computed
by Eq (20) with MC simulations and fithstedy-state empirical
i distribution:
M odel ACF (h) = EP Corr(σt+h, σt) (23)
for a vector of lags h = (0, 1, ...)
• Figure showing ACF of implied vols: A) VIX for the S&P 500 index,
B) MOVE for 10y UST rate, C) OVX for oil ETF, D) BTC ATM vols

29
Remarks on model calibration to time series data of option prices
• One-factor Log-normal SV model is sufficient enough for reproducing
the empirical features of ACF for daily data
• When fitting model parameters to options, it is hard to estimate the
volatility mean-reversion parameters consistently because a higher value of
mean reversion requires higher values of model correlation and convexity
and vice versa
• Instead, we estimate the mean-reversion to the empirical ACF and keep
it fixed when fitting model to options data
• There is a small correction in mean-reversion parameters between P and
Q measures due to risk-premia, which we ignore for time being
• Fitted parameters of steady-state distribution of volatility under Log-
normal SV model
VIX MOVE OVX BTC
θ 0.20 0.91 0.39 0.71
κ1 1.29 0.10 2.78 2.21
κ2 1.93 0.41 2.24 2.18
ε 0.72 0.36 0.83 0.92
30
Calibration of model parameters to Bitcoin implied volatilities
• Analytic solution for MGF provides Fourier-based formulas (Lewis (2000)
and Lipton (2001) for vanilla options of extended to inverse options

• Model provides very good fit to market data within bid/ask

σˆ0 = 0.86, θ̂ = 1.04, κˆ1 = 2.21, κˆ2 = 2.18, β̂ = 0.13, ε̂ = 1.63

31
Analytic solution matches Monte-Carlo simulations
• Compute model implied volatilities under MMA measure (MMA) and
inverse measure (Inverse) for maturity slices used in model calibration
• Dashed lines M C − 0.95ci and M C + 0.95ci are MC 95% CI
• MSE is the mean squared error between model and MC vols

32
Time series of model params fitted to short-dated BTC options
In 2022 and 2023, the model error (the average difference between market
and model implied volatility) became less than 1% most of the times

33
Python library
Python implementation https://github.com/ArturSepp/StochVolModels
See stochvolmodels/examples/run_lognormal_sv_pricer

34
Conclusions

1. Introduce log-normal stochastic volatility model with quadratic drift to

enforce martingality under money-market account and inverse measures

2. Develop closed-form affine expansion for accurate valuation of vanilla

and inverse options

3. Discuss rough log-normal SV model and the impact of roughness on

model auto-correlation function

4. Apply model for empirical fit to time series of option markets on

Bitcoin with close fit to market data

35
Further reading

• Sepp A and Rakhmonov P (2023) Log-normal Stochastic Volatility

Model with Quadratic Drift, International Journal of Theoretical and Ap-
plied Finance, 2023, 26(8) https://www.worldscientific.com/doi/reader/
10.1142/S0219024924500031

• Lucic V and Sepp A (2024) Valuation and Hedging of Cryptocurrency

Inverse Options, Quantitative Finance, 2024, 24(7), 851-869, https://
ssrn.com/abstract=4606748

• Sepp A and Rakhmonov P (2023) What Is a Robust Stochastic Volatility

Model, SSRN preprint https://ssrn.com/abstract=4647027

• Sepp A, Rakhmonov P, Motuzenko (2024) Rough Log-normal Stochas-

tic Volatility Model, in preparation

36
Disclosure

These slides and discussion represent my personal views.

These views do not represent an official view of my current and last

employers.

These views and discussion are not an investment advice in any possible
form.

Volatility products and cryptocurrencies are associated with high risk.

37