March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071

2450003

OPEN ACCESS
International Journal of Theoretical and Applied Finance
Vol. 26, No. 8 (2023) 2450003 (63 pages)
c The Author(s)
DOI: 10.1142/S0219024924500031

LOG-NORMAL STOCHASTIC VOLATILITY

MODEL WITH QUADRATIC DRIFT
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

ARTUR SEPP ∗

Clearstar Labs AG
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

Glärnischstrasse 36, 8027 Zürich, Switzerland

[email protected]

PARVIZ RAKHMONOV
Marex
155 Bishopsgate, London EC2M 3TQ, United Kingdom
[email protected]

Received 9 July 2023

Revised 24 December 2023
Accepted 2 January 2024
Published 28 February 2024

In this paper, we introduce the log-normal stochastic volatility (SV) model with a
quadratic drift to allow arbitrage-free valuation of options on assets under money-market
account and inverse martingale measures. We show that the proposed volatility process
has a unique strong solution, despite non-Lipschitz quadratic drift, and we establish
the corresponding Feynman–Kac partial differential equation (PDE) for computation of
conditional expectations under this SV model. We derive conditions for arbitrage-free
valuations when return–volatility correlation is positive to preclude the “loss of mar-
tingality”, which occurs in many traditional SV models. Importantly, we develop an
analytic approach to compute an affine expansion for the moment generating function of
the log-price, its quadratic variance (QV) and the instantaneous volatility. Our solution
allows for semi-analytic valuation of vanilla options under log-normal SV models closing
a gap in existing studies.
We apply our approach for solving the joint valuation problem of vanilla and inverse
options, which are popular in the cryptocurrency option markets. We demonstrate the
accuracy of our solution for valuation of vanilla and inverse options.a By calibrating
the model to time series of options on Bitcoin over the past four years, we show that
the log-normal SV model can work efficiently in different market regimes. Our model can

∗ Corresponding author.
This is an Open Access article published by World Scientific Publishing Company. It is distributed
under the terms of the Creative Commons Attribution 4.0 (CC BY) License which permits use,
distribution and reproduction in any medium, provided the original work is properly cited.
a Github project https://github.com/ArturSepp/StochVolModels provides Python code with the

implementation of option valuation using the affine expansion and examples of model calibration
to implied volatility data applied in this paper.

2450003-1
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

be well applied for modeling of implied volatilities of assets with positive return–volatility
correlation.

Keywords: Log-normal stochastic volatility; nonaffine models; closed-form solution;

moment generating function; cryptocurrency derivatives; quadratic variance.

JEL Classification: G13, C63

1. Introduction
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Empirical studies strongly support evidence that the volatility of price returns is
itself a stochastic process (see, for example, Shephard (2005) for a comprehensive
survey). It is accepted that the celebrated model by Black & Scholes (1973) and
Merton (1973), which assumes constant volatility, cannot explain implied volatil-
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

ity surfaces observed in option markets, which are inhomogeneous in strike and
maturity dimensions. In contrast, stochastic volatility (SV) models are able to fit
to implied volatility surfaces and their dynamics.

1.1. Evidence of log-normality of implied and realized volatilities

Applications of log-normal SV models, which are based either on the log-normal
dynamics for the instantaneous volatility or on the Gaussian dynamics for the loga-
rithm of the volatility, are widespread in practice. Predominantly, SABR model by
Hagan et al. (2002) is widely used among practitioners for fitting implied volatil-
ity surfaces. However, for modeling the dynamics of volatility surfaces, we need
to account for the mean-reversion of the volatility process like in conventional SV
models. The mean-reversion ensures that the distribution of the volatility has a sta-
tionary long-run distribution and volatility paths simulated in Monte Carlo (MC)
do not explode. In another widely used specification, the so-called Exp-OU specifi-
cation, the SV process is defined for the logarithm of the volatility.b
In an excellent study Christoffersen et al. (2010) examine the empirical per-
formance of Heston (see Heston (1993)), log-normal and 3/2 SV models using
three sources of market data: the VIX index, the implied volatility for options
of the S&P500 index, and the realized volatility of returns on the S&P500 index.
They found that, for all three sources of data, the log-normal SV model outper-
forms its alternatives. Tegner & Poulsen (2018) report similar finding using realized
volatilities.

1.2. Analytical tractability

Log-normal SV models cannot be handled by semi-analytic Fourier methods avail-
able for affine SV models. The analytical intractability may have lead to a slow

b Exp-OU SV models are studied in Fouque et al. (2000), Detemple & Osakwe (2000), Masoliver
& Perelló (2006), Perelló et al. (2008), Muhle-Karbe et al. (2022), Drimus (2012). Exp-OU SV
models are widespread in industry as a basis for local SV models (Lipton (2002)), see Bergomi
(2015), Bühler (2006), Ren et al. (2007), Henry-Labordère (2009), Bergomi & Guyon (2012).

2450003-2
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

adaptation of log-normal SV models in spite of their empirical support. On the

one hand, the log-normal SV model with the linear drift does not belong to the
family of affine diffusions because the so-called admissibility conditions are not sat-
isfied for the variance term, see Duffie et al. (2003), Filipović & Mayerhofer (2009).
On the other hand, the log-normal SV model with the quadratic drift does not
belong to a general family of polynomial diffusions, see Filipović & Larsson (2016),
Cuchiero et al. (2012). Carr & Willems (2019) introduce measure change to relate
the model to a class of polynomial diffusions, which comes at cost of introducing an
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

additional state variable. Authors then approximate the payoff function using an
orthogonal polynomial expansion, see Ackerer & Filipović (2020), for option pricing
applications. A related approach for computing moments under SV models using
expansions is developed by Alòs et al. (2020).
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

We contribute to these studies by developing a closed-form approximate and very

accurate solution for the log-normal SV model which enables analytical intractabil-
ity of this model. We develop an affine expansion to derive closed-form solution for
the moment generating function (MGF) of the log-price process, volatility and its
quadratic variance (QV). We show that our solution to the MGF produces valid
probability density functions of the state variables, which are matched with MC
simulations.

1.3. Properties of log-normal SV model

In a related research, Lewis (2018) introduces a log-normal SV model with a
quadratic drift without constant term. Lewis (2018) suggests to apply the model
only when return–volatility correlation is negative because, otherwise, the model
leads to the loss of martingality. Carr & Willems (2019) introduce the log-normal
SV model with the quadratic drift, similar to our model. In our paper we augment
the log-normal SV model with the linear drift term introduced in Karasinski &
Sepp (2012) to include the quadratic drift term. We contribute to studies of Lewis
(2018) and Carr & Willems (2019) by defining the martingale conditions under
both MMA and inverse martingale measures. We also formulate the Feynman–Kac
theorem linking the model dynamics with the quadratic drift to the partial differ-
ential equation (PDE) approach, because the classic Feynman–Kac does not apply
to non-Lipschitz coefficients. We then introduce the MGF as the solution to the
model PDE and develop an analytic transform-based approach for the valuation of
vanilla and inverse options.

1.4. Assets with positive return–volatility correlation

Carr & Wu (2017) propose the following three economic factors behind the negative
return–volatility correlation, which most typically observed in the dynamics of stock
prices and stock indices: aggregate financial leverage, systematic risk, positive auto-
correlation of negative shocks. However, in many markets we actually observe price
dynamics with positive return–volatility correlation either on a permanent basis

2450003-3
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

(for VIX index, short and leveraged short exchange-traded funds (ETFs)) or on a
transitory regime-based basis (for some commodities, foreign currencies, and cryp-
tocurrencies). In fact, exp-ou and log-normal SV models with the linear drift lead to
the “loss of martingality” of the price process when the return–volatility correlation
is positive (see, for example, Lewis (2000, 2018) and Lions & Musiela (2007)). We
contribute by establishing conditions on model parameters of log-normal SV model
which produce true martingale dynamics.
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

1.5. Inverse martingale measure

For inverse options, both option premium and final payoff are quoted in the units
of the underlying asset. In particular, for options on cryptocurrencies, the advan-
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

tage of inverse payoffs is that all option-related transactions can be handled using
units of underlying cryptocurrencies, such as Bitcoin or Ether, without using fiat
currencies (see Lucic (2021) and Lucic & Sepp (2023) for a connection between the
valuation of the inverse cryptocurrency options and FX options). In case of cryp-
tocurrencies, cash settled options would require using a stablecoin as a tradable
equivalent of United States Dollar (USD) or any other fiat currency. Given inherent
and frequent depegging risk of stablecoins, Alexander et al. (2023) argue for wider
adoptation of inverse and quanto options. We also suggest that similar payoff can
be advantageous for some commodity markets (such as gold) where option holders
may opt for physical settlement of positive option premium through the delivery of
the underlying commodity.
We contribute to the literature by incorporating the inverse measure in our
valuation framework using log-normal SV model. We formulate and solve the joint
valuation problem of vanilla and inverse options using Fourier transform methods.

1.6. Organization of the paper

In Sec. 2, we provide the framework for valuation under money-market account
(MMA) and inverse measures. In Sec. 3, we introduce the log-normal beta SV model
with quadratic drift and establish important properties of the volatility process. We
further characterize martingale property for the drift-adjusted price process and its
inverse under MMA and inverse measures, respectively. We also study moments of
the volatility process by presenting an approximate solution of the finite-dimensional
linear system of ordinary differential equations (ODEs). Finally, we present drift-
implicit Euler scheme for MC simulation that preserves the positivity of the volatil-
ity process and study its convergence properties. In Sec. 4, we present first- and
second-order affine expansion for the MGF of the log-price process, the QV and the
volatility. In Sec. 5, we apply affine expansion to derive closed-form solutions for
option valuation problem under MMA and inverse measures. In Sec. 6, we apply
our method for model calibration to market data and demonstrate the accuracy of
both the model specification and our proposed solution methods. We conclude in
Sec. 7. Technical proofs are provided in Appendix A.

2450003-4
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

2. Framework and Definitions

2.1. Vanilla and inverse payoffs
The payoff of (cash-settled) vanilla options is settled in the units of cash. In con-
trast, the payoff of an inverse option is settled in the units of the underlying asset
using the spot price observed at the option maturity. We denote the spot price at
time T by ST and option strike price by K, with both quoted in United States
Dollars (USD). We consider vanilla call and put payoffs settled in USD cash at
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

maturity time T and denote them by

ucall (ST ) = max{ST − K, 0}, uput (ST ) = max{K − ST , 0}. (2.1)
Payoffs of inverse call and put options are converted to units of the underlying asset
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

at maturity T as follows:
1 1
rcall (ST ) = max{ST − K, 0}, rput (ST ) = max{K − ST , 0}. (2.2)
ST ST

2.2. Valuation under equivalent MMA and inverse measures

We consider a continuous-time market with a fixed horizon date T ∗ > 0 and
uncertainty modeled on probability space (Ω, F, P) equipped with filtration F =
{Ft }0≤t≤T ∗ . We assume that F is right-continuous and satisfies usual conditions.

2.2.1. Spot markets

Assumption 2.1. Risk-free rate r(t) is deterministic with the value of one unit of
T
the MMA M (T ) given by M (T ) = er̄(0,T ) , r̄(t, T ) = t r(s)ds.

We assume that the market is complete and satisfies the so-called No Free Lunch
with Vanishing Risk (NFLVR) and No Dominance (ND) conditions (see Jarrow &
Larsson (2012)). By choosing M as a numéraire, we consider an equivalent martin-
gale measure Q, induced by M . Accordingly, the time-t value of an option, denoted
by U (t, S), with payoff function u(ST ) at time T , equals
 
1
U (t, S) = M (t)E u(ST )|Ft , (2.3)
M (T )
where expectation E is taken under the MMA martingale measure Q.
Next, by choosing spot price S as a numéraire, we consider an inverse martingale
measure Q̃, induced by S. The option value function, denoted by Ũ (t, S), with the
payoff paid in units of S, equals
 
1
Ũ (t, S) = St Ẽ u(ST )|Ft , (2.4)
ST
where expectation Ẽ is taken under the inverse martingale measure Q̃.

Theorem 2.1. We assume that market is complete and satisfies NFLVR and ND
conditions. We further assume that both MMA measure Q and inverse measure Q̃

2450003-5
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

are equivalent martingale measures. Then the values of options under MMA measure
in Eq. (2.3) and under inverse measure in Eq. (2.4) are unique and satisfy
   
1 1
M (t)E u(ST )|Ft = St Ẽ u(ST )|Ft . (2.5)
M (T ) ST

Proof. As shown in Theorem 3.2 of Jarrow & Larsson (2012) and Theorem 2.17
of Herdegen & Schweizer (2018), the market satisfies NFLVR and ND conditions
if and only if Q is a true martingale measure. Hence, asset prices, discounted by
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

numéraire M are true Q-martingales. Consequently, due to the uniqueness of price,

the equality in Eq. (2.5) holds. To justify the equality of both sides, we apply the
existence and uniqueness of equivalent inverse (true) martingale measure Q̃ for the
numéraire S, see Theorem 4.4 and Corollary 3.11 in Herdegen & Schweizer (2018).
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

2.2.2. Futures markets

Since most option markets for commodities and cryptocurrencies are based on
futures, we also adopt our methodology for futures options. We consider a set of
futures contracts settled at times {Tk }k=1,...,K with prices {FtTk }k=1,...,K .
Assumption 2.2. The convenience yield q(t) of futures prices is deterministic with
the integrated convenience yield given by
 T
q̄(t, T ) = q(s)ds. (2.6)
t

Definition 2.1. Measure Q̃ induced by price FtT with futures maturity at T ,

T

T ∈ {Tk }, chosen as a numéraire is called the inverse T -futures measure.

Corollary 2.1. We assume that market is complete and satisfies NFLVR and ND
conditions. We further assume that both MMA measure Q and inverse futures mea-
sure Q̃T are equivalent martingale measures. Then option values under MMA and
inverse measures are unique and satisfy
   
1 1
M (t)E u(FT )|Ft = Ft ẼT u(FT )|Ft . (2.7)
M (T ) FT
For markets with both vanilla and inverse options being tradable, such as cryp-
tocurrency markets, any SV model applied for joint valuation of options across
several exchanges and contract specifications must satisfy the conditions (2.5) and
(2.7). For the proposed log-normal SV dynamics in Eq. (3.12), we must impose
certain conditions on model parameters, which we address in Theorem 3.7.

2.2.3. Put–call parity

We will discuss an important topic of the put–call parity for inverse options under
a generic SV model with SV driver σt . We establish that the put–call parity is valid
if volatility σt is not explosive.

2450003-6
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Lemma 2.1. We assume that volatility process σt is nonexplosive under Q.

Then put–call parity for vanilla call option, C(t, S, K), and put option, P (t, S, K),
defined by
C(t, S, K) − P (t, S, K) = S − Ke−r̄(t,T ) (2.8)
holds irrespective whether the price process is a true martingale under Q.
We assume that volatility process σt is nonexplosive under Q̃. Put–call parity
for inverse call option, C̃(t, S, K), and inverse put option, P̃ (t, S, K), defined by
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

C̃(t, S, K) − P̃ (t, S, K) = 1 − S −1 Ke−r̄(t,T ) (2.9)

holds irrespective whether (drift-adjusted) inverse price process is a true martingale
under Q̃.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

Proof. We only provide a sketch of the proof and refer to Theorems 9.2 and 9.3 in
Lewis (2000) and to Sin (1998) for details. As vanilla call option payoff is unbounded,
the price of the call option is only local martingale under Q. Taking into account
“correction” term (so-called martingale defect) reflecting the difference between true
martingale and strict local martingale properties, yields Eq. (2.8). Relationship in
Eq. (2.9) follows from the same argument, but applied to Q̃.

In Theorem 3.2, we show that volatility process σt does not explode under the
proposed log-normal SV dynamics in Eq. (3.12).

3. Log-Normal Beta SV Model

3.1. Dynamics under P measure
We start with statistical P measure and we consider the log-normal beta SV model
introduced by Karasinski & Sepp (2012). We augment the dynamics with the
quadratic drift in the volatility process following Lewis (2018) and Carr & Willems
(2019). We first introduce the dynamics of price process St and volatility process
σt under statistical measure P as follows:
(0)
dSt = μt St dt + σt St dWt ,
(3.1)
(0) (1)
dσt = (κ1 + κ2 σt )(θ − σt ) dt + βσt dWt + εσt dWt ,

where μt is statistical drift, W (0) , W (1) are uncorrelated Brownian motions under
P, κ1 > 0 and κ2 ≥ 0 are linear and quadratic mean-reversion rates, respectively,
θ > 0 is the mean of the volatility, β ∈ R is the volatility beta which measures the
sensitivity of the volatility to changes in the spot price, and ε > 0 is the volatility
of residual volatility.
Empirically, Bakshi et al. (2006) find that the presence of nonlinear terms,
including the quadratic term, in the volatility drift is significant when using econo-
metric estimation of the dynamics of the VIX index.

2450003-7
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

We assume that statistical measure P and MMA measure Q are related by

   t 
dQ 
ζ(t) := =E − (0) (1)
λ0 (u)dWu + λ1 (u)dWu , (3.2)
dP Ft 0

where E(·) denotes Itô exponential, and λ0 (t) and λ1 (t) are equity and volatility
risk-premias, respectively.
Assuming that process ζ(t) is a true martingale, the following processes
 t  t
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

(0) (0) (1) (1)

Ŵt = Wt + λ0 (u)du, Ŵt = Wt + λ1 (u)du (3.3)
0 0

define independent Brownian motions under Q by Girsanov theorem (we establish

Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

necessary and sufficient conditions for equivalence of measures in Theorem 3.1).

Thus, the dynamics of price process in Eq. (3.1) under MMA measure Q becomes
(0)
dSt = (μt − λ0 (t)σt )St dt + σt St dŴt .

As discounted price process must be a martingale under Q, we need to require that

μt − rt
μt − λ0 (t)σt = rt ⇔ λ0 (t) = . (3.4)
σt
As a result, the dynamics of volatility process under MMA measure Q become

dσt = [κ1 θ − (κ1 − κ2 θ)σt − κ2 σt2 − (βλ0 (t) + ελ1 (t))σt ]dt
(0) (1)
+ βσt dŴt + εσt dŴt , (3.5)

where Ŵ (0) , Ŵ (1) are independent Brownian motions under Q.

As customary,c we assume that the market price of risk is proportional to the
volatility as follows:

λ0 (t) = λ̄0 σt , λ1 (t) = λ̄1 σt , (3.6)

where λ̄0 and λ̄1 are constants. As a result, we rewrite Eq. (3.5) as follows:
 
dσt = κ1 θ − (κ1 − κ2 θ)σt − (κ2 + β λ̄0 + ελ̄1 )σt2 dt
(0) (1)
+ βσt dŴt + εσt dŴt . (3.7)

To retain the functional form of the volatility process, we rescale model param-
eters θ̂, κ̂1 , κ̂2 under Q as follows:

κ1 θ = κ̂1 θ̂, κ1 − κ2 θ = κ̂1 − κ̂2 θ̂, κ2 + β λ̄0 + ελ̄1 = κ̂2 . (3.8)

c Such linear specification is the simplest functional form that allows to establish simple relationship
between gains on a delta-hedged option portfolio and level of the market volatility. As shown in
Bakshi & Kapadia (2003), such form allows to get precise information about the sign and strength
of the risk premium.

2450003-8
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Analytic inversion of Eq. (3.8) yields risk-transformed model parameters under Q:

κ̂2 = κ2 + β λ̄0 + ελ̄1 ,

1
θ̂ = (−(κ1 − κ2 θ) + (κ1 − κ2 θ)2 + 4θκ1 κ̂2 ), (3.9)
2κ̂2
1
κ̂1 = ((κ1 − κ2 θ) + (κ1 − κ2 θ)2 + 4θκ1 κ̂2 ).
2
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Thus, under martingale measure Q, the dynamics of price and volatility pro-
cesses become
(0)
dSt = rt St dt + σt St dŴt ,
(3.10)
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(0) (1)
dσt = (κ̂1 + κ̂2 σt )(θ̂ − σt ) dt + βσt dŴt + εσt dŴt .

Theorem 3.1. We assume that κ1 ≥ 0, θ > 0 . Then measures P and Q are

equivalent, iff

κ2 ≥ max{−β λ̄0 − ελ̄1 , 0}. (3.11)

Proof. By analogy to the proof of Theorem 3.6.

As a result, we obtain that the functional form of log-normal SV model with

the quadratic drift is invariant under the change of measure from P to Q. We note
that the log-normal volatility model with linear drift does not share such invariance
property due to a quadratic term arising under Q. The model invariance under the
change of measure is an important feature shared by the log-normal SV model with
quadratic drift. In contrast, Sepp & Rakhmonov (2023) show that most conventional
SV models do not satisfy the invariance of P-model under the change of measure
to Q. In the following, we will consider model dynamics under Q and drop the hat
notation in Eq. (3.10).

3.2. Dynamics under MMA measure

We consider price dynamics with the log-normal beta SV model in Eq. (3.1) under
the MMA measure Q. The joint dynamics are specified for the spot price St , the
instantaneous volatility σt and the QV It under the MMA measure Q as follows:
(0)
dSt = r(t)St dt + σt St dWt , S0 = S,
(0) (1)
dσt = (κ1 + κ2 σt )(θ − σt )dt + βσt dWt + εσt dWt , σ0 = σ, (3.12)

dIt = σt2 dt, I0 = I,

(0) (1)
where r(t) is the deterministic risk-free rate; Wt and Wt are uncorrelated Brow-
nian motions, and the model parameters are the same as defined in Eq. (3.1). Using

2450003-9
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

volatility beta β and volatility-of-volatility ε, we introduce the total instantaneous

variance of the volatility process ϑ2 as follows:
ϑ2 = β 2 + ε2 . (3.13)
The mean-reversion of the volatility process is specified using the mean level of
the volatility θ, θ > 0, and the mean-reversion speed which is the linear function
of the volatility κ1 + κ2 σt , κ1 ≥ 0, κ2 ≥ 0. As a result, the quadratic drift of the
volatility process is given by
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

μ(σ) ≡ (κ1 + κ2 σ)(θ − σ) = κ1 θ − (κ1 − κ2 θ)σ − κ2 σ 2 . (3.14)

Rt  
We introduce the zero-drift stochastic driver Zt , Zt = e− 0
r(t )dt
St and its
inverse Rt , Rt = Zt−1 , with the following dynamics:
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(0)
dZt = Zt σt dWt , Z 0 = S0 ,
(3.15)
(0)
dRt = σt2 Rt dt − Rt σt dWt , R0 = 1/Z0 .
Next, we introduce the log of the zero-drift price process Xt = ln Zt :
1 (0)
dXt = − σt2 dt + σt dWt , X0 = ln S0 . (3.16)
2
Using XT under Assumption 2.1, we represent the spot price ST by
ST = er̄(0,T ) eXT . (3.17)
Similarly, under Assumption 2.2, using Eq. (3.17), we represent the futures settled
at time T in Q as
fT (T ) = er̄(0,T )−q̄(0,T ) eXT , (3.18)
where the integrated convenience rate q̄(0, T ) is defined in Eq. (2.6). Accordingly,
for modeling purposes, we focus on the zero-drift price dynamics Zt and its log-price
process Xt defined in Eq. (3.16).

3.2.1. Properties of log-normal SV process

We establish important properties related to the regularity of the volatility process.
We consider the volatility process in model dynamics (3.12) represented as follows:
(∗)
dσt = μ(σt )dt + v(σt )dWt , σ0 = σ,
(3.19)
μ(σ) = (κ1 + κ2 σ)(θ − σ), v(σ) = ϑσ,
(∗)
where Wt is a standard Brownian motion and ϑ is the total volatility of volatility.
We notice that the process σt is not a polynomial diffusion, see Filipović & Lars-
son (2016), Lemma 2.2, because the drift coefficient is a second-order polynomial
in σt . Since the process in Eq. (3.19) has super-linearly growing drift, standard
results for the uniqueness of strong solution are no longer applicable, see Sec. 5 of

2450003-10
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Karatzas & Shreve (1991) for details. We will establish the existence of the unique
strong solution in Theorem 3.3. We first provide a boundary classification for the
volatility process to show that log-normal volatility process σt in dynamics (3.19)
cannot reach 0 or explode to +∞, which are important properties of the volatility
dynamics. Our proof is based on existence of solutions for general stochastic differen-
tial equations (SDEs) with so-called monotonic coefficients, so that it is potentially
applicable to wider class of diffusion processes.
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Assumption 3.1. We assume that model parameters κ1 , κ2 , θ, ϑ satisfy

κ1 ≥ 0, κ2 ≥ 0, θ > 0, ϑ ≥ 0. (3.20)
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

Theorem 3.2 (Regularity). Under Assumption 3.1, the boundary points {0, +∞}
are unattainable for process σt in SDE (3.19).

Proof. Our proof is based on Feller boundary classification for one-dimensional

diffusion processes with technical details given in Sec. A.1.

Theorem 3.3 (Existence and uniqueness of solution). Under Assump-

tion 3.1, SDE in Eq. (3.19) has the unique strong solution.

Proof. We will refer to following theorem that guarantees the existence and unique-
ness of general SDEs with monotonic coefficients.

Theorem 3.4 (Gyöngy & Krylov (1980)). Let at (x) : [0, T ] × R → R, bt (x) :
[0, T ] × R → R be functions, continuous in x, satisfying following conditions:
T
(1) 0 sup|x|≤R (|at (x)| + |bt (x)|2 )dt < ∞ for any T ≥ 0, R ≥ 0,
(2) for x, y, z ∈ R such that |x| ≤ R, |y| ≤ R, |z| ≤ R following conditions hold:
(a) (monotonicity condition) :

2(x − y)(at (x) − at (y)) + (bt (x) − bt (y))2 ≤ Kt (R)(x − y)2 , (3.21)

(b) (growth condition) : 2zat (z) + (bt (z))2 ≤ Kt (1)(1 + z)2 ,

t
where function Kt (R) : [0, T ] × R+ → R+ satisfies 0 Ks (R)ds < ∞ for any
T ≥ 0, R ≥ 0.

Then there exists unique solution of the SDE

 t  t
xt = x0 + as (xs )ds + bs (xs )dWs , t ≤ T. (3.22)
0 0

We show that SDE (3.19), where κ1 , κ2 , θ, ϑ satisfy (3.20), does indeed satisfy
conditions 1–3. We emphasize that the positivity of the process demonstrated in

2450003-11
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

Theorem 3.2 is important. Since in our case at (x) = (κ1 + κ2 x)(θ − x), bt (x) = ϑx,
we get
2(x − y)(at (x) − at (y)) + (bt (x) − bt (y))2
= (ϑ2 − 2(κ1 − κ2 θ))(x − y)2 − 2κ2 (x + y)(x − y)2
≤ (ϑ2 − 2(κ1 − κ2 θ) + 2κ2 R)(x − y)2
t
when |x| ≤ R, |y| ≤ R. As for any R > 0 we have 0 [ϑ2 −2(κ1 −κ2 θ)+2κ2 R]ds < ∞,
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

we conclude that 2a is satisfied. For the growth condition, we have

2zat (z) + bt (z)2 = 2κ1 θz + [ϑ2 − 2(κ1 − κ2 θ)]z 2 − 2κ2 z 3 . (3.23)
Using that zero is unattainable boundary and simple inequalities z ≤ 1 + z 2 ,
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

3
z < 0 valid for z > 0, we can estimate the right hand side of (3.23) as follows:
2κ1 θ(z 2 + 1) + [ϑ2 − 2(κ1 − κ2 θ)]z 2 ≤ 2κ1 θ + [2κ1 θ + ϑ2 − 2(κ1 − κ2 θ)]z 2 .
We conclude that the growth condition 2b is also satisfied.

Proposition 3.1 (Bounded moments). Assume that κ1 θ > 0, κ2 > 0 and

remaining parameters κ2 , θ, ϑ satisfy conditions (3.20). Let |p| ≥ 1 and T > 0.
Then we have

E sup σtp < ∞.

t∈[0,T ]

Proof. First we prove a slightly weaker result.

Proposition 3.2. Assume that κ1 θ > 0, κ2 > 0 and κ2 , θ, ϑ satisfy conditions

(3.20). Then, for every |p| ≥ 1 : supt∈[0,T ] Eσtp < ∞ for any T > 0.

Proof. Let us first consider the case p ≥ 1. We use same approach as in Lions &
Musiela (2007). Applying Itô’s lemma to σtp , we have
 
ϑ2
dσtp = pσtp−1 κ1 θ − κ1 − κ2 θ − p(p − 1) σt − κ2 σt2 dt + pϑσtp dWt .
2
The function U (τ ) = E(σtp ) then solves the problem
∂U
− + L(σ) U = 0, U (0, σ) = σ p ,
∂τ
(3.24)
∂ 1 2 2 ∂2
L = (κ1 θ − (κ1 − κ2 θ)σ − κ2 σ )
(σ) 2
+ ϑ σ .
∂σ 2 ∂σ 2
Let us find supersolution Ū to (3.24) using ansatz Ū (σ) = σ p exp(aσ + bτ ), a > 0,
b > 0. Inserting it into (3.24), we see that its left-hand side equals

ϑ2
σ p exp(aσ + bθ) −b + aκ1 θ − p(κ1 − κ2 θ) + p(p − 1) + pκ1 θσ −1
2

ϑ2
+ (pϑ2 − a(κ1 − κ2 θ) − κ2 p)σ + −κ2 a + a2 σ2 .
2

2450003-12
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Since the term in the brackets is a second-order polynomial in σ, Ū is a supersolution

as soon as κ2 > 0.
The case p ≤ −1 is treated similarly: we denote q := −p ≥ 1 and introduce
σ̃t := σt−1 which is well-defined due to Theorem 3.2. We find that Ũ (τ ) := E(σ̃tq )
solves the problem
∂ Ũ
− + L̃(σ) Ũ = 0, Ũ (0, σ) = σ̃ q ,
∂τ
(3.25)
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

∂ 1 ∂2
L̃ = (κ2 + (κ1 − κ2 θ + ϑ )σ̃ − κ1 θσ̃ )
(σ) 2
+ ϑ2 σ̃ 2 2 . 2
∂ σ̃ 2 ∂ σ̃
Inserting ansatz into (3.25), we see that its left-hand side equals
σ̃ q−1 exp(aσ̃ + bθ)[−b + (· · ·) + (· · ·)σ̃ + (−2aκ1 θ + a2 ϑ2 )σ̃],
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

where we omitted the terms not depending on b and σ̃ for brevity. Since the term in
square brackets is a second-order polynomial in σ̃, Ũ is a supersolution if κ1 θ > 0.

Similarly to the proof of Proposition 3.2, we apply Itô’s formula to σtp

 t  t
p p ϑ2
σt = σ0 + p−1
σu (κ1 + κ2 σu )(θ − σu ) + p(p − 1)σu dt + pϑ
p
σup dWu(∗) .
0 2 0
We obtain
 t 
 p−1 ϑ2 
E sup σtp ≤ σ0p + E sup  p
pσu (k0 + k1 σu )(θ − σu ) + 2 p(p − 1)σu  dt
t∈[0,T ] t∈[0,T ] 0

 t
+ pϑE sup σup dWu(∗) .
t∈[0,T ] 0

Using Burkhölder–Davis–Gundy inequality and Proposition 3.2, we establish

 u  1/2  1/2
T T
p 2p
E sup p (∗)
σu dWu ≤ CE σt dt ≤C Eσt .
t∈[0,T ] 0 0 0

3.3. Dynamics under inverse measure

First we define the model dynamics (3.12) under the inverse measure Q̃. As a by-
product, we show that in SDE (3.26) the quadratic term is important to preserve
the functional specification of volatility drift under Q̃.

Theorem 3.5. The dynamics of the volatility σt in (3.12) under Q̃ are given by
t(0) + εσt dW
dσt = (κ1 θ − (κ1 − κ2 θ)σt − (κ2 − β)σt2 )dt + βσt dW t(1) . (3.26)
MMA measure Q and inverse measure Q̃ are related by the density process
Λt = Et [dQ̃/dQ] = Zt /Z0 , Λ0 = 1, (3.27)
where zero-drift stochastic driver Zt is defined in Eq. (3.15).

2450003-13
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

Proof. We consider Rt = Zt−1 . Applying Itô’s lemma, we see that Rt satisfies

(0) (0)
dRt = σt2 Rt dt − σt Rt dWt = −σt Rt (dWt − σt dt).
Using Girsanov’s theorem, we switch to the Q̃ under which the processes
 measure (1)
t(0) and W
W t(1) defined by W
t(0) = Wt(0) − t σs ds, W
t = Wt(1) are uncorrelated
0
Brownian motions. Then the dynamics of volatility process σt under Q̃ become
t
dσt = (κ1 + κ2 σt )(θ − σt )dt + βσt (dW
(0) t
+ σt dt) + εσt dW
(1)
(3.28)
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

and Eq. (3.26) follows. We note that Q̃ and Q are related by the density Λt as
follows:
 t 
1 t 2
dΛt /Λt = σt dWt ⇒ Λt := exp σs dWs − σ ds = Zt /Z0 . (3.29)
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

0 2 0 s

We highlight that the idea of using change of measure to study the martingale
property is widely used in the literature, see Ruf (2015) and references therein. We
now derive sufficient conditions for the equivalence of measures Q ∼ Q̃ which is
required for the application of Girsanov theorem in the proof of Theorem 3.5.

Theorem 3.6. Under Assumption 3.1, measures Q and Q̃ are equivalent if and
only if κ2 ≥ β.

Proof. By Theorem 3.5, the density Et [dQ̃/dQ] = Zt /Z0 . Therefore, the martingale
property for Zt under Q guarantees equivalence Q ∼ Q̃. Using Theorem 3.7(1)
concludes the proof.

Theorem 3.7 (Martingality of Log-normal SV model with quadratic

drift). Under Assumption 3.1, the price processes Zt and Rt in Eq. (3.15) for
model dynamics (3.12) have following properties:
(1) The process Zt 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.

Proof. (1) Theorem 9.2 in Lewis (2000) or Sin (1998) shows that Zt is Q-martingale
if and only if the process

dσt = ((κ0 + κ1 σt )(θ − σt ) + βσt2 )dt + ϑσt dW
(∗)
t

has unique strong solution under Q̃. Using Theorem 3.3, we immediately obtain
that the uniqueness holds iff κ2 ≥ β.
(2) According to Theorem 3.5, the dynamics of {σt , Rt } under Q̃ become
t
dσt = ((κ1 + κ2 σt )(θ − σt ) + βσt2 )dt + βσt dW
(0) t ,
+ εσt dW
(1)

(3.30)
t(0) .
dRt = (−σt )Rt dW

2450003-14
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Using Theorems 2.4(i), 2.4(ii) of Lions & Musiela (2007), Rt is a martingale if

lim sup(βξ 2 + (κ1 + κ2 ξ)(θ − ξ) + βξ 2 )ξ −1 < ∞,
ξ↑+∞

which holds if κ2 ≥ 2β. Rt is not a martingale if for some smooth, positive, increasing
function ϕ we have
lim inf (βξ 2 + (κ1 + κ2 ξ)(θ − ξ) + βξ 2 )ϕ(ξ)−1 > 0, (3.31)
ξ↑+∞
 +∞
ϕ(ξ)−1 dξ < ∞, ε > 0. Choosing ϕ(ξ) = ξ 2 , (3.31) holds if κ < 2β.
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

where ε

Corollary 3.1 (Martingality of Log-normal SV model with linear drift).

For dynamics (3.12) with κ2 = 0 we obtain the following:
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(1) The process Zt is martingale under Q iff β ≤ 0.

(2) The process Rt is martingale under Q̃ iff β ≤ 0.

Proof. Follows by setting κ2 = 0 in Theorem 3.7.

Importantly we conclude that the log-normal SV model with linear drift leads
to arbitrages when the return–volatility correlation is positive.

3.4. Joint dynamics of state variables under Q and Q̃

We introduce the mean-adjusted volatility process Yt
  (∗)
dYt = −κYt − κ2 Yt2 dt + ϑ (Yt + θ) dWt , Y0 ≡ Y = σ0 − θ,
(3.32)
κ = κ1 + κ2 θ
on the domain (−θ, ∞). The marginal dynamics of Yt using SDE (3.19) become
(∗)
dYt = (−κYt − κ2 Yt2 )dt + ϑ(Yt + θ)dWt , Y0 ≡ Y = σ0 − θ, (3.33)
(∗)
where Wt is a standard Brownian motion.
Corollary 3.2 (Dynamics under the MMA measure Q). The joint dynamics
of log-price Xt , the mean-adjusted volatility process Yt , and the QV It under the
MMA measure Q are driven by
1 (0)
dXt = − (Yt + θ)2 dt + (Yt + θ)dWt , X0 = X,
2
(0) (1)
dYt = (−κYt − κ2 Yt2 )dt + β(Yt + θ)dWt + ε(Yt + θ)dWt , Y0 = Y,
dIt = (Yt + θ)2 dt, I0 = I. (3.34)
Itô’s lemma, applied to Eq. (3.33), yields marginal dynamics of Y under Q̃:
t(∗) ,
dYt = (λ̃ − κ̃Yt − κ̃2 Yt2 )dt + ϑ(Yt + θ)dW Y0 = Y,
(3.35)
λ̃ = βθ2 , κ̃ = κ1 − κ2 θ + 2(κ2 − β)θ, κ̃2 = κ2 − β,
t(∗) is a standard Brownian motion under Q̃.
where W

2450003-15
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

Corollary 3.3 (Dynamics under the inverse measure Q̃). The joint dynamics
of log-price Xt , the mean-adjusted volatility process Yt , and the QV It under the
inverse measure Q̃ are driven by
1 t(0) ,
dXt = (Yt + θ)2 dt + (Yt + θ)dW X0 = X,
2
t
dYt = (λ̃ − κ̃Yt − κ̃2 Yt2 )dt + β(Yt + θ)dW
(0) t ,
+ ε(Yt + θ)dW
(1)
Y0 = Y,
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

dIt = (Yt + θ)2 dt, I0 = I. (3.36)

3.5. Steady-state distribution of volatility process

Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

It is important to establish that the steady-state density (SSD) of the volatility

process σt in SDE (3.19) exists. The SSD, denoted by G(σ), solves the following
ODE:
1 2 ∂2 2 ∂
ϑ (σ G) − ((κ1 + κ2 σ)(θ − σ)G) = 0. (3.37)
2 ∂σ 2 ∂σ
The solution is Generalized Inverse Gaussian distribution (Jørgensen (1982)):
 q 
G(σ) = cσ η−1 exp − + bσ , σ > 0,
σ
(3.38)
κ1 θ κ2 κ2 θ − κ1 (b/q)η/2
q = 2 2 , b = 2 2, η = 2 − 1, c(b) = √ ,
ϑ ϑ ϑ2 2Kη (2 qb)
where b > 0, q ≥ 0, η ∈ R and Kη is a modified Bessel function of the second kind.
For κ2 = 0 we use c(0) = q −η Γ(−η), where Γ(x) is the gamma function, and the
steady state PDF is the inverse gamma function. The rth moment m(r) associated
to the density in Eq. (3.38) is given by
⎧ r/2 √
⎪
⎪ q Kη+r (2 qb)
√
⎪
⎪ , b > 0, r ∈ R+ ,
⎪
⎪ b Kη (2 qb)
⎨
m(r) = Γ(−η − r) (3.39)
⎪
⎪ qr , b = 0, r < −η,
⎪
⎪ Γ(−η)
⎪
⎪
⎩
∞, b = 0, r ≥ −η.
We see that when κ2 = 0, the volatility moments exist only to the order r less than
r < 1 + 2κ1 /ϑ2 .
In Fig. 1(a), we show the SSD for the three sets of model parameters. The case
(κ1 = 4, κ2 = 0) corresponds to the linear mean-reversion; the case (κ1 = 4, κ2 = 4)
(generally the case with κ2 = κ1 /θ) corresponds to the pure quadratic drift; the case
(κ1 = 4, κ2 = 8) corresponds to the quadratic drift. The linear model with κ2 = 0
implies the heavy right tail for the distribution of the volatility. In Fig. 1(b), we
show the skewness of the volatility as function of κ2 . As κ2 increases, the skewness
of the volatility declines.

2450003-16
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(a) Steady state distribution of the volatility (b) Skewness of volatility as function of k2

(c) Excess kurtosis of log-returns as function of

k2

Fig. 1. (a) The steady state PDF of the volatility computed using Eq. (3.38) with fixed κ1 = 4
and κ2 = {0, 4, 8}. (b) and (c) The skewness of the volatility computed using Eq. (3.39) and the
excess kurtosis of the unconditional returns distribution computed using Eq. (3.44), 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.

3.5.1. Unconditional distribution of returns

We make an interesting connection between the skewness of the volatility process
and the kurtosis of returns distribution. Following Chap. 9 in Barndorff-Nielsen
& Shiryaev (2015), we assume that the distribution of returns over period δt is
Gaussian conditional on σ:
1 x2
q(x|σ) = √ exp − √ . (3.40)
2πσ 2 δt 2σ δt

2450003-17
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

The unconditional distribution of returns under the SDD of the volatility is

 ∞
p(x) = q(x|σ)G(σ)dσ. (3.41)
0
The above integral cannot be computed explicitly. Instead, we compute the central
moments of returns under the unconditional distribution
 ∞  ∞  ∞ 
m(n) = xn p(x)dx = xn q(x|σ)dx G(σ)dσ, (3.42)
−∞ 0 −∞
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

where we exchange the integration order. It is clear that for odd n, m(n) = 0. For
the second and the fourth moments, we have
 ∞  ∞
2 2
m(2) = δt σ G(σ)dσ, m(4) = 3(δt) σ 4 G(σ)dσ. (3.43)
0 0
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

We compute the excess kurtosis of returns distribution using k = 3m(4)/m2 (2)−

3 with Eq. (3.39) as follows:
⎧ √ √
⎪ 3K (2 qb)Kη (2 qb)
⎪ η+4 √ − 3, κ2 > 0,
⎪
⎪
⎪
⎪ (Kη+2 (2 qb))2
⎨
k= q2 (3.44)
⎪
⎪ 3 − 3, κ2 = 0, κ1 /ϑ2 > 3/2,
⎪
⎪ η 2
⎪
⎪
⎩
∞, κ2 = 0, κ1 /ϑ2 ≤ 3/2.
It follows that the kurtosis of the unconditional distribution in Eq. (3.44) is
maximal when b = 0 (κ2 = 0). The kurtosis is finite only if liner mean-reversion
parameter κ1 is higher than the total volatility of variance. The kurtosis declines
when rate b (κ2 ) increases. In Fig. 1(c), we show the kurtosis for the three choices
of κ1 = 1, 4, 8. We see that for the each choice, the parameter κ2 can be thought
as a dampening parameter to reduce both the skewness of the steady-state PDF of
the volatility and the kurtosis of the unconditional PDF of returns.

3.6. Moments of volatility process

Computation of moments for the log-normal SV with the quadratic drift is an open
problem with no exact solution. We show that there exists a closed-form solution
based on a truncation method, which we further adopt for the affine expansion of
the MGF. We consider the mean-adjusted volatility process defined in Eq. (3.32).
(n) (n) (n)
We define its power function mt , mt = Ytn , and the mth moment, mτ , n =
0, 1, 2, . . . , as follows:
(n)
mt (τ ) = E[m(n)
τ ]. (3.45)
(n)
Applying Itô’s lemma to mt under the dynamics in Eq. (3.33), we obtain
(n) (∗)
dmt = (−κYt − κ2 Yt2 )nYtn−1 dt + c(n)(Yt + θ)2 Ytn−2 dt + ϑ(Yt + θ)nY n−1 dWt
(3.46)
(n)
with m0 = Y0n and c(n) = 12 ϑ2 n(n − 1). We notice a pattern of the powers of n
which allows for a recursive solution as follows.

2450003-18
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Proposition 3.3 (Moments of volatility process). The solution to moments in

Eq. (3.45) can be presented as a matrix equation for an infinite-dimensional vector
∂τ M (0,∞) (τ ) = Λ(0,∞) M (0,∞) (τ ), (3.47)
where ∂τ is the derivative w.r.t. τ and
M (0,∞) (τ ) = (m(0) (τ ), m(1) (τ ), m(2) (τ ), m(3) (τ ), . . .)T ,
M (0,∞) (0) = (1, Y0 , Y02 , Y03 , . . .)T ,
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

⎛ ⎞
0 0 0 0 0 0 ···
⎜ 0 −κ −κ2 0 0 0 · · ·⎟
⎜ ⎟
⎜c(2)θ2 2c(2)θ (c(2) − 2κ) −2κ2 0 0 · · ·⎟
Λ (0,∞)
=⎜
⎜ 0
⎟.
⎜ c(3)θ 2 2c(3)θ (c(3) − 3κ) −3κ 0 · · ·⎟
⎟
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

2
⎝ 0 0 c(4)θ 2 2c(4)θ (c(4) − 4κ) −4κ2 · · ·⎠
···
An approximate solution to ODEs (3.47) is obtained by truncating the number
∗
of terms to k ∗ and defining a finite dimensional vector of mth moments M (1,k ) ,
m = 1, . . . , k ∗ , as follows:
∗ ∗ ∗ ∗
∂τ M (1,k )
= Λ(1,k ) M (1,k )
+ C (1,k ) ,
∗ ∗
M (1,k ) (0) = (Y0 , Y02 , . . . , Y0k )T ,
∗ ∗
C (1,k )
= (0, c(2)θ2 , 0, 0, . . . , −k ∗ κ2 Y0k +1 T
) ,
⎛ ⎞
−κ −κ2 0 0 0 ···
⎜2c(2)θ (c(2) − 2κ) −2κ2 0 0 ··· ⎟
⎜ ⎟
⎜c(3)θ2 2c(3)θ (c(3) − 3κ) −3κ2 0 ··· ⎟
=⎜ ⎟.
∗
Λ(1,k )
⎜ 0 ⎟
⎜ c(4)θ 2 2c(4)θ (c(4) − 4κ) −4κ2 ··· ⎟
⎝ ··· ⎠
··· 0 0 c(k ∗ )θ 2 2c(k ∗ )θ (c(k ∗ ) − k ∗ κ)
(3.48)
The analytic solution to ODE system (3.48) is given by
∗ ∗ ∗ ∗
M (1,k ) (τ ) = expm{Λ(1,k ) t} · M (1,k ) (0) + (Λ(1,k ) )−1
∗ ∗ ∗
· (expm{Λ(1,k ) t} − I (k ) ) · C (1,k ) , (3.49)
−1
where expm() is the matrix exponent, · and are the matrix product and inverse,
∗
respectively, and I (k ) is k ∗ × k ∗ identity matrix.

Proof. We first present Eq. (3.46) as follows:

(n)
dmt = (−nκYtn − nκ2 Ytn+1 )dt + c(n)(Ytn + 2θY n−1 + θ2 Y n−2 )dt
(∗)
+ ϑ(Yt + θ)Y n−1 dWt
= [c(n)θ2 Y n−2 + 2c(n)θY n−1 + (c(n) − nκ)Ytn − nκ2 Ytn+1 ]dt
(∗)
+ ϑ(Yt + θ)mY n−1 dWt .

2450003-19
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

(n)
Thus, moment mτ solves recursive equation (omitting subscript t)

∂τ m(n) = c(n)θ2 m(n−2) + 2c(n)θm(n−1) + (c(n) − nκ)m(n)

+ (2θc(n) − nκ2 )m(n+1) . (3.50)

Using the fact that m(0) = 1 and c(1) = 0 we can present Eq. (3.50) for n =
0, 1, . . . as a matrix equation for infinite-dimensional column vector in Eq. (3.47).
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Clearly, that under regularity conditions (real parts of eigenvalues of Λ(0,∞) are
negative), we must have limk→∞ ∂τ m(k) = 0. Thus, to make a finite-dimensional
approximation of Eq. (3.47), we fix k ∗ and assume
∗ ∗ ∗
∂τ m(k +1)
= 0 ⇒ m(k +1)
= Y0k +1
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

. (3.51)
∗
Finally, by substituting fixed values for m(0) and m(k +1) , we present (3.47) for
finite-dimensional vector of mth moments, m = 1, . . . , k ∗ , in Eq. (3.48). Solution in
Eq. (3.49) is obtained by integration.

In Fig. 2, we plot first four moments of the volatility process computed using
Eq. (3.49) with truncation order k ∗ = 4 (Fig. 2(a)) and k ∗ = 8 (Fig. 2(b)). For
comparison, we show estimates and 95% confidence intervals obtained by MC sim-
ulations. We see that the low-order truncation with k ∗ = 4 is consistent with MC
estimates for the first and second moments. The high-order truncation with k ∗ = 8
is consistent with the first four moments compared to MC estimates.

Remark 3.1. We note that if κ2 = 0, the system (3.48) becomes lower tridiagonal
so that it can be solved exactly for all moments up to k ∗ by the sequential integration
from n = 1 to n = k ∗ .

3.7. Properties of quadratic variance process

We consider the QV process It defined by dIt = dσt2 dt. The QV is an important
quantity for option pricing. In particular, the expected QV equals model-based
fair value of the continuous-time variance swap, which could be used for model
calibration.

Proposition 3.4 (Bounded moments). Under Assumption 3.1 and κ1 θ > 0, the
QV is well defined under the MMA measure and E[(It )p ] < ∞ for any p, |p| ≥ 1.
If κ2 ≥ β, then the QV is well defined under the inverse measure and Ẽ[(It )p ] < ∞
for any p, |p| ≥ 1.

t
Proof. As E[(It )p ] = E[( 0 σs )p ] ≤ tp E[sups∈[0,t] σsp ] < ∞, the first statement fol-
lows from Proposition 3.1. The second statement follows using the volatility dynam-
ics under Q̃ in Eq. (3.28).

2450003-20
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(a) Volatility of moments with k ∗ = 4

(b) Volatility of moments with k ∗ = 8

Fig. 2. The first four moments of mean-adjusted volatility computed using Eq. (3.49) with the
truncation order k ∗ = 4 (a) and k ∗ = 8 (b) as functions of τ . The model parameters σ0 = 1.5,
θ = 1.0, κ1 = κ2 = 4, ϑ = 1.5. Dots and error bars denote the estimate and 95% confidence interval,
respectively, computed using MC simulations of model dynamics using scheme in Eq. (3.59) with
number of path equal 100 000 and daily time steps.

2450003-21
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

We define the annualized expected QV as follows:

 τ   τ 
! 1 1
Iτ = E σt dt = E
2 2
(Yt + θ) dt , I!0 = 0. (3.52)
τ 0 τ 0

Corollary 3.4 (Expected value). The expected value of the QV using k ∗ th order
truncation in Proposition 3.3 is given by

1 τ (2) 
I!τ ≡ (m0 (τ ) + 2θm0 (τ  ))dτ  + θ2
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

(1)
τ 0
(3.53)
1 ! (2) ! (1) 2 ∗
= (m (τ ) + 2θm (τ )) + θ + O(k ),
τ
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

where m !(1)
and m
(2)
! are the first and second elements in vector M "(1,k ) (τ ) computed
∗

using Eq. (3.54), and O(k ∗ ) is the truncation error. The integrated moments of the
volatility with the k ∗ -order truncation are computed by
 τ
"
M (1,k∗ )
(τ ) ≡
∗ ∗ ∗ ∗
M (1,k ) (τ ) = (Λ(1,k ) )−1 · (expm{Λ(1,k ) t} − I) · M (1,k ) (0)
0
∗ ∗ ∗ ∗
+ (Λ(1,k ) )−1 · ((Λ(1,k ) )−1 · (expm{Λ(1,k ) t} − I) − τ I) · C (1,k ) .
(3.54)

The approximation term O(k ∗ ) includes only the truncation error O(k ∗ ) because
Eq. (3.54) is computed analytically using matrix exponentiation. In Fig. 3, we
present comparison of the analytical solution for the expected QV using Eq. (3.53) to
the estimate and 95% confidence interval based on MC simulations. The truncation-
based analytic solution is consistent with MC estimates.

Fig. 3. The expected QV computed using Eq. (3.53) with k ∗ = 4. The top and bottom three
lines correspond to σ0 = 1.5 and σ0 = 0.5, respectively. The dot and the error bars denote the
estimate and 95% confidence interval computed using MC simulations of model dynamics with
scheme in SDE (3.59) with number of path equal 100 000 and daily time steps. The model cases of
mean-reversion correspond to κ1 = 4 and three choices κ2 = {0, 4, 8} with other model parameters
set to θ = 1.0, ϑ = 1.5.

2450003-22
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

3.8. Monte Carlo discretization

We consider the SDE (3.19) for the volatility process σt and introduce the log-
volatility process Lt = ln σt . Applying Itô’s formula, we obtain the following dynam-
ics for Lt :
(∗)
dLt = ζ(Lt )dt + ϑdWt , L0 = ln σ0 ,
1 (3.55)
ζ(L) = −κ1 + κ2 θ − ϑ2 + κ1 θe−L − κ2 eL .
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

2
T
We consider a time horizon T > 0 and an equidistant time grid tk = kΔ, Δ = n,
0 ≤ k ≤ n. Backward Euler–Maruyama (BEM) scheme for Lt is defined by
(∗) (∗)
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

L̂tk+1 = L̂tk + ζ(L̂tk+1 )(tk+1 − tk ) + ϑ(Wtk+1 − Wtk ). (3.56)

BEM scheme (also known as drift implicit Euler–Maruyama scheme) is based on
Lamperti transform where original SDE (3.19) is transformed into an SDE (3.55)
with constant diffusion coefficient and nonlinearity in the diffusion coefficient shifted
into the drift of the process. We refer to Neuenkirch & Szpruch (2014), Alfonsi
(2013), and Chassagneux et al. (2016) for further details.
We prove that SDE (3.56) has a unique solution in Ltk+1 in order to make
BEM scheme well-defined. Although BEM scheme in Eq. (3.56) cannot be solved
analytically in L̂tk+1 , we still can solve Eq. (3.56) numerically due to the smoothness
and monotonicity of function G(L) := L−ζ(L)Δ, using just few iterations of Newton
algorithm, see Proposition 3.5.

Proposition 3.5. We introduce G(l) := l − ζ(l)Δ. Then, for every c ∈ R, the

equation G(l) = c has a unique solution in l ∈ R.

Proof. As G(l) is continuous, G (l) > 0 and liml→±∞ G(l) = ±∞, the result
follows.

Theorem 3.8. Backward Euler–Maruyama scheme for log-volatility process Lt in

SDE (3.55) has a strong convergence rate of 1.

Proof. The proof of the theorem is based on Proposition 3 of Alfonsi (2013). We

present it in Proposition 3.6.

Proposition 3.6. We fix p ≥ 1 and assume that

 T  p  T p/2
  ϑ2  
E  ζ (L )ζ(L ) + ζ (L ) du + E ζ  (Lu )2 du < ∞. (3.57)
 u u
2
u 
0 0

Then, there exists a constant Kp > 0 such that (E supt∈[0,T ] |L̂t − Lt |p )1/p ≤ Kp Δ.

Instead of Eq. (3.57), we prove slightly stronger estimate in Eq. (3.58). The
latter follows from the combination of Hölder inequality and Proposition 3.1. We

2450003-23
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

omit straightforward details to obtain

 p
  ϑ2  

E sup ζ (Lu )ζ(Lu ) + ζ (Lu ) + E sup |ζ  (Lu )|p < ∞. (3.58)
u∈[0,T ] 2 u∈[0,T ]

Corollary 3.5. MC discretization scheme of the model dynamics (3.12) under the
MMA measure using log-price Xt in SDE (3.16) is given by
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

1 (0) (0)
Xtk+1 = Xtk − σt2k (tk+1 − tk ) + σtk (Wtk+1 − Wtk ),
2
(0) (0) (1) (1)
Ltk+1 = Ltk + ζ(Ltk+1 )(tk+1 − tk ) + β(Wtk+1 − Wtk ) + ε(Wtk+1 − Wtk ),
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

Itk+1 = Itk + σt2k (tk+1 − tk ),

σtk+1 = exp(Ltk+1 ) (3.59)

with Xt0 = ln S0 , Lt0 = ln σ0 , It0 = I0 , and where W (0) and W (1) are independent
Brownian motions.

We note that the first two dynamics are defined in (−∞, +∞), which does not
require extra handling of boundary conditions, in contrast to some affine models
for the variance process.

4. Affine Expansion of Moment Generating Function

We consider the valuation problem of a derivative security with payoff function
u(X, I) settled at maturity time T using log-price process XT defined in Eq. (3.16).
We denote current time by t and introduce time-to-maturity variable τ = T − t. We
denote the undiscounted value function of this derivative security by U (τ, X, I, Y ).
To solve the general valuation problem under both Q and Q̃, we parametrize
the value function using binary parameter p, p ∈ {1, −1}, respectively, as follows:
#
E[u(XT , IT )|Xt = X, It = I, Yt = Y ], p = 1,
U (τ, X, I, Y ; p) = (4.1)
Ẽ[u(XT , IT )|Xt = X, It = I, Yt = Y ], p = −1,

where the expectation E under the MMA measure Q is computed using dynam-
ics (3.34) and the expectation Ẽ under the inverse measure Q̃ is computed using
dynamics (3.36). Here we apply the fact that the dynamics are Markovian under
the both measures.

Theorem 4.1. The value function U (τ, X, I, Y ; p) solves the following PDE on the
domain [0, T ] × R × R+ × (−θ, ∞)

−Uτ + (L(Y ; p) + L(X; p) + L(I; p) )U = 0, U (0, X, I, Y ) = u(X, I), (4.2)

2450003-24
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

where diffusive operators L(Y ; p) , L(X; p) , and L(I; p) are defined as follows:

1 2 (p)
L(Y ; p) U = ϑ (Y + θ)2 UY Y + (λ(p) − κ(p) Y − κ2 Y 2 )UY ,
2
 
2 1
L(X; p)
U = (Y + θ) (UXX − pUX ) + βUXY ,
2

L(I; p) U = (Y + θ)2 UI , (4.3)

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

(p) (p)
κ(p) = κ1 − κ2 θ + 2κ2 θ, λ(p) = (κ2 − κ2 )θ2 ,
#
(p) κ2 , p = 1,
κ2 =
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

κ2 − β, p = −1.

Proof. We note that the classic Feynman–Kac formula (for an example, see Sec.
A.17 in Musiela & Rutkowski (2009)) cannot be applied directly to Eq. (4.1) because
the model dynamics (3.34) and (3.36) do not satisfy the linear growth condition. In
Sec. A.2 we prove that the existence and the uniqueness is guaranteed by Theorem
A.1. This theorem states that if the covariance matrix of diffusive operator L ≡
L(Y ; p) + L(X; p) + L(I; p) is nonnegative definite and the state variables neither
explode nor leave an open domain, then there exists a unique solution to the PDE
representation of the conditional expectation under corresponding model dynamics.
Now using the diffusive operators in the valuation PDE (4.2), we obtain that
the diffusion matrix a in Eq. (A.14) equals
⎛ ⎞
(Yt + θ)2 β(Yt + θ)2 0
⎜ ⎟
a=⎜
⎝β(Yt + θ)
2
ϑ2 (Yt + θ)2 0⎟
⎠ (4.4)
0 0 0

with eigenvalues given by

λ1 = 0, λ2,3 = (1 + ϑ2 ± 1 + 4β 2 − 2ϑ2 + ϑ4 )(Yt + θ)2 .

Since β < ϑ, we observe that 0 = λ1 < λ2 < λ3 , hence matrix a is only nonnegative
definite and its smallest eigenvalue is always zero.
We now set the domain D of state variables {Xt , Yt , It } in PDE (4.2) by D =
R×(−θ, +∞)×R+ . As we showed in Theorem (3.2), under the MMA measure (3.34)
and the inverse measure (3.36), the boundary points {0, +∞} are unattainable for
the volatility process σt , hence solution for mean-adjusted volatility process Yt and
QV It cannot explode or leave D before T . Same applies to Xt as {0, +∞} are
unattainable for σt . Thus, condition 1 of Theorem A.1 is satisfied. As boundedness
condition A.1 is satisfied by construction, the statement of Theorem 4.1 follows
from Theorem A.1 stated in Sec. A.2.

2450003-25
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

4.1. Moment generating function

We introduce the MGF of the three model variables: the log-price Xt , the QV It ,
and the mean-adjusted volatility driver Yt = σt − θ. We extend our analysis using
the transform variable in Yt to generalize our solution method for all the three
state variables in the model. We denote the MGF by G(τ, X, I, Y ; Φ, Ψ, Θ; p) using
respective complex-valued transform variables Φ, Ψ, Θ ∈ C

G(τ, X, I, Y ; Φ, Ψ, Θ; p)
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

#
E[e−ΦXτ −ΨIτ −ΘYτ |X0 = X, I0 = I, Y0 = Y ], p = 1,
= (4.5)
Ẽ[e−ΦXτ −ΨIτ −ΘYτ |X0 = X, I0 = I, Y0 = Y ], p = −1.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

Using model dynamics (3.34) under the MMA measure and model dynamics
(3.36) under the inverse measure, respectively, the MGF solves the following PDE
on the domain [0, T ] × R × R+ × (−θ, ∞):

− Gτ + (L(Y ; p) + L(X; p) + L(I; p) )G = 0,

G(0, X, I, Y ; Φ, Ψ, Θ; p) = e−ΦX−ΨI−ΘY (4.6)

with operators defined in Eq. (4.3). First, we establish a sufficient condition for the
existence of the MGF for the three state variables.

Theorem 4.2. 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
3.7(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

3.7(2), the necessary condition is κ2 ≥ 2β.

Proof. As ZT is Q-martingale, the result immediately follows from Jensen’s

−1
inequality, as the function g(y) := y |ΦR | is convex for |ΦR | ∈ (0, 1). Result for the
inverse measure follows analogously.

Theorem 4.3. Given transform variable Ψ = ΨR + iΨI ∈ C, the MGF of QV Iτ

G(τ, I; Ψ; p = 1) = E[e−ΨIτ ]
√
exists for ΨR < 0 if κ2 > ϑ −2ΨR .

2450003-26
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Proof. We denote Kt = eIt which satisfies SDE dKt = Kt dIt = Kt σt2 dt and
set p = −ΨR . Feynman–Kac formula allows to recast the solution U (τ, σ, K) =
E(KTp |Kt = K, σt = σ) as the solution of the Cauchy problem
∂U
− + L(σ,K) U = 0, U (0, σ, K) = K, (4.7)
∂τ
where operator L(σ,K) is defined by
∂ 1 ∂2 ∂
L(σ,K) = (κ1 + κ2 σ)(θ − σ) + ϑ2 σ 2 2 + Kσ 2 .
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

∂σ 2 ∂σ ∂K
We build a supersolution Ū (τ, σ, K) to (4.7) using the following ansatz:
Ū = K p u(τ, σ), u(τ, σ) = eaσ+bτ , (4.8)
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

which yields
 
1 2 2
−uK p
−b + κ1 θa + (κ2 θ − κ1 )aσ + ϑ a + p − κ2 a σ .
2
2
We show there exists a ≥ 0 such that
1 2 2
lim −b + κ1 θa + (κ2 θ − κ1 )aσ + ϑ a − κ2 a + p σ 2 < +∞. (4.9)
σ↑+∞ 2
We notice that inequality (4.9) holds if there exist a > 0 such that 12 ϑ2 a2 −κ2 a+p <
0. As latter defines a parabola in a, we can always find required positive a, if
√ √
κ2 > ϑ 2p = ϑ −2ΨR .

Theorem 4.4. Given the transform variable Θ = ΘR + iΘI ∈ C, the MGF of the
mean-adjusted volatility process Yτ
#
E[e−ΘYτ |Y0 = Y ], p = 1,
G(τ, Y ; Θ; p) =
Ẽ[e−ΘYτ |Y0 = Y ], p = −1
ϑ2 ϑ2
exists for ΘR < 0, if κ2 > 2 |ΘR |, p = 1 and κ2 − β > 2 |ΘR |, p = −1.

Proof. We set c = −ΦR and consider

#
E[ecYτ |Y0 = Y ], p = 1,
U (τ, Y ; c; p) := (4.10)
Ẽ[ecYτ |Y0 = Y ], p = −1.
Using Feynman–Kac formula we can rewrite U (τ, Y ; c; p) as the solution of the
Cauchy problem
∂U
− + L(Y ; p) U = 0, U (0, Y ; c; p) = ecY , (4.11)
∂τ
where operator L(Y ; p) is defined by Eq. (4.3). We build a supersolution Ū (τ, Y ; c; p)
to (4.11) using ansatz Ū = eaY +bτ , a ≥ c, b ≥ 0. Inserting Ū into Eq. (4.11),
we have
 
∂ Ū (p) 2 1 2
− +L (Y ; p)
Ū = −Ū b − (λ − κ Y − κ2 Y )b − ϑ (Y + θ) b ,
(p) (p) 2 2
∂τ 2
(4.12)

2450003-27
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

where coefficients are defined in (4.3). We show there exists a ≥ c such that
(p) 1
lim b − (λ(p) − κ(p) Y − κ2 Y 2 )a − ϑ2 (Y + θ)2 a2 < ∞.
Y ↑+∞ 2
(p)
We notice that last inequality is satisfied if there exist a ≥ c > 0 such that −κ2 a +
1 2 2 (p) ϑ2
2 ϑ a < 0. Dividing by a > 0 we conclude that it is satisfied if κ2 > 2 c. Rewriting
the last condition in terms of κ2 , β using (4.3), completes the proof.
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

4.2. Affine expansion

Given that coefficients in PDE (4.6) are nonaffine in state variable Y , the PDE
cannot be solved by assuming an exponential affine-solution with a finite number
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

of coefficients. Instead, we can show that the solution to the PDE (4.6) can be rep-
resented by an exponential-affine ansatz E [∞] using an infinite dimensional vector
A(τ ) = {A(k) (τ ; Φ, Ψ, Θ; p)}, k = 0, 1, . . . , ∞, as
# ∞
%
$
E (τ, X, I, Y ; Φ, Ψ, Θ; p) = exp −ΦX − ΨI +
[∞] (k)
A (τ ; Φ, Ψ, Θ; p)Y k
,
k=0
(4.13)

where vector function A(τ ) solve an infinite-dimensional system of quadratic ODEs:

A(k)
τ =A M
(k)
A + (L(k) ) A + H (k) , k = 0, 1, . . . , ∞, (4.14)

where M (k) , L(k) , H (k) are infinite-dimensional sparse matrices and vectors.
We note the similarity between the current ansatz (4.13) for problem (4.6) and
the problem of computing the moments of the volatility process addressed in Propo-
sition 3.3. There we first represent the recursive solution to the moments as an
infinite dimensional vector in Eq. (3.47). We then obtain an approximate solution
by a truncation of infinite series and specifying a boundary condition to reduce the
problem to a finite-dimensional vector (3.48), which can be solved using analytical
methods. Here we follow this insight for solving PDE (4.6) and term our (truncated)
solution as the affine expansion for a given order.

4.2.1. First-order expansion

Theorem 4.5 (First-order affine expansion). The MGF in Eq. (4.5) can be
decomposed as follows:

G(τ, X, I, Y ; Φ, Ψ, Θ; p) = exp{−ΦX − ΨI}[E [1] (τ, Y ; Φ, Ψ, Θ; p)

+ R[1] (τ, Y ; Φ, Ψ, Θ; p)], (4.15)

where E [1] is the leading first-order term and R[1] is the remainder term, such that
E [1] (τ = 0) = 1 and R[1] (τ = 0) = 0. The leading term E [1] is given by the

2450003-28
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

exponential-affine form
# %
$
2
[1] (k) k
E (τ, Y ; Φ, Ψ, Θ; p) = exp A (τ ; Φ, Ψ, Θ; p)Y , (4.16)
k=0

where vector function A(τ ) = {A(k) (τ ; Φ, Ψ, Θ; p)}, k = 0, 1, 2, solve the quadratic

differential system as a function of τ :
A(k)
τ =A M
(k)
A + (L(k) (p)) A + H (k) (p),
⎧⎛ ⎞ ⎛ ⎞⎫
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

⎪
⎪ 0 0 0 ⎛ ⎞ 0 0 0 ⎪
⎪
⎪
⎪ ⎟⎪
⎨⎜ ⎟ 0 ⎜ ⎪
0 0
⎜ 2 2
θ ϑ ⎟ ⎜ ⎟ ⎜ ϑ 2 ⎟ ⎬
M (k)
= ⎜ ⎜ 0 0⎟ ⎜
, ⎝0 θϑ 2 2 2 ⎟
θ ϑ ⎠, ⎜ ⎜ 0 2θϑ 2⎟
,
⎪ ⎟ ⎟
⎪
⎪ ⎝ 2 ⎠ ⎝ 2 ⎠⎪⎪
⎪
⎪
⎩ 0 θ ϑ2 2
0 ⎪
⎭
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

0 0 0 0 2θϑ2 2θ2 ϑ2
⎧⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎫
⎪
⎪ 0 0 0 ⎪
⎪
⎨⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎬
⎜
L (p) = ⎝λ − θ βΦ⎠, ⎝
(k) ⎟ ⎜ −κ − 2θβΦ ⎟ ⎜ (p) ⎟
⎠, ⎝ −βΦ − κ2
(p) 2 (p)
⎪ ⎠ ⎪,
⎪
⎩ ⎪
⎭
2 2
θ ϑ 2(λ + θϑ − θ βΦ)
(p) 2 2
ϑ − 2κ(p) − 4θβΦ
) *
1 2 2 1
H (k) (p) = θ (Φ + pΦ − 2Ψ), θ(Φ2 + pΦ − 2Ψ), (Φ2 + pΦ − 2Ψ)
2 2
(4.17)
with the initial condition A(0) = (0, −Θ, 0) .
The remainder term R[1] solves the following PDE (omitting arguments) :
−Rτ[1] + L(Y ; p) R[1] = −E [1] F [1] (Y, A(1) , A(2) ),
(4.18)
R[1] (0, Y ; Φ, Ψ; p) = 0
with operator L(Y ) defined in Eq. (4.3) and residual F [1] defined by
$
4
[1] (1) (2)
F (Y, A ,A )= C (n) (τ ; A(1) , A(2) )Y n ,
n=3
(4.19)
C (3) (τ ) = 2A(2) (ϑ2 (2θA(2) + A(1) ) − βΦ − κ2 ),

C (4) (τ ) = 2ϑ2 (A(2) )2 .

Proof. We take as an ansatz the leading term E [1] (τ, Y ; Φ, Ψ, Θ; p) in decompo-

sition (4.16) and substitute it into the PDE (4.6). To obtain the ODEs (4.17) for
A(k) , we match terms with Y k , k = 0, 1, 2. To obtain the PDE for the remainder
term R[1] , we account for higher powers of Y k , k = 0, 1, 2. Finally, initial conditions
for ODEs (4.17) and PDE (4.18) are set to reproduce conditions of Eq. (4.6).

Proposition 4.1. We take Ψ = Θ = 0 and Φ = −p/2, p ∈ {−1, 1}. Assuming

that hypotheses 1,2,3 listed in Theorem 4.7 hold, the continuous solution A(τ ) in
Eq. (4.17) exists on [0, +∞).

2450003-29
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(a) R[A(τ ),Φ = −0.50 + 2.00i (b) J[A(τ ),Φ = −0.50 + 2.00i

(c) E [1] [A(τ ),Φ = −0.50 + 2.00i

Fig. 4. The solution of ODEs of the first order expansion in Eq. (4.17) as function of τ for
Φ = −0.5 + 2i, p = 1. We use the model parameters calibrated to Bitcoin options data and
reported in Eq. (6.4). (a) The real part of the solution, (b) the imaginary part of the solution, (c)
the the real and imaginary parts of the exponential term E [1] in Eq. (4.16).

Proof. The proof is identical to that of Theorem 4.7 for the second-order expansion.
We skip it for brevity.

In Fig. 4, we show the real and imaginary parts of ODE solutions A(τ ) and E [1] .
We see that functions A(1) and A(2) quickly reach an equilibrium point, while A(0)
is the nonstationary part which contributes to the leading term E [1] .

2450003-30
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Corollary 4.1 (Estimate of the first-order remainder term). We obtain the

following estimate for the remainder term R[1] in Eq. (4.18):

$
4
|R[1] (τ, Y ; Φ, Ψ, Θ; p)| ≤ C (4) (τ ∗ ) × Mσ(n) (τ ∗ ), (4.20)
n=3

(n)
where Mσ (τ ) is the nth central moment of the volatility defined by
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

Mσ(n) (τ ) ≡ E[(σ(τ ) − θ)n ] = E[Y n (τ )]. (4.21)

Proof. The proof is analogous to the second-order expansion in Proposition 4.3.

Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

Corollary 4.2. First-order affine expansion for the MGF (4.5) is obtained by using
the leading term E [1] in Eq. (4.15):

G(τ, X, I, Y ; Φ, Ψ, Θ; p) = exp{−ΦX − ΨI}E [1] (τ, Y ; Φ, Ψ, Θ; p), (4.22)

where the approximation error arises from the truncation of the infinite dimensional
affine anzats in Eq. (4.13) with the error magnitude estimated using Eq. (4.20).

In Fig. 6, we illustrate that the first-order approximation (4.22) for the MGF
produces valid and accurate PDFs for all three state variables.

4.2.2. Second-order expansion

Theorem 4.6 (Second-order affine expansion). The MGF in Eq. (4.5) can be
decomposed as follows:

G(τ, X, I, Y ; Φ, Ψ, Θ; p) = exp{−ΦX − ΨI}[E [2] (τ, Y ; Φ, Ψ, Θ; p)

+ R[2] (τ, Y ; Φ, Ψ, Θ; p)], (4.23)

where E [2] is the leading term such that E [2] (τ = 0) = 1 and R[2] (τ = 0) = 0 is the
remainder term.
The leading term E [2] (τ, Y ; Φ, Ψ; p) is given by the exponential-affine form
# 4 %
$
E [2] (τ, Y ; Φ, Ψ, Θ; p) = exp A(k) (τ ; Φ, Ψ, Θ; p)Y k , (4.24)
k=0

where vector function A(τ ) = {A(k) (τ ; Φ, Ψ, Θ; p)}, k = 0, . . . , 4, solves the system

the quadratic differential system as a function of τ :

A(k)
τ = A M (k) A + (L(k) (p)) A + H (k) (p),

2450003-31
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

⎧⎛ ⎞
⎪
⎪ 0 0 0 0 0 ⎛ ⎞
⎪
⎪ ⎜ ⎟ 0 0 0 0 0
⎪
⎪ ⎜ 2
θ ϑ 2
⎟ ⎜
⎪
⎪ ⎜0 0 0 0⎟ ⎜0 θϑ2 θ2 ϑ2 0 0⎟
⎪
⎨⎜ ⎟ ⎜ ⎟
⎜
2
⎟ ⎜ ⎟
M (k) = ⎜ ⎟, ⎜0 θ2 ϑ2 0 0 0⎟ ⎟,
⎪⎜0
⎪ 0 0 0 0⎟ ⎜ ⎟
⎪
⎪ ⎜ ⎟ ⎝0 0 0⎠
⎪
⎪ ⎜0 ⎟ 0 0
⎪
⎪ ⎝ 0 0 0 0⎠
⎪
⎩ 0 0 0 0 0
0 0 0 0 0
⎛ ⎞
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

0 0 0 0 0
⎜ ⎛ ⎞
⎜ ϑ2
3 2 2 ⎟ ⎟ 0 0 0 0 0
⎜0 2θϑ2 θ ϑ 0⎟ ⎜
⎜
⎜
2 2 ⎟ ⎜0 0
⎟ ⎜ ϑ2 3θϑ2 2θ2 ϑ2 ⎟ ⎟
⎜ ⎟ ⎜ ⎟
⎟,
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

⎜0 2θϑ2 2θ2 ϑ2 0 0⎟ , ⎜ 0 ϑ 2
4θϑ 2
3θ 2 2
ϑ 0 ⎟
⎜ ⎟ ⎜ ⎟
⎜ 3 2 2 ⎟ ⎝0 3θϑ2 3θ2 ϑ2 ⎠
⎜0 θ ϑ 0 0 0⎟
0 0
⎜ 2 ⎟
⎝ ⎠ 2
0 2θ ϑ 2
0 0 0
0 0 0 0 0
⎛ ⎞⎫
0 0 0 0 0 ⎪
⎪
⎜ ⎟⎪⎪
⎪
⎜0 3 2 2 ⎟⎪
4θϑ ⎟⎪
⎜ 0 0 ϑ ⎪
⎜ 2 ⎟⎪⎪
⎜ ⎟⎪⎬
⎜ 2 2⎟
⎜0 0 2ϑ 2
6θϑ 2
4θ ϑ ⎟ ,
⎜ ⎟⎪
⎜ ⎟⎪⎪
⎟⎪
3 2 9 2 2
⎜0 ϑ 6θϑ 2
θ ϑ 0 ⎪
⎪
⎜ ⎟ ⎪
⎝ 2 2 ⎠⎪⎪
⎪
⎪
⎭
0 4θϑ2 4θ2 ϑ2 0 0
⎧⎛ ⎞ ⎛ ⎞ ⎛ ⎞
⎪
⎪ 0 0 0
⎪
⎪
⎪
⎪⎜λ(p) − θ2 βΦ⎟ ⎜ −κ(p) − 2θβΦ ⎟ ⎜ ⎟
⎟ ⎜ ⎟
(p)
⎨⎜
⎪ ⎟ ⎜ ⎜ −βΦ − κ2 ⎟
⎜ ⎟ ⎜ ⎟
L (p) = ⎜
(k) 2
θ ϑ 2 ⎟, ⎜2(λ + θϑ − θ βΦ)⎟, ⎜ ϑ2 − 2κ(p) − 4θβΦ ⎟
(p) 2 2 ⎜
⎟,
⎪⎜ ⎟ ⎜ ⎟
⎪
⎪⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎪
⎪⎝ 0 ⎠ ⎝ 2
3θ ϑ 2 ⎠ ⎝3(2qϑ2 + λ(p) − θ2 βΦ)⎟
⎜
⎠
⎪
⎪
⎩ 0 0 2 2
6θ ϑ
⎛ ⎞ ⎛ ⎞⎫
0 0 ⎪
⎪
⎜ ⎟ ⎜ ⎟⎪⎪
⎪
⎜ ⎟ ⎜ ⎟ ⎪
⎟⎪
0 0
⎜ ⎟ ⎜ ⎬
⎜ (p) ⎟ ⎜ ⎟
⎜ −2(βΦ + κ2 ) ⎟, ⎜ 0 ⎟ ,
⎜ ⎟ ⎜ ⎟⎪
⎜ 3(ϑ2 − κ(p) − 2θβΦ) ⎟ ⎜ −3(βΦ + κ2 )
(p) ⎟⎪⎪
⎪
⎝ ⎠ ⎝ ⎠⎪⎪
⎪
⎭
4(3θϑ + λ − θ βΦ)
2 (p) 2
2(3ϑ − 2κ − 4θβΦ)
2 (p)

) *
1 2 2 1
H (k) (p) = θ (Φ + pΦ − 2Ψ), θ(Φ2 + pΦ − 2Ψ), (Φ2 + pΦ − 2Ψ), 0, 0
2 2
(4.25)

with initial condition A(0) = (0, −Θ, 0, 0, 0) .

2450003-32
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

The remainder term R[2] solves the following PDE (omitting arguments)

−Rτ[2] + L(Y ; p) R[2] = −E [2] F [2] ,

(4.26)
R[2] (0, Y ; Φ, Ψ, Θ; p) = 0

with operator L(Y ) defined in Eq. (4.3) and function F [2] defined by

$
8
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

F [2] (τ, Y ; A(1) , A(2) , A(3) , A(4) ) = C (k) (τ ; A(1) , A(2) , A(3) , A(4) )Y k ,
k=5

C (5) 2
(τ ) = 3ϑ A (3) (3)
(3θA + 2A (2)
) + 4A(4) (ϑ2 (3θ2 A(3) + 4θA(2) + A(1) ) − κ2 ),
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

1 2
C (6) (τ ) = ϑ (16θ2 (A(4) )2 + 16A(4) (3θA(3) + A(2) ) + 9(A(3) )2 ),
2
C (7) (τ ) = 4ϑ2 A(4) (4θA(4) + 3A(3) ), C (8) (τ ) = 8ϑ2 (A(4) )2 . (4.27)

Proof. The proof follows by analogy to the first-order expansion in Theorem 4.5.

Remark 4.1. We emphasize that sparse matrices M (k) in the quadratic term do
not depend on the measure parameter p, while the linear L(k) (p) and free terms
H (k) (p) depend on p in ODEs (4.17), (4.25) for the first- and the second-order
expansion.

We now present the result on the existence of continuous solution to Eq. (4.25)
in Proposition 4.2 and discuss it further in Sec. 4.2.3.

Proposition 4.2. We take Ψ = Θ = 0 and Φ = −p/2, p ∈ {−1, 1}. Assuming

that hypotheses 1,2,3 listed in Theorem 4.7 hold, the continuous solution A(τ ) in
Eq. (4.25) exists on [0, +∞).

Proof. See Theorem 4.7.

In Fig. 5, we show the real and imaginary parts of A(τ ) for the second-order
expansion and the leading term E [2] . We see that, similarly to the first order expan-
sion illustrated in Fig. 4, functions A(1) , A(2) , A(3) , A(4) reach an equilibrium point,
while A(0) is the nonstationary part which contributes to the leading term E [2] .
By comparing the low order terms A(0) , A(1) , A(2) between the first and the sec-
ond expansions, we notice no visible difference, which suggest that the expansion is
rather recursive and higher order terms make insignificant contribution to the low
order terms. We also notice that terms A(3) and A(4) are very small in magnitude
compared to the first three terms, which suggest that higher order term only provide
a small marginal contribution.

2450003-33
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(a) R[A(τ ),Φ = −0.50 + 2.00i (b) J[A(τ ),Φ = −0.50 + 2.00i

(c) E [2] [A(τ ),Φ = −0.50 + 2.00i

Fig. 5. The solution of ODEs in Eq. (4.25) as function of τ for Φ = −0.5 + 2i, p = 1. We use the
model parameters calibrated to Bitcoin options data and reported in Eq. (6.4). (a) The real part
of the solution, (b) the imaginary part of the solution, (c) the real and imaginary parts of the
exponential term E [2] in Eq. (4.24).

Corollary 4.3 (Estimate of the second-order remainder term). We obtain

the following estimate for the remainder term R[2] which solves Eq. (4.26):

$
8
|R (τ, X; Φ, Ψ, Θ; p)| ≤
[2]
Cn (τ ∗ ) × Mσ(n) (τ ∗ ), (4.28)
n=5

(n)
where Mσ is the nth central moment of the volatility defined by (4.21).

2450003-34
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Proof. The formal solution for the remainder term R[2] solving problem (4.26) is
obtained by applying the Feynman–Kac formula and is given by

R[2] (τ, Y ; Φ, Ψ)
 τ ∞
= [E [2] (τ − t, Y  ; Φ, Ψ)F [2] (τ − t, Y  )]P (t, Y ; Y  )dtdY  , (4.29)
0 −θ

where P (t, Y ; Y ) is PDF of Y  = Y (t) conditional on Y . P solves PDE (4.2),


by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

subject to initial condition P (0, Y ; Y  ) = δ(Y  − Y ). Using Eq. (4.29), we obtain

 τ ∞
|R[2] (τ, Y ; Φ, Ψ)| ≤ |E [2] (τ − t, Y  ; Φ, Ψ)F [2] (τ − t, Y  )P (t, Y ; Y  )|dtdY 
0 −θ
 
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

τ ∞
≤ |F [2] (τ − t, Y  )P (t, Y ; Y  )|dtdY  . (4.30)
0 −θ

Here, we apply the bound on the MGF |E [2] (τ − t, Y  ; Φ, Ψ)| ≤ 1 and integrate
out the log-return X and the QV I. Next we substitute the polynomial function
F [2] in Eq. (4.27). We use the continuity of functions A(k) , k = 1, 2, 3, 4, and apply
the first mean value theorem for definite integrals to estimate the time integral.
Finally, we approximate the expected moments of the mean-adjusted volatility by
(n)
its steady-state nth order central moments Mσ specified in Eq. (4.21).

Corollary 4.4. Second-order affine expansion for the MGF (4.5) is obtained by
using the leading term E [2] in Eq. (4.24) as

G(τ, X, I, Y ; Φ, Ψ, Θ; p) = exp{−ΦX − ΨI}E [2] (τ, Y ; Φ, Ψ, Θ; p), (4.31)

where the approximation error arises from the truncation of the infinite dimensional
affine anzats in (4.13) with the error magnitude estimated by Eq. (4.28).

In Fig. 6, we illustrate that the second-order approximation (4.31) for the MGF
produces valid and accurate PDFs for all three state variables.

4.2.3. Solution of quadratic differential systems

It is established that the global existence problem for the solution of quadratic dif-
ferential systems is nontrivial, see Coppel (1966), Dickson & Perko (1970), Jacobson
(1977), Baris et al. (2008) among others. We can show that the matrices of quadratic
terms in Eqs. (4.17) and (4.25) are indefinite because their eigenvalues are of both
signs. Thus, we can provide a result for global existence that is only conditional by
nature. We focus on important case of option valuation on underlying assets with
Φ = −p/2, p ∈ {−1, 1}, and Ψ = Θ = 0.
Dickson & Perko (1970) show that complex-valued solution of quadratic ODEs
such as A(τ ; Φ) in Eq. (4.17) or Eq. (4.25) has maximal interval of existence
[0, τ+ (Φ; A)), where τ+ (Φ; A) denotes its blow-up time.

2450003-35
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(a) log-returnXτ (b) Quadratic variance lττ

(c) Volatilityστ

Fig. 6. PDFs computed using the inversion of the first order expansion term E [1] in Eq. (4.16)
and the second-order expansion term E [2] in Eq. (4.24) for the state variables under the MMA
measure: (a)–(c) PDFs of log-return, Xτ , of QV normalized by τ , Iτ /τ , of volatility στ , στ = Yτ +θ,
respectively. We use the model parameters for Bitcoin options reported in Eq. (6.4) with time to
maturity of one month, τ = 1.0/12.0. The blue histogram is computed using realizations from MC
simulations of joint model dynamics (3.12) using scheme in Eq. (3.59) with the number of path
equal 400,000 and daily time steps.

Theorem 4.7. Assume following conditions are satisfied:

(1) τ+ (Φ; A) ≥ τ+ ( Φ; A), Φ ∈ C, i.e. real part of A(τ ; Φ) cannot blow up before
A(τ ; Φ).
+2m
k=0 Ak (τ ; Φ; p)Y → +∞ as τ ↑ τ+ (Φ), Φ ∈ R, i.e. leading term of the affine
k
(2)
expansion does not vanish as we approach blow-up time when transform variable
Φ is real.

2450003-36
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

(3) limτ ↑τ+(Φ) R[n] (τ ; Φ) > −∞, Φ ∈ R, i.e. remainder term is uniformly bounded
from below.
If MGF in Eq. (4.6) is finite with G(τ0 ; X, I, Y ; Ψ = Φ, Ψ = 0, Θ = 0) < ∞,
then continuous solution A(τ ; Φ, Ψ = 0, Θ = 0) of Eq. (4.17) and of Eq. (4.25) exists
on [0, τ0 ).

Proof. See Sec. A.4.

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

By Theorem 4.7, we obtain that continuous solution A(τ ) for the first- and
second-order affine expansions in Eqs. (4.16) and (4.24), respectively, exists on
[0, +∞) if 1, 2, 3 hold.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

4.3. Properties of affine expansions

We now consider the key properties of the affine expansion of the MGF. For brevity,
we focus on the properties of the second-order expansion, which is our key develop-
ment. For the first order expansion, martingale conditions stated in Proposition 4.3
hold, and the consistency with the expected values (but not with their variances)
of three state variables is ascertained in the same way as for the second-order
expansion.
Proposition 4.3 (Martingale conditions). Assuming that hypotheses 1, 2, 3
in Theorem 4.7 hold, the second-order expansion of the MGF in (4.31) with E [2]
defined in Eq. (4.24) satisfies the martingale conditions for log-return Xτ :
G[2] (τ, Y ; Φ = 0, Ψ = 0, Θ = 0; p) ≡ E [2] (τ, Y ; Φ = 0, Ψ = 0, Θ = 0; p) = 1,
(4.32)
E[eXτ ] ≡ eX0 E [2] (τ, Y ; Φ = −1, Ψ = 0, Θ = 0; p = 1) = eX0 , (4.33)
Ẽ[e−Xτ ] ≡ e−X0 E [2] (τ, Y ; Φ = 1, Ψ = 0, Θ = 0; p = −1) = e−X0 .

Proof. For Φ = Ψ = Θ = 0, we obtain H (p) ≡ 0 in (4.25). Given zero initial

conditions, A(k) (τ ; Φ = 0, Ψ = 0, Θ = 0; p) ≡ 0 due to the uniqueness of the
solution. Similarly, for Φ = −1 and Θ = Ψ = 0, we obtain that H (1) ≡ 0, hence
A(k) (τ ; Φ = −1, Ψ = 0, Θ = 0; p = 1) ≡ 0. Similarly, for Φ = 1 and Ψ = 0 we obtain
that H (−1) ≡ 0, hence A(k) (τ ; Φ = 1, Ψ = 0, Θ = 0; p = −1) ≡ 0.

Proposition 4.4 (Volatility moments). The MGF in Eq. (4.31) with the second-
order affine term E [2] in (4.24) reproduces exactly the expected value and the vari-
ance of the volatility for κ2 = 0.

Proposition 4.5 (Log-price moments). The MGF in Eq. (4.31) with the second-
order affine term reproduces exactly the expected value and the variance of the log-
price for κ2 = 0.

2450003-37
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

Proposition 4.6 (QV moments). The MGF in Eq. (4.31) with the second-order
affine term reproduces exactly the expected value and the variance of QV for κ2 = 0.

Proof. Propositions 4.4 and 4.5 are proved in Secs. A.3.1 and A.3.2, respectively.
Proof of Proposition 4.6 is similar to the proof of Proposition 4.5.

Remark 4.2. In Remark 3.1 we noted that the system of volatility moments (3.48)
is solved exactly for κ2 = 0, otherwise the solution is approximate. Thus, we cannot
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

generalize above statements for κ2 > 0. However, it is our hypothesis that moments
obtained using (4.24) correspond to moments computed in (3.48) using truncation
order k ∗ = 4 and that truncation error as function of κ2 is very small.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

In Fig. 6, we show PDFs computed using the inversion of first-order solution

E [1] in Eq. (4.16) and second-order solution E [2] in Eq. (4.24) for the three state
variables in model dynamics (3.34). The blue histogram is computed using MC
simulations of the joint model dynamics. We see that both first- and second-order
expansions produce valid PDFs and match consistently histograms produced by the
MC simulations. The first-order expansion may not be very accurate for the QV and
the volatility variables because it is not consistent with variances of these variables.
The second-order expansion is consistent with variances and co-variances of all state
variables and it accurately matches PDFs computed using MC simulations.

5. Option Valuation
We apply the zero-drift log-price process Xt in Eq. (3.16) and Eqs. (3.17) and (3.18)
for spot and futures prices, respectively, to consider option valuation on both spot
and futures underlyings. We denote by PT the price of either spot or futures asset
PT = eμ̄(T ) eXT , X0 = ln S0 , (5.1)
where μ̄(T ) = r̄(0, T ) for spot price and μ̄(T ) = r̄(0, T ) − q̄(0, T ) for futures.
As we conclude in Corollaries 4.2 and 4.4, we apply affine expansion given by
either the first order in Eq. (4.16) or the second order in Eq. (4.24) as analytic
solutions to the MGF under the dynamics (3.16). Given that the MFG is available
analytically, we then apply Lewis–Lipton approach for valuation of vanilla options
using Fourier transform, see Lewis (2000), Lipton (2001, 2002).

5.1. Vanilla calls and puts

Using Eq. (5.1), we represent put and call payoff functions for strike price K using
capped payoff by
ucall (PT , K) = max{PT − K, 0} = PT − min{PT , K} = PT − min{eμ(T )+XT , K},
uput (PT , K) = max{K − PT , 0} = K − min{PT , K} = K − min{eμ(T )+XT , K}.
(5.2)

2450003-38
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Thus, we evaluate capped payoff u(PT ) = min{PT , K} under MMA measure Q by

U (τ, X) = e−r(T ) E[min{eμ(T )+XT , K}|Ft ]. (5.3)

Proposition 5.1. The valuation formula for capped payoff is given by

 ∞
e−r(T ) K ∗
U (τ, X) = e−(iy−1/2)X
π 0

1
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

× E [m]
(τ, Y ; Φ = iy − 1/2, Ψ = 0, Θ = 0; p = 1)dy , (5.4)
y 2 + 1/4
where X ∗ = ln(S0 /K)+μ(T ) is log-moneyness, E [1] and E [2] are given in Eqs. (4.16)
and (4.24), respectively.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

Proof. Using affine expansion for MGF G with either first-order in Eq. (4.22) or
second-order in Eq. (4.31), the PDF of Xτ is computed by Fourier inversion
PDF(τ, X, Y ; X  ; p = 1)
 ∞ 
1
= exp{Φ(X  − X)}E [m](τ, Y ; Φ, Ψ = 0; p = 1)dΦ . (5.5)
π 0

For valuation problem in Eq. (5.3), we obtain

 ∞

−r(T )
U (τ, X) = e min{eμ(T )+X , K}
−∞
  ∞  
1 Φ(X  −X) 
× e E [m]
(τ, Y ; Φ, Ψ = 0; p = 1)dΦ dX .
π 0

Assuming that inner integrals are finite, we exchange the integration order as
 ∞ 
1 −r(T )
U (τ, X) = e !(Φ)E (τ, Y ; Φ, Ψ = 0; p = 1)dΦ ,
u [m]
(5.6)
π 0

!(Φ) is the transformed payoff function

where u
 ∞
 
−ΦX
!(Φ) = e
u eΦX min{eX +μ(T ) , K}dX 
−∞
∗
e(Φ+1)k ∗ 1 ∗
= e−ΦX eμ(T ) − ek +μ(T ) eΦk (5.7)
Φ+1 Φ
1 ∗ 1
= −Ke−Φ(ln(S/K)+μ(T )) = −Ke−ΦX ,
(Φ + 1)Φ (Φ + 1)Φ
where X ∗ = ln(S/K) + μ(T ) is log-moneyness, k ∗ = ln K − μ(T ) with the first
integral being finite for [Φ] > −1 and the second integral being finite for [Φ] < 0.
Integral in Eq. (5.7) is finite for −1 < [Φ] < 0. For Φ = iy − 1/2, we get Eq. (5.4)
∗ 1
!(Φ = iy − 1/2) = Ke−(iy−1/2)X
u . (5.8)
y2 + 1/4

2450003-39
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

As a result, calls and puts on spot underlying are valued using Eqs. (5.2)

U call (τ, S, K) = e−r(T )+μ(T ) S − U (τ, X),

(5.9)
U put (τ, S, K) = e−r(T ) K − U (τ, X).

The value of call option on future underlying, in turn, equals

U call (τ, F, K) = e−r(T ) f0 (T ) − U (τ, X), (5.10)

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

where f0 (T ) is the futures and X ∗ = ln(f0 (T )/K) denotes log-moneyness. Hence,

given a set of futures prices quoted in the market, we can evaluate options on this
set of futures without need to compute the convenience yield explicitly.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

We emphasize that in valuation equations (5.9) we implicitly assume that the

price process Zt = exp(Xt ), t ∈ (0, T ], is Q-martingale under the MMA measure,
otherwise formulas (5.9) are not valid. For that we need a restriction κ2 ≥ β, see
Theorem 3.7.

5.2. Inverse calls and puts

Using generic price process of underlying spot and futures given in Eq. (5.1), we
represent the payoffs of inverse calls and puts by
1
,call (PT , K) =
u max{PT − K, 0}
PT
= 1 − e−XT −μ(T ) min{eμ(T )+XT , K},
1 (5.11)
,put (PT , K) =
u max{K − PT , 0}
PT
K
= − e−XT −μ(T ) min{eμ(T )+XT , K}.
PT

As a result, we need to evaluate option on the inverse capped payoff under Q̃

Ũ (τ, X) = Ẽ[e−XT −μ(T ) min{eμ(T )+XT , K}|Ft ]. (5.12)

Proposition 5.2. The valuation formula (5.6) for the inverse capped payoff
becomes
 ∞
, (τ, X) = 1
U
∗
e−(iy+1/2)X 2
1
E [m]
π 0 y + 1/4
× (τ, Y ; Φ = iy + 1/2, Ψ = 0, Θ = 0, p = −1)dy, (5.13)

where X ∗ = ln(S0 /K)+μ(T ) is log-moneyness, E [1] and E [2] are given in Eqs. (4.16)
and (4.24), respectively.

2450003-40
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

!(Φ) of the inverse capped option

Proof. As in Eq. (5.7), we compute the transform u
 ∞
  
!(Φ) = e−ΦX
u eΦX e−X −μ(T ) min{eX +μ(T ) , K}dX 
−∞
  
k∗ ∞
  
= e−ΦX eΦX dX  + K eΦX −X −μ(T )
dX 
−∞ k∗

∗ 1
= −e−ΦX
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

, (5.14)
Φ(Φ − 1)

where X ∗ = ln(S/K) + μ(T ) is the log-moneyness and k ∗ = ln K − μ(T ), with

the first integral being finite for [Φ] > 0 and the second integral being finite for
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

[Φ] < 1. Integral in Eq. (5.14) is finite for 0 < [Φ] < 1. We set Φ = iy + 1/2,
then
∗ 1
!(Φ = iy + 1/2) = e−(iy+1/2)x
u . (5.15)
y2 + 1/4

As a result, calls and puts are valued using capped payoff (5.12):

, call (τ, S, K) = 1 − U(τ,

U , X), , put (τ, S, K) = Ke−X−μ(T ) − U
U , (τ, X). (5.16)

Hereby, we need to compute

  
1 
f (τ, X) = Ẽ Ft = Ẽ[e−XT −μ(T ) |Ft ] = e−μ(T ) e−X (5.17)
PT 

with the last equality following from Eq. (4.33). Given the term structure of futures
prices f0 (T ) in Eq. (3.18), the put option on the future is

K , (τ, X)
p(τ, F, K) = −U (5.18)
f0 (T )

with log-moneyness X ∗ = ln(f0 (T )/K). We emphasize that Eq. (5.16) holds only if
Q ∼ Q̃, which requires restriction κ2 ≥ β, see Theorem 3.6. In addition, we assume
that the inverse process Rt = exp(−Xt ), t ∈ (0, T ], is a true Q̃-martingale. For this
we need κ2 ≥ 2β by Theorem 3.7.

5.3. Options on quadratic variance

We consider a call option on QV with the payoff function
1
u(I) = max{I(T ) − T K, 0}, (5.19)
T
where K is the strike in terms of the annualized variance.

2450003-41
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

Proposition 5.3. The value of the call option on the QV under Q is given by
U (τ, X, Y, I) = e−r̄(T ) E[u(IT )|Ft ]

e−r̄(T ) 1 +∞
= [û(Ψ)E [m] (τ, Y ; Φ = 0, Ψ; Θ = 0; p = 1)]dΨ,
T π 0
(5.20)
where
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

e−ΨI ΨT K
û(Ψ) = e (5.21)
Ψ2
provided [Ψ] < 0, with E [1] and E [2] given in Eqs. (4.16) and (4.24), respectively.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

Proof. Consider an option on QV IT with terminal payoff u(IT ). Using Sepp

(2008), we find that option value is given by Eq. (5.20), where û(Ψ) is the trans-
formed payoff function
 +∞

û(Ψ) = u(I  )eΨ(I −I) dI  . (5.22)
0

For a call option on QV with strike K, provided [Ψ] < 0, Eq. (5.22) becomes
 +∞

û(Ψ) = e−ΨI u(I  )eΨI dI 
0
 +∞
 e−ΨI ΨT K
= e−ΨI max{I  − T K, 0}eΨI dI  = e .
0 Ψ2

Sepp (2008) derives transforms û(Ψ) for typical payoff on the QV including puts
and calls on the square root of the QV. Assuming that the inverse measure Q̃ is a
martingale measure for generic spot and futures underlying Pt defined in Eq. (5.1),
the inverse option on the QV is given under Q̃ by
  
u(IT ) 
Ũ (τ, X, Y, I) = Ẽ Ft . (5.23)
PT 
Similarly to Eq. (2.5), values Ũ(τ, X, Y, I) and U (τ, X, Y, I) in Eqs. (5.23) and (5.20),
respectively, are related by
U (τ, X, Y, I) = Pt Ũ (τ, X, Y, I).

Proposition 5.4. The value of the call option on the QV under the inverse measure
Q̃ is given by

e−r̄(T ) 1 +∞
Ũ (τ, X, Y, I) = [û(Ψ)eX−μ(T ) E [m]
T π 0
× (τ, Y ; Φ = 1, Ψ, Θ = 0; p = 1)]dΨ, (5.24)
[m]
provided [Ψ] < 0, with E and û(Ψ) defined as in Eq. (5.21).

2450003-42
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Proof. Using affine expansion for MGF G with either first-order in Eq. (4.22) or
second-order in Eq. (4.31), we compute the joint density of {Xτ , Iτ } using inverse
Fourier transform by

PDFX,I (τ, X, Y, I; X  , I  )
 +∞  +∞ 
1 Φ(X  −X)+Ψ(I  −I) [m]
= 2 e E (τ, Y ; Φ, Ψ)dΦ dΨ .
4π −∞ −∞
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

As a result, we obtain
 +∞  +∞
e−r̄(T ) 
Ũ (τ, X, Y, I) = e−X −μ̄(T ) u(I  )
T −∞ 0
 +∞  +∞ 
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

1 Φ(X  −X)+Ψ(I  −I) [m]

× 2 e E (τ, Y ; Φ, Ψ)dΦ dΨ dX  dI  .
4π −∞ −∞
Exchanging the integration order, we have
 +∞  +∞
e−r̄(T ) 1
Ũ (τ, X, Y, I) = [E [m] (τ, Y ; Φ, Ψ)û(Φ, Ψ)]dΦ dΨ,
T 4π 2 −∞ −∞
(5.25)
where û(Φ, Ψ) is the transformed payoff function
 +∞  +∞
  
û(Φ, Ψ) = e−ΦX−ΨI e−X −μ̄(T ) u(I  )eΦX +ΨI dX  dI 
−∞ 0
 +∞  +∞
−ΦX−ΨI −μ̄(T ) (Φ−1)X   
=e e e dX u(I  )eΨI dI 
−∞ 0
 +∞

= 2πe−ΦX−μ̄(T ) δ0 (Φ − 1)û(Ψ), û(Ψ) = e−ΨI u(I  )eΨI dI  ,
0
(5.26)

provided [Ψ] < 0. We integrated out the complex exponential in Eq. (5.26) using
results of Sepp (2007). Thus, Eq. (5.25) becomes
 +∞
e−r̄(T ) 1
Ũ (τ, X, Y, I) = [e−X−μ̄(T ) E [m] (τ, Y ; Φ = 1, Ψ)û(Ψ)]dΨ.
T 2π −∞
(5.27)

6. Model Implementation and Calibration to Options Data

6.1. Implementation
In practical applications, either for model calibration or for options valuation, we
need to compute call and put option prices on a grid consisting of several maturities
and strikes, also referred to as an option chain. We consider grid with M maturities

2450003-43
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

and J strikes. For option valuation using the MGF, we apply the affine expansion
given by either the first order in Eq. (4.16) or the second order in Eq. (4.24).
First, we fix the space grid with size P (typically, P ≈ 500) of transform variable
Φ for the log-price. We solve the system of ODEs numerically for either the first-
order expansion in Eq. (4.16) or for the second-order expansion in Eq. (4.24) in
time up to the last maturity time. The computational cost of this step is
C (1) = O(P × Nmax ), (6.1)
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

where Nmax is the number of time steps for ODE solver (typically we have one-two
steps per one week) and O is the number of arithmetic (elementary) operations.
Second, we compute values of the MGF on the grid of Φ for each maturity slice.
We then compute values of capped payoffs for the MMA measure in Eq. (5.4) or
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

for the inverse measure in Eq. (5.13) using Simpson rule for numerical integration.
We then value all options in a given chain. The computational cost of this step is
C (2) = O(J × M × P ). (6.2)
The main difference with numerical implementation of affine models is that the
first step can be computed with cost close to O(P × M ) because the solution to the
MGF can be computed analytically without solving ODEs numerically. The cost
ratio between the implementation of the log-normal SV model and an affine SV
model compares favorably when the number of maturities becomes large with M
close to Nmax , and when the number of strikes J becomes large as well.

6.2. Model calibration

To illustrate that the log-normal SV model can handle different market regimes,
we use time series of option prices on Bitcoin observed on Deribit exchange from
April 2019 to October 2023. We focus on short-dated options with expiries including
one and two weeks and one month, because these are the most liquid and actively
traded expiries. We use time series of options chains traded on Deribit. We fix
calibration every week with option prices samples each Friday at 10:00 UTC. We
include puts and calls with absolute values of option deltas between [−0.10, −0.50]
and [0.10, 0.50], respectively.
For the stability of calibration, we estimate mean-reversion parameters κ1 and
κ2 beforehand. The reason is that mean-reversion parameters and volatility-of-
volatility with volatility beta are co-dependent so that it is possible to produce
similar smiles and skews using different values of these parameters. We follow the
procedure in Sepp & Rakhmonov (2023) for estimation of parameters κ1 and κ2
by fitting the model implied auto-correlation of the log-normal volatility process to
the empirical auto-correlation. This is efficient because the model auto-correlation
function is determined solely by mean-reversion parameters. The estimated values
of κ1 and κ2 are the following κ̂1 = 2.21 and κ̂2 = 2.18, which are kept fixed through
the entire period (in practice we would adjust these estimate regularly using most
recent data).

2450003-44
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

Thus, we need to estimate four model parameters σ0 , θ, β, ε. Parameters σ0 and

θ control the level and the slope of at-the-money (ATM) implied volatilities, and
β and ε control the skewness and the convexity of implied volatilities, respectively.
We apply a stronger constraint κ2 ≥ 2β that guarantees martingale property for the
inverse process Rt under Q̃ using Theorem 3.7. To compute model prices we apply
the second-order affine expansion in Eq. (5.9). For each set of sampled option prices,
we minimize the weighted mean squared error (WMSE) between model implied
volatilities, σnmodel (T, K), and market mid-point implied volatilities, σnimplied(T, K),
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

where K and T are option’s strike and maturity, respectively, as followsd :

$
WMSE = wn (T, K)(σnmodel(T, K) − σnimplied(T, K))2 (6.3)
n
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

with weight wn (T, K) = Vn (T, K) set to option Black–Scholes vega Vn (T, K).
In Fig. 7(a), we show the time series with the quality of the model fit which is
computed using the square root of average squared differences between model and
market implied volatilities. As a benchmark, we show the average spread between

(a) Average mean-squared error and bid-ask (b) Initial and mean volatilities
volatility spread

(c) Volatility beta (d) Volatility-of-volatility

Fig. 7. Time series of the quality of the model calibration and of calibrated model parameters
using weekly model calibrations to prices of option on Bitcoin traded on Deribit exchange from
April 2019 to October 2023. (a) Time series of the mean squared error (MSE) between model and
market implied volatilities and the average bid-ask (Bid/Ask) spread of quoted implied volatilities.
(b)–(d) Time series of estimated initial volatility σ0 and mean volatility θ, volatility beta β, and
volatility-of-volatility ε, respectively.

d We apply scipy implementation of Sequential Least Squares Programming (SLSQP) algorithm.

2450003-45
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

implied volatilities of ask and bid quotes for options in the calibration set. We
observe that most of the times, the average model mis-pricing is within the average
bid-ask spread. We see that year 2020 of COVID-19 pandemic is characterized by
high volatility and high spreads, yet the model is able to fit these markets well. In
recent years of 2022 and 2023, bid-ask spreads declined while the model error (the
average difference between market and model implied volatility) became less than
1% most of the times.
In Figs. 7(b)–7(d), we show the time series of calibrated values of initial volatility
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

σ0 and mean volatility θ, volatility beta β, and volatility-of-volatility ε, respectively.

We observe downward trend in the level of implied volatility, falling from average
levels of 80% in years 2021 and 2022 to most recent averages of about 40%. At the
same time, volatility beta and volatility-of-volatility remain range bound suggesting
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

than relative value of calls and puts is intact despite falling level of the overall
volatility. The volatility beta fluctuates between positive and negative values most
of the times following the market sentiment. When the sentiment is bullish, calls are
bid and volatility beta becomes positive and, vice versa, during bearish sentiment,
puts are bid and the volatility beta becomes negative. Thus, as we discuss in Sec. 3.3,
the conventional SV models may not be arbitrage-free for the valuation of options
on cryptocurrencies when return–volatility correlation become positive.
In Fig. 8, we show the quality of model fit to Bitcoin options traded on Deribit
exchange on 20 June 2023. On this day, fitted model parameters are the following:
σ̂0 = 0.41, θ̂ = 0.38, β̂ = 0.50, ε̂ = 3.06, κ̂1 = 2.21, κ̂2 = 2.18. (6.4)
In Fig. 8, we show the quality of model fit on this date. We observe positive
skewness of implied volatilities, which results in positive volatility beta. Continuous
black line shows model implied volatilities computed using the second-order affine
expansion in Eq. (5.9). Bid and ask displays the market bid and ask quotes, the
ATM is the ATM mid-point volatility. MSE denotes the mean squared error between
the model implied volatilities and the mid of market bid-ask implied volatilities. We
see that the log-normal SV model calibrated to Bitcoin options is able to capture
the market implied skew very well across most liquid maturities with only four
parameters. The average MSE is about 0.5% for longer maturities, which is within
the bid-ask spread. Calibration to the ATM region can be further improved using
a term structure of the mean volatility θ. In a practical setting, the SV model
dynamics can be augmented with a local volatility (Lipton (2002)) to accurately fit
the implied volatility surface.

6.3. Comparison of valuation under MMA and inverse measures

Since Bitcoin options, which are traded on the Deribit exchange, have inverse pay-
offs, it is important that there is no-arbitrage between the MMA and the inverse
measures. In Fig. 9, we compare the model implied volatilities computed from prices
of vanilla and inverse options valued under the MMA and inverse measures, respec-
tively. We use the second-order affine expansion with model parameters reported

2450003-46
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(a) (b)

(c)

Fig. 8. Quality of model fit to Bitcoin options on 20 June 2023 reported in Eq. (6.4) using the three
most liquid slices: one week (a), two weeks (b), one month (c). Continuous black line is model
implied volatilities computed using the second-order affine expansion in Eqs. (5.9) and (5.16). Bid
and ask displays the market bid and ask quotes, the ATM is the ATM mid-point volatility. Model
implied volatility is displayed using the continuous black line. MSE is the mean squared error
between the model and the mid of market bid-ask volatilities.

2450003-47
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(a) slice - 1w: 14Jul2023 (b) slice - 2w: 21Jul2023

(c) slice - 1m: 28Jul2023

Fig. 9. Implied volatilities of Bitcoin options which correspond to the slices used in model calibra-
tion and a uniform range of strikes and which are computed using model parameters reported in
Eq. (6.4) under the MMA measure (MMA) and the inverse measure (Inverse). Continuous aqua
and pink lines show model implied volatilities computed using the second-order affine expansion
in Eqs. (5.9) and (5.16), respectively. Dashed lines labeled by MC − 0.95ci and MC + 0.95ci are
MC 95% confidence intervals computed using scheme in Eq. (3.59) with number of path equal
400, 000 and daily time steps. MSE is the mean squared error between BSM volatilities inferred
from model values and the MC estimates.

2450003-48
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(a) slice - 1w (b) slice - 2w

(c) slice - 1m

Fig. 10. Implied volatilities of call options on QV of Bitcoin using model parameters in Eq. (6.4)
and for the same maturities as in model calibration in Fig. 8. We use the valuation under the MMA
measure (MMA) and the inverse measure (Inverse). BSM volatilities are implied from model prices
using expected QV Ibτ = {0.18, 0.20, 0.23} computed using Eq. (3.53). Continuous aqua and pink
lines show model implied volatilities computed using the second-order affine expansion under MMA
measure in Eq. (5.20) and under inverse measure in (5.24), respectively. Dashed lines labeled by
MC−0.95ci and MC+0.95ci are MC 95% confidence intervals computed using scheme in Eq. (3.59)
with number of path equal 400, 000 and daily time steps. MSE is the mean squared error between
BSM volatilities inferred from model values and the MC estimates.

2450003-49
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

in Eq. (6.4) for computing model implied volatilities for the same maturities as in
Fig. 8. As benchmark, we use the MC simulation of the model dynamics (3.12)
using scheme in Eq. (3.59) with the number of paths equal 400, 000 and daily time
steps. We show the 95% confidence interval of MC estimates as dashed lines like
bid/ask lines
We observe that option values computed using the second order affine expansion
under the MMA and inverse measures are very close to each other with the differ-
ences being within the numerical accuracy of an ODE solver and Fourier inversion
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

(of order 10−4 ). The difference in terms of implied volatilities does not exceed 10−4 .
In Fig. 10, we show model prices of call options on the QV computed using the
second-order affine expansion under the MMA and inverse measures, and the com-
parison with MC estimates. We use the same expiries as for the model calibration
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

and the same parameters reported in Eq. (6.4). We also observe that the solution
under the both measures are consistent and agree with MC simulations. Interest-
ingly, we observe that the implied volatility skew of options on the QV is upward
sloping, which is observed in market data of options on the QV and listed options on
VIX index. It is known that Heston model, even augmented with jumps, produces
downward sloping skews on options on the QV (see for an example Drimus (2012)
and Sepp (2012)). As a result, the log-normal SV model implies realistic patterns
of implied volatility-of-volatility.

7. Conclusion
We have considered the log-normal SV model with the quadratic drift which plays an
important role for the existence of martingale measures. We have shown that the
log-normal volatility process has the strong solution in spite of having quadratic
drift. We have derived admissible bounds of model parameters for the existence
of both the MMA measure and the inverse measures. Based on that, we applied
proposed SV model for modeling implied volatility surfaces of assets with positive
return–volatility correlation.
Since the log-normal SV model is not affine, there is no analytical solution for
the MGF in this model. To circumvent this, we have developed an affine expansion
approach under which the joint MGF of three state variables (log-price, the QV, and
the volatility) is decomposed into a leading term, which has exponential-affine form,
and a residual term, whose estimate depends on the higher order moments of the
volatility process. We have proved that the second-order leading term is consistent
with the expected values and the variances of the state variables. We have applied
Fourier inversion techniques for valuation of vanilla and inverse options on spot and
futures underlying. By comparison of model values computed using our method
with estimates obtained from MC simulations, we have shown that the second-order
leading term is very accurate for option valuation. We have demonstrated that our
model fits implied volatilities of assets with positive return–volatility correlation.
We have shown that the log-normal SV model fits well different market regimes by

2450003-50
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

calibrating model parameters to time series of options on Bitcoin for the past four
years.
We have derived backward Euler–Maruyama scheme for discretization of the
log-volatility process and show that it has a strong convergence rate of 1. Proposed
scheme for the simulation of the log-normal SV model is robust, computationally
efficient and does not require any special treatment in discretization.
As an extension of the frameworke we shall consider the applications to the
rough SV models proposed by Gatheral et al. (2018) using log-normal volatility. We
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

shall also consider a path-dependent model of the log-normal volatility following

approaches in Guyon & Lekeufack (2023) and Lipton and Reghai (2023).f
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

Acknowledgments
The authors are thankful to Vladimir Lucic for valuable insights and comments.
We thank Alan Lewis, Johannes Muhle-Karbe, Blanka Horvath, Kalev Pärna and
participants at Imperial College Finance and Stochastics Seminar, TU Munich
Oberseminar, ETH Seminar, EAJ Conference 2022 in Tartu for useful discussion
and comments. We are grateful for anonymous referees for helpful comments.

Appendix A. Proofs
A.1. Proof of Theorem 3.2
We consider the interval (l, r) and define the scale function and its density, denoted
by S(x) and s(x), respectively, as follows (see Chap. IV.15 in Borodin (2017)):

)  x *  x
2μ(σ)
s(x) = exp − dσ , S(x) = s(y)dy, (A.1)
c v(σ)2 c

where x ∈ (l, r) and c is an arbitrary but fixed interior point of (l, r). We define the
speed function and its density, denoted by M (x) and m(x), respectively, by

) x *  x
2 2μ(σ)
m(x) = exp dσ , x ∈ (l, r), M (x) = m(y)dy. (A.2)
v(x)2 c v(σ)2 c

e Sepp & Rakhmonov (2023a) apply the developed solution for modeling of SV in Cheyette-type
interest rate model and obtain closed-form solution for valuation of swaptions under this model.
f The volatility beta is an interesting variable to make path-dependent as well. We observe that in

many markets the implied options skewness follows most recent price performance. For example,
in options on Bitcoin, the skewness implied from option prices becomes positive following strong
performance of Bitcoin.

2450003-51
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

x κ1 −κ2 θ
Choosing c such that c 2μ(σ)v(σ)2 dσ = 0 and setting η = 2 ϑ2 , we obtain
) * ) *
2κ1 θ 1 2κ2 2 −η−2 2κ1 θ 1 2κ2
s(x) = xη exp + x , m(x) = x exp − − x .
ϑ2 x ϑ2 ϑ2 ϑ2 x ϑ2
 x  x ) *
2κ1 θ 1
S(x) = s(y)dy ≤ y η exp dy, x ↓ 0+,
c c ϑ2 y
 x  x ) *
2κ2
S(x) = s(y)dy ≥ y η exp y dy, x ↑ +∞. (A.3)
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

c c ϑ2
Using upper and lower bounds in (A.3), we find that
S(0+) = lim S(x) = −∞, S(+∞) = lim S(x) = +∞. (A.4)
x↓0+ x↑+∞
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

Consequently, boundaries l = 0, r = +∞ are not attractive.

Now we show that boundary r = +∞ is unattainable. We define Σ(l), Σ(r) by
 
Σ(l) := s(z)m(y)dzdy, Σ(r) := s(z)m(y)dzdy, (A.5)
l<z<y<x x<y<z<r
 r  r  r  y
Σ(r) = s(z)dz m(y)dy = m(z)dz s(y)dy
x y x x
 r  y ) *
2 −η−2 2κ1 θ 1 2κ2
= 2
z exp − 2 z
− 2
z dz y η
x x ϑ ϑ ϑ
) *
2κ1 θ 1 2κ2
× exp + y dy. (A.6)
ϑ2 y ϑ2
We note that the
 rright
 yhand side of
) (A.6) is* finite if and)only if*
! −η−2 2κ2 2κ2
Σ(r) := z exp − 2 z dz y η exp y dy (A.7)
x x ϑ ϑ2
is finite. We split inner integral into two integrals, over [x, +∞) and [y, +∞)
 y ) *  +∞  +∞
2κ2
z −η−2 exp − 2 z dz = · · · dz − · · · dz = I1 − I2 . (A.8)
x ϑ x y

We see that I1 is finite for all κ1 , κ2 > 0. Using asymptotics of incomplete gamma
function (see Eq. (6.5.32) in Abramowitz & Stegun (1972)), we obtain for y → +∞
 +∞ ) * −1
−η−2 2κ2 2κ2
I2 = z exp − 2 z dz = y −η−2
y ϑ ϑ2
) *
2κ2 1
× exp − 2 y 1+O . (A.9)
ϑ y
Substituting asymptotic expansion (A.9) into Eqs. (A.8) and (A.7), we have
 r ) * −1  r  r
! 2κ2 2κ2 −2
Σ(r) = I1 y η exp y dy − y dy + O y −3 dy ,
x ϑ2 ϑ2 x x

y → +∞.

2450003-52
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

!
We conclude that Σ(+∞) = +∞. Hence, boundary r = +∞ is unattainable.
Next, we show that boundary l = 0 is unattainable as well. By definition of Σ(l)
 x  y  x  x
Σ(l) = s(z)dz m(y)dy = m(z)dz s(y)dy
l l l y
 x  x ) *
2 −η−2 2κ1 θ 1 2κ2
= 2
z exp − 2 − 2 z dz y η
l y ϑ ϑ z ϑ
) *
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

2κ1 θ 1 2κ2
× exp + 2 y dy.
ϑ2 y ϑ
We make an reciprocal transformation {z → 1/z, y → 1/y} to obtain
 1/l  y * 
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

)
2 η 2κ1 θ 2κ2 1
Σ(l) = 2
z exp − 2 z − 2 dz y −η−2
1/x 1/x ϑ ϑ ϑ z
) *
2κ1 θ 2κ2 1
× exp y + dy.
ϑ2 ϑ2 y
The right-hand side of is finite if and only if
 1/l  y ) *  ) *
! 2κ1 θ 2κ1 θ
Σ(l) := z exp − 2 z dz y −η−2 exp
η
y dy (A.10)
1/x 1/x ϑ ϑ2

is finite. Splitting inner integral into two, over [1/x, +∞) and [y, +∞), respectively
 y ) *  +∞  +∞
2κ1 θ
z η exp − 2 z dz = · · · dz − · · · dz = I1 − I2 , (A.11)
1/x ϑ 1/x y

we see that I1 if finite for all κ1 , κ2 > 0. As before, using asymptotic expansion of
incomplete gamma function we have for y → +∞:
 +∞ ) * ) *
2κ1 θ ϑ2 2κ1 θ 1
I2 = z exp − 2 z dz =
η
y exp − 2 y
η
1+O .
y ϑ 2κ1 θ ϑ y
(A.12)

Substituting Eq. (A.12) into Eqs. (A.11) and (A.10), we have for y → +∞
 1/l ) *  1/l  
1/l
! −η−2 2κ1 θ ϑ2 −2 −3
Σ(l) = I1 y exp y dy − y dy + O y dy .
1/x ϑ2 2κ1 θ 1/x 1/x

!
We conclude that Σ(+∞) = +∞. Hence, boundary l = 0 is unattainable.

A.2. Proof of Theorem 4.1

We provide a version of Feynman–Kac formula in general multidimensional setting,
building on results in Heath & Schweizer (2000). Let T > 0 be a fixed time horizon

2450003-53
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

and D be a domain in Rd . We consider Cauchy problem

∂U
+ LU = 0, (t, x) ∈ Q, Q = (0, T ) × D,
∂t (A.13)
U (T, x) = φ(x), x ∈ D,

where bt : D → Rd , at : D → Rd×d are continuous functions and operator L is

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

$
d
∂U 1 $ ik
d
∂2U
LU = bit (x) (t, x) + at (x) (t, x). (A.14)
i=1
∂xi 2 ∂xi ∂xk
i,k=1
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

We assume that matrix (aij t ) is nonnegative definite for every t. Under certain
restrictions on the coefficients of operator L and growth conditions on φ, it is possi-
ble to represent the solution of (A.13) by means of conditional expectation known
as a Feynman–Kac formula. However, standard results require uniform ellipticity
of the operator L, see Friedman (1975), which is not satisfied in our case, because
the QV It is degenerate. Thus, problem (A.13) becomes degenerate parabolic one.
We consider the multi-dimensional SDE

dXs = bs (Xs )ds + σs (Xs )dWs , Xt = x ∈ D, (A.15)

where W = (W (1) , . . . , W (d) ) is d-dimensional Brownian motion and σt (x) =

(σtij (x)) denotes square root of matrix at (x): aik ik
t (x) = (σt (x)σt (x) ) . We assume
∗
that (A.15) admits unique strong solution and define U : [0, T ] × D → [0, +∞]
by U ∗ (τ, x) = E[φ(XT )|Xt = x] which is well defined in [0, +∞] if X does not
explode or leave D before T . We give sufficient conditions to ensure that U ∗ solves
the problem (A.13).

Theorem A.1 (Existence). Suppose that following conditions hold:

(1) Solution X of (A.15) neither explodes nor leaves D before T.

P sup |Xs | < ∞ = P(Xs ∈ D, ∀ s ∈ [t, T ]) = 1.

t≤s≤T

(2) φ(x) is bounded in D.

Let U ∈ C(Q) ∩ C 2 (Q) be the solution of problem (A.13). Then U (t, x) =

∗
U (t, x) for any (t, x) ∈ Q.

Proof. Consider sequence of bounded domains {Dn }∞ n=1 contained in D such that
-∞
n=1 Dn = D and Dn ⊆ {x ∈ R : x < n}. We fix x ∈ D and find n such that
d

x ∈ Dn . Let η = inf{s ≥ t : Xs ∈ D}, ηn = inf{s ≥ t : Xs ∈ Dn } denote exit time

2450003-54
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

from domains D and Dn . By Itô’s formula

 ηn ∧T  
∂U
u(ηn ∧ T, Xηn ∧T ) = u(x) + − LU (s, Xs ) ds
t ∂s
 ηn ∧T
∂U
+ σij (Xs ) dWs(j) .
t ∂xi
As domain Dn is bounded and U ∈ C 2 (Q), we have
 ηn ∧T
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

∂U
E t,x
σij (Xs ) dWs(j) = 0 ⇒ u(x) = Et,x [u(ηn ∧ T, Xηn ∧T )].
t ∂xi

(A.16)
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

−2
We show that P [τn ≤ T ] ≤ c1 (1 + x )n
t,x 2
. Using Markov inequality and
estimate from Karatzas & Shreve (1991), p. 306:

Pt,x [τn ≤ T ] ≤ Pt,x sup Xs  ≥ n ≤ Et,x sup |Xs |2 n−2 ,

s∈[t,T ] s∈[t,T ]

Et,x sup |Xs |2 ≤ c(1 + x2 ).

s∈[t,T ]

Thus limn→+∞ u(ηn ∧ T, Xηn∧T ) = φ(XT ) a.s. It implies that u(t, x) is bounded
by the maximum principle. Finally, by the dominated convergence theorem:
lim E[u(ηn ∧ T, Xηn ∧T )] = E[φ(XT )]. (A.17)
n→+∞

Combining (A.16) and (A.17) concluded the proof.

A.3. Consistency of moment for second-order affine expansion

We apply the following lemma relating moments to derivatives of MGF.

Lemma A.1. The MGF defined by the following equation:

G(Φ) = E[e−ΦZ ], (A.18)
where Z is a random variable with finite MGF has the following properties:
G(0) = 1,
∂Φ G(Φ)|Φ=0 ≡ GΦ (0) = −E[Z], (A.19)
2
∂Φ G(Φ)|Φ=0 ≡ GΦΦ (0) = E[Z 2 ].
Corollary A.1. Given that MGF assumes an exponential-affine form
G(Φ) = ef (Φ) (A.20)
with f (0) = 0, we have
fΦ (0) = −E[Z], fΦΦ (0) = Var[Z]. (A.21)

2450003-55
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

A.3.1. Proof of Proposition 4.4

We set Φ = Ψ = 0 in the second-order expansion in Proposition 4.4 and suppress
arguments for brevity. We consider R[2] (τ, Y ; Θ) in Eq. (4.26). Differentiating it
[2]
w.r.t. Θ, we see that remainder term RΘ solves the following problem:

[2] [2]
$
8
(k) [2]
−RΘ,τ + L(Y ) RΘ = − Y k [CΘ (τ ; Θ)E [2] + C (k) (τ ; Θ)EΘ ], (A.22)
k=5
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

[2]
subject to boundary condition RΘ (0, Y ; Θ) = 0, where C (k) (τ ; Θ) are given by
Eq. (4.27). We note that A(k) (τ ; Θ) vanish at Θ = 0.
A(k) (τ ; Θ = 0) = 0. (A.23)
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(k)
Calculating CΘ (τ ; Θ) and taking into account conditions (A.23) at Θ = 0
# (4)
(k) −4κ2 AΘ (τ ; Θ)|Θ=0 , k = 5,
CΘ (τ ; Θ)|Θ=0 = (A.24)
0, k ≥ 6.

Using condition C (k) (τ ; Θ = 0) ≡ 0, we simplify problem (A.22) at Θ = 0:

[2] [2] (4) [2]
−RΘ,τ + L(Y ) RΘ = 4κ2 AΘ Y 5 , RΘ (0, Y ; Θ) = 0. (A.25)
[2]
We see that RΘ (τ, Y ; Θ = 0) ≡ 0 when κ2 = 0. Hence, second order decompo-
sition E [2] (τ, Y ; Θ) is consistent with expected value of Yt for κ2 = 0.
To prove the consistency for variance, we first differentiate (4.25) w.r.t. Θ to
(k)
obtain that AΘ (τ, Θ) solve the system of ODEs (A.26) at Θ = 0 as functions of τ .
(0) (2) (1)
AΘ,τ = θ2 ϑ2 AΘ + λ(p)AΘ ,
(1) (1) (2) (3)
AΘ,τ = −κAΘ + 2(θϑ2 + λ(p))AΘ + 3θ2 ϑ2 AΘ ,
(2) (1) (2) (3) (3) (4)
AΘ,τ = −κ2 AΘ + (ϑ2 − 2κ)AΘ + 6θϑ2 AΘ + 3λ(p)AΘ + 6θAΘ , (A.26)
(3) (2) (4) (4) (4)
AΘ,τ = −2κ2 AΘ + 3(ϑ2 − κ)AΘ + 12θϑ2 AΘ + 4λ(p)AΘ ,
(4) (3) (4)
AΘ,τ = −3κ2 AΘ + 2(3ϑ2 − 2κ)AΘ
(k )
with boundary condition AΘ (τ ; Θ)|Θ=0 = −1, AΘ (τ ; Θ)|Θ=0 = 0, k  = 1.
(1)

We verify that last two functions in solution of Eq. (A.26) are zero for κ2 = 0:
A(3) (τ ; Θ) ≡ A(4) (τ ; Θ) ≡ 0. (A.27)
[2]
Now, taking the second derivative w.r.t. Θ in (4.26), we obtain for RΘΘ
[2] [2]
−RΘΘ,τ + L(Y ) RΘΘ

$
8
(k) (k) [2] [2]
=− {Y k [CΘΘ (τ ; Θ)E [2] + 2CΘ (τ ; Θ)EΘ + C (k) (τ ; Θ)EΘΘ ]}
k=5

2450003-56
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

[2]
with RΘΘ (0, Y ; Θ) = 0. Taking the second derivative of C (k) (τ ; Θ) in (4.27) w.r.t.
Θ, and accounting for conditions (A.23) and (A.27) at Θ = 0, we find that
# (4)
(k) −4κ2 AΘ (τ ; Θ)|Θ=0 , k = 5,
CΘΘ (τ ; Θ)|Θ=0 = (A.28)
0, k ≥ 6.

We note that following straightforward relationships are valid:

by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

[2]
E [2] (τ ; Θ)|Θ=0 = 1, EΘ (τ ; Θ)|Θ=0 = −Y. (A.29)

Combining Eqs. (A.28), (A.24), and (A.29), we simplify Eq. (A.25) at Θ = 0:

[2] [2] (4) (4) [2]
−RΘΘ,τ + L(Y ) RΘΘ = 4κ2 (Y 5 AΘΘ − 2Y 6 AΘ ), RΘΘ (0, Y ; Θ) = 0. (A.30)
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

[2]
Thus, when κ2 = 0, RΘΘ (τ, Y ; Θ = 0) ≡ 0, and E [2] (τ ; Θ) reproduces the variance
of Yt .

A.3.2. Proof of Proposition 4.5

We proceed as in Proposition 4.4. We set Θ = Ψ = 0 and consider the remain-
[2]
der term R[2] (τ, Y ; Φ) in (4.26). Differentiating it w.r.t. Φ, we see that RΦ solves
following problem:

∂ $ (k)
8
[2] [2]
−RΦ,τ + L(Y ) RΦ = − C (τ ; Φ)E [2] Y k
∂Φ
k=5

$
8
(k) [2] (A.31)
=− Y k [CΦ (τ ; Φ)E [2] + C (k) (τ ; Φ)EΦ ],
k=5

[2]
RΦ (0, Y ; Φ) = 0,

where C (k) (τ ; Φ) are given by (4.27). We note that A(k) (τ ; Φ) vanish at Φ = 0:

A(k) (τ ; Φ)|Φ=0 = 0. (A.32)

Differentiating C (k) (τ ; Φ) w.r.t. Φ and considering (A.32) at Φ = 0, we find

# (4)
(k) −4κ2 AΦ (τ ; Φ)|Φ=0 , k = 5,
CΦ (τ ; Φ)|Φ=0 = (A.33)
0, k ≥ 6.

Due to condition C (k) (τ ; Φ = 0) = 0, we simplify Eq. (A.31) at Φ = 0 as

[2] [2] (4) [2]
−RΦ,τ + L(Y ) RΦ = 4κ2 AΦ Y 5 , RΦ (0, Y ; Φ) = 0. (A.34)
[2]
We see that RΦ (τ, Y ; Φ = 0) ≡ 0 when κ2 = 0. Hence, the second-order expan-
sion term E [2] (τ, Y ; Φ) is consistent with the expected value of Xτ for κ2 = 0.

2450003-57
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

To show that the second-order expansion reproduces the variance of the log-
(k)
price, we differentiate Eq. (4.25) with respect to Φ and find that AΦ (τ, Φ) solve
the system of ODEs (A.35) at Φ = 0 as functions of τ :

(0) (2) (1) 1

AΦ,τ = θ2 ϑ2 AΦ + λ(p)AΦ + θ2 ,
2
(1) (1) (2) (3)
AΦ,τ = −κAΦ + 2(θϑ2 + λ(p))AΦ + 3θ2 ϑ2 AΦ + θ,
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

(2) (1) (2) (3) (3) (4) 1 (A.35)

AΦ,τ = −κ2 AΦ + (ϑ2 − 2κ)AΦ + 6θϑ2 AΦ + 3λ(p)AΦ + 6θAΦ + ,
2
(3) (2) (4) (4) (4)
AΦ,τ = −2κ2 AΦ + 3(ϑ2 − κ)AΦ + 12θϑ2 AΦ + 4λ(p)AΦ ,
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

(4) (3) (4)

AΦ,τ = −3κ2 AΦ + 2(3ϑ2 − 2κ)AΦ

with boundary condition

#
(k) −1, k = 1,
AΦ (τ ; Φ)|Φ=0 = (A.36)
0, k = 1.
[2]
Now, taking the second derivative in (4.26) w.r.t. Φ, we obtain for RΦΦ

∂ 2 $ (k)
8
[2] [2]
−RΦΦ,τ +L (Y )
RΦΦ =− 2 C (τ ; Φ)E [2] Y k
∂Φ
k=5

$
8
(k) (k) [2] [2]
=− {Y k [CΦΦ (τ ; Φ)E [2] + 2CΦ (τ ; Φ)EΦ + C (k) (τ ; Φ)EΦΦ ]},
k=5

[2]
× RΦΦ (0, Y ; Φ) = 0. (A.37)
(k)
Calculating CΦΦ (τ ; Φ) in (4.27) and taking into account boundary conditions
(A.32) at Φ = 0, we find that
# (4)
(k) −4κ2 AΦ (τ ; Φ)|Φ=0 , k = 5,
CΦΦ (τ ; Φ)|Φ=0 = (A.38)
0, k ≥ 6.

We note that following straightforward relationships are valid:

[2]
E [2] (τ ; Φ)|Φ=0 = 1, EΦ (τ ; Φ)|Φ=0 = 0. (A.39)

Combining conditions (A.38), (A.33), and (A.39), we are able to simplify prob-
lem (A.34) at Φ = 0 as follows:
[2] [2] (4) (4) [2]
−RΦΦ,τ + L(Y ) RΦΦ = 4κ2 (Y 5 AΦΦ − 2Y 6 AΦ ), RΦΦ (0, Y ; Φ) = 0. (A.40)
[2]
Thus, when κ2 = 0, RΦΦ (τ, Y ; Φ = 0) ≡ 0, so that E [2] (τ, Y ; Φ) is consistent with
the variance of Xt .

2450003-58
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

A.4. Proof of Theorem 4.7

We consider following system of d nonlinear ODEs:

∂τ Ai (τ ; Γ) = A(τ ; Γ) Mi A(τ ; Γ) + Li A(τ ; Γ) + Hi , Ai (0, u) = Ai,0 (A.41)

for i = 1, . . . , d. We assume that Γ ∈ C3 and each matrix Mi is symmetric real-

valued, and vectors Li and Hi can be complex-valued. Hence solutions Ai (τ ; Γ),
i = 1, . . . , d are complex-valued functions. It is well known that complex-valued
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

solution A(τ ; Γ) is defined on maximal interval of existence [0, τ+ (Γ; A)) such that
either τ+ (Γ; A) = +∞ or τ+ (Γ; A) < +∞ and A(τ ; Γ) → +∞ as τ ↑ τ+ (Γ; A).
We combine d nonlinear ODEs in (A.41) into single (nonlinear) vector ODE:
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

∂τ A(τ ; Γ) = M (A(τ ; Γ)) + LA(τ ; Γ) + H, A(0, u) = A0 , (A.42)

where M : Rd → Rd is a vector-valued function of argument A ∈ Rd . For conve-

nience, we denote the right-hand side of (A.42) as follows:

R(A) := M (A) + LA + H. (A.43)

Lemma A.2. There exists finite and nonnegative function g such that
 t Rs
A(τ ; Γ) ≤ A0  + A0 
2 2 2
h(s)e 0 h(r)dr ds, h(s) = g( (A(τ ; Γ)).
0
(A.44)

Proof. The proof is essentially contained in Keller-Ressel & Mayerhofer (2015) and
based on the application of Gronwall’s inequality.

Lemma A.3. Assume that following conditions are satisfied:

(1) HI = HI (u) = (0, . . . , 0) ∈ Rd when Γ is a real vector,

(2) LI = LI (u) is zero d × d matrix when Γ is a real vector,
(3) zero initial condition A(0, Γ) = (0, . . . , 0) ∈ Rd .

Then solution A(τ ; Γ) of system (A.42) remains real for real Γ ∈ Rd .

Proof. The proof is available upon request.

We now consider the problem of continuity of solutions of quadratic differential

systems arising in option valuation problem. It is important to highlight that linear
L(p) and free terms H(p) satisfy assumptions of Lemma (A.3) when we set p = 1 and
restrict Φ = −1/2, Φ = Ψ = 0 for pricing vanilla options under MMA measure.
We use same argument for the inverse options when we set p = −1 and restrict
Φ = 1/2, Φ = Ψ = 0. We omit straightforward calculations.
We use τ+ (Φ; A) and τ+ ( Φ; AR ) to denote blow-up times τ+ (Γ; A) and
τ+ ( Γ; AR ) when Γ = (Φ, Ψ = 0, Θ = 0) and Γ = ( Φ, Ψ = 0, Θ = 0).

2450003-59
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

Theorem 4.7. We assume first that Φ ∈ R. We denote

D(τ ) := {Φ ∈ R : τ < τ+ (Φ; A)}, M (τ ) := {Φ ∈ R : G(τ ; X, I, Y ; Φ) < ∞}.
We argue by contradiction. We assume that solution A(τ ; Φ) blows up before τ0 :
τ+ (Γ; A) < τ0 . (A.45)
We set α∗ := sup{α ≥ 0 : αΓ ∈ D(τ )}. Then assumption (A.45) implies that
∗
α < 1. On the one hand
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

G(τ ; X, I, Y ; αΦ, Ψ = 0, Θ = 0) ≤ G(τ ; X, I, Y ; αΦ, Ψ = 0, Θ = 0)α < ∞ (A.46)

for all α < α∗ < 1 due to Jensen inequality. On the other hand, we have
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

G(τ ; X, I, Y ; αΦ, Ψ = 0, Θ = 0)
# %
$2n
= exp −αΦX + Ak (τ ; αΦ, Ψ = 0, Θ = 0)Y k

k=0

+ exp{−αΦX}R[2](τ ; X, I, Y ; αΦ, Ψ = 0, Θ = 0).

Thus, we obtain
lim G(τ ; X, I, Y ; αΦ, Ψ = 0, Θ = 0) = +∞ (A.47)
α↑α∗

under assumptions 2 and 3. Comparing (A.46) and (A.47) we get contradiction, so

τ+ (Φ; A) ≥ τ0 . (A.48)
Now, we let Φ ∈ C be complex. Due to assumption 4.7 and Lemma A.2, we
have τ+ (Φ; A) ≥ τ+ (Φ; AR ) ≥ τ ( Φ; A). Repeating the first part of the proof for
Φ, leads to (A.48), so that τ+ ( Φ; A) ≥ τ0 . Hence, solution A(τ ; Φ, Ψ = 0, Θ = 0)
must remain continuous on [0, τ0 ).

ORCID
Artur Sepp https://orcid.org/0000-0002-7038-1748
Parviz Rakhmonov https://orcid.org/0000-0001-9571-7378

References
M. Abramowitz & I. A. Stegun (1972) Handbook of Mathematical Functions with Formulas,
Graphs, and Mathematical Tables, Applied Mathematics Series. National Bureau of
Standards.
D. Ackerer & D. Filipović (2020) Option pricing with orthogonal polyolynomial expansions,
Mathematical Finance 30 (1), 47–84.
C. Alexander, D. Chen & A. Imeraj (2023) Crypto quanto and inverse options, Mathe-
matical Finance 33(4), 1005–1043.
E. Alòs, J. Gatheral & R. Radoicic (2020) Exponentiation of conditional expectations
under stochastic volatility, Quantitative Finance 20, 13–27.

2450003-60
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

A. Alfonsi (2013) Strong order one convergence of a drift implicit Euler scheme: Application
to the CIR process, Statistics & Probability Letters 83(2), 602–607.
G. Bakshi & N. Kapadia (2003) Delta-hedged gains and the negative market volatility risk
premium, The Review of Financial Studies 16(2), 527–566.
J. Baris, P. Baris & B. Ruchlewicz (2008) Blow-up solutions of quadratic differential sys-
tems, Journal of Mathematical Sciences 149 (4), 1369–1375.
O. E. Barndorff-Nielsen & A. N. Shiryaev (2015) Change of Time and Change of Measure,
Vol. 21. World Scientific Publishing Company.
G. Bakshi, J. Ju & H. Ou-Yang (2006) Estimation of continuous-time models with an appli-
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

cation to equity volatility dynamics, Journal of Financial Economics 82(1), 227–249.

L. Bergomi (2015) Stochastic Volatility Modeling. CRC Press.
L. Bergomi & J. Guyon (2012) Stochastic volatility orderly smiles, Risk Magazine 25(5)
117–123.
F. Black & M. Scholes (1973) The pricing of options and corporate liabilities, The Journal
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

of Political Economy 81, 637–659.

A. N. Borodin (2017) Stochastic Processes. Springer.
H. Bühler (2006) Volatility Markets: Consistent Modeling, Hedging and Practical Imple-
mentation. PhD Thesis, TU Berlin.
P. Carr & S. Willems (2019) A lognormal type stochastic volatility model with quadratic
drift. arXiv:1908.07417 .
P. Carr & L. Wu (2017) Leverage effect, volatility feedback, and self-exciting market
disruptions, Journal of Financial and Quantitative Analysis 52(5), 2119–2156.
J.-F. Chassagneux, A. Jacquier & I. Mihaylov (2016) An explicit Euler scheme with strong
rate of convergence for financial SDEs with non-Lipschitz coefficients, SIAM Journal
on Financial Mathematics 7(1), 993–1021.
P. Christoffersen, K. Jacobs & K. Mimouni (2010) Models for S&P 500 dynamics: Evidence
from realized volatility, daily returns and options prices, The Review of Financial
Studies 23(9), 3141–3189.
W. Coppel (1966) A survey of quadratic systems, Journal of Differential Equations
2(3), 293–304.
C. Cuchiero, M. Keller-Ressel & J. Teichmann (2012) Polynomial processes and their
applications to mathematical finance, Finance and Stochastics 16(4), 711–740.
J. Detemple & C. Osakwe (2000) The valuation of volatility options, Review of Finance
4(1), 21–50.
R. Dickson & L. Perko (1970) Bounded quadratic systems in the plane, Journal of Differ-
ential Equations 7(2), 251–273.
G. G. Drimus (2012) Options on realized variance by transform methods: A non-affine
stochastic volatility model, Quantitative Finance 12(11), 1679–1694.
D. Duffie, D. Filipović & W. Schachermayer (2003) Affine processes and applications in
finance, The Annals of Applied Probability 13(3), 984–1053.
D. Filipović & M. Larsson (2016) Polynomial diffusions and applications in finance,
Finance and Stochastics 20(4), 931–972.
D. Filipović & E. Mayerhofer (2009) Affine diffusion processes: Theory and applications,
Advanced Financial Modelling 8, 1–40.
J.-P. Fouque, G. Papanicolaou & K. R. Sircar (2000) Mean-reverting stochastic volatility,
International Journal of Theoretical and Applied Finance 3(1), 101–142.
A. Friedman (1975) Stochastic Differential Equations and Applications, Vol. 1. Academic
Press.
J. Gatheral, T. Jaisson & M. Rosenbaum (2018) Volatility is rough, Quantitative Finance
18(6), 933–949.

2450003-61
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

A. Sepp & P. Rakhmonov

J. Guyon & J. Lekeufack, (2023) Volatility is (mostly) path-dependent, Quantitative

Finance 23(9), 1221–1258.
I. Gyöngy & N. V. Krylov (1980) On stochastic equations with respect to semimartin-
gales I, Stochastics: An International Journal of Probability and Stochastic Processes
4(1), 1–21.
P. S. Hagan, D. Kumar, A. S. Lesniewski & D. E. Woodward (2002) Managing smile risk,
The Best of Wilmott 1, 249–296, 333.
D. Heath & M. Schweizer (2000) Martingales versus PDEs in finance: An equivalence result
with examples, Journal of Applied Probability 37(4), 947–957.
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

P. Henry-Labordère (2009) Calibration of local stochastic volatility models to market

smiles, Risk Magazine 22(9) 112–117.
M. Herdegen & M. Schweizer (2018) Semi-efficient valuations and put–call parity, Mathe-
matical Finance 28(4), 1061–1106.
S. L. Heston (1993) A closed-form solution for options with stochastic volatility with appli-
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

cations to bond and currency options, The Review of Financial Studies 6(2), 327–
343.
D. H. Jacobson (1974) Extensions of Linear-Quadratic Control; Optimalization and Matrix
Theory. London: Academic Press.
B. Jørgensen (1982) Statistical Properties of the Generalized Inverse Gaussian Distribution.
Springer Science & Business Media.
P. Karasinski & A. Sepp (2012) Beta stochastic volatility model, Risk Magazine 22(9)
66–71.
I. Karatzas & S. E. Shreve (1991) Brownian Motion and Stochastic Calculus, Graduate
Texts in Mathematics. New York: Springer-Verlag.
M. Keller-Ressel & E. Mayerhofer (2015) Exponential moments of affine processes, The
Annals of Applied Probability 25(2), 714–752.
R. A. Jarrow & M. Larsson (2012) The meaning of market efficiency, Mathematical
Finance: An International Journal of Mathematics, Statistics and Financial Eco-
nomics 22(1), 1–30.
A. L. Lewis (2000) Option Valuation Under Stochastic Volatility. Finance Press.
A. L. Lewis (2018) Exact solutions for a GBM-type stochastic volatility model having a
stationary distribution. arXiv:1809.08635.
P.-L. Lions & M. Musiela (2007) Correlations and bounds for stochastic volatility models,
Annales de l’Institut Henri Poincaré C, Analyse non linéaire 24(1), 1–16.
A. Lipton (2001) Mathematical Methods for Foreign Exchange: A Financial Engineer’s
Approach. World Scientific.
A. Lipton (2002) The vol smile problem, Risk Magazine 61–65.
A. Lipton and A. Reghai (2023), ‘SPX, VIX and Scale-Invariant LSV Local Stochastic
Volatility’, Wilmott Magazine, February 78–84.
V. Lucic (2021) BTC inverse call and the standard FX framework, SSRN,
https://ssrn.com/abstract=4113726.
V. Lucic & A. Sepp (2023) Valuation and hedging of cryptocurrency inverse
options: With backtest simulations using Deribit options data, SSRN, https://
ssrn.com/abstract=4606748.
J. Masoliver & J. Perelló (2006) Multiple time scales and the exponential Ornstein–
Uhlenbeck stochastic volatility model, Quantitative Finance 6(5), 423–433.
R. C. Merton (1973) Theory of rational option pricing, The Bell Journal of Economics
and Management Science 4(1) 141–183.
M. Musiela & M. Rutkowski (2009) Martingale Methods in Financial Modelling, Vol. 36.
Springer Science & Business Media.

2450003-62
 March 6, 2024 15:25 WSPC/S0219-0249 104-IJTAF SPI-J071
2450003

Log-Normal Stochastic Volatility Model with Quadratic Drift

J. Muhle-Karbe, O. Pfaffel & R. Stelzer (2012) Option pricing in multivariate stochastic

volatility models of OU type, SIAM Journal on Financial Mathematics 3(1), 66–94.
A. Neuenkirch & L. Szpruch (2014) First order strong approximations of scalar SDEs
defined in a domain, Numerische Mathematik 128(1), 103–136.
J. Perelló, R. Sircar & J. Masoliver (2008) Option pricing under stochastic volatility:
The exponential Ornstein–Uhlenbeck model, Journal of Statistical Mechanics 2008
(6), P06010.
Y. Ren, D. Madan & M. Q. Qian (2007) Calibrating and pricing with embedded local
volatility models, Risk Magazine 20(9) 138–143.
by 130.0.122.228 on 10/26/24. Re-use and distribution is strictly not permitted, except for Open Access articles.

J. Ruf (2015) The martingale property in the context of stochastic differential equations,
Electron. Commun. Probab 20. 1–10.
A. Sepp (2007) Variance swaps under no conditions, Risk Magazine 20(1) 82–87.
A. Sepp (2008) Pricing options on realized variance in the Heston model with jumps in
returns and volatility, Journal of Computational Finance 11(4), 33–70.
Int. J. Theor. Appl. Finan. 2023.26. Downloaded from www.worldscientific.com

A. Sepp (2012) Pricing options on realized variance in the Heston model with jumps in
returns and volatility — Part II: An approximate distribution of discrete variance,
Journal of Computational Finance 16(22), 3–32.
A. Sepp & P. Rakhmonov (2023) A robust stochastic volatility model for interest rates,
Risk Magazine, September 2023, 1–6, https://ssrn.com/abstract=4315906.
A. Sepp & P. Rakhmonov (2023) What is a robust stochastic volatility model, SSRN,
https://ssrn.com/abstract=4647027.
N. Shephard (2005) Stochastic Volatility: Selected Readings. Oxford University Press.
C. A. Sin (1998) Complications with stochastic volatility models, Advances in Applied
Probability 30(1), 256–268.
M. Tegner and R. Poulsen, Volatility is log-normal — But not for the reason you think,
Risk Magazine 6(2), 1–16.

2450003-63