Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

Contents lists available at ScienceDirect

Communications in Nonlinear Science and

Numerical Simulation
journal homepage: www.elsevier.com/locate/cnsns

Research paper

Lattice-based model for pricing contingent claims under

mixed fractional Brownian motion
∗,1
Massimo Costabile 1 , Ivar Massabó 1 , Emilio Russo , Alessandro Staino 1
Department of Economics, Statistics and Finance, University of Calabria, Ponte Bucci cubo 1 C, 87036, Rende (CS), Italy

article info a b s t r a c t

Article history: We propose a lattice-based model to approximate the dynamics of an asset with
Received 30 August 2022 diffusion driven by a mixed fractional Brownian motion. Being it defined as the sum of a
Received in revised form 14 November 2022 fractional Brownian motion and an independent Brownian motion, the method starts by
Accepted 5 December 2022
discretizing separately the two processes. For the first process, we develop a binomial
Available online 9 December 2022
approach that is able to replicate the process variance at each discrete epoch, while
Keywords: a recombining binomial lattice is used to discretize the Brownian motion. Given the
Mixed fractional Brownian motion fractional Brownian motion path-dependency, the binomial approach is not recombining.
American options Hence, to reduce the computational complexity, we establish a grid of representative
Binomial algorithms values whose elements cover the range of possible asset values at each time slice.
Discrete-time models
For each value of the grid, the algorithm identifies four successors in a bivariate
environment. The successors may not appear among the generated asset values at the
next epoch. In such cases, interpolation techniques are needed when solving backward
to compute the initial value of the contingent claim. The proposed discretization allows
to evaluate both European and American derivatives and numerical experiments confirm
its accuracy and efficiency.
© 2022 Elsevier B.V. All rights reserved.

1. Introduction

Since its appearance in the 1970s, several modifications and extensions of the Black and Scholes [1] model have been
presented in the financial literature in order to overcome the evident biases induced in contingent claim prices by the
Black–Scholes model. They aim at providing new models that would be more consistent with the stylized facts showed
by financial returns as fat tails, volatility clustering, excess of kurtosis, no-linearity, self-similarity, no-independence or,
better, long-range dependence, to name just a few. To give some examples, we may refer to stochastic volatility models (cf.,
Wiggins [2], Hull and White [3], Scott [4], Johnson and Shanno [5], Chesney and Scott [6], Stein and Stein [7], Heston [8],
Ball and Roma [9], Schöbel and Zhu [10]), to jump diffusion models introduced by Merton [11] and generalized in different
ways also by including regime-switching (cf., Yuen and Yang [12], Ramponi [13], Costabile et al. [14]), or to models based
on Lévy processes for option pricing (cf., Madan and Seneta [15], Shirzad et al. [16]). Unfortunately, in some cases, when
changing the basic Black–Scholes assumptions, the tractability of the problem for what concerns the valuation of financial
derivatives complicates a lot and explicit form solutions could no longer be available.
A similar problem occurs when dealing with no-independence, no-linearity, self-similarity and, in particular, long-
range dependence that affect the asset return time series as evidenced, among others, in Bentes et al. [17], Berg and

∗ Corresponding author.
E-mail address: emilio.russo@unical.it (E. Russo).
1 All the authors contributed equally to this work.

https://doi.org/10.1016/j.cnsns.2022.107042
1007-5704/© 2022 Elsevier B.V. All rights reserved.
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

Lyhagen [18], and Podobnik et al. [19]. The management of such features represents a hot topic of both empirical and
theoretical finance. Indeed, to take properly into account all the cited stylized facts, it has been proposed to use the
fractional Brownian motion (fBm) process. A fBm, introduced by Kolmogorov in the 1940s and for which Mandelbrot and
Van Ness [20] have provided a representation theorem, is a Gaussian process characterized by stationary increments, self-
similarity, and long- or short-range dependence according to the value assumed by the Hurst parameter H ∈ (0, 1): H > 12
represents the long-range dependence case, the increments are positively correlated, and time series show persistent
behaviour; H < 21 represents the short-range dependence case, the increments are negatively correlated, and time series
show anti-persistent behaviour.2
Unfortunately, the fBm process turns out to be problematic for derivative pricing. Indeed, if on the one hand, it is
able to capture the long-range dependence of the asset returns, on the other hand, it is neither a Markov nor a semi-
martingale. Hence, unlike what happens under the Black–Scholes assumptions, it is not possible to use the classical Itô
calculus and to construct a self-financing strategy providing the risk-neutral price of financial derivatives. To solve this
issue, while still taking into account the long-range dependence of the asset returns, a generalization of the fBm has
been introduced by Cheridito [21] and Bender et al. [22] through the proposal of the so-called mixed fractional Brownian
motion (mfBm), which is a linear combination of a fBm and of an independent Brownian motion (Bm). The literature on
the adoption of the mfBm for financial applications is really vast and, for the sake of brevity, hereafter we concentrate on
its use in option pricing. In this sense, among others, contributions have been presented by: Murwaningtyas et al. [23],
who obtain an explicit formula for European plain vanilla options following the arguments already used by Hu and
Oksendal [24] and Necula [25] when dealing only with a fBm; Sun [26], who provide a pricing model for currency options;
Xiao et al. [27], who manage equity warrants; Ahmadian and Ballestra [28], who price geometric Asian rainbow options;
Ballestra et al. [29], who provide an approach for pricing barrier options. The reason of this wide diffusion of the mfBm
is due to fact that, when the Hurst parameter H in the fBm component is greater than 1/2, the mfBm turns out to be
a long-memory process of Gaussian type, thus being particularly suitable for describing the logarithmic returns of the
financial assets. Cheridito [30] has also proved that, for H ∈ (3/4, 1), the mixed model is equivalent to the Bm, and
therefore it is arbitrage-free. Furthermore, Bender et al. [22] show the absence of arbitrage for any value of H ∈ (0, 1)
when suitable replicating portfolios are considered.
The models cited above leave unsolved the evaluation problem of American-style contingent claims written on assets
with dynamics driven by a mfBm for which approximation methods need to be developed due to the unknown distribution
of the optimal exercise time. The main problem to manage is the presence in the mfBm of the fBm that needs accurate
approximations, while the presence of the Bm does not entail any difficulty since it has been already discretized in
several efficient ways. As already evidenced in Lindstrom [31], several discrete approximations of a fBm have been
proposed during the last decades, but not all of them are suitable to be adapted for pricing American options. Indeed,
approximation methods for American options are more difficult to deal with than their European counterpart. In addition,
it is fundamental to develop accurate and efficient methods to price American-style securities because of their wide
diffusion in financial markets. This is the little explored area in which we would like to propose the contribution of
this paper that is mainly focused on the development of a flexible lattice-based model for evaluating both European and
American-style derivatives.
The proposed methodology starts considering separately the Bm and the fBm component in the mfBm process
driving the underlying asset dynamics. The Bm is simply discretized through a recombining binomial lattice. The fBm
is approximated by a novel discrete-time model based on a binomial structure that ensures the matching of the fBm
variance at each observation epoch. The latter represents a crucial aspect because, as reported in ( Lindstrom [31], who
refers to Theorem 1 in Konstantopoulos and Sakhanenko [32], when the Hurst parameter H ∈ 21 , 1 , the matching of
)
the variance guarantees the convergence of the discrete-time model to the continuous fBm as the number of the discrete
epochs tends to infinity. Hence, the proposed discretization captures the long-range dependence affecting asset dynamics
when H > 12 and, as showed in the section devoted to numerical experiments, it provides accurate results even when
H < 12 representing the short-range dependence case. At this point, to capture the asset dynamics, the discrete process
generated for the fBm must be combined with the discrete process approximating the Bm. Unfortunately, the strong path-
dependent structure characterizing the fBm process makes complex the construction of a discrete approximation for the
mfBm (and, consequently, for the asset dynamics) obtained by coupling at the same discrete epoch each value generated
to approximate the fBm with each value of the recombining lattice for the Bm. Indeed, starting from inception, the
supposed binomial structure discretizing the fBm presents a non-recombining shape that makes the evaluation problem
computationally unmanageable. To reduce the complexity, we propose to apply an idea that resorts to the forward
shooting grid method proposed by Hull and White [33] and then generalized by Barraquand and Pudet [34]. In detail,
we propose to discretize the asset dynamics by generating at each time step of the discrete distribution a grid of buckets
(hereafter also referred to as ‘‘representative’’ asset values) whose elements cover the range of possible asset values at
each time slice. Once the grid has been constructed, for each bucket we identify four successors in a bivariate environment,
where both the fBm and the Bm may show an upward or a downward movement. Being the grid made up of representative
values, the successors might not appear among the discrete process values generated at the next time step. In such

2 When H = 1
the fBm reduces to a standard Brownian motion having independent increments.
2

2
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

cases, interpolation techniques are needed when solving backward to compute the initial value of the contingent claim.
The proposed discretization is supported by a numerical analysis confirming that the model computes highly accurate
prices for all the considered cases having, in addition, the appealing feature of being of immediate application to evaluate
American-style derivatives.
The paper proceeds as follows. In the next section, we describe the continuous-time mfBm model. In Section 3, we
present the discrete approximation and the backward recursion scheme for pricing European and American contingent
claims. Section 4 is concerned with numerical results and, finally, Section 5 concludes the paper.

2. The framework

We consider a continuous-time market model where trading takes place in the interval [0, T ], which is characterized
by two investment possibilities. The first one is represented by a money market account with evolution described by

dA(t) = rA(t)dt , A(0) = 1,

where A(t) denotes the value at time t of one unit of money available at time 0, and r is the instantaneous rate of return.
The second investment is represented by an asset having the following dynamics

dS(t) = µS(t)dt + σ S(t)dB̃H

t ,

t the increment of a mfBm process with Hurst parameter H, µ the

where S(t) denotes the asset price at time t, dB̃H
instantaneous rate of return, and σ the instantaneous volatility of the risky asset. The mfBm process B̃H
t is a stochastic
process defined as

t = W̃t + W̃t ,
B̃H H

with H ∈ (0, 1)
{ }
where W̃tH is a fBm with Hurst parameter H and W̃t an independent Bm. In particular, the fBm W̃tH t ≥0
is a continuous and centred Gaussian process with initial value equal to zero, mean Ẽ(W̃tH ) = 0, variance Ṽ (W̃tH ) = t 2H ,
and covariance
1(
Ẽ(W̃tH W̃sH ) = |t |2H + |s|2H − |t − s|2H .
)
2
Working under a risk-neutral measure, the asset price presents the dynamics

dS(t) = rS(t)dt + σ S(t)dWtH + σ S(t)dWt , (1)

and, as reported in Murwaningtyas et al. [23], the following solution may be obtained
( ) ( )
rt − 12 σ 2 t 2H +t +σ WtH +Wt
S(t) = S(0)e ,
t + W̃t . In particular, for ∆t ̸ = 0, we
r −µ r −µ
with WtH = σ
t + W̃tH and Wt = σ
may write

r ∆t − 21 σ 2 (t +∆t )2H −t 2H +∆t +σ WtH+∆t −WtH +(Wt +∆t −Wt )

{[ ] } [( ) ]
S(t + ∆t) = S(t)e . (2)

The main contribution of the present paper is to provide a discrete version of the presented continuous-time
framework. The basic idea is to create a discretization of the mfBm by combining a binomial approximation of the fBm that
is able to match its variance at each discrete observation epoch with a recombining binomial lattice that approximates the
Bm evolution. The discretization is suitable for evaluating both European and American-style contingent claims because
it is based on the bivariate binomial dynamics presented hereafter.

3. The discretization

As usual when developing discrete-time models, we start to divide the entire time horizon [0, T ] into n subintervals
of equal length ∆t = T /n, so that the discrete observation epochs may be identified by i∆t , i = 0, . . . , n. The first
step of the proposed discretization is to consider separately the two components of the mfBm, i.e., the fBm and the
independent Bm. Starting from the Bm, we propose to discretize this diffusion by a recombining lattice (depicted in Fig. 1
with n = 4 time steps) defined through the discrete version of the Bm independent increment in each discrete interval
[k∆t , (k + 1)∆t), k = 0, . . . , n − 1, i.e.,
{ √
∆t with probability 12
∆Wk∆t = √ .
− ∆t with probability 1
2

3
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

Fig. 1. The structure of the lattice approximating the Bm process.

On the interval [0, i∆t ], the Bm value is computed as

i−1
∑
Wi∆t = ∆Wk∆t , with W0 = 0, and i = 1, . . . , n,
k=0

and, clearly, its mean is E(Wi∆t ) = 0 and its variance is V (Wi∆t ) = i∆t.3
More problematic is the development of an efficient approximation of the fBm. We propose a binomial approach that
defines the discrete version of the fBm increment in the interval [k∆t , (k + 1)∆t), k = 0, . . . , n − 1, as follows
{ √
(k + 1)2H − k2H ∆t H with probability 12
∆WkH∆t = √ .
− (k + 1)2H − k2H ∆t H with probability 1
2

As a consequence, the fBm value on the interval [0, i∆t ] is computed as

i−1
∑
WiH∆t = ∆WkH∆t , with W0H = 0, and i = 1, . . . , n,
k=0

and, imposing independence among the fBm increments, by little algebra it may be easily proved that the mean of the
proposed discrete version of the fBm is E(WiH∆t ) = 0, while its variance is given by V (WiH∆t ) = (i∆t)2H , thus replicating
their continuous-time counterparts. In this way, we obtain a discrete-time process that is able to capture two main
characteristics of the fBm continuous-time process, even though it lacks to capture the fBm covariance structure detailed
in Section 2. Nonetheless, extensive numerical experiments presented in the next section evidence that the obtained fBm
approximation, used to create the discrete-time version of the mfBm detailed hereafter, allows us to obtain very accurate
results when valuing financial derivatives.
Unfortunately, if we apply straightforwardly the proposed binomial approach, we obtain the non-recombining shape
u
depicted in Fig. 2. Indeed, starting from W0H = 0, the fBm at time ∆t may assume value W∆Ht = ∆t H with probability 1/2
d
if an up step occurs (denoted by u), or W∆Ht = −∆t H with probability 1/2 if a down step takes place (denoted by d). After
u d
two time steps, both the fBm values W∆Ht and W∆Ht may show upward and downward movements with equal probability
1/2. Consequently, each node presents two successors at time 2∆t that do not recombine
√ each others. In other words,
u uu
starting from W∆Ht , the binomial dynamics provides the values W2H∆t = ∆t H + 22H − 1∆t H (in Fig. 2, it is associated
ud √ d
with node D), and W2H∆t = ∆t H − 22H − 1∆t H (in Fig. 2, it is associated with node C ). Similarly, starting from W∆Ht ,
du √
the binomial dynamics provides the values W2H∆t = −∆t H + 22H − 1∆t H (in Fig. 2, it is associated with node B), and

3 The mean and the variance are computed under the risk-neutral measure.

4
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

Fig. 2. The non-recombining shape of the fBm binomial approximation.

dd √
W2H∆t = −∆t H − 22H − 1∆t H (in Fig. 2, it is associated with node A). Each trajectory starting from time 0 and reaching
time 2∆t has occurrence probability 1/4. At time 3∆t, we have 8 different fBm values, given by:
uuu
√ √
W3H∆t = ∆t H + 22H − 1∆t H + 32H − 22H ∆t H ,

uud
√ √
W3H∆t = ∆t H + 22H − 1∆t H − 32H − 22H ∆t H ,

udu
√ √
W3H∆t = ∆t H − 22H − 1∆t H + 32H − 22H ∆t H ,

udd
√ √
W3H∆t = ∆t H − 22H − 1∆t H − 32H − 22H ∆t H ,

duu
√ √
W3H∆t = −∆t H + 22H − 1∆t H + 32H − 22H ∆t H ,

dud
√ √
W3H∆t = −∆t H + 22H − 1∆t H − 32H − 22H ∆t H ,

ddu
√ √
W3H∆t = −∆t H − 22H − 1∆t H + 32H − 22H ∆t H ,

ddd
√ √
W3H∆t = −∆t H − 22H − 1∆t H − 32H − 22H ∆t H .
Proceeding in this way, at the generic ith time step, we have 2i nodes, i.e, 2i different trajectories to capture the fBm
dynamics. Hence, as Fig. 2 clearly shows, this aspect makes unmanageable to track every possible values arising from the
construction of a binomial lattice for the fBm approximation.
In addition, we have to remark that, at each time step, the nodes of the non-recombining structure of the fBm binomial
approximation (depicted in Fig. 2) must be combined with the nodes of the recombining lattice approximating the Bm
(see Fig. 1) to generate a bivariate tree that might approximate the mfBm process driving the underlying asset dynamics
in (1). The strong path-dependent structure highlighted in the fBm discretization does not simplify matters under this
perspective, because it reflects on the asset value (2). As a consequence, the main scope is to propose an algorithm able to
manage and reduce the exponential complexity evidenced above. This is obtained by using a methodology that generates
at each time step of the discrete distribution a grid of buckets for the asset price whose elements cover the range of
possible asset values at each time slice. In correspondence of such a limited number of buckets, we compute the contingent
claim value. The essence of such a procedure is to generate the grid values at each time step i∆t in order to form a vector
−→
S(i) of representative asset values. We present the procedure step by step.
5
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

Step 1. At each time i∆t, the maximum and the minimum asset values are computed coherently with (2) as
{[ ] } (√ √ )
r ∆t − 21 σ 2 (i∆t )2H −((i−1)∆t )2H +∆t +σ i2H −(i−1)2H ∆t H + ∆t
Smax (i) = Smax (i − 1)e ,
with Smax (0) = S(0), and
{[ ] } (√ √ )
r ∆t − 12 σ 2 (i∆t )2H −((i−1)∆t )2H +∆t −σ i2H −(i−1)2H ∆t H + ∆t
Smin (i) = Smin (i − 1)e ,
with Smin (0) = S(0).

Step 2. Span the interval [Smin (i), Smax (i)] through a bucketing procedure that generates a limited number of buckets of
−→
asset prices, i.e., the elements of the vector S(i) are:

−→
• the first component in S(i) is the maximum asset value Smax (i), and it is denoted by S(i; 0) = Smax (i);
• the other representative asset prices are computed as
S(i; l) = Smax (i)M (σ , ∆t )l ,

where
1 − λσ ∆t
M (σ , ∆t ) = , (3)
1 + λσ ∆t
and λ is a positive parameter that preserves accuracy in the bucketing procedure. The exponent l assumes all
the integer values in the interval [1, L(i) − 1], with L(i) the smallest integer assuring that Smax (i)M (σ , ∆t )L(i) ≤
Smin (i), i.e.,
⌈ ⌉
ln(Smin (i)) − ln(Smax (i))
L(i) = . (4)
ln (M (σ , ∆t ))
The choice of the form reported in (3) for the bucketing function M (σ , ∆t ) has been suggested by the
construction detailed in Step 1, according to which the ratio between Smin (i) and Smax (i) is the following:
(√ √ )
i−1 −σ (k+1)2H −k2H ∆t H + ∆t
Smin (i) ∏ e
= (√ √ ) ≈
Smax (i) σ (k+1)2H −k2H ∆t H + ∆t
ek=0
(√ √ )
i−1 1 − σ (k + 1)2H − k2H ∆t H + ∆t
∏
≈ (√ √ ).
k=0 1+σ (k + 1)2H − k2H ∆t H + ∆t

We observe that, for 0 < λ ≤ 1 and k = 0, . . . , i − 1, since

(√ √ )
λσ ∆t < σ (k + 1)2H − k2H ∆t H + ∆t ,

when ∆t is sufficiently small (∆t ≤ 1), we have

(√ √ )
1−σ (k + 1)2H − k2H ∆t H + ∆t 1 − λσ ∆t
√ ) < .
1 + λσ ∆t
(√
1+σ (k + 1)2H − k2H ∆t H ∆t

Now, let ηk be a positive real number such that

(√ √ )
1−σ (k + 1)2H − k2H ∆t H + ∆t [
1 − λσ ∆t
]ηk
√ ) = ;
1 + λσ ∆t
(√
1+σ (k + 1)2H − k2H ∆t H ∆t

hence,
i−1
∑
ηk
1 − λσ ∆t
[ ]
Smin (i)
= k=0 ,
Smax (i) 1 + λσ ∆t
⌈∑ ⌉
i−1
and the integer L(i) in (4) coincides with k=0 ηk ;
−→
• the minimum asset value Smin (i) is the last component in S(i) and it is denoted by S(i; L(i)) = Smin (i).
6
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

Fig. 3. The scenarios for asset bucket S(i; l).

−→
In this way, the vector S(i) is made up of L(i) + 1 components spanning the interval [Smin (i), Smax (i)]. The choice
of the bucketing function guarantees that the difference between two consecutive buckets is of order O(∆t). This
aspect also allows to obtain a grid spanning the interval [Smin (i), Smax (i)] at each time slice and, coupled with the
choice of the parameter λ, to produce accurate results as showed in the next section. It is worth mentioning that the
algorithm is suitable to accommodate different forms of the bucketing function useful to the scope of any evaluation
problem without additional efforts.

Once the bucketing procedure has been used to generate a vector of ‘‘representative’’ asset prices at each time slice,
the next step of the algorithm is to define the backward recursion scheme needed to compute the contingent claim
value at inception. In detail, for each asset bucket S(i; l), with i = 0, . . . , n, and l = 0, . . . , L(i), we generate a bivariate
model and identify four possible scenarios arising by combining the two successors identified in the binomial structures
approximating the fBm and the Bm appearing in the mfBm. As depicted in Fig. 3, they are labelled with an ordered pair
where the first element indicates whether the fBm process shows an upward movement, denoted by ‘‘u’’, or a downward
movement, denoted by ‘‘d’’, and the second element of the pair indicates the Bm movement. For example, the pair uu
is used to denote the branch associated with an upward movement in both the processes, and the pair ud denotes the
branch associated with an upward movement in the fBm discrete dynamics and a downward movement in the discrete Bm
dynamics. The occurrence probability of each scenario is computed by the product of the marginal probabilities assigned to
each movement in the discrete fBm and Bm processes, which are equal to 21 , according to the hypothesized independence
between the fBm and the Bm appearing in the mfBm. To summarize, the four scenarios, namely S uu (i; l), S ud (i; l), S du (i; l),
and S dd (i; l), are characterized by the following asset values having each one occurrence probability equal to 1/4:
⎧ {[ ] } (√ √ )
uu r ∆t − 12 σ 2 ((i+1)∆t )2H −(i∆t )2H +∆t +σ (i+1)2H −i2H ∆t H + ∆t
S (i; l) = S(i ; l)e
⎪
⎪
⎪
⎪
⎪
} (√
⎪
√ )
⎪
⎪ {[ ]
r ∆t − 12 σ 2 ((i+1)∆t )2H −(i∆t )2H +∆t +σ (i+1)2H −i2H ∆t H − ∆t
⎪
⎨ S ud (i; l) = S(i; l)e
⎪
⎪
} ( √ .
{[ ] √ )
r ∆t − 12 σ 2 ((i+1)∆t )2H −(i∆t )2H +∆t +σ − (i+1)2H −i2H ∆t H + ∆t
⎪
du
⎪
⎪
⎪
⎪ S (i ; l) = S(i; l)e
⎪
⎪
} ( √
⎪
⎪ {[ ] √ )
r ∆t − 12 σ 2 ((i+1)∆t )2H −(i∆t )2H +∆t +σ − (i+1)2H −i2H ∆t H − ∆t
⎪
⎪
⎩ dd
S (i; l) = S(i; l)e
At this point, we have to observe that the asset value generated for each scenario could not be components of the
−−−−→
vector S(i + 1) since it contains representative asset values. To overcome this obstacle, we resort to linear interpolation
techniques, as detailed hereafter, even though the algorithm is suitable to accommodate also different and more complex
interpolation techniques like cubic spline interpolation. For instance, suppose to consider the asset value following
scenario uu, i.e., S uu (i; l). We detect the greatest bucket smaller than S uu (i; l), S(i + 1; luu
1 ), and the smallest bucket greater
−−−−→
than or equal to S uu (i; l), S(i + 1; luu
2 ), in S(i + 1), so that

S uu (i; l) = (1 − ωuu )S(i + 1; luu

1 ) + ω S(i + 1; l2 ),
uu uu
with ωuu ∈ [0, 1].

Similarly, we operate upon S ud (i; l), S du (i; l), and S dd (i; l) to identify the quantities ωud , ωdu , and ωdd , respectively, all
belonging to the interval [0, 1].4
The contingent claim price at inception is computed through a backward recursion scheme. For a detailed description,
let us consider a European call option and denote by c(i; l) its value in correspondence with the asset value S(i; l) being

4 It is worth noting that others than the linear interpolation scheme proposed above, like cubic spline interpolation function, may be easily
included in the proposed framework without any additional effort. However, numerical experiments do not evidence a significant effect of different
interpolation techniques on the precision of the pricing algorithm.

7
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

−→
the lth component of the vector S(i). The option payoff at maturity T = n∆t, for each value S(n; l), l = 0, . . . , L(n), is
known and given by c(n; l) = max(S(n; l) − K , 0).
Starting from the option payoff c(n; l) and applying the backward procedure, we compute the option values at the
(n − 1)th step, then at the (n − 2)th step, and so on up to inception, applying the following formula that here is detailed
for the generic state (i; l):
{ [
−r ∆t 1 ]
c(i; l) = e 2 ) + (1 − ω )c(i + 1; l1 ) +
ωuu c(i + 1; luu uu uu
4
1[ ]
ωud c(i + 1; lud
2 ) + (1 − ω ud
)c(i + 1; lud
1 ) +
4
1 du
[ ]
ω c(i + 1; ldu2 ) + (1 − ω )c(i + 1; l1 ) +
du du
4
1 [ dd ]}
ω c(i + 1; l2 ) + (1 − ω )c(i + 1; l1 ) ,
dd dd dd
4
where the quantities ωuu , ωud , ωdu , and ωdd , are indicated above.
In presence of an American put option, starting from the payoff at maturity P(n; l) = max(K − S(n; l), 0), the option
value P(i; l) in the generic state (i; l) is computed by
{ { [
−r ∆t 1 uu ]
P(i; l) = max e ω c(i + 1; luu
2 ) + (1 − ω )c(i + 1; l1 ) +
uu uu
4
1[ ]
ωud c(i + 1; lud
2 ) + (1 − ω ud
)c(i + 1; lud
1 ) +
4
1 du
[ ]
ω c(i + 1; ldu2 ) + (1 − ω )c(i + 1; l1 ) +
du du
4
1 [ dd ]} }
ω c(i + 1; l2 ) + (1 − ω )c(i + 1; l1 ) , K − S(i; l) .
dd dd dd
4

4. Numerical results

We test the pricing model presented in Section 2 by computing the prices of both European and American-style options
when the underlying asset value depends upon a mfBm. Numerical results are generated by choosing the parameter λ = 12
in the bucketing function (3) to control the fineness of the grid, even though different values for λ may also be chosen.
Clearly, the smaller is λ, the greater is the value assumed by the bucketing function (3) and, consequently, the greater is
the number of representative values generated at each time slice. This aspect has to be taken properly into account to
keep the pricing algorithm efficient from a computational point of view.
Initially, to assess the goodness of the proposed approximation model, in Table 1, we provide comparison between
European call option prices under mfBm and the ones obtained using the explicit formula provided by Murwaningtyas
et al. [23] and reported in the last column (Explicit). We report the option prices computed by the approximation method
when varying the number of time steps n for different levels of the strike price K and time to maturity T , whereas the
initial volatility has value σ = 0.2, the initial asset price is S0 = 100, the risk-free interest rate is r = 0.05, and the Hurst
parameter is H = 0.7 in order to consider the long-range dependence case.
As evidenced in Table 1, the approximation method computes call option prices really close to the values supplied
by the formula of Murwaningtyas et al. [23]. To give evidence of the computational efficiency of the proposed algorithm
when choosing the bucketing parameters λ = 12 and n = 200 in the test case presented in Table 1, we highlight that
option prices are computed in 10 s, approximately, on a laptop pc equipped with an Intel Core i7 with 2.20 GHz and 16 GB
of RAM. Clearly, increasing the value of n, the computational time increases but it is worth mentioning that already with
n = 200 we obtain an accurate call option price approximation since the relative error is really small being of order 10−4
in the worst case. Hence, the choice λ = 21 permits to obtain a good balance between algorithm accuracy and efficiency.
Investigating further the impact of the value assumed by the parameter λ on the algorithm efficiency and accuracy,
in Table 2 we consider, for instance, the at-the money options of the test cases presented in Table 1, i.e., K = 100,
and consider descendent values for λ, smaller than 21 . For each λ-value, we report the value assumed by the bucketing
function (3), M (σ , ∆t ), the number of representative values generated at the nth time step, L(n), when fixing n = 1000,
the European call option price computed by the proposed algorithm with the same number of time step (Price), and the
computational time (Time) in seconds needed to compute the option price.
From Table 2, it is evident that decreasing the λ-value, the value assumed by the bucketing function (3) increases and,
consequently, the number of representative values increases. Furthermore, when decreasing λ, the option price is slightly
more accurate with respect to the value supplied by the explicit formula of Murwaningtyas et al. [23] reported in Table 1,
but at the cost of an increasing computational time that will be greater and greater if we augment the number of time
steps. Clearly, this aspect is more evident when increasing T because, being n = 1000 fixed, the greater ∆t = Tn , the
smaller the bucketing function value and, consequently, the number of representative values (i.e., the grid of buckets is
8
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

Table 1
European call option prices under the mfBm model: the long-range dependence case.
σ = 0.2, S0 = 100, r = 0.05, H = 0.7
T K n = 200 n = 400 n = 600 n = 800 n = 1000 Explicit
0.0833 90 10.515216 10.515339 10.515377 10.515395 10.515408 10.515457
95 6.156218 6.156111 6.156069 6.156045 6.156035 6.155987
100 2.902972 2.902625 2.902500 2.902434 2.902400 2.902256
105 1.054867 1.054731 1.054678 1.054649 1.054636 1.054578
110 0.289437 0.289581 0.289625 0.289646 0.289661 0.289716
0.25 90 12.221846 12.221609 12.221557 12.221537 12.221527 12.221443
95 8.551252 8.550669 8.550512 8.550441 8.550402 8.550196
100 5.618423 5.617664 5.617455 5.617359 5.617305 5.617021
105 3.459473 3.458821 3.458646 3.458568 3.458524 3.458284
110 1.996874 1.996529 1.996450 1.996418 1.996402 1.996286
0.5 90 14.747252 14.746790 14.746540 14.746376 14.746366 14.746087
95 11.454926 11.454220 11.453869 11.453647 11.453619 11.453225
100 8.678861 8.678039 8.677640 8.677389 8.677354 8.676887
105 6.417900 6.417141 6.416758 6.416513 6.416485 6.416039
110 4.636874 4.636317 4.636006 4.635799 4.635790 4.635443
1 90 19.156975 19.155135 19.154733 19.154682 19.154489 19.153895
95 16.198799 16.136663 16.196185 16.196115 16.195890 16.195184
100 13.584279 13.581983 13.581469 13.581395 13.581153 13.580377
105 11.303942 11.301633 11.301122 11.301059 11.300815 11.300051
110 9.339272 9.337099 9.336632 9.336591 9.336359 9.335633
5 90 42.740525 42.734648 42.733660 42.733886 42.733447 42.731828
95 40.717404 40.711139 40.710008 40.710241 40.709758 40.707988
100 38.801578 38.794813 38.793617 38.793843 38.793325 38.791435
105 36.987240 36.980244 36.978969 36.979205 36.978661 36.976669
110 35.269375 35.262027 35.260706 35.260959 35.260391 35.258324

Table 2
The impact of the λ-value on the algorithm efficiency when n = 1000.
σ = 0.2, S0 = 100, K = 100, r = 0.05, H = 0.7, n = 1000
T λ M (σ , ∆t ) L(n) Price Time
0.0833 0.5 0.999983 350571 2.902400 558.18
0.4 0.999987 438214 2.902397 1255.22
0.3 0.99999 584285 2.902395 2185.72
0.25 0.5 0.99995 224014 5.617305 348.94
0.4 0.99996 276267 5.617275 785.94
0.3 0.99997 368356 5.617265 1370.48
0.5 0.5 0.9999 166219 8.677354 261.38
0.4 0.99992 207773 8.677285 588.16
0.3 0.99994 277031 8.677241 1023.18
1 0.5 0.9998 125607 13.581153 197.14
0.4 0.99984 157009 13.580971 446.46
0.3 0.99988 209345 13.580943 776.03
5 0.5 0.999 66764 38.793325 98.62
0.4 0.9992 83455 38.792438 222.36
0.3 0.9994 111273 38.791435 387.40

less fine), and the option value computation is faster. All these aspects reflect on the option valuation that is less precise
increasing T , but it gains accuracy by decreasing λ (once T has been fixed) at the cost of an increment of the computational
time. To sum up, it is fundamental to choose the value of λ in order to have a good balance between option price accuracy
and algorithm computational efficiency.
To provide an application to American-style option, we consider the put version of the option contract already analysed
in Table 1. In Table 3, for each strike price value, we report two rows: in the upper one, we provide a comparison between
European put option prices and the ones obtained using the explicit formula provided by Murwaningtyas et al. [23] and
reported in the last column (Explicit); in the lower one, we show the prices of the corresponding American put contracts.
As evidenced in Table 3, the approximation method computes European put option prices really close to the values
supplied by the formula of Murwaningtyas et al. [23] and, when considering the American-style contracts, it computes
larger values than the European ones, as expected.
To present numerical experiments also in the short-range dependence case, i.e., H < 21 , in Table 4, we provide an
additional comparison between European call option prices computed through the proposed model and the ones obtained
9
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

Table 3
European (first row) and American (second row) put option prices under the mfBm model: the long-range dependence case.
σ = 0.2, S0 = 100, r = 0.05, H = 0.7
T K n = 200 n = 400 n = 600 n = 800 n = 1000 Explicit
0.0833 90 0.141146 0.141268 0.141307 0.141325 0.141337 0.141387
0.141951 0.142069 0.142105 0.142122 0.142134
95 0.761366 0.761259 0.761217 0.761193 0.761183 0.761134
0.767278 0.767149 0.767097 0.767049 0.767056
100 2.487338 2.486992 2.486866 2.486800 2.486767 2.486622
2.514358 2.513962 2.513818 2.513742 2.513702
105 5.618452 5.618316 5.618263 5.618233 5.618221 5.618162
5.705522 5.705332 5.705258 5.705217 5.705198
110 9.832240 9.832384 9.832428 9.832449 9.832464 9.832519
10.052595 10.052705 10.052739 10.052754 10.052765
0.25 90 1.103854 1.103614 1.103561 1.103540 1.103530 1.103445
1.119170 1.118927 1.118870 1.118846 1.118834
95 2.371148 2.370563 2.370405 2.370334 2.370294 2.370087
2.411259 2.410673 2.410510 2.410436 2.410394
100 4.376208 4.375447 4.375237 4.375140 4.375082 4.374801
4.466878 4.466135 4.465924 4.465826 4.465771
105 7.155148 7.154492 7.154317 7.154238 7.154195 7.153954
7.337397 7.336805 7.336643 7.336569 7.336527
110 10.630437 10.630089 10.630010 10.629978 10.629968 10.629844
10.964497 10.964289 10.964246 10.964230 10.964222
0.5 90 2.525171 2.524695 2.524441 2.524275 2.524263 2.523979
2.584683 2.584279 2.584039 2.583877 2.583869
95 4.109395 4.108675 4.108319 4.108095 4.108066 4.107667
4.221530 4.220927 4.220600 4.220386 4.220364
100 6.209879 6.209044 6.208640 6.208387 6.208351 6.207878
6.406147 6.405501 6.405147 6.404915 6.404892
105 8.825467 8.824695 8.824307 8.824060 8.824032 8.823579
9.148682 9.148193 9.147889 9.147681 9.147673
110 11.920992 11.920421 11.920106 11.919896 11.919886 11.919534
12.427224 12.427059 12.426872 12.426727 12.426748
1 90 4.767762 4.765853 4.765428 4.765365 4.765165 4.764543
4.957324 4.955710 4.955372 4.955353 4.955174
95 6.565733 6.563528 6.563026 6.562945 6.562714 6.561980
6.856509 6.854720 6.854343 6.854324 6.854123
100 8.707361 8.704995 8.704458 8.704372 8.704123 8.703319
9.136405 9.134611 9.134248 9.134246 9.134043
105 11.183171 11.180793 11.180258 11.180183 11.179933 11.179141
11.795691 11.794078 11.793778 11.793814 11.793628
110 13.974648 13.972405 13.971915 13.971862 13.971623 13.970869
14.824928 14.823686 14.823502 14.823592 14.823442
5 90 12.840834 12.830839 12.828478 12.828017 12.827166 12.823899
14.748913 14.741229 14.739623 14.739651 14.739017
95 14.711717 14.701334 14.698890 14.698376 14.697481 14.694063
17.019720 17.012079 17.010607 17.010497 17.009976
100 16.689895 16.679012 16.676443 16.675982 16.675051 16.671513
19.449970 19.442404 19.440922 19.441115 19.440504
105 18.769561 18.758447 18.755798 18.755348 18.754392 18.750751
22.037294 22.030076 22.028714 22.029013 22.028438
110 20.945700 20.934234 20.931540 20.931106 20.930126 20.926410
24.779981 24.773164 24.771977 24.772395 24.771867

using the explicit formula of Murwaningtyas et al. [23] (Explicit). To do this, we fix H = 0.2 and leave unchanged the
other parameter values used in Table 1. Again, we report the option prices computed by the approximation method when
varying the number of time steps n for different levels of the strike price K and time to maturity T .
Table 4 evidences that the approximation method computes accurate call option prices with respect to the values
supplied by the formula of Murwaningtyas et al. [23], also in the short-range dependence case, even though when
increasing the time to maturity T , the algorithm seems to require a greater number of time steps to reach the full
convergence to the benchmark.
To provide an application to American-style option in the short-range dependence case, we consider the put version of
the option contract analysed in Table 4 and, in Table 5, we report two rows for each strike price value: in the upper one,
we provide a comparison between European put option prices and the ones obtained using the explicit formula provided
by Murwaningtyas et al. [23] (Explicit); in the lower one, we show the prices of the corresponding American put contracts.
As evidenced in Table 5, the approximation method computes again European put option prices really close to the
values supplied by the formula obtained by Murwaningtyas et al. [23] also in the short-range dependence case, even
10
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

Table 4
European call option prices under the mfBm model: the short-range dependence case.
σ = 0.2, S0 = 100, r = 0.05, H = 0.2
T K n = 200 n = 400 n = 600 n = 800 n = 1000 Explicit
0.0833 90 11.871207 11.870231 11.869856 11.869652 11.869524 11.868843
95 8.362425 8.360075 8.359184 8.358703 8.358400 8.356822
100 5.574174 5.571333 5.570259 5.569678 5.569314 5.567416
105 3.512264 3.510116 3.509300 3.508859 3.508582 3.507131
110 2.093049 2.092318 2.092034 2.091879 2.091782 2.091267
0.25 90 13.780067 13.778323 13.777692 13.777358 13.777148 13.776023
95 10.533845 10.531216 10.530260 10.529753 10.529434 10.527737
100 7.834301 7.831376 7.830312 7.829748 7.829392 7.827504
105 5.672327 5.669780 5.668858 5.668369 5.668062 5.666408
110 4.002410 4.000757 4.000167 3.999857 3.999661 3.998609
0.5 90 15.868979 15.867117 15.866342 15.865895 15.865692 15.864413
95 12.771166 12.768704 12.767693 12.767114 12.766841 12.765157
100 10.115173 10.112501 10.111402 10.110774 10.110477 10.108638
105 7.890232 7.887766 7.886743 7.886154 7.885884 7.884177
110 6.066663 6.064764 6.063953 6.063478 6.063279 6.061942
1 90 19.159556 19.156729 19.155905 19.155636 19.155299 19.153895
95 16.201968 16.198618 16.197623 16.197285 16.196884 16.195184
100 13.587714 13.584092 13.583024 13.582657 13.582226 13.580377
105 11.307259 11.303680 11.302630 11.302282 11.301855 11.300051
110 9.342178 9.338879 9.337941 9.337655 9.337263 9.335639
5 90 36.344541 36.333265 36.330782 36.330568 36.329653 36.325978
95 33.850174 33.837665 33.834908 33.834629 33.833614 33.829505
100 31.503691 31.490127 31.487106 31.486784 31.485672 31.481185
105 29.301340 29.286851 29.283625 29.283279 29.282092 29.277274
110 27.238341 27.223139 27.219854 27.219396 27.218141 27.213066

though when increasing the time to maturity T , the algorithm seems to require a greater number of time steps to
reach the full convergence to the benchmark, as already evidenced before. Finally, we remark that when considering
the American-style contracts, the algorithm provides larger values than the European ones, as expected.

5. Conclusions

We have proposed a lattice-based model to approximate the dynamics of an asset with diffusion part represented by
a mfBm that is given by the sum of a fBm and an independent Bm. The Bm is discretized through a recombining binomial
lattice. The fBm is approximated by a discrete-time model based on a binomial structure that ensures the matching of
the fBm variance. To capture the asset dynamics, the discrete process generated for the fBm has to be combined with the
discrete process approximating the Bm. Due to the strong path-dependent structure characterizing the fBm process that
makes the problem unmanageable, we propose to reduce the computational complexity by discretizing the asset dynamics
through the generation of a grid of buckets covering the range of possible asset values at each time slice. Once the grid
has been constructed, for each bucket we identify four successors in a bivariate environment, where both the fBm and
the Bm may show an upward or a downward movement. Being the grid made up of representative values, the successors
might not appear among the discrete process values at the next time step. In such a case, interpolation techniques are
needed when solving backward to compute the initial value of the contingent claim. Numerical experiments reported
for both European and American options confirm that the proposed algorithm is efficient and computes accurate prices
in comparison to the benchmark values. Future works will aim, at first, at developing a discrete-time model that is
able to capture not only the mean and the variance of the fBm but also its covariance structure, despite numerical
experiments confirm that capturing only such two main characteristics allows us to obtain accurate results with respect
to the benchmark values. A second aspect we would like to investigate is relative to the valuation under mfBm of more
complex derivatives, like exotic contingent claims, and of the premiums of security linked insurance products whose
values is strictly dependent upon the asset market evolutions.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have
appeared to influence the work reported in this paper.

Data availability

No data was used for the research described in the article.

11
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

Table 5
European (first row) and American (second row) put option prices under the mfBm model: the short-range dependence case.
σ = 0.2, S0 = 100, r = 0.05, H = 0.2
T K n = 200 n = 400 n = 600 n = 800 n = 1000 Explicit
0.0833 90 1.497177 1.496184 1.495803 1.495595 1.495465 1.495772
1.508730 1.507722 1.507335 1.507124 1.506992
95 2.967614 2.965246 2.964349 2.963864 2.963559 2.961970
2.993925 2.991504 2.990587 2.990091 2.989779
100 5.158580 5.155723 5.154642 5.154058 5.153691 5.151782
5.210573 5.207607 5.206485 5.205879 5.205499
105 8.075889 8.073724 8.072902 8.072457 8.072178 8.070716
8.167137 8.164804 8.163920 8.163441 8.163141
110 11.635892 11.635145 11.634855 11.634695 11.634597 11.634070
11.780762 11.779802 11.779433 11.779231 11.779106
0.25 90 2.662170 2.660384 2.659737 2.659395 2.659179 2.658025
2.718118 2.716362 2.715723 2.715384 2.715170
95 4.353837 4.351166 4.350194 4.349678 4.349353 4.347628
4.456197 4.453517 4.452535 4.452012 4.451682
100 6.592182 6.589214 6.588135 6.587562 6.587201 6.585284
6.764097 6.761059 6.759943 6.759348 6.758972
105 9.368097 9.365508 9.364570 9.364073 9.363760 9.362077
9.636429 9.633694 9.632688 9.632151 9.631812
110 12.636069 12.634374 12.633768 12.633449 12.633248 12.632167
13.029130 13.027217 13.026512 13.026135 13.025897
0.5 90 3.647056 3.646116 3.644312 3.643849 3.643636 3.642305
3.790540 3.788746 3.787983 3.787538 3.787336
95 5.425792 5.423253 5.422212 5.421618 5.421335 5.419599
5.661076 5.658655 5.657642 5.657058 5.656778
100 7.646348 7.643599 7.642471 7.641827 7.641520 7.639629
8.009262 8.006576 8.005452 8.004805 8.004489
105 10.297958 10.295414 10.294361 10.293757 10.293477 10.291718
10.828776 10.826217 10.825134 10.824508 10.824203
110 13.350938 13.348961 13.348121 13.347631 13.347421 13.346033
14.091976 14.089887 14.088979 14.088447 14.088195
1 90 4.770563 4.767583 4.766701 4.766401 4.766045 4.764543
5.124136 5.121574 5.120822 5.120581 5.120259
95 6.569123 6.565618 6.564567 6.564198 6.563778 6.561980
7.098660 7.095609 7.094686 7.094368 7.093980
100 8.711016 8.707240 8.706114 8.705717 8.705266 8.703319
9.470391 9.467089 9.466077 9.465714 9.465289
105 11.186708 11.182974 11.181868 11.181489 11.181402 11.179141
12.234958 12.231684 12.230663 12.230292 12.229860
110 13.977773 13.974321 13.973326 13.973009 13.972598 13.970869
15.376626 15.373609 15.372658 15.372312 15.371901
5 90 6.439336 6.426791 6.423864 6.423420 6.422363 6.418048
8.787446 8.777796 8.775899 8.776109 8.775349
95 7.838973 7.825195 7.821993 7.821484 7.820329 7.815579
10.873491 10.863462 10.861479 10.861691 10.860887
100 9.386494 9.371661 9.368196 9.367644 9.366391 9.361263
13.229693 13.219901 13.217868 13.218085 13.217249
105 11.078147 11.062389 11.058718 11.058141 11.056814 11.051356
15.859619 15.849473 15.847454 15.847670 15.846821
110 12.909151 12.892681 12.888851 12.888263 12.886868 12.881153
18.760828 18.751348 18.749427 18.749631 18.748782

References

[1] Black F, Scholes M. The pricing of options and corporate liabilities. J Political Econ 1973;81:637–54.
[2] Wiggins JB. Option values under stochastic volatility: Theory and empirical estimates. J Financ Econ 1987;19(2):351–72.
[3] Hull J, White A. The pricing of options on assets with stochastic volatility. J Finance 1987;42(2):281–300.
[4] Scott LO. Option pricing when the variance changes randomly: Theory, estimation and an application. J. Financ Quant Anal 1987;22(4):419–38.
[5] Johnson H, Shanno D. Option pricing when volatility is changing. J Financ Quant Anal 1987;22(2):143–52.
[6] Chesney M, Scott L. Pricing European currency options: A comparison of the modified Black–Scholes model and a random variance model. J
Financ Quant Anal 1989;24(3):267–84.
[7] Stein E, Stein J. Stock price distribution with stochastic volatility: An analytic approach. Rev Financ Stud 1991;4(4):727–52.
[8] Heston SL. A closed-form solution for options with stochastic volatility with applications to bond and currency options. Rev Financ Stud
1993;6(2):327–43.
[9] Ball CA, Roma A. Stochastic volatility option pricing. J Financ Quant Anal 1994;29(4):589–607.
[10] Schobel R, Zhu J. Stochastic volatility with an Ornstein–Uhlenbeck process: An extension. Eur Finance Rev 1999;3(1):23–46.

12
M. Costabile, I. Massabó, E. Russo et al. Communications in Nonlinear Science and Numerical Simulation 118 (2023) 107042

[11] Merton RC. Option pricing when underlying stock returns are discontinuous. J Financ Econ 1976;3:125–44.
[12] Yuen FL, Yang H. Option pricing in a jump-diffusion model with regime-switching. Astin Bull 2009;39(2):515–39.
[13] Ramponi A. Fourier transform methods for regime-switching jump-diffusions and the pricing of forward starting options. Int J Theor Appl
Finance 2012;15(5).
[14] Costabile M, Leccadito A, Massabó I, Russo E. Option pricing under regime-switching jump-diffusion models. J Comput Appl Math
2014;256:152–67.
[15] Madan DB, Seneta E. The variance gamma model for share market returns. J Bus 1990;63(4):511–24.
[16] Shirzad SM, Dehghn M, Bastani AF. On the pricing of multi-asset options under jump–diffusion processes using mesh-free moving least squares
approximation. Commun Nonlinear Sci Numer Simul 2020;84.
[17] Bentes SR, Menezes R, Mendes DA. Long memory and volatility clustering: Is the empirical evidence consistent across stock markets? Phys A
2008;387:3826–30.
[18] Berg L, Lyhagen J. Short and long-run dependence in Swedish stock returns. Appl Financial Econ 1998;8:435–43.
[19] Podobnik B, Wang D, Horvatic D, Grosse I. Time-lag cross-correlations in collective phenomena. Europhys Lett 2010;90.
[20] Mandelbrot BB, Van Ness JW. Fractional Brownian motions, fractional noises and applications. SIAM Rev 1968;10:422–37.
[21] P. Cheridito. Mixed fractional Brownian motion. Bernoulli 2001;6:913–34.
[22] Bender C, Sottinen T, Valkeila E. Arbitrage with fractional Brownian motion? Theory Stoch Process 2007;13(29).
[23] Murwaningtyas CE, Kartiko SH, Gunardi, Suryawan HP. European option pricing by using a mixed fractional Brownian motion. J Phys Conf Ser
2018;1097.
[24] Hu Y, Oksendal B. Fractional white noise calculus and applications to finance, infinite dimensional analysis. Quantum Probab Relat Top
2003;6:1–32.
[25] Necula C. Option pricing in a fractional Brownian motion environment. Pure Math 2002;2:63–8.
[26] Sun L. Pricing currency options in the mixed fractional Brownian motion. Phys A 2013;392(16):3441–58.
[27] Xiao WL, Zhang WG, Zhang XL. Pricing model for equity warrants in a mixed fractional Brownian environment and its algorithm. Physica A
2012;391(24):6418–31.
[28] Ahmadian D, Ballestra LV. Pricing geometric Asian rainbow options under the mixed fractional Brownian motion. Phys A 2020;555.
[29] Ballestra V, Pacelli G, Radi D. A very efficient approach for pricing barrier options on an underlying described by the mixed fractional Brownian
motion. Chaos Solitons Fractals 2016;87:240–8.
[30] P. Cheridito. Arbitrage in fractional Brownian motion models. Finance Stoch 2003;7:533–53.
[31] T. Lindstrom. A weighted random walk approximation to fractional Brownian motion. Pure Math 2007;11.
[32] Konstantopoulos T, Sakhanenko AI. Convergence and convergence rate to fractional Brownian motion for weighted random sums. Sibirskie
Ehlektronnye Mat Izvestiya 2004;1:47–63.
[33] Hull J, White A. Efficient procedures for valuing European and American path-dependent options. J Deriv 1993;1(1):21–31.
[34] Barraquand J, Pudet T. Pricing of American path-dependent contingent claims. Math Finance 1996;6(1):17–51.

13