Toward an efficient hybrid method for pricing

barrier options on assets with stochastic volatility

arXiv:2202.07849v1 [q-fin.CP] 16 Feb 2022

Alexander Lipton∗, Artur Sepp†

February 17, 2022

Abstract
We combine the one-dimensional Monte Carlo simulation and the semi-
analytical one-dimensional heat potential method to design an efficient
technique for pricing barrier options on assets with correlated stochastic
volatility. Our approach to barrier options valuation utilizes two loops.
First we run the outer loop by generating volatility paths via the Monte
Carlo method. Second, we condition the price dynamics on a given volatil-
ity path and apply the method of heat potentials to solve the conditional
problem in closed-form in the inner loop. We illustrate the accuracy and
efficacy of our semi-analytical approach by comparing it with the two-
dimensional Monte Carlo simulation and a hybrid method, which com-
bines the finite-difference technique for the inner loop and the Monte Carlo
simulation for the outer loop. We apply our method for computation of
state probabilities (Green function), survival probabilities, and values of
call options with barriers. Our approach provides better accuracy and is
orders of magnitude faster than the existing methods. s a by-product of
our analysis, we generalize Willard’s (1997) conditioning formula for val-
uation of path-independent options to path-dependent options and derive
a novel expression for the joint probability density for the value of drifted
Brownian motion and its running minimum.
Keywords: barrier options; stochastic volatility; Heston model; heat
potentials; semi-analytical solution; Volterra equation; Willards’s formula;
2010 MSC: 91G20; 91G60; 91G80; 47G10; 47G40; 35Q79;

1 Introduction
By expressing prices of European calls and puts in terms of the price of the
underlying asset and its volatility, Black and Scholes (1973) and Merton (1973)
started the quantitative finance revolution. Their formula, which is known as
the Black-Scholes-Merton formula, is based on two assumptions: (a) the risky
∗ Abu Dhabi Investment Authority, 211 Corniche, PO Box 3600, Abu Dhabi, United Arab

Emirates
† Sygnum Bank, Uetlibergstrasse 134a, 8045 Zürich, Switzerland

1
price dynamics can be delta-hedged so that options can be valued using the
risk-neutral measure under which the underlying grows at the risk-neutral rate;
(b) price evolution is driven by a geometric Brownian motion of the form:

dSt
= rdt + σdWt , S0 = s, (1)
St
where Wt is a Brownian motion, r and σ are constant risk-neutral interest rate
and volatility of returns, respectively. Thus, at time T , the price ST has the
log-normal distribution:
1 2
ST = e(r− 2 σ )T +σWT S0 , (2)

Under these assumptions, it is very easy to derive the price of, say, a call option
with maturity T and strike K, with payoff of the form (ST − K)+ :

C BS (0, S0 , T, K; r, σ) = S0 N (d+ ) − e−rT KN (d− ) ,

±σ 2 T /2 (3)
d± = − ln(K/S0σ)+rT
√
T
,

where N(x) is the cdf for the standard (0, 1) normal random variable. Lipton
(2002) showed that in many situations it is more convenient to write C BS as a
Fourier integral:

C BS (0, S0 , T, K; r, σ) !
1
R∞ e(iχ+1/2)(ln(K/S0 )−rT )−(χ2 +1/4)σ2 T /2 (4)
= S0 1− 2π (χ2 +1/4) dχ .
−∞

However, it became clear in a few years that the Black-Scholes formula,

taken literally, is not as helpful as initially thought since it could not reproduce
option market prices by using constant volatility for pricing options with differ-
ent strikes and maturities. The volatility skew (or smile) effect (i.e., the need to
use different volatilities to reproduce market prices of European calls and puts
with different strikes and maturities) is observed in all markets. Therefore, it
is of great interest to practitioners and academics alike. Over the thirty years,
many models were developed to describe the smile effect.
Eventually, two approaches emerged - a reduced approach replacing the con-
stant volatility, central to the Black-Scholes approach, by the maturity- and
strike-dependent implied volatility, used to match the corresponding market
prices. Specifically, instead of a constant σ for all T, K in Eq. (3), a function
σ (T, K) is used. The corresponding call price has the form

C(0, S0 , T, K) = C BS (0, S0 , T, K; r, σ (T, K)). (5)

While useful in some situations, this approach is conceptually fairly limited,

since it does not explain why volatility is not constant. Hence, a structural
approach replacing the geometric Brownian motion with a different, but still
risk-neutral, driver for the underlying was quickly developed. For instance,

2
Derman and Kani (1994), Dupire (1994), and Rubinstein (1994) simultane-
ously and independently developed the local volatility model. Merton (1976),
Andersen and Andreasen (2000), Lewis (2001) and many others developed
jump-diffusion models. Stochastic volatility models were developed by Hull and
White (1987), Scott (1987), Wiggins (1987), Stein and Stein (1991), Heston
(1993), Lewis (2000), Bergomi (2015), and others. Various combinations of
the above were proposed, for example by Dupire (1996), Jex et al. (1999),
Hagan et al. (2002), Lipton (2002), which culminated in Lipton’s universal
volatility model; see Lipton (2002). The above-mentioned models are compared
and contrasted in Lipton (2002).
Lipton and McGhee (2002) explained that the actual worth of a structural
model is not in its ability to price vanilla options, which all structural models
worth their salt can do well, but in producing consistent prices for both vanillas
and first-generation exotics. Despite more than thirty years of strenuous efforts,
finding a proper theoretical framework and implementing it in practice remains
a significant challenge.
Pricing of exotic options in the presence of a smile is usually difficult and
seldom can be done analytically. Asymptotic methods developed by Hull and
White (1987), Hagan et al. (2002), Lipton (1997), Lipton (2001), among
others, proved to be very useful for solving the corresponding pricing problems.
Numerical methods, such as the Monte Carlo simulation (MCS) and the finite
difference method (FDM), are equally useful; see Glasserman (2004), Achdou
and Pironneau (2005). Adding new methods to the classical ones is definitely
worth the effort. One can reduce many problems we wish to solve to the initial-
boundary value problems (IBVPs) for one-dimensional parabolic partial differ-
ential equations (PDEs) with moving boundaries and (or) time-dependent coef-
ficients. Such problems appear naturally in various areas of science and technol-
ogy. Finding their semi-analytical solutions requires using sophisticated tools,
such as the method of heat potentials (MHP) and a complementary method of
generalized integral transforms (MGIT). Such methods were actively developed
by the Russian mathematical school in the 20th century; see Kartashov (2001)
and references therein.
In the context of mathematical finance, A. Lipton and his coauthors actively
utilized the MHP; see Lipton (2001), Lipton et al. (2019) Lipton and Kaushan-
sky (2020) and references therein. In addition, A. Itkin and his coauthors used
the MGIT to price barrier and American options in the semi-closed form; see,
e.g., Carr et al. (2020), Itkin and Muravey (2021), Carr et al. (2022), Itkin
and Muravey (2022). In principle, the MHP and MGIT can be generalized and
used for any linear differential operator with time-independent coefficients.
The MHP and MGIT boil down to solving linear Volterra equations of the
second kind and representing option prices as one-dimensional integrals. Itkin
et al. (2021) described the MHP and MGIT in the recent comprehensive book
and showed that they are much more efficient and provide better accuracy and
stability than the existing methods, such as the backward and forward FDM or
MCS.
This paper revisits the classical problem of pricing barrier options on assets

3
with stochastic volatility. We show that by using the concept of conditional
independence, we can reduce it to solving an initial-boundary value problem
with time-dependent coefficients and subsequent averaging over the space of
variance trajectories. Based on this observation, we develop an efficient method
that combines the MHP and MCS and provides a fast and accurate solution to
the problem at hand. Our method is very general and can handle all known
stochastic volatility models, as well as models with rough volatility.1
The paper is organized as follows. In Section 2, we introduce generic stochas-
tic volatility models and describe the splitting method, which allows one to study
the dynamics of the log-price Xt = ln (St /S0 ), as a conditionally-independent
one-dimensional processes. We specialize these equations for the Heston and
Stein-Stein models. We also present the exact Lewis-Lipton and conditional
Willard formulas for vanilla options, such as European calls and puts on assets
with stochastic volatility and compare the corresponding prices. In Section 3,
we introduce barrier options on assets with stochastic volatility, which are the
main object of our study. We derive a conditional valuation formula for such
options, which generalizes the Willard formula for vanilla options. In Section 4,
we describe a hybrid method for pricing barrier options, which relies on the con-
ditional independence decomposition. The method consists of the outer Monte
Carlo loop and the inner loop, which requires solving the advection-diffusion
problem for the drifted Brownian motion with time-dependent coefficients on a
semi-axis. We solve the latter problem via two complementary methods: the
FDM and the MHP. Results produced by both methods are in perfect agree-
ment. However, as expected, the second method is orders of magnitude faster
than the first one. In Section 4, we use the MHP to solve an old problem in
probability theory and show how to find the joint probability density for the
value of drifted Brownian motion and its running minimum via the MHP. We
draw our conclusions in Section 6.

2 Stochastic volatility models

2.1 Conditionally independent dynamics
We consider the joint evolution of the asset price, St , and its stochastic variance,
Vt , as follows:
dSt
√  p 
St = rdt + Vt ρdBt + 1 − ρ2 dWt , S0 = s,
(6)
dVt = Φ (Vt ) dt + Ψ (Vt ) dBt , V0 = v,
where Bt and Wt are independent Brownian motions. Given that Eqs (6) are
scale invariant with respect to St , we can write them in terms of Xt = ln (St /S0 )
and Vt :
√  p 
dXt = r − 21 Vt dt + Vt ρdBt + 1 − ρ2 dWt , X0 = 0,


(7)
dVt = Φ (Vt ) dt + Ψ (Vt ) dBt , V0 = v,
1 We intend to cover the latter topic in a separate publication.

4
Alternatively, we can study the joint evolution of the price St and its volatility
σt :  
dSt
p
= rdt + σ ρdB + 1 − ρ 2 dW , S0 = s,
St t t t
(8)
dσ t = φ (σ t ) dt + ψ (σ t ) dBt , σ 0 = σ.
Given that Eqs (8) are scale invariant with respect to St , we can write them in
terms of Xt = ln (St /S0 ) and σ t :
 p 
dXt = r − 21 σ 2t dt + σ t ρdBt + 1 − ρ2 dWt , X0 = 0,


(9)
dσ t = φ (σ t ) dt + ψ (σ t ) dBt , σ 0 = σ,
which is often more convenient. From now on, we shall concentrate of studying
the dynamics of Xt .
In the general case, we can write dBt in the form
dVt − Φ (Vt ) dt
dBt = , (10)
Ψ (Vt )
and obtain the following conditionally-independent dynamics for the log-price
Xt :
 √  √
Vt ρ Vt Φ (Vt ) ρ Vt dVt p p
dXt = r − − dt + + 1 − ρ2 Vt dWt , X0 = 0.
2 Ψ (Vt ) Ψ (Vt )
(11)
Similarly, we can write dBt in the form
dσ t − φ (σ t )
dBt = , (12)
ψ (σ t )
and get the following dynamics for Xt :
 
1 ρσφ (σ t ) ρσ t dσ t p
dXt = r − σ 2t − dt + + 1 − ρ2 σ t dWt , X0 = 0. (13)
2 ψ (σ t ) ψ (σ t )
Assuming that the variance or volatility paths are given, Eqs (11), (13) describe
drifted arithmetic Brownian motion with time-dependent drift and volatility.
For the well-known Heston model, Heston (1993), we have
√
Φ (Vt ) = κ (θ − Vt ) , Ψ (V
√t ) = ε Vt , (14)
dVt = κ (θ − Vt ) dt + ε Vt dBt ,
so that Eq. (11) has the form
   
ρκθ 1 ρκ ρ p p
dXt = r − − − Vt dt + dVt + 1 − ρ2 Vt dWt . (15)
ε 2 ε ε
Thus, Xt is the so-called drifted Brownian motion driven by the stochastic
differential equation of the form
dX
 t = µ (t) dt + ν (t)
dWt ,
µ (t) = r − ρκθ 1 ρκ
ε − 2 − ε Vt +
ρ dVt
ε dt , (16)
p √
ν (t) = 1 − ρ2 Vt .

5
 For the Stein-Stein model, Stein and Stein (1991), we have
 
φ (σ t ) = κ̂ θ̂ − σ t , ψ (σ t ) = ε̂,
  (17)
dσ t = κ̂ θ̂ − σ t dt + ε̂dBt ,

so that Eq. (13) becomes

  !
ρκ̂θ̂ 1 ρκ̂ ρ p
dXt = r − σt − − σ 2t dt + σ t dσ t + 1 − ρ2 σ t dWt . (18)
ε̂ 2 ε̂ ε̂

Traditionally, the Heston model is viewed as better describing the market

than the Stein-Stein model since the latter does allow for zero variance. In our
opinion, this is not a particularly important issue, outweighed by many advan-
tages of the Stein-Stein model, such as a very straightforward way of simulating
the evolution of the volatility path. The Heston model gained popularity due
to the simple fact that it is exactly solvable for vanilla options. The explicit
solution can be calculated via the original Heston (1993) formula. However, the
Lewis-Lipton formula is much more efficient; see Lewis (2000), Lipton (2001),
Lipton (2002), Lipton and Sepp (2008), and Schmeltze (2010).
In this paper, we consider the Heston model with constant coefficients to
follow a long-established tradition, even though it is not necessarily our preferred
model. We emphasize that our method is general and can handle any reasonable
stochastic volatility model.

2.2 Analytical valuation formula for vanilla options

The popularity of the Heston model stems from the fact that one can write
its solution in the closed-form; see Heston (1993). However, experience has
shown that using the original formula is difficult due to several technical draw-
backs. Therefore, for benchmarking purposes, here we present the Lewis-Lipton
formula, which is easy to implement and use in practice:
C H (0, S0 , K, T ; r, ρ, κ, θ, ε, V0 ) !
1
R∞e
(iχ+1/2)(ln(K/S0 )−rT )+α(T ,χ)−(χ2 +1/4)β(T ,χ)V0
= S0 1− 2π (χ2 +1/4) dχ ,
−∞
h  i
ψ − +ψ + exp(−ζT )
α (T, χ) = − κθ
ε2 ψ + T + 2 ln 2ζ , (19)
1−exp(−ζT )
β (T, χ) = ψ − +ψ + exp(−ζT ) ,

q ψ ± = ∓ (iρεχ + κ̂) + ζ,
ε2
ζ= ε2 (1 − ρ2 ) χ2 + 2iερκ̂χ + κ̂2 + 4 ,

where κ̂ = κ − ρε/2. Further details are given in Lewis (2000), Lipton (2001),
Lipton (2002), and Schmeltze (2010).2 It is clear that Eq. (19) is a general-
ization of Eq. (4).
2 There is a typo in Lipton (2002) - a minus sign in front of β. This typo is corrected in

Lipton (2018), Chapter 10.

6
2.3 Conditional valuation formula for vanilla options
Unfortunately, with very few exceptions, finding a closed-form solution for bar-
rier or other exotic options on assets with stochastic volatility is not possible,
even if such a solution exists for vanilla options. Hence, more general volatility
models for barrier options are as good (or bad) as the more traditional Heston
and Stein-Stein models, which enjoy closed-form solutions for vanilla options.
We express the log-return process as a linear combination of the two pro-
cesses:
    
ρκθ 1 ρκ ρ
Xt = Yt + r− t− − It + (Vt − V0 ) ≡ Yt + Mt , (20)
ε 2 ε ε

where
p Rt√
1 − ρ2 0 Vt dWt , Y0 = 0,
Yt =
Rt 0
  It = 0 Vt0 dt , I0 = 0, (21)
ρκθ
Mt = r − ε t − 2 − ε It + ρε (Vt − V0 ) ,
1 ρκ


M0 = 0.

Accordingly, conditionally on the filtration generated by the variance process

Vt , we can represent the solution to the price process given by Eq. (6) as follows:

St = eMt +Yt S0 , (22)

where Mt is interpreted as a time-deterministic cumulative drift, and Yt is a

martingale with a deterministic time-dependent quadratic variance.
It is clear that pricing for path-independent options, such as European calls
and puts, simplifies to the Black-Scholes-Merton formula, provided that either
a variance path { Vt | 0 ≤ t ≤ T } (or a volatility path { σ t | 0 ≤ t ≤ T }) is given.
To be concrete, consider a call option with maturity T and strike K. Eq. (6)
yields
   p
dSt 1 p p
= r − Vt dt + ρ Vt dBt + 1 − ρ2 Vt dWt , , S0 = s. (23)
St 2

Thus, √ 2√  1 2 
2
ST = erT − 2 (1−ρ )IT + 1−ρ IT η e− 2 ρ IT +ρJT S0 ,
1
(24)

where the non-dimensional random variables IT , JT , are given by

Z T Z T p
IT = Vt dt, JT = Vt dBt , (25)
0 0

and η is the standard (0, 1) normal variable. Accordingly for a particular tra-
jectory { Vt | 0 ≤ t ≤ T }, we obtain the following expression
r !
BS − 12 ρ2 IT +ρJT
p
2
IT
C=C 0, e S0 , T, K; r, 1 − ρ , (26)
T

7
where the values (IT , JT ) are assumed to be known. The unconditional price is
obtained by averaging over all possible (IT , JT ):

 C H (0, S0 , K, T ; r, ρ, κ, θ, ε, v0q
) 
R∞R∞ − 12 ρ2 IT +ρJT
p I
= 0 0
C BS 0, e S0 , T, K; r, 1 − ρ 2
T Ξ (IT , JT ) dJT dIT ,
(27)
where Ξ (IT , JT ) is the joint probability density function (pdf) for the pair
(IT , JT ); see Willard (1997) and Romano and Touzi (1997). Thus, Eq. (27)
splits the calculation of the call option price into two stages. First, the condi-
tional price is found analytically via the standard Black-Scholes formula. Second,
this conditional price is averaged according to the particular choice of the pro-
cess for the variance Vt . Of course, the first stage is trivial. The usefulness of
this formula depends on how easy (or difficult) it is to find the pdf for (I, J)
and complete the second stage.
Two approaches have been used in practice - the standard Monte-Carlo
method for calculating { Vt | 0 ≤ t ≤ T }, I, J, and a more advanced (but much
harder) method based on the augmentation principle described in Section (13.2)
Lipton (2001). Specifically, the augmented dynamic equation for Vt yields the
following system of degenerate PDEs for the triple (V, I, J):

dVt = Φ (Vt ) dt + Ψ (Vt ) dBt , V0 = v,

dIt =√Vt dt, I0 = 0, (28)
dJt = Vt dBt , J0 = 0.

The corresponding Green’s function G (t, V, I, J) is governed by the degenerate

Fokker-Planck equation of the form
√ 
Gt + (Φ (V ) G)V + V GI − 12 Ψ2 (V ) G V V − − 21 (V G)JJ = 0,


V Ψ (V ) G
VJ
G (0, V, I, J) = δ (V − v) δ (I) δ (J) .
(29)
In general, solving this equation is complicated; however, for the Heston model,
Lipton (2001) found a closed-form solution.
In Figure 1 we compare prices of European call options given by analytical
formula (19), and Monte Carlo formula (27). As expected, both formulas agree,
although the latter formula is much slower than the former.

3 Barrier options
3.1 Formulation
Our task is to price a barrier option written on an asset with stochastic volatility.
For brevity, we consider barrier options with the lower barrier B < S0 only.
Considering other possibilities, such as pricing options with the upper barrier
or popular double-no-touch options, is left to the reader. The corresponding

8
Figure 1: Implied volatilities of European call options. We obtain the corre-
sponding prices by using Eqs. (19) and (27). Here, and throughout the paper
we use the following parameters: S0 = 1, V0 = 0.25, T = 1.0, r = 0.03, κ = 1.0,
θ = 0.2, ρ = −0.3, ε = 0.4.

9
IBVP can be written in the form
Pt + r − 12 V PX + κ (θ − V ) PV + 12 V (PXX + 2ρPXV + PV V ) − rP = 0,

P (T, X, V ) = Π (X) ,
P (t, ξ, V ) = 0,
(30)
where ξ = ln (B/S0 ) < 0, Π(X) is the terminal payoff function. Typical exam-
ples include the no-touch, call and put payoffs defined by:

Π (XT ) = 1, Π (XT ) = S0 eXT − ek + , Π (XT ) = S0 ek − eXT + ,

(31)

where k = ln (K/S0 ). Alternatively, we can write the adjoint problem for

Green’s function G:
Gt + r − 21 V G X + (κ (θ − V ) G)V

− 21 ((V G)XX + 2ρ (V G)XV + (V G)V V ) + rG = 0,

(32)
G (0, X, V ) = δ (X) δ (V − V0 ) ,
G (t, ξ, V ) = 0,

Once the corresponding Green’s function is calculated, we can find P via simple
integration:
Z ∞Z ∞
P (0, 0, V0 ) = G (0, 0, V0 , T, XT , VT ) Π (XT ) dVT dXT . (33)
ξ 0

While such options can be priced via either FDM or MCS, both are notoriously
slow. Therefore, we want to design a much faster method, enjoying equal or
higher accuracy than the classical alternatives.
As far as analytical solutions are concerned, only one is known. It was
discovered by Lipton (1997), see also Lipton (2001), Lipton and McGhee
(2002), and Andreasen (2001). Lipton (1997) observed that in the special case
r = 0, ρ = 0, IBVPs (30), (32) are symmetric with respect to the transformation
X → −X. Hence, the classical method of images is applicable, and solutions to
these problems can be presented as a linear combination of solutions without
barriers. Of course, one can use this approach for options in the presence of an
upper barrier, as well as for double-barrier options.
Recently, there were several unsuccessful attempts to solve the pricing prob-
lem with r2 + ρ2 > 0. For example, De Gennaro Aquino and Bernard (2019)
presented a solution, relying on an explicit expression for the joint distribution
of the value of a Brownian motion with time-dependent drift and its maximum
and minimum; it was quickly shown by one of the present authors that their
calculation is erroneous; see De Gennaro Aquino and Bernard (2021).3 He and
Lin (2021) presented a “solution”, which relies on the unsubstantiated replace-
ment of the time-dependent drift by a constant. Their approach is so arbitrary
and frivolous that its detailed repudiation is not warranted.
3 Finding the joint distribution for a Brownian motion with time-dependent drift and its

maximum and minimum is challenging. We present its solution in Section 5.

10
3.2 Conditional valuation formula for barrier options
It is hard to extend interesting formula (27) for barrier options. However, it is
not impossible! Following Section13.3 in Lipton (2001), we express the value of a
path-dependent option as an integral in the functional space of price trajectories:
Z
P (S0 ) = e−rT F(ω)dD(ω) (34)
Ω

where F(ω) is a functional mapping of the space of trajectories into payoffs, and
D(ω) is the risk-neutral probability measure.
Further, by applying the augmentation principle, we introduce the functional
Λt to represent the path-dependent variable linked to the evolution of the spot
price St . We then consider evaluation of a derivatives security with the terminal
pay-off function f (S, Λ). Finally, we extend the joint dynamics in Eq. (6) with
the dynamics of augmented variable Λt = min0≤t0 ≤t St :
√  p 
St = rSt dt + Vt St ρdBt + 1 − ρ2 dWt , S0 = s,
dVt = Φ(Vt )dt + Ψ(Vt )dBt , V0 = v, (35)
dΛt = θ (Λt − St ) (dSt )− , Λ0 = S0 .

The payoff function f (S, Λ) is given by:

f (ST , ΛT ) = 1{ΛT >B} Π (ST ) (36)

The joint dynamic, given by Eqs (35), is Markovian. Accordingly, we can

reduce the general formula (34) to the form:
Z ∞ Z ∞ Z ∞
−rT
P (0, S0 ) = e f (S, Λ)G(T, V, S, Λ; 0, V0 , S0 , Λ0 )dΛdSdV.
V =0 S=0 Λ=0
(37)
Here G(T, V, S, Λ; 0, V0 , S0 , Λ0 ) is the risk-neutral probability density function
for the joint evolution of state variables in SDE (35). We emphasize that while
the payoff function does not depend on the variance V , the valuation problem
has three spatial variables, including variance, in addition to the time variable.
Below, we introduce a novel method to solve the valuation problem (37)
semi-analytically. Using Eq. (22), the density of the price St is log-normal
with time-dependent drift and variance conditional on the filtration generated
by stochastic variance Vt , FV . Therefore, we represent the valuation formula
(37) as follows:
Z ∞ Z ∞ 
−rT
P (0, S0 ) = e EV f (S, Λ)Γ ( T, S, Λ; 0, S0 , Λ0 | V (ω)) dΛdS , (38)
S=0 Λ=0

where Γ is the Gaussian density of St and Λt conditional of the variance path

V (ω), and the expectation is computed over all paths of variance process Vt .
The term in the square brackets is the value of a path-dependent option on Λ
and S, computed in closed-form or numerically.

11
 We apply the MHP to compute the inner integral for barrier options in the
closed-form. Thus, we have generalized Eq (27), valid for path-independent
options, to path-dependent options, and apply the new result to value barrier
options in the semi-closed form.

4 Mixed MHP-MCS approach to solving IBVPs

We consider the IBVP (30). In the spirit of Section 3.2, we split our algorithm
in two steps: the outer MCS loop, which generates a bunch of trajectories for
the stochastic variance vt , and the inner loop, which solves the one-dimensional
IBVP for Xt . We discuss two approaches for solving the latter problem: the
finite difference approach developed by Loeper and Pironneau (2009), and the
MHP approach inspired by Lipton et al. (2019), Lipton and Kaushansky
(2020). We start with the inner loop. The outer loop is relatively straight-
forward. It requires to average the inner price by using the equation for Vt .
This will provide solution of the form given by Eq. (2) in Lipton and McGhee
(2002). The simplest way of doing this averaging is via MCS, although in some
exceptional cases other approaches can be envisaged.

4.1 One-dimensional IBVP

Eq. (15), conditional on the variance path, can be written as an SDE of the
form
dXt = µ (t) dt + ν (t) dWt , Xt ≥ ξ, (39)
where Xt the drifted Brownian motion with µ, ν given by Eqs (16).
While studying Xt on the entire axis (−∞, ∞) is almost trivial, dealing with
the same process in a semi-bounded domain (ξ, ∞) might be quite hard. This
paper shows how to do it by using the FDM and the MHP.
We can always rescale time t by introducing υ, such that dυ = ν (t) dt. As
a result, the governing SDE becomes

dXυ = λ (υ) dυ + dWυ , (40)

Rt
where λ (υ) = µ (t) /ν (t), υ (t) = 0 v (t0 ) dt0 .
Using the decomposition for Xt given by Eq. (20) we present the valuation
problems follows:

dXt = dYt + dMt , Y0 = 0, M0 = 0, Xt ≥ ξ, (41)

where Yt is stochastic part and Mt is deterministic part, and ξ < 0. Then we

can express the problem (20) in terms of stochastic variable Yt as follows:
p p
dYt = 1 − ρ2 Vt dWt , Y0 = 0, Yt ≥ ξ − Mt . (42)

We use a new time variable

υ (t) = 1 − ρ2 It ,


(43)

12
and write
dYυ = dWυ , Y0 = 0, Yυ ≥ ξ − Nυ , (44)
where
Nυ = Mt(υ) . (45)
Thus, we have one of the two venues to explore: (A) studying the processes
Xt or Xυ given by Eq. (39) and Eq. (40), respectively; (B) dealing with the
processes Yt or Yυ given by Eq. (42) and Eq. (44). In case (A) there is non-zero
time-dependent drift and flat boundary; in case (B) there is no drift but the
boundary is time-dependent.
Given a variance trajectory, we need to discuss how to calculate µ, ν, λ, M ,
and N . Let {vk |k = 0, 1, ..., K}, v0 = v, be a particular path generated via
discretization of Eqs (14) with homogeneous time-step ∆t = T /K. This equa-
tion can be discretized in various ways, for example, via the Euler-Maryama
scheme or the Milstein scheme. For special cases such as the Feller process cor-
responding to the Heston model, there are clever schemes tailored to the specific
process at hand. However, we are not pursuing them here since it is unnecessary
to achieve our objective. We have
√
v0 = v, vk = vk−1 + κ (θ − vk−1 ) ∆t + ε ∆tη,
(46)
I0 = 0, Ik = Ik−1 + ∆t 2 (vk + vk−1 ) ,

ρκθ ρ ((vk −vk−1 ))

− 12p− ρκ


µk = r − ε ε v√ k + ε ∆t ,
νk = 1  2
− ρ vk , (47)
Mk = kT r − ρκθ − 21 − ρκ ρ


M0 = 0, K ε ε Ik + ε (vk − v) .

It is clear that
υ k = 1 − ρ2 Ik , Υ = 1 − ρ2 IK ,



(48)
where {υ k |k = 0, 1, ..., K} is an inhomogeneous partition of the interval [0, Υ].
For our purposes, we treat the sequences {λk = µk /ν k |k = 0, 1, ..., K} and
{Mk |k = 0, 1, ..., K} as functions of υ k , which is possible because υ k is a mono-
tonically increasing sequence, and interpret them accordingly.
We illustrate the corresponding functions in Figures 2, 3. We emphasize
that µ and λ are very irregular, since they depend on the white noise process
dWt /dt, so that we have to deal with random terms of order ∆t−1/2 . At the
same time, the moving boundary Mt is much more regular, and depends on
random terms of order ∆t1/2 .

4.2 Solving IBVPs via Crank-Nicolson method

Let us describe how to price a conditional barrier option via the FDM; see
Loeper and Pironneau (2009).4 Specifically, we want to solve the following
4 Surprisingly, Loeper and Pironneau (2009) do not comment on the irregular nature of

the drift coefficient.

13
 (a)

(b)

Figure 2: Time-dependent coefficients (a) advection, (b) diffusion. The corre-

sponding parameters are the same as in Figure 1. Here Nt = 52 - one step per
week.

14
 (a)

(b)
.
Figure 3: (a) t (Υ), (b) λ (Υ)

15
problem on the semi-axis (ξ, ∞):

Pt (t, X) + µ (t) PX (t, X) + 12 ν (t) PXX (t, X) − κ (t) P (t, X) = 0,

(49)
P (T, X) = Π (X) , P (t, ξ) = 0.

We introduce P̄ , such that

κ (t0 )dt0
RT
P (t, X) = e− t P̄ (t, X) , (50)

and get the following problem for P̄ (t, X):

P̄t (t, X) + µ (t) P̄X (t, X) + 12 ν (t) P̄XX (t, X) = 0,

(51)
P̄ (T, X) = Π (X) , P̄ (t, ξ) = 0.

In the context we are interested in, κ (t) = r, so that P = exp (r (t − T )) P̄ .

We choose a uniform spatial grid { xm = ξ + m (U − ξ) /M | m = 0, 1, ..., M },
where U is a sufficiently remote upperboundary such that 0 belongs to the spa-
tial grid, xo = 0, and a temporal grid tn | t0 < t1 < ... < tN = T , which is not
necessarily uniform. Then, we apply the usual Crank-Nicolson method for the
advection-diffusion diffusion and get the following system of matrix equations
n
for P̄m :
n+1 n+1
n+1 n


P̄m − P̄m + αn+1/2 P̄m+1 − P̄m−1 n
+ P̄m+1 n
− P̄m−1
(52)
+β n+1/2 P̄m+1
n+1 n+1
n+1 n n n


− 2P̄m + P̄m−1 + P̄m+1 − 2P̄m + P̄m−1 = 0.
where
µn+1/2 ∆tn+1/2 ν n+1/2 ∆tn+1/2
αn+1/2 = 4∆x , β n+1/2 = 4∆x2 . (53)
We emphasize that given the extreme irregularity of µ, using more complicated
approaches for treading the drift term is not warranted.
We can write the system of equations in matrix form.
n+1/2 n+1/2
A(α,β) P̄ n = A(−α,−β) P̄ n+1 ,
N
(54)
P̄m = Π (xm ) ,
where
 
1 0 0 0 0 0

 α−β 1 + 2β −α − β 0 0 0 

 . . . . . . 
A(α,β,γ) = .

 . . . . . . 

 0 0 0 α−β 1 + 2β −α − β 
0 0 β −α − 4β 4α + 5β 1 − 3α − 2β
(55)
For brevity, the superscripts are omitted. At infinity we apply one-sided deriva-
tives and use the PDE itself to formulate the boundary condition.
Alternatively, we can rescale time, reduce the problem (29) to the form:

P̄υ + λ (υ) P̄x + 21 P̄xx = 0,

(56)
P̄ (Υ, x) = Π (x) , P̄ (υ, ξ) = 0,

16
and apply the Crank-Nicolson method to problem (56).
Of course, we can attack the pricing problem by solving the corresponding
Fokker-Planck equation for Green’s function G:

Gt (t, X) + µ (t) GX (t, X) − 21 ν (t) GXX (t, X) + κ (t) G(t, X) = 0,

(57)
G (0, X) = δ (X) , G(t, ξ) = 0,

and write
R∞
P (0, 0) = G (0, 0; T, x) Π (x) dx. (58)
0

We introduce Ḡ, such that

κ (t0 )dt0
Rt
G (t, X) = e− 0 Ḡ (t, X) . (59)

Accordingly,

Ḡt (t, X) + µ (t) ḠX (t, X) − 21 ν (t) ḠXX (t, X) = 0,

(60)
Ḡ (0, X) = δ (X) , Ḡ(t, ξ) = 0,
RT
κ (t0 )dt0
R∞
P (0, 0) = e− 0 Ḡ (0, 0; T, x) Π (x) dx. (61)
0

As before, we apply the Crank-Nicolson method for the advection-diffusion and

get the following system of matrix equations for Ḡnm :

− Ḡnm + αn+1/2 Ḡn+1 n+1

Ḡn+1
m
n n
m+1 − Ḡm−1 + Ḡm+1 − Ḡm−1

(62)
−β n+1/2 Ḡn+1
m+1 − 2Ḡm
n+1
+ Ḡn+1 n n n
m−1 + Ḡm+1 − 2Ḡm + Ḡm−1 = 0.

In the matrix form we get

n+1/2 n+1/2
A(α,β) Ḡn+1 = A(−α,−β) Ḡn ,
(63)
Ḡ0 = δ o,m .

Once ḠN is found, we can represent Pk0 as

M
Pk0 = e−rT ∆x ḠN
P
m Πm . (64)
m=1

4.3 Solving IBVPs via MHP

4.3.1 Analytic results
We wish to find Green’s function for process (44), or, equivalently, to solve the
following IBVP in a bounded domain with a moving boundary:
2
∂
∂υ Ḡ (υ, Y ) = 21 ∂Y
∂
2 Ḡ (υ, Y ) ,

Ḡ (0, Y ) = δ (Y ) , Ḡ (υ, ξ − Nυ ) = 0, Ḡ (υ, Y → ∞) → 0, (65)

ξ − Nυ ≤ Y < ∞, 0 ≤ υ ≤ Υ.

17
 Figure 4: A typical lower boundary ξ − Nυ as a function of υ.

For a representative Monte Carlo path, the corresponding boundary is shown

in Figure 4.
We split Ḡ, and represent it in the form:
Ḡ (υ, Y ) = H (υ, Y ) − F (υ, Y ) , (66)
where H (υ, Y ) is the standard heat kernel,
 2

exp − Y2υ
H (υ, Y ) = √ , (67)
2πυ
and F (υ, Y ) solves the following problem:
2
∂
∂υ F(υ, Y ) = 21 ∂Y
∂
2 F (υ, Y ) , ξ − Nυ ≤ Y < ∞,
(68)
F (0, Y ) = 0, F (υ, ξ − Nυ ) = f (υ) , F (υ, Y → ∞) → 0,
where
f (υ) = H (υ, ξ − Nυ ) . (69)
The MHP allows one to represent F (υ, Y ) in the form
 
Z υ (Y − ξ + Nυ0 ) exp − (Y −ξ+N0υ0 )2
2(υ−υ )
F (υ, Y ) = q φ (υ 0 ) dυ 0 , (70)
0 0 3
2π (υ − υ )

where φ (υ) solves the Volterra equation of the second kind:

Z υ
Θ (υ, υ 0 ) Ξ (υ, υ 0 )
φ (υ) + p φ (υ 0 ) dυ 0 = f (υ) , (71)
0 2π (υ − υ 0 )

18
and  
υ −Nυ 0 (υ−υ0 )Θ2 (υ,υ0 )
Θ (υ, υ 0 ) = − N(υ−υ 0) , Ξ (υ, υ 0 ) = exp − 2 ,
(72)
Θ (υ, υ) = − dN
dυ ,
υ
Ξ> (υ, υ) = 1.
Assuming that φ (υ) is known, we can represent Ḡ (Υ, Y ) as follows:

Ḡ (Υ, Y ) = H (Υ, Y ) − F (Υ, Y ) , (73)

where F (Υ, Y ) is given by Eq. (70). Finally, returning back to the original
variables, we get

Ḡ (T, X) = H (Υ (T ) , X − MT ) − F (Υ (T ) , X − MT ) . (74)

When Z = X − ξ → 0, the corresponding integral has to be dealt with

carefully due to a singularity at υ 0 = υ. We have

(Z−Nυ +Nυ0 )2
 
(Z−Nυ +Nυ0 ) exp −
Rυ 2(υ−υ 0 )
F (υ, X − Nυ ) = 0
√ φ (υ 0 ) dυ 0
  2π(υ−υ 0 )3
Z2
Z exp − R υ I (1) (υ,υ0 ) 0 R υ I (2) (υ,υ0 ) 0
Rυ 2(υ−υ 0 )
ZN 0 (υ)
=e φ (υ) 0
√ dυ 0 + 0 √ dυ + 0 √ dυ
2π(υ−υ 0 )3 (υ−υ 0 ) (υ−υ 0 )
(1) 0 (2) 0
υ I (υ,υ ) υ I (υ,υ )
0
 
= 2eZN (υ) φ (υ) N − √Zυ + 0 √ dυ 0 + 0 √ dυ 0 ,
R R
0 0
(υ−υ ) (υ−υ )
(75)
where
 
Z2 0

Z exp − exp(−ZΘ(υ,υ 0 ))Ξ(υ,υ 0 )φ(υ 0 )−eZN (υ) φ(υ)
2(υ−υ 0 )
I (1) (υ, υ 0 ) = √
0
,
 2π(υ−υ ) 2
 (76)
( Z−N υ +N 0 )
Θ(υ,υ 0 ) exp − υ φ(υ 0 )
2(υ−υ 0 )
I (2) (υ, υ 0 ) = √
2π
.

It is clear that integrals in Eq. (75) have weak singularities and hence are easy
to handle.

4.3.2 Numerics
There are numerous well-known approaches to solving Volterra equations; see,
Linz (1985), among many others. We choose the most straightforward approach
and show how to solve the following archetypal Volterra equation with weak
singularity numerically:
Z υ
K(υ, υ 0 )
φ(υ) + √ φ(υ 0 ) dυ 0 = f (υ), (77)
0 υ − υ0
where K(υ, υ 0 ) is a non-singular kernel. We write
√
Z υ Z υ
K(υ, υ 0 )φ (υ 0 ) 0 0 0
√ dυ = −2 K(υ, υ )φ (υ ) d υ − υ0 . (78)
0 υ − υ0 0

19
We map this equation to a grid 0 = υ 0 < υ 1 < . . . < υ N = υ. We introduce the
following notation:

fk = f (υ k ), φk = φ (υ k ) , Kk,l = K(υ k , υ l ),
∆k,l = υ k − υ l , Πk,l = √
∆l,l−1
√ . (79)
( ∆k,l−1 + ∆k,l )

Then, Eq. (77) can be approximated by the trapezoidal rule as

k
X

φk + Πk,l Kk,l φl + Kk,l−1 φl−1 = fk , (80)
l=1

so that

 p Pk−1

fk − ∆k,k−1 Kk,k−1 φk−1 − l=1 Πk,l Kk,l φl + Kk,l−1 φl−1
φk = p
. (81)
1 + ∆k,k−1 Kk,k

Thus, φk can be found by induction starting with φ0 = f0 .

After φk are determined, Fk can be written in the form
  X k k
0 Z 
(1) (1)
 X 
(2) (2)

Fk = 2eZNk φk N − √ + Πk,l Ik,l + Ik,l−1 + Πk,l Ik,l + Ik,l−1 ,
υk
l=1 l=1
(82)

4.3.3 Example
Let us consider the special case of constant drift λ; the corresponding boundary
is linear, ξ − λυ, where υ = t. Then Eq. (71) becomes
 2 
− λ (υ−υ 0 )
Z υ exp (ξ−λυ)2
2
0 0 e− 2υ
φ (υ) − λ p φ (υ ) dυ = √ . (83)
0 2π (υ − υ 0 ) 2πυ

Lipton and Kaushansky (2020) show that

(ξ−λυ)2
e−
 
2υ ξ + λυ
φ (υ) = √ + λe2ξλ N √ , (84)
2πυ υ
 2

exp 2ξλ − (Y −2ξ)
2υ
F (υ, Y ) = √ . (85)
2πυ
Accordingly,

Ḡ (T, X) = H (T, X − λT ) − e2ξλ H (T, X − λT − 2ξ) . (86)

It is easy to see that Ḡ (T, ξ) = 0, as it should. Figure 5 illustrates our findings.

20
Figure 5: φ (υ) computed analytically and by solving the Volterra equation.
The difference between the corresponding functions is less that 10( − 4). The
parameters are ξ = −0.5, λ = 0.5.

4.3.4 Semi-analytical solution of pricing problems

Green’s function is given by Eq. (74). In Figure 6 we compare Green’s functions
obtained via the FDM and the MHP. The figure shows that these functions are
reassuringly close. The MHP is clearly preferred since it is much faster.
Once Ḡ (T, X) is found all barrier problems can be solved analytically.
For instance, we can calculate the P (0, T ; ξ) of the no-touch option for the
barrier level ξ < 0:
R∞
P (0, T ; ξ) = e−rT ξ (H (Υ, X 0 − NΥ ) − F (Υ, X 0 − NΥ )) dX 0
  2
 
  R exp − (ξ−N Υ +Nυ 0 )
2 Υ−υ 0 (87)
= e−rT N −(ξ−N √ Υ
)
− 0
Υ
√ ( 0) φ (υ 0 ) dυ 0  .
Υ 2π(Υ−υ )

The relative price C (0, T, K; ξ) /e−rT S0 of the barrier call struck at K = S0 ek ,

where k ≥ ξ, has the form:

21
Figure 6: G (T, x) computed via the FDM and the MHP. The absolute difference
between the corresponding functions is less that 10−3 . The parameters are the
same as in Figure 1. The log-barrier is ξ = −0.5.

C(0,S0 ,T,K;ξ)
e−rT S0  0 
R∞
= k
(H (Υ, X 0 − NΥ ) − F (Υ, X − NΥ )) eX − ek dX 0
0

η2 η2
−
e− 2Υ
R∞ +η
k ∞
= eNΥ k−NΥ e√2Υ
R
2πΥ
dη − e √
k−NΥ 2πΥ
dη
(η−ξ+Nυ0 )2
 
RΥR∞ (η−ξ+Nυ0 ) exp − 0 +η
2(Υ−υ )
−eNΥ 0 k−NΥ
√
0 )3
dηφ (υ 0 ) dυ 0
2π(Υ−υ
(η−ξ+Nυ0 )2
 
RΥR∞ (η−ξ+Nυ0 ) exp − 0
2(Υ−υ )
+ek √ dηφ (υ 0 ) dυ 0
0 k−NΥ
 2π(Υ−υ  0)
3
  (88)
NΥ + Υ N +Υ NΥ
=e 2 N √Υ
Υ
− ek N √ Υ
(k−NΥ −ξ+Nυ0 )2
 
exp −
Υ 2(Υ−υ 0 )
−ek 0 √ φ (υ 0 ) dυ 0
R
2π(Υ−υ 0 )
 
RΥ (Υ−υ0 ) k−NΥ −ξ+Nυ0 −(Υ−υ 0 )
− 0
e NΥ +ξ−Nυ0 + 2 N − √ φ (υ 0 ) dυ 0
(Υ−υ 0 )
(k−NΥ −ξ+Nυ0 )2
 
exp − 0
Υ
√ ( 0)
2 Υ−υ
+ek 0 φ (υ 0 ) dυ 0 .
R
2π(Υ−υ )

We found that it more efficient to price these options using the backward
induction. For example, to price the no-touch option backward, we introduce
the new time variable $ = Υ − υ, and the boundary O$ = NΥ−υ , and write

22
P (0, T ; ξ) in the form

P (0, T ; ξ) = e−rT (1 − Q (Υ; ξ)) . (89)

Here
 2

Z Υ (−ξ + O$0 ) exp − (−ξ+O $0 )
2(Υ−$ 0 )
Q (Υ; ξ) = q ψ (N T ) ($0 ) d$0 , (90)
0 0 3
2π (Υ − $ )

where ψ (N T ) ($) solves Eq. (71) with f ($) = 1.

By the same token, we can represent C (0, S0 , T, K; ξ) as follows

C (0, S0 , T, K; ξ) = e−rT S0 (D (Υ, k) − E (Υ, k; ξ)) , (91)

where
 Υ
   
D (Υ, k) = ek eO0 −k+ 2 N O0 −k+Υ
√
Υ
− N 0 −k
O√
Υ
,
(−ξ+O$0 )2
 
(−ξ+O$0 ) exp − 0
(92)
RΥ 2(Υ−$ )
E (Υ, k; ξ) = 0
√ ψ (C) ($0 ) d$0 ,
2π(Υ−$ 0 )3

and ψ (C) ($0 ) solves Eq. (71) with

  
−k+Υ
f ($) = ek eξ−O$0 +O0 −k+ 2 N ξ−O$0 +O
Υ
√ 0
Υ
  (93)
0 −k
−N ξ−O$√0 +O
Υ
.

In Figures 7, 8 we compare prices of no-touch and call options obtained via

the FDM and the MHP. The figure shows that the corresponding prices are very
close. As before, the MHP is clearly much faster than the FDM.

4.4 External loop: averaging over all variance paths

After the corresponding solution is found and expressed in the original variables,
we produce a set of random sequences {vk |k = 0, 1, ..., K} and repeat steps.
Once a sufficiently large number of paths is generated, we perform averaging
and obtain the solution.
In Figures 9, 10, 11,we show the averaged value of Green’s function, as well
as prices of the no-touch and call options.

5 Joint probability density for the value of drifted

Brownian motion and its running minimum
This section, which serves as a mathematical aside, shows that the MHP al-
lows solving a very complex problem of finding the joint probability distribu-
tion for a Brownian motion with time-dependent drift and volatility and its

23
Figure 7: P (T, x) computed via the FDM and the MHP. The absolute difference
between the corresponding functions is less that 10−3 . The parameters are the
same as in Figure 1. The log-barrier is ξ = −0.5.

Figure 8: C (T, x) computed via the FDM and the MHP. The absolute difference
between the corresponding functions is less that 10−4 . The parameters are the
same as in Figure 1. The log-barrier is ξ = −0.5.

24
 (a)

(b)

Figure 9: Green’s function averaged over Monte Carlo paths. The number of
MC path for the MHP and FDM is 10, 000; the number of path for the classical
MCS is 100, 000.

25
 (a)

(b)

Figure 10: Price of the no-touch with the log-barrier = ξ = −0.5: figure (a),
table (b). The number of MC paths for the MHP and MFD is 10, 000; the
number of patha for the MCS is 100, 000.

26
 (a)

(b)

Figure 11: (a) The price of the down-and-out call with the barrier at S = 0.9
as a function of the strike K; (b) a summary table. The number of MC paths
for the MHP and MFD is 10, 000; the number of paths for the MCS is 100, 000.

27
minimum. Without loss of generality, we can restrict ourselves to the case of
time-dependent drift and unit volatility by scaling time.5
To put things in perspective, we start with the standard Brownian motion
and consider the following problem:
Gt − 12 Gxx = 0,
(94)
G (0, x) = δ (x) , G(t, a) = 0,

where a is the lower bound, a < 0. It is clear that G (T, b; a) db is the probability
of the Brownian motion ending in the interval (b − db/2, b + db/2) and having
its minimum on the interval [a, 0]. Thus, the corresponding joint pdf has the
form
∂
π (T, a, b) = − G (T, b; a) . (95)
∂a
The method of images yields

G (T, b; a) = H (T, b) − H (T, b − 2a) , (96)

so that
2
(b − 2a) H (T, b − 2a) .
π (T, a, b) = (97)
T
For the drifted Brownian motion the problem can be written as follows:
Gt + λGx − 21 Gxx = 0,
(98)
G (0, x) = δ (x) , G(t, a) = 0,

G (T, b; a) = H (T, b − λT ) − e2λa H (T, b − λT − 2a) , (99)

∂
π (T, a, b) = − ∂a G (T, b; a)
2 2λa (100)
= T (b − 2a) e H (T, b − λT − 2a) .
Needless to say that for zero drift, λ = 0, we recover Eq. (97).
Expressions (97) and (100) are very well-known, even though their derivation
is usually somewhat convoluted. However, to the best of our knowledge, what is
not known, is a similar expression when the drift λ depends on the (scaled) time
υ. We shall use the MHP to derive the corresponding formula. The problem of
interest has the form
Gυ + λ (υ) Gx − 21 Gxx = 0,
(101)
G (0, x) = δ (x) , G(υ, a) = 0,

G (Υ, b; a) = H (Υ, b − NΥ ) − F (Υ, b; a) , (102)

 2

(b+Θ(Υ,υ0 )(Υ−υ0 ))
Z Υ (b + Θ (Υ, υ 0 ) (Υ − υ 0 )) exp − 2(Υ−υ 0 )
F (Υ, b; a) = q φ (υ 0 ; a) dυ 0 .
3
0 2π (Υ − υ 0 )
(103)
5 Despite this being a classical problem, its correct solution is not presented in the literature,

except for the simple case of constant drift. At the same time, several incorrect solutions have
been proposed.

28
 υ
Θ (υ, υ 0 ) Ξ (υ, υ 0 )
Z
φ (υ; a) + p φ (υ 0 ; a) dυ 0 = f (υ; a) , (104)
0 2π (υ − υ 0 )
f (υ; a) = H (υ, −Nυ − a) = H (υ, Nυ + a) . (105)
Thus,
∂
π (Υ, a, b) = − ∂a G (Υ, b; a) 
(Υ,υ0 )(Υ−υ0 ))2

(b+Θ
RΥ (b+Θ(Υ,υ0 )(Υ−υ0 )) exp − 2(Υ−υ 0 )
(106)
= 0
√ 3
ψ (υ 0 ; a) dυ 0 ,
2π(Υ−υ 0 )

where
υ
Θ (υ, υ 0 ) Ξ (υ, υ 0 )
Z
(Nυ + a)
ψ (υ; a) + p ψ (υ 0 ; a) dυ 0 = f (υ, a) . (107)
0 2π (υ − υ ) 0 υ

6 Conclusions
This paper introduced a new hybrid MHP/MCS technique for pricing barrier op-
tions on assets with stochastic volatility. The idea is to decompose the solution
process into the inner step, which solves a barrier problem for the condition-
ally independent process, and the outer step, which averages the corresponding
solutions over the one-dimensional stochastic volatility dynamics.
Our methodology is general and can manage all known stochastic volatility
models equally efficiently. Besides, relatively simple extensions (which will be
described elsewhere) can also handle rough volatility models. With minimal
changes, one can use the method to price popular double-no-touch options and
other similar instruments.
While several authors used hybrid techniques before, see, e.g., Loeper and
Pironneau (2009), their methods use the FDM and are still relatively slow, al-
though undeniably faster than the standard two-dimensional MCS. Our method
reduces the inner barrier problem to solving a linear Volterra equation of the
second kind. It is very efficient and is an order of magnitude faster than other
hybrid methods with the same (or better) accuracy. Our results are a natural
generalization of Willard’s formula, see Willard (1997), for barrier options.
As a byproduct of our analysis, we derived a new expression for the joint
pdf for the value of a drifted Brownian motion and its running minimum or
maximum in the case of time-dependent drift.
Acknowledgement 1 We are grateful to Andrey Itkin and Dmitry Muravey
for several useful discussions of the topics covered in this paper. Numerous
conversations with Marcos Lopez de Prado and Alexey Kondratiev were very
helpful.

References
Achdou Y., Pironneau O. Computational Methods for Option Pricing, SIAM
Series: Frontiers in Mathematics 2005; vol. 30, Philadelphia.

29
Andersen L., Andreasen J. Jump-Diffusion Processes: Volatility Smile Fitting
and Numerical Methods for Option Pricing , Review of Derivatives Research
2000; 4, 231–262.
Andreasen J. Behind the mirror, Risk Magazine 2001; 14(11), 109–110.
Bergomi L. Stochastic volatility modeling 2015; CRC press.
Black F., Scholes M. The pricing of options and corporate liabilities , Journal
of Political Economy 1973; 81(3), 637-659.
Carr P., Itkin A., Muravey D. Semi-closed form prices of barrier options in
the time-dependent CEV and CIR models, The Journal of Derivatives 2020,
28(1), 26-50.
Carr P., Itkin A., Muravey D. Semi-analytical pricing of barrier op-
tions in the time-dependent Heston model 2022. Working Paper,
https://arxiv.org/abs/2202.06177.
De Gennaro A.L., Bernard C. Semi-analytical prices for lookback and barrier
options under the Heston model, Decisions in Economics and Finance 2019;
42(2), 715-741.
De Gennaro A.L., Bernard C. Correction to: Semi-analytical prices for lookback
and barrier options under the Heston model, Decisions in Economics and
Finance 2021; 1-3.
Derman E., Kani I. Riding on a smile, Risk Magazine 1994; 7(2), 32-39.
Dupire B. Pricing with a smile , Risk Magazine 7 (1994), no. 1, 18-20.
Dupire B. A unified theory of volatility, Working Paper 1996.
Glasserman P. Monte-Carlo Methods in Financial Engineering, vol. 53 2004;
Stochastic Modeling and Applied Probability, Springer-Verlag, New York.
Hagan P., Kumar D., Lesniewski A., Woodward D. Managing smile risk.
Wilmott Mag. 2002; September 84–108.
He X.-J., Lin. S. An analytical approximation formula for barrier option prices
under the Heston model, Computational Economics 2021; 1-13.
Heston S. L. A closed-form solution for options with stochastic volatility with
applications to bond and currency options, Review of Financial Studies 1993;
6, 327-343.
Hull J., White A. The pricing of options on assets with stochastic volatilities,
Journal of Finance 1987; 42, 281-300.
, Itkin A., Muravey D. Semi-analytical pricing of barrier options
in the time-dependent λ-SABR model 2021. Working Paper,
https://arxiv.org/abs/2109.02134.

30
Itkin A., Muravey D. Semi-analytic pricing of double barrier options with time-
dependent barriers and rebates at hit. Frontiers of Mathematical Finance
2022; 1(1), 53–79.
Itkin A., Lipton A., Muravey, D. Generalized Integral Transforms in Mathemat-
ical Finance 2021; World Scientific.

Jex M., Henderson R., Wang D. Pricing exotics under the smile. Risk Mag.
1999; 12(11), 72–75.
Kartashov E. Analytical Methods in the Theory of Heat Conduction in Solids
2001; Vysshaya Shkola, Moscow.

Lewis A. Option Valuation under Stochastic Volatility with Mathematica Code

2000; Finance Press.
Lewis A. A simple option formula for general jump-diffusion and other expo-
nential Lévy processes 2001; Working Paper.

Linz P. Analytical and Numerical Methods for Volterra Equations 1985; SIAM.
Lipton A. Analytical valuation of barrier options on assets with stochastic
volatility 1997; Working paper, Bankers Trust.
Lipton A. Mathematical Methods For Foreign Exchange: A Financial Engineer’s
Approach 2001; World Scientific.

Lipton A. The volatility smile problem. Risk Mag. 2002; 15(2), 61–65.
Lipton A. Financial Engineering: Selected Works of Alexander Lipton 2018;
World Scientific.
Lipton A., Kaushansky V. On the first hitting time density
for a reducible diffusion process. Quantitative Finance 2020;
https://doi.org/10.1080/14697688.2020.1713394.
Lipton A., Kaushansky, V., Reisinger, C. Semi-analytical solution of a McKean–
Vlasov equation with feedback through hitting a boundary. Euro. Jnl of Ap-
plied Mathematics 2019; 1–34, doi:10.1017/S0956792519000342.

Lipton A., McGhee W. Universal barriers. Risk Mag. 2002; 15(5), 81–85.
Lipton A., Sepp A. Stochastic volatility models and Kelvin waves. Journal of
Physics A: Mathematical and Theoretical 2008; 41(34), p.344012.
Lipton A., Sepp A. Credit value adjustment for credit default swaps via the
structural default model. The Journal of Credit Risk 2009; 5(2) pp.127-150.
Loeper G., Pironneau, O. A mixed PDE/Monte-Carlo method for stochastic
volatility models, C. R. Acad. Sci. Paris, Ser. I 347 2009; 559–563.

31
Merton R. C. Theory of rational option pricing, Bell Journal of Economics and
Management Science 1973; 4, 141-183.
Merton R. C. Option pricing when underlying stock returns are discontinuous,
Journal of Financial Economics 1976; 3, 125-144.

Romano M., Touzi N. Contingent Claims and Market Completeness in a

Stochastic Volatility Model, Mathematical Finance 1997; 7(4), 399-412.
Rubinstein M. Implied binomial trees, Journal of Finance 1994; 49, 771-818.
Scott L. O. Option pricing when variance changes randomly: theory, estimation
and an application, Journal of Financial and Quantitative Analysis 1987; 22,
419-438.
Schmelzle M. Option pricing formulae using Fourier transform: Theory and
application 2010; Preprint, http://pfadintegral.com.
Stein E. M., Stein J. C. Stock price distributions with stochastic volatility: an
analytic approach, Review of Financial Studies 1991; 4, 727-752.
Wiggins J. B. Option values under stochastic volatility: theory and empirical
evidence, Journal of Financial Economics 1987; 19, 351-372.
Willard A. G. Calculating Prices and Sensitivities for Path-Independent Deriva-
tives Securities in Multifactor Models, The Journal of Derivatives 1997; 5(1),
45-61.

32