Accepted Manuscript

Pricing American options under the constant elasticity of variance model: An

extension of the method by Barone-Adesi and Whaley

Ballestra Luca Vincenzo, Cecere Liliana

PII: S1544-6123(15)00061-6
DOI: http://dx.doi.org/10.1016/j.frl.2015.05.017
Reference: FRL 385

To appear in: Finance Research Letters

Please cite this article as: Vincenzo, B.L., Liliana, C., Pricing American options under the constant elasticity of
variance model: An extension of the method by Barone-Adesi and Whaley, Finance Research Letters (2015), doi:
http://dx.doi.org/10.1016/j.frl.2015.05.017

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Pricing American options under the constant elasticity of variance
model: An extension of the method by Barone-Adesi and Whaley

Ballestra Luca Vincenzo*, Cecere Liliana

Dipartimento di Economia,

Seconda Università di Napoli,

81043, Capua, ITALY

email: lucavincenzo.ballestra@unina2.it

Abstract

We consider the problem of pricing American options on an underlying described by the constant elasticity of

variance (CEV) model. Such a problem does not have an exact closed-form solution, and therefore some kind of

approximation is required. In this paper we extend the approach proposed by Barone-Adesi and Whaley [1997],

which allows us to obtain a direct semi-analytical approximate solution. Numerical experiments are presented

showing that the proposed method is satisfactorily accurate and computationally very fast.

KeyWords: CEV model, American option, Barone-Adesi and Whaley, free boundary, option pricing

JEL classification codes: G13 C63 C02

1. Introduction

One of the most famous models for pricing financial derivatives is that introduced by Black and Scholes

[1973]. The main assumption on which this model stands is that asset prices follow a log-normal diffusion

process with constant volatility, which yields a considerable amount of mathematical tractability and allows

one to price European vanilla options using simple analytical formulae. Nevertheless, as revealed by several

empirical studies, the volatility of log-returns is far from being constant, and therefore, in order to take into

account this fact, several improvements of the Black-Scholes model have been developed. In particular, Cox

[1975] and Cox and Ross [1976] have proposed the so-called constant elasticity of variance (CEV) model,

according to which the volatility is specified as a function of the price of the option’s underlying asset.

The CEV model is still nowadays quite popular among researchers and practitioners (see, for example,

Ballestra and Pacelli [2011], Hsu et al. [2008], Thakoor et al. [2013], Thakoor et al. [2014], Ahmadian and

*Corresponding author. Email: lucavincenzo.ballestra@unina2.it; phone number: +393479154562

Preprint submitted to Elsevier May 27, 2015

Ballestra [2014]), as it offers the following advantages: first of all, the volatility is modeled using a simple and

parsimonious specification, without introducing any additional stochastic process; second, the leverage effect,

i.e. the inverse relationship that is frequently observed between prices and volatility, can be adequately taken

into account.

In this paper, we consider the problem of pricing American options on an underlying described by the CEV

model; in the literature, this topic has already been addressed in some works: Wong and Zhao [2008] and Zhao

and Chang [2014] have proposed the use of enhanced finite difference schemes; Ruas et al. [2013] and Chung

et al. [2013] have developed static-hedge approaches; Xu and Knessl [2011] and Nunes [2009] have reduced the

problem to an integral equation and have established certain asymptotic properties of the American option

price; Nunes [2009] have employed an optimal stopping procedure for solving a jump to diffusion extended

CEV model; Wong and Zhao [2010] obtains a fast approximation of the early exercise boundary and of the

American option price based on the Laplace-Carson transform and on a Gaussian quadrature formula; Pun

and Wong [2013] use an asymptotic series expansion to improve the approach by Wong and Zhao [2010],

which turns out to be unstable and inefficient for certain set of parameter values; Zhao and Wong [2012]

apply homotopy approximation techniques for pricing American options on underlying described by the CEV

model and other diffusion processes.

In the present manuscript, we extend and test the method that has been proposed by Barone-Adesi and

Whaley [1997] for pricing American options under the Black-Scholes model. Such an approach, which, to the

best of our knowledge has never been applied to the CEV model, has the advantage that the American option

price can be estimated by simply solving a single algebraic equation (with no differential terms). Numerical

experiments are presented showing that the proposed method is satisfactorily accurate and computationally

very fast: the errors obtained are often of order 10−3 , and using a normal personal computer the option price

is always computed in just a few thousandths of a second.

The paper is organized as follows: in Section 2 we introduce the problem of pricing American options under

the CEV model; in Section 3 we extend the method by Barone-Adesi and Whaley to the case of the CEV

model; finally in Section 4 we present and discuss the numerical results obtained.

2
2. Pricing American options under the constant elasticity of variance model

According to the constant elasticity of variance (CEV) model (Cox [1975], MacBeth and Merville [1980]),

the price of an asset, which we denote S(t), satisfies (under the risk neutral measure) the stochastic differential

equation:
β
dS(t) = (r − q)S(t)dt + δS 2 (t)dW (t), (1)

where r and q are the (constant) interest rate and the (constant) dividend yield, respectively and δ and β are

positive constants. It follows immediately that the above process is an extension of the well-known geometric

Brownian motion to the case where the volatility is price dependent and equal to:

β
σ(S) = δS 2 −1 . (2)

Therefore, when β = 2 the volatility (2) is constant and the CEV model (1) reduces to the geometric Brownian

motion on which the famous Black-Scholes model stands.

For the sake of brevity, in this paper we describe the option pricing method by considering the case of

American Call options only, however the case of Put options can be treated analogously and just requires

straightforward modifications. Let C̃(S, t) denote the price of an American Call option on an underlying S

described by (1), with maturity T and strike price E. Let us apply the change of variables:

τ = T − t, C(S, τ ) = C̃(S, T − t). (3)

As is well-known (Wilmott et al. [1995]), there is a function B(τ ), which is commonly referred to as optimal

exercise boundary, such that for S > B(τ ) the option is exercised and:

C(S, τ ) = S − E. (4)

Instead, for S < B(τ ), the American option price satisfies the following partial differential equation:

∂C(S, τ ) 1 2 β ∂ 2 C(S, τ ) ∂C(S, τ )

− δ S 2
− (r − q) + rC(S, τ ) = 0, (5)
∂τ 2 ∂S ∂S

with boundary conditions:


∂C(S, τ ) 
C(0, τ ) = 0, = 1, (6)
∂S S=B(τ )

and initial condition:

3
 C(S, 0) = max [S − E, 0]. (7)

Moreover, the optimal exercise boundary is implicitly defined by the equation:

C(B(τ ), τ ) = B(τ ) − E. (8)

The free-boundary partial differential problem (5)-(8) does not have an exact closed-form solution, and thus

some numerical approximation is required. In this paper, we extend the method that has been originally

proposed by Barone-Adesi and Whaley [1997] for the case of the Black-Scholes model, i. e. for the case β = 2.

3. Extending the method by Barone-Adesi and Whaley

For the sake of clarity, this section is divided into two subsections: in subsection 3.1 we extend the method

by Barone-Adesi and Whaley to the case of the CEV model, whereas in subsection 3.2 we show how to

practically compute the non-elementary functions involved.

3.1. The Barone-Adesi and Whaley method

Let c(S, τ ) denote the price of a European Call option on the underlying described by (1) with maturity

T and strike price E. As described in Schroder [1989], c(S, τ ) can be obtained as follows:


    
 2 2
Se−qτ 1 − P 2x, − Ee−rτ P 2y,

, 2y , 2x , β > 2,


β−2 β−2
c(S, τ ) =      (9)

 −qτ 2 −rτ 2
Se 1 − P 2y, 2 + , 2x − Ee P 2x, , 2y , β < 2,


2−β 2−β
where:
2(r − q)
y = k E 2−β , x = kSt2−β e(r−q)(2−β)τ , k= , (10)
δ 2 (2 − β)(e(r−q)(2−β)τ − 1)

and P (x, v, λ) is the non central chi-squared probability function at x with non-centrality parameter v and

degrees of freedom λ (Cox [1975]). Note that in (9) the case β = 2 is not considered, as it corresponds to the

standard Black and Scholes model, which is already dealt with by Barone-Adesi and Whaley [1997].

Let us consider the early exercise premium:

εC (S, τ ) = C(S, τ ) − c(S, τ ). (11)

4
It can be easily shown (see Barone-Adesi and Whaley [1997]) that for S < B(τ ), εC satisfies the partial

differential equation:

∂εC (S, τ ) 1 2 β ∂ 2 εC (S, τ ) ∂εC (S, τ )

− δ S 2
− (r − q) + rεC (S, τ ) = 0, (12)
∂τ 2 ∂S ∂S

with boundary conditions:

 
∂εC (S, τ )  ∂c(S, τ ) 
=1− , εC (0, τ ) = 0, (13)
∂S 
S=B(τ ) ∂S S=B(τ )

and initial condition:

εC (S, 0) = 0. (14)

Moreover, using (11), equation (8) is transformed as follows:

εC (B(τ ), τ ) = B(τ ) − E − c(B(τ ), τ ). (15)

Now, let us consider the following change of variables:

εC (S, K)
f (S, K) = , K = 1 − e−rτ . (16)
K

Using (16), the partial differential equation (12) and the boundary condition (13) are rewritten as follows:

1 β ∂ 2 f (S, K) ∂f (S, K) r ∂f (S, K)

S 2
+ (r − q)S − f (S, K) − r(1 − K) = 0, (17)
2 ∂S ∂S K ∂K
   
∂f (S, K)  1 ∂c(S, τ ) 
= 1− , f (0, K) = 0, (18)
∂S 
S=B(τ ) K ∂S S=B(τ )

respectively. In order to obtain a closed-form approximation of the function f , we neglect the last term at

the left hand side of (17) (such an approximation is nicely motivated in Barone-Adesi and Whaley [1997]):

1 β ∂ 2 f˜(S, K) ∂ f˜(S, K) r
S 2
+ (r − q)S − f˜(S, K) = 0. (19)
2 ∂S ∂S K

Equation (19) is actually an ordinary differential equation and can be solved using an exact analytical formula.

In particular, following Davydov and Linetsky [2001] and using the boundary condition (18), we have:

f˜(S, τ ) = aψ ψl (S), (20)

where

5
   
 β−1
(d + 12 )u 12 +m 1
S e u M + m − κ, 1 + 2m, u , β < 2, r 6= q,

 2 2



 2

  
 β−1 d 1 1 1
S 2 e( 2 + 2 )u u 2 +m U

+ m − κ, 1 + 2m, u , β > 2, r 6= q,


2
ψ(S) = r  (21)

 1 r
S 2 Ip 2 w , β < 2, r = q,


K





  r 

S 12 K
 r
 p 2 w , β > 2, r = q,
K

|(r − q)| −(β−2) 1 β−2

u=  S , w= S− 2 , (22)
2 β − 2  |β − 2|
 
δ 
2 

 
β−2 1
d = sign (r − q) , m= , (23)
2 2|β − 2|

 
1 1 r 1
κ=d + − , p= , (24)
2 2(β − 2) |K(r − q)(β − 2)| |β − 2|

where K is given by (16) and aψ is a constant to be determined. Moreover, in (21) M (a, b, u) and U (a, b, u) are

the so-called Kummer’s confluent hypergeometric functions, and Ip and Kp are the modified Bessel functions

of the first kind and of the second kind, respectively (Abramowitz and Stegun [1972]). Now, from (11), (16)

and (20) it follows that:

C(S, τ ) = c(S, τ ) + Kaψ ψ(S). (25)

Then, by differentiating (25) with respect to S and by setting S = B(τ ), we obtain:

  −1    
∂ψ(S)  ∂C(S, τ )  ∂c(S, τ ) 
aψ = K − . (26)
∂S S=B(τ ) ∂S S=B(τ ) ∂S S=B(τ )

Moreover, from (16) and (25) we have:

c(B(τ ), τ ) + Kaψ ψ(B(τ )) = B(τ ) − E. (27)

Substitution of (26) in (27) yields:

    −1
∂c(S, τ )  ∂ψ(S) 
B(τ ) − E = c(B(τ ), τ ) + 1 − ψ(B(τ )). (28)
∂S S=B(τ ) ∂S S=B(τ )

The above equation allows us to determine the uknown free boundary B(τ ). However, an exact closed-form

solution of (28) is not available, and thus B(τ ) must be computed by numerical approximation. In this paper

we employ the well-known secant method (see, for example, Quarteroni et al. [2007]), which is satisfactorily

6
fast and does not require us to compute the derivatives of the Kummer functions appearing in (21).

Then, once B(τ ) is obtained, according to (11), (16) and (20), the Barone-Adesi and Whaley approximation

of the price of an American Call option, which we denote CBAW , is computed as follows:


c(S, τ ) + Kaψ ψ(S), S < B(τ ),


CBAW (S, τ ) = (29)


S − E,

S ≥ B(τ ),

where aψ is given by (27) and the second of (29) comes from (4).

3.2. Computing the functions P , M , U , Ip and Kp

Let us observe that in order to use the approximate formula (29) one has to evaluate the non-central chi-

squared probability function P (needed in (9)) and either the Kummer’s confluent hypergeometric functions

M and U (needed in (21) if r 6= q) or the modified Bessel functions Ip and Kp (needed in (21) if r = q). To

this aim, we proceed as follows.

The calculation of M , U , Ip and Kp is performed using the numerical algorithms developed in Abramowitz

and Stegun [1972], Amos [1985], Amos [1986]. Instead, as far as the evaluation of P is concerned, we calculate

the function P using the method described below. According to Sankaran [1959], if v > 100, the following

approximation is used:

P (x, v, λ) ∼
= N (x; ε, σ), (30)

where N (x, ε, σ) is the probability function at x of a normally distributed variable with mean ε and standard

deviation σ and:
k2 k2
ε = 1 + h(h − 1) 2 − h(h − 1)(2 − h)(1 − 3h) 24 , (31)
2k1 8k1
1  
hk22 (1 − h)(1 − 3h)
σ= 1− k2 , (32)
k1 4k12
k1 k3
where ks = 2s−1 (s − 1)!(v + sλ) and h = 1 − .
3k22
By contrast, if v ≤ 100, following Schroder [1989], we employ:

∞   n  
X v x X λ
P (x, v, λ) = g n+ , g i, , (33)
n=1
2 2 i=1 2

where g is the so-called gamma density function (Abramowitz and Stegun [1972]). To efficiently compute the

infinite series in (33) we use the ingenuous iterative method that has been proposed by Schroder [1989]; for

7
the description of such an algorithm the interested reader is reminded to Schroder [1989].

4. Numerical Results

The numerical simulations are performed on a personal computer with an Intel Core i7 3.40 GHZ 8.00 GB

Ram and the software programs are written in Matlab 7.0.

For comparison purposes, the option price is also calculated by finite difference approximation. In particular,

in the partial differential equation (5) the derivatives with respect to S are discretized using the standard

(centered) three-point finite difference scheme, whereas the derivative with respect to τ are computed using

the implicit Euler time-stepping, whose accuracy is enhanced (up to second-order) by Richardson extrapola-

tion (see Chang et al. [2007], Tavella and Randall [2000]).

Let CBAW (S, τ ) denote the American option price computed using the Barone-Adesi-Whaley method de-

scribed in Section 3 and and let CF D (S, τ ) denote the American option price computed by finite difference

approximation. The (relative) errors on CBAW (S, τ ) and CF D (S, τ ) are thus:

|C(S, τ ) − CBAW (S, τ )| |C(S, τ ) − CF D (S, τ )|

ErrBAW = , ErrF D = . (34)
C(S, τ ) C(S, τ )

Note that the exact value C(S, τ ) needed in the above expression is not available. Therefore, a very accurate

estimation of it is obtained using the finite difference approximation with a very large number of discretization

nodes along both the S and the τ directions. Moreover, we also evaluate the (relative) error on the optimal

exercise boundary B(τ ), which is computed using relations similar to (34).

The numerical experiments described below are carried out by considering option data as in Barone-Adesi

and Whaley [1997]. Moreover, the CEV parameters are selected as follows: for the parameter β we consider

four different values, β = 1.33, 1.66, 2.33, 2.66. Instead, for the CEV parameter δ we choose values with

only two decimal significant digits and such that the volatility of the underlying asset price, which can be

computed according to (2), is in the range [0.4, 0.7]. Finally, to provide a direct comparison between the

approach proposed in this paper and the finite difference method (as well as to save space), the only finite

difference simulations which we report are those carried out using a number of discretization nodes (along

both the S and the τ directions) which is chosen such that the errors ErrF D and ErrBAW on the American

option price are roughly the same for τ = 3 (which is the largest of the times to maturity which we consider).

The results experienced are shown in Tables 1-4. Precisely, Table 1 and Table 3 concern the calculation of

8
the option price, whereas Table 2 and Table 4 concern the calculation of the optimal exercise boundary

(CP U T imeBAW and CP U T imeF D refers to the computer times required in order to compute the option

price using the approach by Barone-Adesi-Whaley and the finite difference approximation, respectively). As

we can observe, the method by Barone-Adesi and Whaley [1997] yields a satisfactorily accurate approximation

of the American option price. In fact, as shown in Table 1 and Table 3, the errors on C(S, τ ) are always smaller

than 7.81 × 10−2 , and in addition they are often of order 10−3 and 10−4 .

Moreover, the method is very fast from the computational standpoint, as the American option price is

always obtained in just a few thousands of a second. Finally, judging from the results reported in Table 1

and Table 3, we can conclude that, as far as the calculation of the option price is concerned, the approach

proposed in this paper is approximately 3-5 times faster than the finite difference scheme (if ErrBAW and

ErrF D are nearly the same, CP U T imeBAW is 3-5 times smaller than CP U T imeF D ).

Moreover, looking at Table 2 and Table 4, we may note that the errors on the optimal exercise boundary

are about one order of magnitude bigger than the errors on the option prices. In particular, errors of the

order 10−1 are experienced in correspondence of β = 2.33 and β = 2.66 and times to maturities τ ≤ 1. This

is due to the fact that the optimal exercise boundary is usually quite difficult to approximate numerically,

especially for small time to maturities, as B(τ ) is not smooth at τ = 0 (see Wilmott et al. [1995]). In this

respect, it is also worth noting that the finite difference approximation of the optimal exercise boundary is

not particularly accurate as well (ErrF D is often of order 10−2 or 10−1 , see Table 2 and Table 4). Finally,

we may note that, as far as the calculation of B(τ ) for a given (single) value of τ is concerned, in most of the

cases the method by Barone-Adesi and Whaley is about three times faster than the finite difference scheme.

It should be observed, however, that using the finite difference scheme the option price and the early exercise

boundary can be computed simultaneously for several times to maturity. Instead, the method by Barone-Adesi

and Whaley requires one to run a different computer simulation each time that τ is varied. This is clearly

an advantage in favor of the finite difference method, which thus turns out to be faster than the method by

Barone-Adesi and Whaley if the calculation of the option price and of the early exercise boundary has to be

done for several times to maturity. Nevertheless, the approach by Barone-Adesi and Whaley yields a direct

and highly tractable semi-analytical solution (29), which makes it rather appealing and suitable for practical

uses (by contrast, financial practitioners often consider the finite difference method not so straightforward to

implement, as it requires the generation of a mesh of nodes and it entails solving a set of linear systems).

9
 β = 1.33, δ = 3.0 β = 1.66, δ = 1.0
τ S ErrBAW CP U T imeBAW ErrF D CP U T imeF D ErrBAW CP U T imeBAW ErrF D CP U T imeF D
−3 −3 −1 −2 −3 −3
80 4.86 × 10 2.85 × 10 s 4.48 × 10 1.44 × 10 s 8.17 × 10 2.35 × 10 s 1.20 × 10 0
1.46 × 10−2 s
90 2.60 × 10−3 2.87 × 10−3 s 3.92 × 10−1 1.44 × 10−2 s 3.49 × 10−3 2.16 × 10−3 s 7.03 × 10−1 1.46 × 10−2 s
100 9.29 × 10−4 2.75 × 10−3 s 3.41 × 10−1 1.44 × 10−2 s 5.98 × 10−4 2.02 × 10−3 s 5.04 × 10−1 1.46 × 10−2 s
0.25
110 3.66 × 10−4 2.73 × 10−3 s 1.92 × 10−1 1.44 × 10−2 s 1.34 × 10−3 2.43 × 10−3 s 2.25 × 10−1 1.46 × 10−2 s
120 1.34 × 10−3 2.78 × 10−3 s 1.09 × 10−1 1.44 × 10−2 s 2.46 × 10−3 2.15 × 10−3 s 1.08 × 10−1 1.46 × 10−2 s
80 9.02 × 10−3 2.75 × 10−3 s 2.09 × 10−1 1.44 × 10−2 s 1.24 × 10−2 2.14 × 10−3 s 4.73 × 10−1 1.46 × 10−2 s
90 6.16 × 10−3 2.78 × 10−3 s 2.06 × 10−1 1.44 × 10−2 s 7.10 × 10−3 2.18 × 10−3 s 4.04 × 10−1 1.46 × 10−2 s
100 3.92 × 10−3 2.65 × 10−3 s 1.94 × 10−1 1.44 × 10−2 s 3.39 × 10−3 2.13 × 10−3 s 3.45 × 10−1 1.46 × 10−2 s
0.5
110 2.07 × 10−3 2.64 × 10−3 s 1.31 × 10−1 1.44 × 10−2 s 6.79 × 10−4 2.09 × 10−3 s 1.86 × 10−1 1.46 × 10−2 s
120 5.68 × 10−4 2.62 × 10−3 s 8.83 × 10−2 1.44 × 10−2 s 1.22 × 10−3 2.13 × 10−3 s 9.89 × 10−2 1.46 × 10−2 s
80 1.71 × 10−2 2.53 × 10−3 s 1.05 × 10−1 1.44 × 10−2 s 2.10 × 10−2 1.63 × 10−3 s 2.05 × 10−1 1.46 × 10−2 s
90 1.34 × 10−2 2.58 × 10−3 s 9.63 × 10−2 1.44 × 10−2 s 1.45 × 10−2 1.67 × 10−3 s 2.04 × 10−1 1.46 × 10−2 s
100 1.03 × 10−2 2.65 × 10−3 s 8.76 × 10−2 1.44 × 10−2 s 9.58 × 10−3 1.60 × 10−3 s 1.93 × 10−1 1.46 × 10−2 s
1
110 7.62 × 10−3 2.58 × 10−3 s 6.72 × 10−2 1.44 × 10−2 s 5.72 × 10−3 1.59 × 10−3 s 1.29 × 10−1 1.46 × 10−2 s
120 5.37 × 10−3 2.51 × 10−3 s 5.09 × 10−2 1.44 × 10−2 s 2.73 × 10−3 1.61 × 10−3 s 8.64 × 10−2 1.46 × 10−2 s
80 3.05 × 10−2 2.33 × 10−3 s 5.64 × 10−2 1.44 × 10−2 s 3.57 × 10−2 2.34 × 10−3 s 1.09 × 10−1 1.46 × 10−2 s
90 2.56 × 10−2 2.31 × 10−3 s 4.69 × 10−2 1.44 × 10−2 s 2.75 × 10−2 2.33 × 10−3 s 1.01 × 10−1 1.46 × 10−2 s
100 2.14 × 10−2 2.25 × 10−3 s 3.92 × 10−2 1.44 × 10−2 s 2.08 × 10−2 2.20 × 10−3 s 9.26 × 10−2 1.46 × 10−2 s
2
110 1.77 × 10−2 2.25 × 10−3 s 3.31 × 10−2 1.44 × 10−2 s 1.54 × 10−2 2.22 × 10−3 s 7.32 × 10−2 1.46 × 10−2 s
120 1.45 × 10−2 2.25 × 10−3 s 2.79 × 10−2 1.44 × 10−2 s 1.09 × 10−2 2.20 × 10−3 s 5.77 × 10−2 1.46 × 10−2 s
80 3.98 × 10−2 3.04 × 10−3 s 3.09 × 10−2 1.44 × 10−2 s 4.60 × 10−2 3.56 × 10−3 s 7.43 × 10−2 1.46 × 10−2 s
90 3.43 × 10−2 3.09 × 10−3 s 2.55 × 10−2 1.44 × 10−2 s 3.66 × 10−2 3.21 × 10−3 s 6.16 × 10−2 1.46 × 10−2 s
100 2.94 × 10−2 3.03 × 10−3 s 2.09 × 10−2 1.44 × 10−2 s 2.89 × 10−2 3.14 × 10−3 s 4.13 × 10−2 1.46 × 10−2 s
3
110 2.51 × 10−2 2.96 × 10−3 s 1.83 × 10−2 1.44 × 10−2 s 2.25 × 10−2 3.23 × 10−3 s 4.13 × 10−2 1.46 × 10−2 s
120 2.12 × 10−2 2.92 × 10−3 s 1.58 × 10−2 1.44 × 10−2 s 1.70 × 10−2 3.05 × 10−3 s 3.27 × 10−2 1.46 × 10−2 s
Table 1: Numerical approximation of the option price, r = 0.08, q = 0.12, E = 100.

β = 1.33, δ = 3.0 β = 1.66, δ = 1.0

τ ErrBAW CP U T imeBAW ErrF D CP U T imeF D ErrBAW CP U T imeBAW ErrF D CP U T imeF D
−2 −3 −2 −2 −2 −3 −2
0.25 2.31 × 10 2.75 × 10 s 1.11 × 10 1.44 × 10 s 1.45 × 10 2.16 × 10 s 3.89 × 10 1.46 × 10−2 s
0.5 2.04 × 10−2 2.24 × 10−3 s 7.56 × 10−2 1.44 × 10−2 s 1.72 × 10−2 2.13 × 10−3 s 4.76 × 10−2 1.46 × 10−2 s
1 1.45 × 10−2 2.51 × 10−3 s 4.14 × 10−2 1.44 × 10−2 s 1.19 × 10−2 2.54 × 10−3 s 1.20 × 10−1 1.46 × 10−2 s
2 3.96 × 10−3 3.32 × 10−3 s 2.48 × 10−2 1.44 × 10−2 s 5.58 × 10−3 3.39 × 10−3 s 4.17 × 10−2 1.46 × 10−2 s
3 4.07 × 10−3 3.04 × 10−3 s 3.70 × 10−2 1.44 × 10−2 s 2.41 × 10−3 3.21 × 10−3 s 3.03 × 10−2 1.46 × 10−2 s
Table 2: Numerical approximation of the early exercise boundary, r = 0.08, q = 0.12, E = 100.

10
 β = 2.33, δ = 0.2 β = 2.66, δ = 0.1
τ S ErrBAW CP U T imeBAW ErrF D CP U T imeF D ErrBAW CP U T imeBAW ErrF D CP U T imeF D
80 8.33 × 10−3 2.96 × 10−3 s 1.58 × 10−0 1.43 × 10−2 s 7.61 × 10−3 2.40 × 10−3 s 1.24 × 10−0 1.44 × 10−2 s
90 1.12 × 10−2 2.99 × 10−3 s 7.84 × 10−1 1.43 × 10−2 s 1.11 × 10−2 2.58 × 10−3 s 6.91 × 10−1 1.44 × 10−2 s
100 1.47 × 10−2 2.87 × 10−3 s 5.36 × 10−1 1.43 × 10−2 s 1.42 × 10−2 2.51 × 10−3 s 5.04 × 10−1 1.44 × 10−2 s
0.25
110 1.89 × 10−2 2.87 × 10−3 s 2.35 × 10−1 1.43 × 10−2 s 1.76 × 10−2 2.46 × 10−3 s 2.25 × 10−1 1.44 × 10−2 s
120 2.38 × 10−2 2.88 × 10−3 s 1.14 × 10−1 1.43 × 10−2 s 2.16 × 10−2 2.45 × 10−3 s 1.20 × 10−1 1.44 × 10−2 s
80 1.80 × 10−2 2.75 × 10−3 s 5.48 × 10−1 1.43 × 10−2 s 1.91 × 10−2 3.40 × 10−3 s 4.18 × 10−1 1.44 × 10−2 s
90 2.26 × 10−2 2.79 × 10−3 s 4.48 × 10−1 1.43 × 10−2 s 8.59 × 10−3 3.33 × 10−3 s 3.83 × 10−1 1.44 × 10−2 s
100 2.77 × 10−2 2.66 × 10−3 s 3.80 × 10−1 1.43 × 10−2 s 3.80 × 10−3 3.25 × 10−3 s 3.45 × 10−1 1.44 × 10−2 s
0.5
110 3.33 × 10−2 2.71 × 10−3 s 2.02 × 10−1 1.43 × 10−2 s 8.87 × 10−4 3.24 × 10−3 s 1.99 × 10−1 1.44 × 10−2 s
120 3.95 × 10−3 2.70 × 10−3 s 1.08 × 10−1 1.43 × 10−2 s 1.00 × 10−3 3.44 × 10−3 s 1.17 × 10−1 1.44 × 10−2 s
80 3.93 × 10−2 2.74 × 10−3 s 2.10 × 10−1 1.43 × 10−2 s 2.04 × 10−2 3.33 × 10−3 s 1.52 × 10−1 1.44 × 10−2 s
90 4.65 × 10−2 2.74 × 10−3 s 2.26 × 10−1 1.43 × 10−2 s 1.38 × 10−2 3.53 × 10−3 s 1.84 × 10−1 1.44 × 10−2 s
100 5.41 × 10−2 2.71 × 10−3 s 2.20 × 10−1 1.43 × 10−2 s 9.00 × 10−3 3.46 × 10−3 s 1.90 × 10−1 1.44 × 10−2 s
1
110 6.21 × 10−2 2.72 × 10−3 s 1.45 × 10−1 1.43 × 10−2 s 5.45 × 10−3 3.83 × 10−3 s 1.32 × 10−1 1.44 × 10−2 s
120 7.03 × 10−2 2.88 × 10−3 s 9.59 × 10−2 1.43 × 10−2 s 2.75 × 10−3 3.80 × 10−3 s 9.16 × 10−2 1.44 × 10−2 s
80 5.58 × 10−2 2.35 × 10−3 s 9.80 × 10−2 1.43 × 10−2 s 3.43 × 10−2 3.29 × 10−3 s 6.56 × 10−2 1.44 × 10−2 s
90 3.28 × 10−2 2.29 × 10−3 s 1.03 × 10−1 1.43 × 10−2 s 2.58 × 10−2 3.49 × 10−3 s 7.37 × 10−2 1.44 × 10−2 s
100 2.73 × 10−2 2.31 × 10−3 s 1.02 × 10−1 1.43 × 10−2 s 1.94 × 10−2 3.35 × 10−3 s 7.64 × 10−2 1.44 × 10−2 s
2
110 2.10 × 10−2 2.27 × 10−3 s 7.81 × 10−2 1.43 × 10−2 s 1.43 × 10−2 4.32 × 10−3 s 5.93 × 10−2 1.44 × 10−2 s
120 1.50 × 10−2 2.31 × 10−3 s 5.95 × 10−2 1.43 × 10−2 s 1.04 × 10−2 3.38 × 10−3 s 4.52 × 10−2 1.44 × 10−2 s
80 4.56 × 10−2 3.46 × 10−3 s 6.54 × 10−2 1.43 × 10−2 s 4.43 × 10−2 3.36 × 10−3 s 4.45 × 10−2 1.44 × 10−2 s
90 3.53 × 10−2 3.41 × 10−3 s 5.78 × 10−2 1.43 × 10−2 s 3.47 × 10−2 3.39 × 10−3 s 3.82 × 10−2 1.44 × 10−2 s
100 2.75 × 10−2 3.40 × 10−3 s 5.24 × 10−2 1.43 × 10−2 s 2.72 × 10−2 3.31 × 10−3 s 3.46 × 10−2 1.44 × 10−2 s
3
110 2.08 × 10−2 3.43 × 10−3 s 4.04 × 10−2 1.43 × 10−2 s 2.12 × 10−2 3.28 × 10−3 s 2.78 × 10−2 1.44 × 10−2 s
120 1.57 × 10−2 3.38 × 10−3 s 3.08 × 10−2 1.43 × 10−2 s 1.64 × 10−2 3.21 × 10−3 s 2.22 × 10−2 1.44 × 10−2 s
Table 3: Numerical approximation of the option price, r = 0.08, q = 0.12, E = 100.

β = 2.33, δ = 0.2 β = 2.66, δ = 0.1

τ ErrBAW CP U T imeBAW ErrF D CP U T imeF D ErrBAW CP U T imeBAW ErrF D CP U T imeF D
−1 −3 −2 −2 −1 −3 −3
0.25 1.19 × 10 2.88 × 10 s 3.89 × 10 1.43 × 10 s 1.33 × 10 2.45 × 10 s 6.18 × 10 1.44 × 10−2 s
0.5 1.60 × 10−1 2.71 × 10−3 s 4.76 × 10−2 1.43 × 10−2 s 1.54 × 10−2 3.23 × 10−3 s 1.06 × 10−1 1.44 × 10−2 s
1 2.18 × 10−1 2.58 × 10−3 s 1.18 × 10−1 1.43 × 10−2 s 3.55 × 10−3 2.52 × 10−3 s 6.43 × 10−3 1.44 × 10−2 s
2 1.42 × 10−3 3.33 × 10−3 s 2.48 × 10−2 1.43 × 10−2 s 2.69 × 10−2 3.20 × 10−3 s 5.74 × 10−2 1.44 × 10−2 s
3 1.77 × 10−3 3.42 × 10−3 s 1.23 × 10−3 1.43 × 10−2 s 5.31 × 10−2 3.26 × 10−3 s 2.36 × 10−2 1.44 × 10−2 s
Table 4: Numerical approximation of the early exercise boundary, r = 0.08, q = 0.12, E = 100.

11
References

M. Abramowitz and I. A. Stegun. Handbook of Mathematical Functions with Formulas, Graphs, and Mathe-

matical Tables. Dover Publications, 1972.

D. Ahmadian and L. Ballestra. A numerical method to price discrete double barrier options under a constant

elasticity of variance model with jump diffusion. International Journal of Computer Mathematics, 1:1–19,

2014.

D. Amos. Algorithm 644: A portable package for Bessel functions of a complex argument and nonnegative

order. ACM Transactions on Mathematical Software (TOMS), 12:265–273, 1986.

D. E. Amos. Subroutine package for Bessel functions of a complex argument and nonnegative order. Technical

report, Sandia National Labs., Albuquerque, NM (USA), 1985.

L. V. Ballestra and G. Pacelli. The constant elasticity of variance model: Calibration, test and evidence from

the Italian equity market. Applied Financial Economics, 21:1479–1487, 2011.

G. Barone-Adesi and R. E. Whaley. Efficient analytic approximation of American option values. The Journal

of Finance, 42:301–320, 1997.

F. Black and M. Scholes. The pricing of options and corporate liabilities. Journal of Political Economy, 81:

637–659, 1973.

C. C. Chang, S. L. Chung, and R. C. Stapleton. Richardson extrapolation techniques for pricing American-

style options. Journal of Futures Markets, 27:791–817, 2007.

S.-L. Chung, P.-T. Shih, and W.-C. Tsai. Static hedging and pricing American knock-in put options. Journal

of Banking and Finance, 37:191 – 205, 2013.

J. Cox. Notes on option pricing I: Constant elasticity of diffusions. Unpublished paper, 1975.

J. C. Cox and S. A. Ross. The valuation of options for alternative stochastic processes. Journal of Financial

Economics, 3:145 – 166, 1976.

D. Davydov and V. Linetsky. Pricing and hedging path-dependent options under the CEV process. Manage-

ment Science, 47:949–965, 2001.

Y. Hsu, T. Lin, and C. Lee. Constant elasticity of variance (CEV) option pricing model: Integration and

detailed derivation. Mathematics and Computers in Simulation, 79:60 – 71, 2008.

J. MacBeth and L. Merville. Test of the Black-Scholes and Cox call option valuation models. Journal of

12
 Finance, 8:285–301, 1980.

J. P. V. Nunes. Pricing American options under the constant elasticity of variance model and subject to

bankruptcy. Journal of Financial and Quantitative Analysis, 44:1231–1263, 2009.

C. S. Pun and H. Wong. CEV asymptotics of American options. Journal of Mathematical Analysis and

Applications, 403:451–463, 2013.

A. Quarteroni, R. Sacco, and F. Saleri. Numerical Mathematics. Springer Verlag, 2007.

J. P. Ruas, J. C. Dias, and J. P. V. Nunes. Pricing and static hedging of American-style options under the

jump to default extended CEV model. Journal of Banking and Finance, 37:4059 – 4072, 2013.

M. Sankaran. On the non-central chi-squared distribution. Biometrika, 46:235–237, 1959.

M. Schroder. Computing the constant elasticity of variance option pricing formula. The Journal of Finance,

44:211–219, 1989.

D. Tavella and C. Randall. Pricing Financial Instruments: The Finite Difference Method. John Wiley &

Sons, 2000.

N. Thakoor, D. Tangman, and M. Bhuruth. A new fourth-order numerical scheme for option pricing under

the CEV model. Applied Mathematics Letters, 26:160 – 164, 2013.

N. Thakoor, D. Y. Tangman, and M. Bhuruth. Efficient and high accuracy pricing of barrier options under

the CEV diffusion. Journal of Computational and Applied Mathematics, 259:182 – 193, 2014.

P. Wilmott, S. Howison, and J. Dewynne. The Mathematics of Financial Derivatives. Cambridge University

Press, 1995.

H. Wong and J. Zhao. An artificial boundary method for American option pricing under the CEV model.

SIAM Journal on Numerical Analysis, 46:2183–2209, 2008.

H. Wong and J. Zhao. Valuing American options under the CEV model by Laplace-Carson transforms.

Operations Research Letters, 38:474–481, 2010.

M. Xu and C. Knessl. On a free boundary problem for an American put option under the CEV process.

Applied Mathematics Letters, 24:1191 – 1198, 2011.

J. Zhao and H. Wong. A closed-form solution to American options under general diffusion processes. Quan-

titative Finance, 12:725–737, 2012.

Y. Zhao and T. Chang. The differential algorithm for American put option with transaction costs under

CEV model. Journal of Systems Science and Information, 2:401–410, 2014.

13
• The problem of pricing American options under the CEV model is considered

• The method by Barone-Adesi and Whaley is extended to the CEV model

• A direct option pricing formula is obtained, so that computational lattices are avoided

• The practical computation aspects are dealt with

• The proposed method is accurate and computationally very fast