On a free boundary problem for an American put option under the CEV

process

Miao Xu, Charles Knessl

Department of Mathematics, Statistics and Computer Science, University of Illinois at Chicago, 851 S Morgan St, Chicago, IL
60607-7045

Abstract
We consider an American put option under the CEV process. This corresponds to a free boundary problem for
a PDE. We show that this free boundary satisfies a nonlinear integral equation, and analyze it in the limit of
small ρ = 2r/σ 2 , where r is the interest rate and σ is the volatility. We use perturbation methods to find that
the free boundary behaves differently for five ranges of time to expiry.

1. Introduction

The pricing and hedging of options has its origins in the Nobel prize winning work of Black, Scholes, and
Merton [2], who assume that the price of an underlying asset S(t) follows a geometric Brownian motion with
constant volatility. The price of a European call option at time t for an asset with price S, strike K, and
expiry T is then readily established, and is presented in terms of the normal distribution function. However,
there is sufficient empirical evidence [4] to suggest that in many cases the assumption of constant volatility does
not match well to the observed market data. Rather, evidence points out that the implied volatility, which is
obtained by equating the model price of an option to its market price and solving for the unknown volatility
parameter, varies with the strike price across a wide range of markets. This phenomenon is known as the
volatility smile or frown, depending on the shape of the curve, and is not captured by the Blacks-Scholes model
with a constant volatility. As a result, there have been various ideas as to how to modify and extend the basic
Black-Scholes framework, to account for this phenomenon. One of these is the constant elasticity of variance
(CEV) diffusion model, which was introduced by Cox and Ross [3] in the context of European options. Studies
using data from real markets [4], which include both equity and index options, suggest that the CEV models
are a better fit than the Black-Scholes as they lead to smaller smiles/frowns.
Other work on European options under a CEV process include Emanuel and Macbeth [5], Hu and Knessl [9]
and Lo, et. al. [13]. However, there exists little or no analytic work for the valuation of American options under
a CEV process. The analysis of these options are more difficult than the corresponding European options in
that the American options may be exercised prior to the expiration dates. Mathematically the American options
lead to partial differential equations (PDE) with free boundaries, which can only rarely be solved exactly. In
this paper, we apply asymptotic analysis to a CEV model to examine the behavior of the free boundary under
different scaling regimes for the time to expiry, in the limit of small ρ = 2r/σ 2 , where r is the interest rate, and
σ is the volatility. This limit has a small interest rate and/or large volatility, and is of particular relevance to
the financial status of the current economy. We will employ singular perturbation methods, including matched
asymptotic expansions. The main result is the derivation of a nonlinear integral equation that is satisfied by
the free boundary, from which we shall analyze its asymptotic structure for five different ranges of time. The
main results are summarized in section 2 and derived in section 3.

Email addresses: [email protected] (Miao Xu), [email protected] (Charles Knessl)

Preprint submitted to Elsevier January 21, 2011

 Asymptotic analysis and singular perturbation methods have been recently employed in the context of both
European and American options, and this work includes Knessl [10, 11], Howison [8], Kuske and Keller [12],
Addison, et al. [1], Evans, et al. [6], Fouque, et al. [7], and Widdicks, et al. [14].

2. Problem Statement and Summary of Results

We let P (S, T0 ) denote the price of an American put option for an asset with price S at some time T0 prior
to expiry TF . We assume that S satisfies the stochastic differential equation
√
dS = µS dt + σ S dWt . (1)

where Wt is a standard Brownian motion, σ is the volatility of the underlying asset, and µ = r is the risk-free
interest rate. We note that unlike Black-Scholes, this model only guarantees non-negativity of S (S ≥ 0), so
the chance of absorption at 0, i.e., bankruptcy, occurs with positive probability.
Introducing the new variables
σ2 2r
t= (TF − T0 ), ρ = 2 , (2)
2 σ
we find that P satisfies the following boundary value problem

Pt = SPSS + ρSPS − ρP ; t > 0, S > α(t) (3)

P (S, 0) = max(K − S, 0) (4)

P (α(t), t) = K − α(t), PS (α(t), t) = −1, P (∞, t) = 0, (5)
where α(t) is the free boundary in the new time variable. We also have P (S, t) = K − S for 0 < S < α(t), and
α(0) = K. For S ≤ α(t) the option should be exercised, and for S > α(t) it should be held.
We convert (3) − (5) into an integral equation by first making a change in coordinates, letting

P (S, t) = K − S + P̃ (V, t), V = S − α(t) (6)

where V ≥ 0. Then P̃ satisfies the PDE

P̃t − α0 (t)P̃V = [V + α(t)]P̃V V + ρ[V + α(t)]P̃V − ρK − ρP̃ ; V, t > 0 (7)

with the initial and boundary conditions

P̃ (V, 0) = V, P̃ (0, t) = P̃V (0, t) = 0. (8)

We introduce the Laplace transform

Z ∞
Q(θ, t) = e−θV P̃ (V, t) dV. (9)
0

Using (9) in (7) and (8) then yields

ρK
Qt + (θ2 + ρθ)Qθ = [α0 (t)θ + (θ2 + ρθ)α(t) − (2θ + 2ρ)]Q − . (10)
θ
with the initial condition Q(θ, 0) = 1/θ2 . Using the method of characteristics (the characteristics are c =
θe−ρt /(θ + ρ) where c is a constant) it can be shown that the only acceptable solution to (10) is

Kρ α(t)θ ∞ 1
Z    
−1 θ+ρ z
Q(θ, t) = 2 e exp −ρzα t + ρ log dz. (11)
θ θ/ρ z + 1 θ z+1

We note that the most general solution to (11) corresponds to replacing the upper limit on the integral by the
arbitrary function f (c) where c indexes the family of characteristics. But we must take f (c) = ∞ in order for
the integral to decay as θ → ∞, which must happen to offset the exponentially growing factor eα(t)θ in (11).
The next result readily follows.
Theorem 1. The option price P (S, t) for the CEV model has the integral representation
2
 Z
1
P (S, t) = K − S + eθV Q(θ, t) dθ, (12)
2πi Br

where <(θ) > 0 on the Bromwich contour, and Q(θ, t) is given by (11).
Moreover, after setting t = 0 and using α(0) = K in (11), it follows that α(t; ρ) satisfies the nonlinear
integral equation (IE):
Z ∞
e−Kθ
   
1 −1 θ+ρ z
= exp −ρzα ρ log dz. (13)
Kρ θ/ρ z + 1 θ z+1
In the next section we use asymptotic methods to analyze this IE for five different scales of time t, in the
limit of small ρ. We let ρ = e−λ so that λ = − log ρ → ∞. The final results for the free boundary α(t; ρ) are
listed below, and we sketch the derivations in section 3.

(i) t = ω/λ = O(λ−1 ), 0 < ω < K:

√ √
√ √ 4πK 2 ω
 
log λ ω − Kω 1 Kω − ω
α(t; ρ) = ( ω − 2
K) + + log √ + o(λ−1 ), (14)
λ 2 λ 2 K − Kω

(ii) t = K/λ + O(λ−2 ), λ2 t − λK = λ(ω − K) = Λ:

1
α(t; ρ) ∼ F(Λ), (15)
λ2
where F(·) satisfies the nonlinear IE
∞
e−Kν
Z   
1 1
= exp −ξF −νK 2 − dξ, −∞ < ν < ∞. (16)
K 0 ξ ξ

For Λ → ±∞, we have

Λ2 Λ Λ
F(Λ) ∼ + log(−Λ) − log(8πK 3 ), Λ → −∞ (17)
4K 4 4
  
−γ 1 Λ
F(Λ) ∼ Λe exp − exp , Λ → +∞ (18)
K K
where γ is the Euler constant.
(iii) t = ω/λ = O(λ−1 ), K < ω < ∞:
 
ω − K −γ 1 K/ω−1
α(t; ρ) ∼ e exp − ρ , (19)
λ K

(iv) t = O(1), 0 < t < ∞:    

1 1
α(t; ρ) ∼ te−γ exp − + e−K/t , (20)
2 ρK
(v) t = v/ρ = O(ρ−1 ), v > 0:
   
1 1 1
α(t; ρ) ∼ e−γ exp (1 − e −v
) exp − . (21)
ρ ev − 1 ρK

We note that in four of the five cases the expression for α(t; ρ) is completely explicit, and only in case (ii)
must we solve a nonlinear IE, which is somewhat simpler than the one in (13). We can easily compute P (S, t)
as t → ∞, which corresponds to the perpetual American option, where the problem reduces to solving an
ordinary differential equation. Setting P (S, ∞) = P ∞ (S) and using α(∞) to denote the limiting value of the
free boundary, we obtain from (3) − (5)
Z ∞
1 −zρS
P ∞ (S) = Keρα(∞) e dz, (22)
1 z2
3
where α(∞) satisfies Z ∞
1 −ρα(∞)z
Kρ e dz = e−ρα(∞) . (23)
1 z
For ρ → 0 we have  
1 −γ 1
α(∞) = e exp − [1 + O(ρ)], (24)
ρ ρK
which is exponentially small.

3. Analysis

3.1. Analysis for t = ω/λ, 0 < ω < K

We first examine (13) on the t = O(λ−1 ) scale, for small ρ. Recalling that λ = − log ρ, we let

λ(β + x)
θ = λβ, z = , α(t; ρ) ∼ α0 (ω; ρ) (25)
ρ

for ω = (− log ρ)t = O(1). Then (13) can be approximated by

∞
eλ
Z
1
= eλΦ(x;β,ρ) [1 + O(e−λ )] dx, (26)
K 0 β+x
x
where Φ(x; β, ρ) = Kβ − (β + x)α0 ( β(x+β) ; ρ). For large λ and fixed β, we evaluate the right hand side of (26)
by an implicit form of the Laplace method, assuming for now that there is a saddle point where
   
∂Φ x 1 0 x
= −α0 − α = 0. (27)
∂x β(x + β) x + β 0 β(x + β)

Let us denote x = x∗ (β) as the solution to (27). It follows that at x = x∗ , Φ ∼ 1 so that

 
x∗
1 = Kβ − (β + x∗ )α0 . (28)
β(x∗ + β)

x∗ α00 (ω)
Now let ω = β(x∗ +β) . Then from (27) we have β = ωα00 (ω)−α0 (ω) which we use to eliminate β in (28) to obtain
the ODE

[1 − α00 (ω)][ωα00 (ω) − α0 (ω)] = Kα00 (ω). (29)

Dividing (29) by 1 − α00 (ω), we recognize this as the Clairaut equation. The solutions consist of a one-parameter
family of lines and the singular solution √ √
α0 (ω) = ( ω − K)2 (30)
KC
which is the envelope of this family. The linear solutions α0 (ω) = ωC − 1−C must be rejected, since these lead
to α0 (0) 6= K. The above analysis applies only for 0 < ω < K, since the solution (30) vanishes as ω approaches
K. Hence we expect different asymptotics for ω ≈ K.
We next analyze some higher order terms in the expansion of α0 . We evaluate (26) by using the Laplace
method, which gives
s
eλ 1 2π
= eλΦ(x∗ ;β,ρ) [1 + O(λ−1 )], (31)
K β + x∗ −λΦxx (x∗ ; β, ρ)
and expand α0 as
log λ 1
α0 (ω; ρ) = α0 (ω) + α1 (ω) + α2 (ω) + o(λ−1 ). (32)
λ λ
p
In order to balance the two sides of (31), we need α1 to cancel the 1/λ factor. Hence,
  
1 x∗
√ exp −(β + x∗ )(log λ) α1 = 1. (33)
λ β(x∗ + β)
4
Writing (33) in terms of ω we obtain
1 √
(ω − Kω).
α1 (ω) = (34)
2
To find the second order term α2 , we balance the O(1) terms in (31), so that
r   
1 1 2π x∗
= exp −(β + x∗ )α2 . (35)
K β + x∗ −Φxx β(x∗ + β)
1 3 −3 3
It can be shown that Φxx (x∗ ; β, ρ) ∼ − 12 K 2 β 2 x∗ 2 (x∗ + β)− 2 and then
√
4πK 2 ω
 
Kω − ω
α2 (ω) = log √ . (36)
2 K − Kω
With (30), (32), (34), and (36) we have established (14).

3.2. Analysis for t = ω/λ, ω ≈ K

We return to (13) and introduce the scaling

1 ν Λ
θ = λβ, β = + , ω = λt = K + (37)
K λ λ
with    
1 1 K
α(t; ρ) = 2
F(Λ; ρ) = 2 F λ2 t − ;ρ . (38)
λ λ λ
Then we have e−θK = ρe−Kν . By setting z = (θ + y)/ρ in (13) this equation becomes
Z ∞   
1 −Kν 1 1 1
e = exp −(θ + y)α − + O(ρ); ρ dy. (39)
K 0 θ+y+ρ θ θ+y

We use (37) and (38), scale y as y = λ2 ξ and note that ρ = e−λ is exponentially small, thus obtaining
       
2 1 1 −λ 2 K 1 2 1 −2 2 1
λ α − + O(e ); ρ = λ α − 2 νK + + o(λ ); ρ ∼ F −νK − (40)
θ θ+y λ λ ξ ξ

where F(Λ) is the leading term in an expansion of F(Λ; ρ). Letting ρ → 0 (λ → ∞) we obtain the limiting IE
in (16). It does not seem possible to solve (16) explicitly for F(Λ). But we can infer the behavior as Λ → −∞
(ν → +∞) by evaluating the integral in (16) by an implicit Laplace type expansion, similarly to what we did
in section 3.1. This will verify the asymptotic matching between the ω-scale (for ω < K) and the Λ-scale, and
leads to (17). Now consider the limit Λ → +∞. For ν < 0, we rewrite (16) as
∞ Z 1 
eK|ν| 1 F(|ν|K 2 (1 − u)) 1 F(|ν|K 2 )
Z      
1 1 2 1
= exp − F(|ν|K − η) dη + exp − − exp −
K |ν|K 2 η η 0 u u |ν|K 2 u |ν|K 2
∞
e−v
Z
+ dv.
F (|ν|K 2 ) v
|ν|K 2

(41)

Here we broke up the integral over (0, ∞) into the two ranges (0, |ν|K 2 ) and (|ν|K 2 , ∞) and made some
Λ2
elementary substitutions. Now, for Λ → −∞ we have F(Λ) ∼ 4K so that the first integral in the right hand
side of (41) will vanish as ν → −∞. If F(Λ) → 0 as Λ → +∞ the second integral in (41) will also vanish, and
the third may be approximated by using
Z ∞ −v
e
dv = − log ε − γ + O(ε), ε → 0+ . (42)
ε v

5
Hence (41) can be replaced by the asymptotic relation

eK|ν| F(|ν|K 2 )
 
∼ − log −γ (43)
K |ν|K 2

which upon exponentiation leads to the asymptotic result given in (18), for F(Λ) as Λ → ∞.

3.3. Analysis for t = ω/λ, K < ω < ∞

In the remaining time ranges, α(t; ρ) will be exponentially small as ρ = e−λ → 0, and our analysis of (13)
will rely heavily on the asymptotic form in (42). We let z = Z/ρ in (13) to obtain
Z ∞
e−Kθ
   
1 θ+ρ Z
= exp −Zα ρ−1 log dZ. (44)
Kρ θ Z +ρ θ Z +ρ

Now we scale Z = λz∗ and θ = λθ∗ , let α(t; ρ) = α̃(λt; ρ) and note that in sections 3.1 and 3.2 we have already
characterized α̃(λt; ρ) for λt = ω < K and ω ∼ K. We also simplify the argument of α(·) in (44) using
 h      
1 ρ  ρ i 1 1 1 1
α log 1 + − log 1 + =α − + O(ρ) = α̃ − + O(e−λ λ) . (45)
ρ θ Z θ Z θ∗ Z∗

When Z = θ we have z∗ = θ∗ and we rewrite the integral in (44) by splitting the range of integration into
z∗ ∈ (θ∗ , θ∗ /[1 − Kθ∗ ]) and z∗ ∈ (θ∗ /[1 − Kθ∗ ], ∞), thus obtaining
θ∗ Z ∞ !
e−Kλθ∗
Z 1−Kθ   
∗ 1 1 1
∼ + exp −λz∗ α̃ − dz∗ . (46)
Kρ θ∗ θ∗
1−Kθ
z∗ θ∗ z∗
∗

√ √
In the first range α̃(ω) ∼ ( K − ω)2 and the first integral will be o(1) as λ → ∞, since θ∗−1 − z∗−1 ≤ K when
z∗ ≤ θ∗ /[1 − Kθ∗ ]. In the second integral α̃ will be exponentially small and the main contribution will come
from very large values of z∗ , where roughly z∗ = O(α̃−1 ). Then we write α̃(θ∗−1 − z∗−1 ) ∼ α̃(θ∗−1 ) and using (42)
we conclude that
Z ∞
e−Kλθ∗
       
1 1 1 θ∗
∼ exp −λz∗ α̃ dz∗ ∼ − log λα̃ − γ − log , (47)
Kρ θ∗
1−Kθ
z∗ θ∗ θ∗ 1 − Kθ∗
∗

with an error that is o(1) as λ → ∞. Then exponentiating (47) and replacing θ∗ by ω −1 we obtain the asymptotic
result in (19).
For ω → K we note that ρω/K−1 = ρ−1 e−K/t = ρ−1 exp [−λK/(K + Λ/λ)] = ρ−1 exp −λ + Λ/K + O(λ−1 ) ∼
 

exp (Λ/K) and (ω − K)/λ = Λ/λ2 , which can be used to verify the asymptotic matching between the Λ-scale
and the ω-scale for ω > K, in the intermediate limit where ω ↓ K and Λ → ∞.

3.4. Analysis for t = O(1), 0 < t < ∞

Next we consider times t = O(1). We scale z = θw/ρ. Since we again expect α(t; ρ) to be very small we
assume a “WKB-type” ansatz of the form
 
1
α(t; ρ) ∼ g(t) exp − f (t) . (48)
ρ

Expanding α(t; ρ) in (13) for fixed θ and ρ → 0, and noting that

1h  ρ  ρ i 1 1 1 ρ 
2 ρ

log 1 + − log 1 + = − − + O ρ , ,
ρ θ θw θ θw 2 θ2 w2
we have
 h          
1 ρ  ρ i 1 1 1 1 0 1
−θwα log 1 + − log 1 + = −θwg exp − f + 2f + O(ρ)
ρ θ θw θ ρ θ 2θ θ

6
 −1 −1
h f (θ
i − (θw)
Here we also used ) ∼ f (θ−1 ), since w will be scaled to be exponentially large. Then setting
ε = θg θ1 exp − ρ1 f θ1 exp 2θ12 f 0 θ1 , scaling w = ε−1 u and using (42), (13) asymptotically becomes




e−Kθ
      
1 1 1 1 1
= f − γ − log θg − 2 f0 + o(1). (49)
Kρ ρ θ θ 2θ θ

From the O(ρ−1 ) terms in (49) we conclude that f (1/θ) = K −1 e−Kθ and then the O(1) terms determine g(·)
from     
1 1 1
θg = e−γ exp − 2 f 0 .
θ 2θ θ
The above along with (48) establishes the asymptotic result in (20). The asymptotic matching between (19)
and (20) is immediate, since ρK/ω−1 = ρ−1 e−K/t , and (ω − K)/λ ∼ ω/λ = t as ω → ∞.

3.5. Analysis for t = v/ρ = O(ρ−1 ), v > 0

We assume that time to expiry for the option is large, with t = v/ρ = O(ρ−1 ). On this time scale we assume
that  
1 1
α(t; ρ) ∼ exp − A(v), (50)
ρ ρK
where A(·) will be determined from (13). After scaling θ = ρW , (13) becomes
Z ∞
e−KρW
   
1 1
− ρK W +1 z
∼ exp −ze A log dz. (51)
Kρ W z+1 W z+1

The major contribution to the integral in (51) will once more come from large values of z, so we approximate
     
W +1 z W +1
A log ∼ A log , (52)
W z+1 W

and then using (52) is (51) along with (42) leads to

e−KρW
   
1
− ρK W +1
∼ − log e A log − γ − log(W + 1) (53)
Kρ W
   
1 W +1
= − γ − log (W + 1)A log + o(1). (54)
Kρ W

Expanding e−KρW = 1 − KρW + O(ρ2 ) we conclude that

  
W +1 1
A log = e−γ eW (55)
W W +1

which determines the function A(·) and establishes (21).

Finally we verify the asymptotic matching between (20) and (21). For v → 0 we have (ev − 1)−1 =
v − 1/2 + O(v) and 1 − e−v ∼ v = ρt. For t → ∞ we have e−K/t = 1 − K/t + O(t−2 ) = 1 − (Kρ)/v + O(ρ2 )
−1

so that −(1/2 + 1/(ρK))e−K/t ∼ −v −1 + 1/2 and the matching follows. As v → ∞ we have A(v) → e−γ and
thus the expansion in (50) agrees with the small ρ expansion of α(∞; ρ), as given in (24).

4. Discussion and Extensions

To summarize, we have given several asymptotic formulas for a free boundary problem. In contrast to
the Black-Scholes model in the same asymptotic limit, the free boundary α(t) moves from S = K to S ≈ 0
on logarithmically small time scales where t = O((log(1/ρ))−1 ). For the BS model this movement occurs on
logarithmically large time scales, with t = O(log(1/ρ)). Note however that the basic expression in (14), where
the boundary decreases from S = K to S = 0, is similar to the corresponding one for the BS model (see [1],
[9]). For times where the free boundary approaches zero, the behavior of the CEV model is much different from
the BS model, as for the former α(t) becomes exponentially small (see (19) - (21)) while for BS α(t) becomes
only O(ρ) as ρ → 0. This implies that for the CEV model it is advantageous to exercise the option only on
7
short time to expiry scales. If such times are O(1) or O(ρ−1 ) then the put option should be exercised only if
the stock price S is very small.
The present model, where S is governed by (1) is sometimes called the square root process and a more
general CEV model corresponds to the SDE dS = rS dt + σS β/2 dW . Here β is known as the elasticity factor,
so that β = 2 for BS and β = 1 for the model here. It seems that the integral equation approach here works
only for these two special cases. We could, with further analysis, use our asymptotic results for α(t) in (11)
and expand the integrals in (11) and (12) to obtain asymptotic results for the option value P (S, t), for ρ → 0.
Preliminary results show that for each of the five time ranges we will furthermore get different expansions for
P (S, t) for several ranges of S. We are also investigating a direct singular perturbation approach to analyzing
(3) − (5) for ρ → 0, and reconciling our results with the IE approach; the perturbation method should extend
to general elasticity factors β.
We comment that while here we took ρ → 0, the behavior of α(t; √ t→
√ ρ) for 0 with ρ
√= O(1) is√essentially
2
contained
√ in formula (14). Then we would replace ω by λt and use ( ω − K) ∼ K − 2 Kω and Kω − ω ∼
Kω. In this limit, however, the BS and CEV models behave similarly.
It is not immediately clear (at least to us) whether the CEV model call option leads to an interesting
problem. For the American call option under the BS model, early exercise is never advantageous. An important
difference between the European CEV (β = 1) and BS models is that the Green’s function for the BS model is
a proper probability distribution that integrates to one. In contrast the Green’s function for the CEV model is
deficient in mass, due to partial absorption at S = 0. For the put option analyzed here the basic PDE (3) does
not apply near S = 0, since α > 0. However, for a call option the partial absorption of the process may have to
be considered.
We have not at this point done numerical studies to compare with the asymptotics (i.e., determine how
small ρ must be). However, since α(t) is exponentially small for t > K/ log (1/ρ) it may become difficult to
locate the free boundary by purely numerical methods, for small values of ρ. Knowing asymptotic properties
of the type here may aid in the development of, e.g, multi-scale numerical methods, in addition to providing
qualitative information on the optimal exercise boundary.

References
[1] J.A Addison, S.D. Howison, J.R. King. Ray methods for free boundary problems. Quart. Appl. Math., 64(1):41-59, 2006.
[2] F. Black and M. Scholes. The pricing of options and corporate liabilities. J. Political Econom., 81:637-659, 1973.
[3] J. Cox and S. Ross. The valuation of options for alternative stochastic processes. J. Financial Econom., 3:145-166, 1976.
[4] D. Davydov and V. Linetsky. The valuation and hedging of barrier and lookback options under the CEV process. Management
Science, 47:949-965, 2001.
[5] D. Emanuel and J. Macbeth. Further results on the constant elasticity of variance call option pricing model. Journal of
Financial and Quantitative Analysis, 4: 53-54, 1982.
[6] J.D. Evans, R.A. Kuske, and J.B. Keller. American options on assets with dividends near expiry. Math. Finance, 12:219-237,
2002.
[7] J.P. Fouque, G. Papanicolaou, R. Sircar, and K. Solna. Singular perturbations in option pricing. SIAM J. Appl. Math.,
63(5):1648-1665, 2003.
[8] S.D. Howison. Matched asymptotic expansions in financial engineering. J. Engrg. Math., 53:385-406, 2005.
[9] F. Hu and C. Knessl. Asymptotics of barrier option pricing under the CEV process. Appl. Math. Finance, 17(3):261-300, 2010.
[10] C. Knessl. A note on a moving boundary problem arising in the American put option. Stud. Appl. Math., 107(2):157-183,
2001.
[11] C. Knessl. Asymptotic analysis of the American call option with dividends. European J. Appl. Math., 13(6):587-616, 2002.
[12] R.A Kuske and J.B. Keller. Optimal exercise boundary for an American put option. Appl. Math. Finance, 5:107-116, 1998.
[13] C.F. Lo, P.H. Yuen, and C.H. Hui. Pricing barrier options with square root process. Int. J. Theor. Appl. Finance, 5:805-818,
2001.
[14] M. Widdicks, P.W Duck, A.D. Andricopoulos, and D.P. Newton. The Black-Scholes equation revisited: Asymptotic expansions
and singular perturbations. Math. Finance, 15(2):373-391, 2005.