Asymptotic Expansions of the Lognormal Implied Volatility : A

Model Free Approach

Cyril Grunspan∗
ESILV, Department of Financial Engineering
92916 Paris La Défense Cedex
cyril.grunspan@devinci.fr

December 5, 2011

Abstract
We invert the Black-Scholes formula. We consider the cases low strike, large strike, short
maturity and large maturity. We give explicitly the first 5 terms of the expansions. A
method to compute all the terms by induction is also given. At the money, we have a
closed form formula for implied lognormal volatility in terms of a power series in call price.

JEL Classification G12 G13 C65

Keywords : Smile asymptotics, implied lognormal volatility.

1 Introduction
1.1 Overview
In a market with no arbitrage, the price of a call option can take two extreme values : its
“intrinsic value” which is equal to the payoff of the option (lower boundary value) and the
spot price (upper boundary value). For simplicity, we assume a market with no interest rate.
Otherwise, we would consider the forward price. We will consider here the case when the
call price is close to its boundary value and we will obtain in that case an approximation
of the corresponding lognormal implied volatility. This case happens in particular when the
maturity of the option is small. To be precise, in the case when T → 0 (resp. T → +∞),
we will obtain an asymptotic expansion of the implied lognormal volatility as a sum of terms
1 1
of the form λi lnj (λ) with j < i and λ = −   (resp. λ = −  )
C(T,K)−(S−K)+ S−C(T,K)
ln S ln S
where C(T, K) denotes the price of a call option with strike K, maturity T and spot price
S (Proposition 5). Note that here, the spot price S is present only to insure that the ratio
C(T, K) − (S − K)+ S − C(T, K)
(resp. )is with no-dimension. The important quantity
S S
is the “time-value” TV(T, K) := C(T, K) − (S − K)+ (resp. “covered call” CC(T, K) :=
S − C(T, K)) The computations involve no complicated formulas except may be a well known
∗
I wish to thank Yann Braouezec, Daniel Gabay, Peter Tankov, Joris Van der Hoeven.

1
Electronic copy available at: http://ssrn.com/abstract=1965977
asymptotic expansion for the incomplete Gamma function (Equation (16)). The interest of
such a formula is twofold. First, it gives quickly an easy approximation of the true implied
lognormal volatility. This can serve as a starting point for the calculus of the exact implied
lognormal volatility using a Newton method for instance. The formula can also be useful to
transform theoretical approximations of a call price into approximations of implied lognormal
volatility. Indeed, asymptotics of call prices can be obtained with the help of stochastic
differential equations of partial differential equations using perturbation methods. Then,
a transformation has to be made to obtain the implied lognormal volatility which is of a
fundamental interest for the practitioner.
All our work is based on a single inversion formula. Explicitly we invert the following
equation for λ  1 and β > 0 (see Note 2):
"N #
1 X 1
v β e− v αk v k + O v N +1 = eγ e− λ


(1)
k=0

The good framework for solving this problem (i.e. obtain v in terms of λ) is the theory of
transseries (see [6]). In the expansion of v in terms of λ coming from (1) it is important to go
up to order 5 (for us, order 0 is λ, order 1 is λ2 ln(λ), ... order 5 is λ3 ) to see α1 (see Lemma
1):

v = λ − βλ2 ln λ + γλ2 + β 2 λ3 ln2 (λ) + β 2 − 2βγ λ3 ln(λ) + γ 2 − βγ − α1 λ3 + o λ3

1.2 Basic definitions

In a Black-Scholes world, the dynamic of a stock (St ) is given by:

dSt = σLN St dWt ,

with initial value S at t = 0. The so-called lognormal volatility σLN is related to the price of
a call BS (S, K, T, σ) struck at K with maturity T by the Black-Scholes formula (See [2]):

BS (S, K, T, σ) = SN (d+ ) − KN (d− ) (2)

with Z x  2
1 u
N (x) = √ exp − du
2π −∞ 2
and
σ2 T
 
S
ln ± LN
K 2
d± = √
σLN T
To simplify matters, we have considered r = 0. Otherwise, we would consider the forward
price Ft = St ert instead of the spot price St . Following Ropper-Rutkowski ([4]), we set:

Definition 1 Let us denote by:

• TV(S, K, K, T ) (or simply TV(T, K) or TV) the time-value of a European call option
struck at strike K with maturity T : TV (S, K, T, σ) := BS (S, K, T, σ) − (S − K)+

• x := ln( K
S ) (the log-moneyness)

2
Electronic copy available at: http://ssrn.com/abstract=1965977
 √
• θ := σLN T (the square root of the time-variance)

The spot price S is assumed to be fixed by the market. We will consider the two following
√ √ K
cases: K is fixed and σ T is small (case 1) and σ T is fixed and is large (case 2). In
S
both cases, we will obtain a similar expression for the asymptotic expansion of the implied
lognormal volatility.

2 Asymptotic expansions of a European call option

First let us assume that x 6= 0.

2.1 Asymptotic expansions of a European call option for x 6= 0.

We note that the expression giving the time-value of a call-option in the case (θ  1 and x
fixed) is very similar to the case (|x|  1 and θ fixed).

Proposition 1 (Case 1.) Let N ∈ N. When θ → 0 and x fixed, the asymptotic expansion of
the time-value T V = C(T, K) − (S − K)+ of a call price is given at order N by:
 32 1 
x − N k
√ e− 2 2θ2 (−1)k x2 2θ2
      
TV 2θ 2 x2
+ O θ2N +5 e− 2θ2
X
4 π = e x2 ak (3)
|x| S x2 2k 8 x2
k=0

with
ak (z) := (2k + 1)!! fk (z) (4)
k
X zj
fk (z) := (5)
j! (2j + 1)!!
j=0

j
Y Y
and for j ∈ Z, (2j + 1)!! := (2l + 1) (with the convention := 1).
l=1 ∅
(Case 2.) Let N ∈ N. When |x| → +∞ (i.e., K → 0 or K → +∞) and θ fixed, the asymptotic
expansion of the time-value of a call price is given at order N by:
1 
x − N  2   2 2k
√ e− 2 2θ2 (−1)k
     
TV 2θ 2 θ 2θ x2
−2N −4 − 2θ2
X
2 2π = e x2 bk +O x e
θ S x2 2k 4 x2
k=0
(6)
with
k
zj
 
X
j k
bk (z) := (2k + 1)!! (−1) (7)
j (2j + 1)!!
j=0
.

(Case 3.) Let N ∈ N. When θ → +∞ and x fixed, the asymptotic expansion of the covered
call CC = S − C(T, K) of a call price is given at order N by:
1 − 1 N  2   k
√ − x CC (−1)k
  
8 2 8 x 8 2
e ( θ2 ) −2N −3 − θ8
X
πe 2 = ck +O θ e (8)
S θ2 2k 8 θ2
k=0

3
with

ck (z) := (2k − 1)!! gk (z) (9)

k
X zj
gk (z) := (10)
j! (2j − 1)!!
j=0

Note that gk0 (z) = fk−1 (z) and c0k (z) = ak−1 (z).

Proof. Case 1. For n ∈ N, we denote by ẽn the function defined by

u2 un
∀ u ∈ R, e−u = 1 − u +
− ... + (−1)n + ẽn (u) (11)
2! n!
Then, it is classical (properties of alternate series) that

un+1
∀ u > 0, |ẽn (u)| ≤ . (12)
(n + 1)!
Now, let us fix N ∈ N. We start from:
 
x2 2
√ − x2 TV
Z θ −1
2 ξ2
+ ξ4
2π e = e dξ (13)
S 0
 
K
with x = ln as before. This formula can be obtained by deriving the Black-Scholes
S
formula with respect to θ and then integrating the result (See [RR], Lemma 3.1). We have:

√ x TV
Z θ
− x2 ξ2
2π e− 2 = e 2ξ2 e− 8 dξ (14)
S 0
So, by (11),
N
√ X (−1)n θ x2 θ x2 ξ2
Z Z  
x TV − −
2π e− 2 = e 2ξ2 ξ 2n dξ + e 2ξ2 ẽN dξ
S n! 8n 0 0 8
n=0

x2
So, with the change of variables u := , we get:
2ξ 2

N
√ x TV X (−1)n |x|2n+1 +∞ −n− 3 −u
Z
2π e− 2 = √ u 2e du + RN (θ)
S n! 16n 2 2 x2
n=0 2θ 2

N n 2n+1 1 x2
 
X (−1) |x|
= √ Γ −n − , 2 + RN (θ) (15)
n! 16n 2 2 2 2θ
n=0

θ x2 ξ2
Z  
−
with RN (θ) := e 2ξ2 ẽn dξ.
0 8
We have:

4
 θ x2 ξ2
Z  
−
|RN (θ)| ≤ e 2ξ2 |ẽN | dξ
0 8
Z θ x2
 
− 1
≤ e 2ξ2 ξ 2(N +1) dξ
0 (N + 1)! 8N +1
Z θ x2
1 2(N +1) − 2ξ2
≤ ξ e dξ
(N + 1)! 8N +1 0
Z +∞  2 N +1
1 x |x| −(N +1)− 3 −u
≤ N +1
√ u 2 e du
(N + 1)! 8 x 2 2 2 2
2θ 2

|x|2N +3 3 x2
 
≤ √ Γ −N − , 2
2 2 (N + 1)! 16N +1 2 2θ

We recall the following asymptotic expansion valid for z → +∞ and m ∈ N (see Formula
6.5.32 in ([1])):

 
a−1 −z a−1 (a − 1)...(a − m)
Γ(a, z) = z e 1+ + ... + + γm (a, z) (16)
z zm

with

z a−1 e−z
 
γm (a, z) = O (17)
z m+1
3 x2
In particular, with a = −N − , z = 2 and m = 0, in the limit when θ → 0, we get:
2 2θ
" −N − 52 #
3 x2 x2
  2
− x2
Γ −N − , 2 = O e 2θ .
2 2θ 2θ2
So, when θ → 0,
 
x2
RN (θ) = O θ2N +5 e− 2θ2 (18)

x2
Moreover, for any n < N , we have by (16) with m = N − n, a = −n − 12 , z = 2θ2
:

5
 −n− 23
1 x2 x2
  
x2
Γ −n − , 2 = e− 2θ2 ×
2 2θ 2θ2
 
1 1 1




−n − − 1 −n − 2 2 − 1 ... −n − 2 − (N − n) 
× 1 + + ... + +

 1  N −n
x2 x2
2θ2 2θ2

1 x2
 
+ γN −n −n − , 2
2 2θ
 2  32 2
2θ − x2
= × e 2θ
x2
" n n+1 N #
2θ2 (−1) (2n + 3)!! 2θ2 (−1)N −n (2N + 1)!! 2θ2
 
× + + ... +
x2 2 (2n + 1)!! x2 2N −n (2n + 1)!! x2
1 x2
 
+ γN −n −n − , 2
2 2θ
 2  32 " N
k−n (2k + 1)!!
 2 k #
2θ x2 (−1) 2θ
× e− 2θ2
X
= 2 k−n
x 2 (2n + 1)!! x2
k=n
1 x2
 
+ γN −n −n − , 2 (19)
2 2θ

Moreover, when θ → 0, we have by (17):

−n− 23  
x2 2
− x2
e 2θ
1 x2
   
 2θ2 
γN −n −n − , 2 = O
  2 N −n+1

2 2θ  x


2θ2
 
x2
2N +5 − 2θ2
= O θ e (20)

Therefore, by (15), (18), (19), (20), we obtain:

x 3 N N k
√ e− 2 T V 2θ2 2 − x22 X (−1)n |x|2n X (−1)k−n (2k + 1)!! 2θ2
 
2π = e 2θ √
|x| S x2 n! 16n 2 2 2k−n (2n + 1)!! x2
n=0 k=n
 
x2
2N +5 − 2θ2
+ O θ e

Hence,
x  23 N  2   2 k
√ e− 2 2θ2 (−1)k
    
TV x2 x 2θ x2
− 2N +5 − 2θ2
X
4 π = e 2θ 2 ak +O θ e
|x| S x2 2k 8 x2
k=0

6
 with
k j
x2 x2
  
X 1
ak := (2k + 1)!!
8 j! (2j + 1)!! 8
j=0

which is exactly Proposition 1 - (3).

Case 2: θ is fixed and |x| → +∞. Set:

Z θ x2 ξ2
−
I(x) := e 2ξ2 e− 8 dξ. (21)
0

θ2
With the help of the change of variables z = , we have:
ξ2
Z +∞ x2
θ
I(x) = e− 2θ2 z h̃(z) dz
2 1
√  
16 2 8z 1
with h̃(z) := 3
f− 3 2
and fα (z) := z α e− z . By induction on n, we show that
θ 2 θ
n  
α−2n − z1
X n
∀n ∈ N, ∀z ∈ R∗+ , fα(n) (z) =z e [α − n + p]p z p (22)
p
p=0

(n) Qk−1
where fα is the nth derivative of fα and with by definition, [u]k := j=0 (u−j) for any real u
(N )
and integer k. In particular, for any (α, N ) ∈ R− × N∗ fixed, fα (z) is uniformly bounded in
z ∈ R∗+ . Therefore, h̃(N ) (z) is also uniformly bounded in z ∈ R∗+ . So h(N ) (z) is also uniformly
bounded in z ∈ R∗+ with h(z) := h̃(z + 1) (the function h is analytic on R+ ), i.e.,

∀N ∈ N ∃ MN ∈ R+ ∀z ∈ R+ , h(N ) (z) ≤ MN (23)

Let us fix N ∈ N. By Taylor-Lagrange, we get:

θ +∞ − x22 (z+1)
Z
I(x) = e 2θ h(z) dz
2 0
N
Z +∞ !
θ − x22 2
− x2 z
X h(k) (0) k
= e 2θ e 2θ z + RN +1 (z) dz
2 0 k!
k=0

z N +1
with RN +1 (z) ≤ MN +1 . So,
(N + 1)!

N
θ − x22 X h(k) (0) +∞ − x22 z k
Z Z +∞
θ − x22 x2
I(x) = e 2θ e 2θ z dz + e 2θ e− 2θ2 z RN +1 (z) dz
2 k! 0 2 0
k=0

Using the fact that Z +∞

n!
∀A ∈ R∗+ , e−Az z n dz = (24)
0 An+1
we get:

7
 N
"  2 k+1 #  2 N +2 !
θ − x22 X h(k) (0) 2θ 2
− x2 θ
I(x) = e 2θ k! + O e 2θ
2 k! x2 x2
k=0
 2 N  2 k  2 N +2 !
θ 2θ x2 X 2θ x2 θ
= 2
e− 2θ2 h̃(k) (1) 2
+ O e− 2θ2
2 x x x2
k=0
N  2 k
θ 2θ2
  2 X
 
2θ x2
− x2 (k) −2N −4 − 2θ2
= e 2θ h̃ (1) +O x e (25)
2 x2 x2
k=0

with
√  
16 2 dk

(k) 8z
h̃ (1) := f 3
θ3 z k − 2 θ2 z=1
√  k  
16 2 8 (k) 8
= 3 2
f− 3
θ θ 2 θ2
√  k  − 3 −2k k      j
16 2 8 8 2 X k 3 8
= − −k+j
θ3 θ2 θ2 j 2 j θ 2
j=0
k   
k X   2 −j
θ2

k 3 θ
= − −k+j
8 j 2 j 8
j=0
k     2 k−j
X k 3 θ
= − − (k − j)
j 2 j 8
j=0
k      2 j
X k 3 θ
= − −j
j 2 k−j 8
j=0
k    j
X k (−1)k−j (2k + 1)!! θ2
=
j 2k−j (2j + 1)!! 8
j=0
k  2 j
(−1)k X
 
j k (2k + 1)!! θ
= (−1) (26)
2k j (2j + 1)!! 4
j=0

Using (14), (21), (25) and (26), this concludes the proof of Case 2.

Case 3. We know turn to the case θ → +∞. We start again from

 
x2 2
√ − x2 C(T, K) − (S − K)+
Z θ −1
2 ξ2
+ ξ4
2π e = e dξ
S 0

When θ → +∞, C(T, K) → S. So,

 
x2 2
√ − x2 CC
Z +∞ − 1
2 ξ2
+ ξ4
2π e = e dξ
S θ

8
 ξ2
with CC := S − C(T, K). By the change of variables η := , we get:
8
√ − x2 CC
Z +∞ x
− 16
2 √ 1

2π e = e η e−η 2η − 2 dη
S θ2
8

So, with the notations of (11) and with N ∈ N∗ ,

Z +∞
√ − x CC 2
− x16 − 12 −η
πe 2 = e η e dη
S θ2
8
Z +∞ "X N k  2 #
(−1)k x2

x 1
= + ẽN η − 2 e−η dη
θ 2 k! 16η 16 η
8 k=0
N  k Z +∞ Z +∞
(−1)k x2
 2 
X
−k− 12 −η x 1
= η e dη + 2 ẽN η − 2 e−η dη
k! 16 θ 2 θ 16 η
k=0 8 8
N  k   Z +∞
(−1)k x2 1 θ2
 2 
X x 1
= Γ −k + , + 2 ẽN η − 2 e−η dη (27)
k! 16 2 8 θ 16 η
k=0 8

Moreover, for θ → +∞, we have by (12) and (17):

Z   2 N +1
 +∞  2 
x  Z +∞
1 x
− 12 −η 1
ẽ η e dη ≤ η − 2 e−η dη
 
N
16 η (N + 1)! 16 η
 2 
 θ  θ2
8 8
 2 N +1 
1 θ2

1 x
≤ Γ −N − ,
(N + 1)! 16 2 8
 2 −N − 32 !
θ θ2
= O e− 8
8
 2

−2N −3 − θ8
= O θ e (28)

On the other hand, with the notations of (16) and N, k ∈ N with N ≥ k,

 
2
 2 −k− 12 1 1 1
−k − 2 (−k − 2 )...(−k − 2 − (N − k) + 1) 
 
1 θ θ θ2 
Γ −k + , = e− 8 1 + θ 2 + ... +  N −k
2 8 8

8 θ2
8
2
 
1 θ
+ γN −k −k + ,
2 8
 
 2 − 12 N Qj−k 1
l=1 (−k + 2 − l)  1 θ2
 
θ 2
− θ8 
X
= e  j  + γN −k −k + , (29)
8 2 8

θ2
j=k 8

with
2  (−k − 21 )...(−k − 12 − (N − k)) − θ2
  
γN −k −k + 1 , θ  ≤

 N −k+1 e 8
 2 8  θ2
8
 2

−2N −3 − θ8
= O θ e (30)

9
Therefore,
 
N 2 k
 1
2 −2 N Qj−k 1
√ (−1)k X l=1 (−k + 2 − l) 
  
x CC X x θ θ2
π e− 2 = e− 8  j
S k! 16 8
 
θ2
k=0 j=k 8
 
θ2
+ O θ−2N −3 e− 8
 
1 N k N j−k 1
(−1)k x2
Q
X l=1 (−k + 2 − l) 
  
8 2 θ2 X
= e− 8  j
θ2 k! 16
 
θ2
k=0 j=k 8
 2

−2N −3 − θ8
+ O θ e
1 N   j  k j−k
8 2 − θ2 X 8 j X (−1)k x2
 Y 1
= 2
e 8
2
(−k + − l)
θ θ k! 16 2
j=0 k=0 l=1
 
θ2
+ O θ−2N −3 e− 8 (31)

We have
j−k
Y 1 (−1)j−k (2j − 1)!!
(−k + − l) =
2 2j−k (2k − 1)!!
l=1

Hence,
1 1 N   j  k
√ 8 2 − ( 82 ) X 8 j X (−1)k x2 (−1)j−k (2j − 1)!!

− x2 CC
πe = e θ
S θ2 θ2 k! 16 2j−k (2k − 1)!!
j=0 k=0
 2

−2N −3 − θ8
+ O θ e (32)
1 − 1 N  2   k
(−1)k
  
8 2 8 x 8 2
e ( θ2 ) −2N −3 − θ8
X
= ck , +O θ e (33)
θ2 2k 8 θ2
k=0

with
k j
x2 x2
  
X 1
ck = (2k − 1)!!
8 j! (2j − 1)!! 8
j=0

This is exactly (8) and it puts an end to the proof of Proposition 1.

2.2 Asymptotic expansions of a European call option for x = 0.

At the money, we have (S − K)+ = 0. So, TV = C and by (14):

√ C
Z θ ξ2
2π = e− 8 dξ
S 0

So,

10
Proposition 2 At the money,
 
θ
C = S erf √ (34)
2 2
Z u
2 2
with erf(u) := √ e−ζ dζ.
π 0

In the same way, we have:

 
CC θ
= erfc √ (35)
S 2 2
Z +∞
2 2
with erfc(u) := √ e−ζ dζ.
π u

Proposition 3 Let N ∈ N. (Case 1.) For θ → 0 and θ 6= 0, we have:

N  2 k
√ C X (−1)k 1 θ
+ O θ2N +3


2π = θ k
. (36)
S 2 (2k + 1) k! 4
k=0

(Case 2.) For θ → +∞, we have:

1 N k
√ CC (−1)k
   
8 2
− θ8
2 X 8 2
−2N −3 − θ8
π = e .(2k − 1)!! +O θ e (37)
S θ2 2k θ2
k=0

Proof. Formula (37) comes from the well known asymptotic expansion of erfc(x) for x large:
N
2 2
!
e−x X (−1)k (2k − 1)!! e−x
erfc(x) = √ +O (38)
x π 2k x2k x2N +3
k=0

Note 1 Equation (37) agrees with Proposition 1 - Equation (8) in the limit when x → 0.

3 Asymptotic expansions of the implied lognormal volatility

This section is intended for people like me who are not familiar with the notion of transseries.
Otherwise, all the results below are supposed to be a simple consequence of the fact that
(λ, ln(λ)) form a transbase (See [6], Theorem 5.12).

We want now to express the time-variance θ2 = σLN

2 T in terms of the time-value T V (resp.

covered call CC) for θ  1 (resp. θ  1).

11
3.1 Asymptotic expansions of the implied lognormal volatility when K 6= S
Let us assume that K 6= S i.e., x 6= 0. We need to invert Equations (3) and (8).
Our main result will be seen as a consequence of the following note.
Note 2 Both Equations (3) and (8) are of the form:
"N #
β − v1 1
X
k N +1
= eγ e− λ


v e αk v + O v (39)
k=0

with:
√ x
!
2 θ2 (−1)k x2 4 πe− 2
 
3
• (Case θ  1) v = 2 , β = , αk = . ak , γ = ln and λ =
x 2 2k 8 |x|
1
− TV .
ln( S )

(−1)k x2 √
 
8 1 x

• (Case θ  1) v = , β = , α k = . ck , γ = ln π e− 2 and λ =
θ2 2 2k 8
1
− .
ln( CC
S )

We are going to invert (39) and thus to obtain an asymptotic expansion of v in terms of
λα ln(λ)β .

Lemma 1 For any (αk ) ∈ RN , γ ∈ R and N ∈ N∗ , the asymptotics expansion of (39) for
0 < λ  1 and β > 0 is given by:

v = λ − βλ2 ln λ + γλ2 + β 2 λ3 ln2 (λ) + β 2 − 2βγ λ3 ln(λ) + γ 2 − βγ − α1 λ3 + o λ3 (40)

Proof.
Order λ
We use the fact that if f ∼ g with lim f = 0+ or +∞, then also ln f ∼ ln g. Therefore,
from
1 1
v β e− v ∼ eγ e− λ (41)
1
and the fact that lim eγ e− λ = 0, we get:
λ→0

1 1
β ln v − ∼γ−
v λ
1
The function g : x 7→ β ln(x) − is non-decreasing and lim g(x) = −∞. So, lim v = 0+ .
x x→0+ λ→0
Moreover, since lim v = 0 and lim v ln v = 0, we get v ∼ λ.
v→0 v→0

Order λ2 ln(λ)
Let us define w by v = λ(1 + w). Necessarily, lim w = 0. Let us also denote by ε the
function such that lim ε = 0 and
1 1
v β e− v (1 + ε) = eγ e− λ

12
We have:
1 1
β ln v − + ln(1 + ε) = γ −
v λ
So,
1 1
− = β ln v + ln(1 + ε) − γ
v λ
The right hand side of the last equality is clearly equivalent to β ln v when λ (and so also v)
goes to 0. So,
w
− ∼ β ln v
v
Thus,
w ∼ −βv ln v ∼ −βλ ln λ
So, we have proved:
v = λ − βλ2 ln λ + o λ2 ln λ


(42)

Order λ2
By (41), we have:
1 1
β ln v − = γ − + o(1) (43)
v λ
Set
v = λ (1 − βλ ln(λ) + z) (44)
with z = o(λ ln(λ)). Then,
ln(v) = ln(λ) + o(1) (45)
and
1 1
= [1 − βλ ln(λ) + z]−1
v λ
1
= (1 + βλ ln(λ) − z + o(λ))
λ
1 z
= + β ln(λ) − + o(1) (46)
λ λ
Therefore, by (43),(45) and (46), we obtain:
z
= γ + o(1)
λ
So,
z ∼ γλ (47)
We have proved:
v = λ − βλ2 ln λ + γλ2 + o λ2


(48)

Order λ3 ln2 (λ)

Let ξ be defined by
v = λ (1 − βλ ln(λ) + γλ + ξ) (49)

13
Then, ξ = o(λ) and

ln(v) = ln(λ) + ln (1 − βλ ln(λ) + γλ + ξ)

= ln(λ) + O (λ ln(λ))
= ln(λ) + o λ ln2 (λ)


(50)

On the other hand,

1 1
= [1 − βλ ln(λ) + γλ + ξ]−1
v λ
1
1 + βλ ln(λ) − γλ − ξ + β 2 λ2 ln2 (λ) + o λ2 ln2 (λ)


=
λ
1 ξ
+ β ln(λ) − γ − + β 2 λ ln2 (λ) + o λ ln2 (λ)


= (51)
λ λ
With N = 1, Equation (39) says that
1 1
β ln(v) − + ln (1 + α1 v + o(v)) = γ − (52)
v λ
We have

ln (1 + α1 v + o(v)) ∼ v
∼ λ
= o λ ln2 (λ)

So, by (50) and (51), we get:

ξ
− β 2 λ ln2 (λ) + o λ ln2 (λ) = 0


(53)
λ
Therefore,
v = λ − βλ2 ln λ + γλ2 + β 2 λ3 ln2 (λ) + o λ3 ln2 (λ)


(54)

Order λ3 ln(λ)
Set φ so that
v = λ 1 − βλ ln(λ) + γλ + β 2 λ2 ln2 (λ) + φ


(55)
with φ = o λ2 ln2 (λ) . We have:

ln(v) = ln(λ) − βλ ln(λ) + o (λ ln(λ)) (56)

and
1 1
−1
= 1 − βλ ln(λ) + γλ + β 2 λ2 ln2 (λ) + φ
v λ
1
1 + βλ ln(λ) − γλ − β 2 λ2 ln2 (λ) − φ + β 2 λ2 ln2 (λ) − 2βγλ2 ln(λ) + o λ2 ln(λ)



=
λ
1 φ
= + β ln(λ) − γ − − 2βγλ ln(λ) + o (λ ln(λ)) . (57)
λ λ

14
So,
1 1 φ
= γ − − β 2 λ ln(λ) + 2βγλ ln(λ) + + o (λ ln(λ))
β ln(v) −
v λ λ
On the other hand, we have:

ln (1 + α1 v + o(v)) ∼ α1 v
∼ α1 λ
= o (λ ln(λ)) (58)

Thus, by (52), we deduce that

φ
= β 2 λ ln(λ) − 2βγλ ln(λ) + o (λ ln(λ))
λ
and so,
φ ∼ β 2 − 2βγ λ2 ln(λ)

Therefore,

v = λ − βλ2 ln λ + γλ2 + β 2 λ3 ln2 (λ) + β 2 − 2βγ λ3 ln(λ) + o λ3 ln(λ)

(59)

Order λ3
Set ψ = o λ2 ln(λ) such that:

v = λ 1 − βλ ln(λ) + γλ + β 2 λ2 ln2 (λ) + β 2 − 2βγ λ2 ln(λ) + ψ .

(60)

Then,

ln(v) = ln(λ) + ln 1 − βλ ln(λ) + γλ + β 2 λ2 ln2 (λ) + β 2 − 2βγ λ2 ln(λ) + ψ

= ln(λ) − βλ ln(λ) + γλ + o(λ) (61)

(62)

Also,
1 1
−1
1 − βλ ln(λ) + γλ + β 2 λ2 ln2 (λ) + β 2 − 2βγ λ2 ln(λ) + ψ


=
v λ
1
1 + βλ ln(λ) − γλ − β 2 λ2 ln2 (λ) − β 2 − 2βγ λ2 ln(λ) − ψ



=
λ
1
β 2 λ2 ln2 (λ) − 2βγλ2 ln(λ) + γ 2 λ2 + o(λ2 )


+
λ
1 ψ
= + β ln(λ) − γ − β 2 λ ln(λ) + γ 2 λ − + o(λ)
λ λ
and

ln(1 + α1 v + o(v)) = α1 v + o(v)

= α1 λ + o(λ) (63)

Therefore,
1 1 ψ
β ln(v) − + ln(1 + α1 v + o(v)) − γ + = βγλ − γ 2 λ + + α1 λ + o(λ)
v λ λ

15
By (52), the left hand side of this equation is 0. So,

ψ
= γ 2 − βγ − α1 λ + o(λ)


λ
and
ψ = γ 2 − βγ − α1 λ2 + o(λ2 )

This put an end to Lemma 1.

By induction on m and n, we can also prove the following generalization of Lemma 1.

Proposition 4 There are ai,j defined for (i, j)2 ∈ N and j < i such that for any (m, n) ∈ N2 ,
with n < m, we have:
v = vm,n + o (λm lnn (λ)) (64)
with
m m−1
X X
vm,n := ai,j λi lnj (λ) (65)
i=1 j=n

• We have λ  λ2 ln(λ)  λ2  λ3 ln2 (λ)  λ3 ln(λ)  λ3  λ4 ln3 (λ)  .... The symbol
 is defined by f  g if and only if g = o(f ) in a neighborhood of 0.
i(i + 1)
• In this sequence, λi lnj (λ) is in position πi,j := − j.
2
• for any k ∈ N, there is a unique (i, j) ∈ N with j < i such that k = πi,j .

• we set vk := vi,j with k = πi,j .

For m ≥ 3, the coeficient am,n can be obtained by induction by the following way:
m−2
!
v  1
πm,n −1 X
k
• We expand ln , and ln αk v and we keep the terms in λm−2 lnn (λ).
λ vπm,n −1
k=0

• We note those terms Am,n , Bm,n and Cm,n respectively.

3
• Then, am,n = Bm,n − Am,n − Cm,n .
2
As an application of Note 2 and Lemma 1, we get:

Proposition 5 (Case 1: short expiry). Let us denote by T V := C(T, K) − (S − K)+ the

time-value of a European call option, σLN its implied !lognormal volatility and T the maturity
√ −x
1 4 πe 2 3 x2
of the option. Set λ := − T V , γ := ln and α1 = − − with x = ln( K
S ).
ln( S ) |x| 2 16
Then, when T → 0, we have the following expansion for the time-variance of the call option:
2
2 T = x v with
σLN
2
   
3 2 2 9 3 2 9 3
− 3γ λ ln(λ) + γ − γ − α1 λ3 + o λ3
3 2


v = λ − λ ln λ + γλ + λ ln (λ) +
2 4 4 2

16
(Case 2: large expiry). Let us denote by CC := S − C(T, K) the covered call of a European
call option, σLN its implied lognormal volatility and T the maturity of the option. Set λ :=
1 √ x
 1 x2
− CC , γ := ln π e− 2 and α1 = − − with x = ln( K
S ). Then, when T → +∞, we
ln( S ) 2 16
have the following expansion for the time-variance of the call option:
 
2 8 1 1 2 γ 
2 2


σLN T = 1 + λ ln(λ) − γλ − λ ln(λ) + + α1 λ + o λ (66)
λ 2 4 2
s
2 ln CC


S
In particular, when T → +∞, σLN ∼ 2 − .
T
Equation (66) is a generalization of [5].
Proof. Case 1 is just an application of Note 2 and Lemma 1. Case 2 follows from the
expansion of v −1 . Indeed, by (40), we have:
1 1
1 + βλ ln(λ) − γλ − β 2 λ2 ln(λ) + (βγ + α1 )λ2 + o λ2


= (67)
v λ
Therefore using Note 2 - Case 2,
2 T
σLN
 
1 1 1 γ
1 + λ ln(λ) − γλ − λ2 ln(λ) + ( + α1 )λ2 + o λ2


=
8 λ 2 4 2
Hence, we get the result.

Note 3 The case x → +∞ and θ fixed (Case 2 of Proposition 1) can be treated exactly in
the same way. It is more or less exactly the same as the!case x fixed and θ → 0 except that β
x
√ e− 2
is now equal to 1, γ has to be replaced by ln 2 2π , and αk (k ∈ N) has to be replaced
θ
(−1)k
 2
θ 3 x2
by k
bk with bk given in (7). Therefore, α1 = − + and the formula for the
2 4 2 8
2 x2
implied lognormal volatility is σLN = v with
2
v = λ − λ2 ln λ + γλ2 + λ3 ln2 (λ) + (1 − 2γ) λ3 ln(λ) + γ 2 − γ − α1 λ3 + o λ3



(68)

3.2 Implied lognormal volatility at the money

It turns out that at the money, there is a closed form formula for implied lognormal volatility
in terms of call price. No asumption on T is made.

Proposition 6 At the money, implied lognormal volatility σLN can be obtained as a power
series in call price C according to the formula:
r ∞  2k
2π C X π k ηk C
σLN = (69)
T S 4k (2k + 1) S
k=0
with ηk given by induction:
k
X ηj ηk−1−j
ηk = (70)
(j + 1)(2j + 1)
j=0

17
Proof. We have the well known expansion of erf −1 (see for instance [1]):
∞ √ 2k+1
−1
X ηk π
erf (x) = x (71)
2k + 1 2
k=0
k
X ηj ηk−1−j
ηk = (72)
(j + 1)(2j + 1)
j=0

So, by (34),
∞ √  √ 2k  2k
√
π CX ηk π C
θ = 2 2
2 S (2k + 1) 2 S
k=0
∞
√ CX π k ηk C 2k
 
= 2π (73)
S 4k (2k + 1) S
k=0

References
[1] Abramowitz M., Stegun IA. (1966), “Handbook of Mathematical Functions”, Dover.

[2] F. Black and M. Scholes (1973), “The Pricing of Options and Corporate Liabilities”, J.
Political Economy 81: 637-654

[3] K. Gao, R. Lee, “Asymptotics of Implied Volatility to Arbitrary Order”, ssrn 1768383.

[4] Roper M., Rutkowski M. (2009), “A Note on the Behaviour of the Black-Scholes Implied
Volatility Close to Expiry”, Int. J. of Theoretical and Applied Finance, 12(4): 427 - 441.

[5] M. Tehranchi (2009), Asymptotics of Implied Volatility Far From Maturity. J. of Applied
Probability, 46(3):629-650.

[6] Van der Hoeven, J. (2006), “Transseries and Real Differential Algebra”. Lecture Notes in
Math. 1888, Springer.

18