Professional Documents
Culture Documents
Abstract
We derive a direct link between local and implied volatilities in the form of a
quasilinear degenerate parabolic partial differential equation. Using this
equation we establish closed-form asymptotic formulae for the implied
volatility near expiry as well as for deep in- and out-of-the-money options.
This in turn leads us to propose a new formulation near expiry of the
calibration problem for the local volatility model, which we show to be well
posed.
In the Black–Scholes–Merton model [4,24], it is assumed that different strikes and otherwise identical have different implied
the price of a non-dividend paying stock St follows the log- volatilities. This phenomenon, usually referred to as the smile
normal stochastic differential equation effect, clearly violates the Black–Scholes–Merton model, since
in this framework a constant σ is supposed to determine the
dSt = St (µ dt + σ dWt ), (1) dynamics of the underlying stock St through (1) regardless of
options, strikes and maturities.
where t is time, µ and σ are constants and Wt is a standard
To overcome this difficulty, the model has to be extended.
Brownian motion. The parameter σ is called the volatility of
One widely used approach is to consider that the volatility
the stock St . It is well-known that the price C(St , t; K, T ) of a
also follows a stochastic diffusion process. Another case is
European call option written on St with strike K and maturity
T satisfies the linear parabolic partial differential equation when the volatility σ is not a constant any more but rather a
(deterministic) function of the underlying asset and the time.
σ2 2 Actually this can be seen as a particular case of the previous
Ct + S CSS + rSCS − rC = 0 in (0, +∞) × (0, T ) approach. These types of models are called local volatility
2
models. For them, the dynamics of the underlying asset is
C(S, T ) = (S − K)+ ,
governed by the stochastic differential equation
(2)
where r is the risk-free short-term interest rate. Such options dSt = µSt dt + σ (St , t)St dWt . (3)
are commonly traded on markets, however σ is not directly
observable. Therefore it is common practice to start from the There are two problems which are relevant in practice. Firstly
observed prices and invert the closed-form solution to (2) in one needs to compute accurately the implied volatilities of
order to find that constant σ —called implied volatility—for option prices—the pricing problem. If one follows a traditional
which the solution to (2) agrees with the market price at today’s approach (e.g. solve the PDE corresponding to each model
value of the stock. It is widely observed that calls having and then invert Black–Scholes formula), this is known to be
computationally difficult, especially near expiry or far from for the implied volatility near expiry. Furthermore, using
the money. this limit, we then show that some new formulation of the
Secondly, one wants to recover the value of the parameter calibration problem is well posed. Another use of the equation
of the model from market data—the calibration problem. One allows us to give asymptotics of the implied volatility for deep
remarkable result, due to Dupire [13], states the following: out-of-the-money options. Most of the results stated here have
should the call options corresponding to all possible strikes been announced in [2].
and maturities be priced on the market in a consistent manner, In a broader perspective we also hope that this nonlinear
the local volatility σ (S, t) would be uniquely determined by PDE approach will prove useful for the challenging problem of
the relation calibrating the local volatility and understanding its qualitative
properties.
CT + rKCK We adopt throughout the paper the reduced variables
(Dupire’s formula) σ (K, T ) = 2 , x = ln(S/K) + rτ , τ = T − t, so that from (2) the transformed
K 2 CKK
(4) prices
where CK , CKK are the first- and second-order derivatives of v(x, τ ) = erτ C(S, T − τ ; K, T )/K
the call price with respect to its strike and CT the derivative formally satisfy
with respect to its maturity. For the reader’s convenience,
vτ = 21 σ 2 (x, τ )(vxx − vx ) in T = R × (0, T )
we present in the appendix a proof of this result using PDE (6)
x
methods. v(x, 0) = (e − 1)+ ,
It turns out that in practice, this approach has two severe
abusing somewhat the notation (see for instance [5]). We shall
shortcomings. Firstly, there is but a finite set of observations.
assume that
Hence some interpolation is needed in order to use (4). It is
nowadays generally acknowledged (see for instance [8, 26]) σ ∈ BUC(T ), 0 < σ σ (x, τ ) σ < ∞, (7)
that it is by no means obvious to figure out how to interpolate
the data set in such a way that the radicand in (4) remains BUC being the space of (globally) bounded uniformly
positive and finite. Further, the result is overly sensitive to continuous functions and σ , σ are constants. In order to make
precise statements let us now specify the technical conditions
the (arbitrary) choice of the interpolation—especially for short
that we impose, and state some definitions of functional spaces
maturities—this resulting in poor robustness of the method.
that we shall need.
Secondly, a (related) difficulty lies in the intrinsic 2,1,p
We require that v ∈ C(T ) ∩ Wloc (), =
indeterminacy of formula (4) in the regions {T − t 1},
R × (0, +∞), satisfy the equation in (6) pointwise almost
{|ln(S0 /K)|
1}, {T − t
1}, where it assumes the form 00 .
everywhere in (strong solution). As is classical when
This is the reason why several other approaches have
studying parabolic problems, we make use of the following
been proposed in the recent years. We shall not attempt
anisotropic Sobolev spaces:
to give any comprehensive survey of this broad subject, but
rather focus on the results that are the closest to our point of
W 2,1,p () = w |wxx |p + |wτ |p + |w|p < ∞
view. Let us mention in this respect the approach by Lagnado (8)
and Osher [20], further extended by Jackson et al [18] and
Berestycki and Crépey [3] (see also [9]). The basic idea is to W 2,0,∞ () = {w||w|∞ + |wxx |∞ < ∞}
introduce a regularized cost functional in the form endowed with their natural norms, together with their local
2,1,p
versions Wloc and Wloc 2,0,∞
.
2
J (σ ) = ε |∇σ |2 + (C − C ∗ )(S0 , t0 ; Ki , Tj ) , (5) Let us denote by u the solution to (6) corresponding to
i,j σ ≡ 1, i.e. satisfying
62
Q UANTITATIVE F I N A N C E Asymptotics and calibration of local volatility models
Proposition 1 (existence). Under assumption (7), there (here and throughout the paper δ(x) is the Dirac mass at the
exists a unique solution v in the class origin), with σ satisfying (7). It is then proved that
2,1,p
Wloc () ∩ C() ∩ {w | ∃C, β > 0, x 2
ds
∀(x, τ ) ∈ |w(x, τ )| Ce βx 2
} −2τ ln v(x, τ ) →
0 σ (s)
for all 1 < p < ∞ to
as τ → 0. This corresponds to the Riemannian metric
vτ − 21 σ 2 (x, τ )(vxx − vx ) = 0 a.e. in associated with the inverse of the diffusion coefficient.
Denoting by U the fundamental solution of ∂t − 21 ∂x2 , namely
= R × (0, +∞) (12) x2
U (x, τ ) = (2π τ )−1/2 exp − 2τ , we can rephrase the result
v(x, 0) = (ex − 1)+ . as follows. If φ is such that v(x, τ ) = U (x, τ φ(x, τ )2 ), noting
Furthermore, v satisfies that by (7) φ is bounded, we deduce that
0 < v(x, τ ) < ex v(x, τ ) (ex − 1)+ 1
1 ds
→ (18)
vτ (x, τ ) > 0 in T . (13) φ(x, τ ) 0 σ (sx)
√ is that (x, τ ) → u(x, λτ )
A trivial yet crucial observation as τ → 0. Hence this problem gives rise to the same change of
(λ > 0) satisfies (12) with σ ≡ λ. This implies that if one metric as in theorem 1 (ii). However, it is not straightforward
defines ϕ 0 so that to derive from (18) the asymptotics in theorem 1 (ii), since we
v(x, τ ) = u(x, τ ϕ 2 (x, τ )) (14) are dealing with different initial conditions and composition
with a different function.
for all (x, τ ) ∈ , then ϕ is clearly the implied volatility of
the corresponding call option, in the sense described above. We also refer the reader to [14] for other large-deviation
That (14) uniquely determines ϕ(x, τ ) 0 follows easily results.
from the fact that (ex − 1)+ v(x, τ ) < ex (see (13) above), We next establish asymptotics for deep in- and out-of-the-
u(x, 0) = (ex − 1)+ , limτ →∞ u(x, τ ) = ex (see (10)), together money options.
with uτ > 0 in .
Our main result gives the implied volatility as the unique Theorem 2. Suppose σ satisfies (7) and
solution to a well-posed degenerate quasilinear parabolic limx→+∞ σ (x, τ ) = σ+ (τ ) (respectively x → −∞, σ− (τ ))
problem. locally uniformly in τ , with σ± continuous. Then
τ 1/2
Theorem 1. Under assumption (7) suppose that the implied 1
volatility ϕ is defined by (14) where v and u are solutions lim ϕ(x, τ ) = σ±2 (s) ds . (19)
x→±∞ τ 0
to (12) and (9) respectively. Then:
2,1,p
(i) The implied volatility ϕ lies in Wloc () for all We point out that theorem 1 (ii) and theorem 2 clarify in
1 < p < ∞ and satisfies particular the indeterminacy in Dupire’s formula in the regions
{T − t 1} and {|ln(S0 /K)|
1} that we have mentioned
ϕx 2 earlier.
2τ ϕϕτ + ϕ − σ (x, τ ) 1 − x
2 2
ϕ It results from theorem 1 (ii) that ϕ can be extended up to
−σ 2 (x, τ )τ ϕϕxx + 41 σ 2 (x, τ )τ 2 ϕ 2 ϕx2 = 0 (15) τ = 0 as a continuous function. That the limit of ϕ(x, τ )
a.e. in . at τ = 0 exists at all is by no means obvious from (14).
(ii) In the limit τ → 0, the implied volatility is the harmonic As a matter of fact, the value of the limit is quite specific to
mean of the local volatility, namely the problem under consideration, as the following proposition
makes clear. Specifically, if the payoff function is strictly
1
1 ds convex in the initial variable S, then the asymptotic behaviour
lim = , (16)
τ →0 ϕ(x, τ ) 0 σ (sx, 0) is dominated by the local volatility.
uniformly in x ∈ R. Proposition 2. If one replaces (ex − 1)+ in (9) by any smooth
2,1,p
(iii) Conversely, if ϕ̃ ∈ Wloc () (for some p > 1), satisfies function ζ (x) satisfying ζxx − ζx > 0, that is, ζSS > 0, then
(15) and (16) then ϕ̃ ≡ ϕ. limτ →0 ϕ(x, τ ) = σ (x, 0).
This result is actually much simpler than theorem 1. The main
Remark. The small time-to-maturity asymptotic in theorem 1 reason is that here uτ > 0 throughout T , so that the implicit
part (ii) has an interesting connection with a classical result function theorem applied to (14) gives bounds for ϕ together
by Varadhan [29], although it does not readily follow from it. with its derivatives near τ = 0. In other words, it is the very
In [29] the fundamental solution of the operator ∂t − 21 σ 2 (x)∂x2 , degeneracy of the original problem near τ = 0 that allows such
is considered, namely an ‘instantaneous averaging’ as that in (16) to take place.
vτ = 21 σ 2 (x)vxx A important feature of (16) is the following. Suppose that
(17) for x = 0 σ vanishes on some interval ω between 0 and x.
v(x, 0) = δ(x) Then ϕ(x, 0) = 0. This is consistent with the probabilistic
63
H Berestycki et al Q UANTITATIVE F I N A N C E
point of view. Indeed, in this limiting case the stock price 1. Proofs
process starting at x will never cross ω and so never reaches the
convexity region (x = 0). This feature was absent from other In the proof of theorem 1 we shall need the following series of
approximation formulae that were proposed, like weighted lemmas.
linear means (see for instance [10] ‘the poor man’s model’). Lemma 4 (maximum principle). Suppose a, b, c are
We refer to the book by Rebonato [26] for a very interesting globally bounded continuous coefficients defined in T , and
and detailed discussion on the qualitative properties of local assume that a(x, τ ) a > 0. Suppose that
and implied volatilities. 2,1,p
z ∈ Wloc (T ) ∩ C(T ) satisfies
The asymptotics in theorem 1 (ii) exhibits a linear relation
between the inverse of the local and implied volatilities. This zt − a(x, τ )zxx + b(x, τ )zx + c(x, τ )z 0 a.e. in T
leads us to propose a new regularization of the calibration
problem. Instead of (5) we introduce the following penalized lim z(x, τ ) 0 in D (R)
τ →0
functional:
z(x, τ ) −Ceβ|x|
2
for some β, C > 0.
2 2
1 1 1 (23)
J ε (σ ) = ε ∇ + − (x , τ ) , (20)
σ i,j
i j Then z 0 in T .
ϕ ϕ∗
(i) For all ε > 0 there exists a unique solution of the uxτ 1 x
(x, ξ ) = −
minimization problem uτ 2 ξ
inf J ε (ξ ), (22) uτ τ 1 x2 1
ξ ∈H 1 (R) (x, ξ ) = − + 2− .
uτ 2ξ 2ξ 8
denoted by ξ ε . Combining the above identities gives the result.
(ii) When ε → 0, ξ ε converges uniformly in R to a solution ξ̂
of the exact asymptotic calibration problem, i.e. such that Lemma 5 applied to w ≡ v, hence to ψ ≡ ϕ, yields
statement (i) in theorem 1.
1
An important tool in the subsequent argument is the fact
ζ̂ (xi , 0) ≡ ξ̂ (sxi ) ds = ζ ∗ (xi ).
0 that this problem satisfies a certain comparison principle that
64
Q UANTITATIVE F I N A N C E Asymptotics and calibration of local volatility models
we now state. For this purpose, let us denote by H the near τ = 0. This implies loss of a priori estimates for ϕ in
quasilinear operator this region. To circumvent this serious difficulty, the essential
idea is to define suitable sub and supersolutions of (12) from
ψx 2
H [ψ] ≡ H (x, τ, ψ, Dψ, D ψ) = 1 − x
2
the formal limiting solution of (15) and prove that actual
ψ convergence takes place through the comparison principle.
+τ ψψxx − 41 τ 2 ψ 2 ψx2 , (27) For the sake of clarity we first treat the case that σ satisfies
and define I (0, T ) to be the class of those functions ψ ∈ the additional regularity assumption
C 2,1 () for which the ‘associated local volatility’
σ ∈ C 2,1 (T ) and σxx ∈ L∞ (T ). (35)
2
1/2
τψ τ Let us define the formal limiting solution of (15) ϕ0 as the
σ [ψ](x, τ ) = (28)
H (x, τ, ψ, Dψ, D2 ψ) unique positive solution of F (x, 0, ϕ 0 , Dϕ 0 , D2 ϕ 0 ) = 0, i.e.
65
H Berestycki et al Q UANTITATIVE F I N A N C E
Now clearly
Proof of proposition 2. We apply identity (26) to w ≡ v
ε
−ω(η) + ε σ − σ ω(η) + ε (43) solution of (6), hence to ψ ≡ ϕ, the implied volatility. In this
case, uτ (x, 0) = 21 (uxx (x, 0) − ux (x, 0)) = 21 (ζxx − ζx ) > 0,
in δ/2 , if ω denotes a modulus of uniform continuity for σ so that the terms uuxττ and uuτττ in (26) remain bounded as τ → 0.
throughout δ . We can then choose η = η(ε) in such a way Furthermore, applying the implicit function theorem to (14),
that ω(η) < ε, so that σ ε σ in δ/2 and σ ε → σ uniformly and using uτ α > 0, we get that φ, φτ , φx , φxx remain
in δ/2 . bounded as τ → 0. Hence, sending τ to 0 in (26) clearly
Let us now define ϕ0ε (x) as the harmonic mean of σ ε (x, 0) yields φ(x, 0) = σ (x, 0).
i.e.
1 −1
ε ds Proof of theorem 2. By an obvious symmetry in the argument,
ϕ0 (x) = ε
(44) we treat only the case x → +∞. As in the proof above, we
0 σ (sx, 0)
shall construct for a given T > 0 a sub- and supersolution
and, correspondingly, ϕ ε (x, τ ) = ϕ0ε (x)(1 + κτ ) together with ψ, ψ ∈ I (0, T ) that have the required behaviour at infinity. To
v ε (x, τ ) = u(x, τ ϕ ε (x, τ )2 ). this purpose, we shall need auxiliary functions whose relevant
Repeating the computations in the first step, we get that properties are summarized in the following lemmas. We refer
ϕ ε ∈ I (0, δ) and the reader to the appendix for the proof.
σ [ϕ ε ] σ ε σ (45)
Lemma 7. Given A > 0, η ∈ (0, 0.1), κ > 1/(1 − η), there
in δ by fixing κ = κ(ε) large enough, and δ = δ(ε) small
exists ψ ∈ C 2 (R) satisfying the following properties:
enough. Lemma 6 thus implies ϕ ϕ ε in T . Similarly,
one constructs σε σ , ϕε0 , ϕ ε = (1 − κτ ), and shows that (i) ψ ∈ W 2,∞ (R) with ψW 2,∞ independent of A
ϕ ε (x, τ ) ϕ(x, τ ) in δ . (ii) ψ(z) 1−η 1
= limξ →+∞ ψ(ξ ) ∀z ∈ R
Summing up, we get that for all ε > 0 there exists (iii) ψ(z) 2κ ∀z ∈ (−∞, A)
(κ, δ) = (κ, δ)(ε) such that zψ (z) 1
(iv) ψ(z) 2 ∀z ∈ R
1 −1
ds zψ (z)
(1 − κτ ) (v) ψ(z)
→ 0, ψ (z) → 0, ψ (z) → 0 as z → +∞.
0 σε (sx, 0)
ϕ(x, τ ) By assumption, σ (x, τ ) → σ+ (τ ) as x → +∞, uniformly
1 −1 in τ ∈ (0, T ). Hence, given η ∈ (0, 0.1) there exists à such
ds
ε
(1 + κτ ) (46) that
0 σ (sx, 0)
σ+ (τ )
for all (x, τ ) ∈ R × (0, δ(ε)). This yields 1−η ∀(x, τ ) ∈ (Ã, +∞) × (0, T ). (49)
1 −1 σ (x, τ )
ds
lim inf ϕ(x, τ ) Besides, using (7) we get κ > 1 for which
0 σε (sx, 0) τ →0
lim sup ϕ(x, τ ) 1 σ+ (τ )
τ →0 κ in R × (0, T ). (50)
1 −1 κ σ (x, τ )
ds
(47) Let us define the quadratic mean limiting volatility
0 σ ε (sx, 0)
for all ε > 0, so that τ 1/2
1
8+ (τ ) = σ+2 (s) ds (51)
1 −1 τ 0
ds
lim ϕ(x, τ ) = . (48)
τ →0 0 σ (sx, 0) and, for ε > 0
That this limit is uniform in x ∈ R is easily seen through (46). ϕ(x, τ ) = 8+ (τ )ψ(εx), (52)
Part (iii) in theorem 1 (uniqueness) results from the where ψ is given in lemma 7, the values of η, κ > 0 being
maximum principle (lemma 4) applied to the difference defined as above, and A, ε to be fixed. A simple computation
function z(x, τ ) = u(x, τ ϕ̃ 2 (x, τ ))−u(x, τ ϕ 2 (x, τ )). Indeed, yields
by (16) z can be extended as a continuous function on T ,
2
with z(x, 0) = 0. Furthermore, by lemma 5 this function ψ
H [ϕ](x, τ ) = 1 − εx (εx) + τ 8+ ε 2 ψψ
satisfies zτ = 21 σ 2 (x, τ )(zxx − zx ) a.e. in T . Since clearly ψ
|z(x, τ )| 2ex , we can infer from the maximum principle 2 2
(lemma 4) that z ≡ 0, and thus ϕ̃ ≡ ϕ since uτ > 0. − 41 τ 2 ε 2 8+2 ψ ψ . (53)
Clearly by (iv) in lemma 7 we can choose ε =
Remark. Note that we retained from (16) only the fact that ε(ψW 2,∞ , T , 8+ ) > 0 independent of A such that
τ ϕ̃ 2 (x, τ ) → 0 as τ → 0. In this sense equation (15) has a
‘built-in’ initial condition, due to its degeneracy at τ = 0.
1
4
H [ϕ] 2 (54)
66
Q UANTITATIVE F I N A N C E Asymptotics and calibration of local volatility models
in T ; now by (v) in lemma 7 there is B > 0 for which for x A and 41 H 2 in T . Now, by (50), (61), (ii),
x B, τ ∈ (0, T ) imply H [ϕ](x, τ ) < 1−η1
. Setting (iii) in lemma 8, it follows that the local volatility associated
A = ε max(B, Ã), we see that (54) holds for all x and with ψ
2
1 τ 8+ (τ )ψ(εx)2 τ
H [ϕ](x, τ ) < (55) σ [ϕ](x, τ ) =
2
(64)
1−η H [ψ](x, τ )
67
H Berestycki et al Q UANTITATIVE F I N A N C E
68
Q UANTITATIVE F I N A N C E Asymptotics and calibration of local volatility models
[10] Derman E 1997 Forecasting implied volatility: rolling along inverse problem J. Comput. Finance 1 13–25
the local volatility surface Derivatives ’97 (Brussels, 1997) [21] Lagnado R and Osher S 1997 Reconciling differences Risk
Communication Mag. 10
[11] Derman E and Kani I 1994 Riding on a smile Risk Mag. 7 32–9 [22] Lasry J M and Lions P-L 1995 Grandes déviations pour des
[12] Derman E, Kani I and Chriss N 1996 Implied trinomial trees processus de diffusion couplés par un processus de sauts
of the volatility smile J. Derivatives 3 C. R. Acad. Sci., Paris I 321 849–54
[13] Dupire B 1994 Pricing with a smile Risk Mag. 7 18–20 [23] Lee R 2001 Implied and local volatilities under stochastic
[14] Fournié E, Lasry J M and Lions P-L 1997 Some nonlinear volatility Int. J. Theor. Appl. Finance 4 45–89
methods for studying far-from-the-money contingent claims [24] Merton R C 1973 Theory of rational pricing Bell. J. Econ.
Publ. Newton Inst. (Cambridge: Cambridge University Management Sci. 4
Press) pp 115–45 [25] Musiela M and Rutkowski M 1997 Martingale Methods in
[15] Friedman A 1964 Partial Differential Equations of Parabolic Financial Modelling (Applications of Mathematics Series)
Type (Englewood Cliffs, NJ: Prentice-Hall) vol 36 (Berlin: Springer)
[16] Hagan P and Woodward D E 1999 Equivalent black volatilities [26] Rebonato R 1999 Volatility and Correlation in the Pricing of
Global Derivatives (Paris) Communication Equity, FX and Interest-Rate Options (Financial Eng.
[17] Isakov V 1991 Inverse parabolic problems with the final Series) (Chichester: Wiley)
overdetermination Commun. Pure Appl. Math XLIV [27] Rubinstein M 1994 Implied binomial trees J. Finance LXIX
185–209 771–818
Isakov V 1998 Inverse problems for PDE Applied [28] Tykhonov M 1963 Regularization of incorrectly posed
Mathematical Sciences vol 127 (New York: Springer) problems Sov. Math. 4 1624–7
[18] Jackson N, Süli E and Howison S 1998 Computation of [29] Varadhan S R S 1967 On the behaviour of the fundamental
deterministic volatility surfaces Report Oxford Univ. Comp. solution of the heat equation with variable coefficients
Lab. Commun. Pure Appl. Math. 20 431–55
[19] Kirsch A 1996 An Introduction to the Mathematical Theory of [30] Wang L 1992 On the regularity theory of fully nonlinear
Inverse Problems (New York: Springer) parabolic equations: I Commun. Pure Appl. Math. XLV
[20] Lagnado R and Osher S 1997 A technique for calibrating 27–76
derivative security pricing models: numerical solution of an [31] Wilmott P 1998 Derivatives (Chichester: Wiley)
69