HAL Id: hal-00678437
https://hal.archives-ouvertes.fr/hal-00678437v4
Submitted on 27 May 2013
HAL
is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire
HAL, estdestinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Monotonicity of Prices in Heston Model
Sidi Mohamed Ould Aly
To cite this version:
Sidi Mohamed Ould Aly. Monotonicity of Prices in Heston Model. International Journal of Theoretical and Applied Finance, World Scientific Publishing, 2013, 16 (3), pp.1350016.
�10.1142/S0219024913500167�. �hal-00678437v4�
c World Scientific Publishing Company
Monotonicity of Prices in Heston Model
S. M. OULD ALY
Universit Paris-Est, Laboratoire d’Analyse et de Mathmatiques Appliques 5, boulevard Descartes, 77454 Marne-la-Valle cedex 2, France
sidi-mohamed.ouldaly@univ-mlv.fr
Received (29 May 2012) Revised (04 October 2012)
In this article, we study the price monotonicity in the parameters of the Heston model for a contract with a convex pay-off function; in particular we consider European put options. We show that the price is increasing in the constant term in the drift of the variance process and decreasing in the coefficient of the linear term in the drift of variance process. We also show that the price is increasing in the correlation for small values of the stock and decreasing for the large values.
Keywords: Heston model; Monotonicity; Put options; Maximum principle; Correlation.
1. Introduction
The main attraction of the Black-Scholes model is the ability to express the price of European options in terms of a volatility parameter. Moreover, for convex pay- offs, these formulas are strictly increasing with respect to the volatility parameter, which can cover the risk associated with this parameter through the purchase or sale of options. However, following the rejection of deterministic volatility assumption by empirical studies, practitioners are increasingly convinced that the best way to model the dynamics of an underlying is to consider a model where the process of instantaneous variance is stochastic.
In a general stochastic volatility model, the variance process does not depend solely on its current value. For example, under the Heston model, the variance pro- cess is given as the unique solution of the following stochastic differential equation
dVt= (a−bVt)dt+σp
VtdWt, V0=v. (1.1) The options prices depend on the initial value of the variance process v and the parameters a, b, σ and ρ. These parameters are often calibrated to market price of derivatives, so they tend to change their values regularly. It is then important to know the impact they have on option prices.
The initial value of the variance process has a positive effect on prices of convex pay-off in a large class of stochastic volatility models. See for example, Bergman et al. (1996), Hobson (1998), Janson and Tysk (2002) and Kijima (2002). When
1
the volatility process is stochastic but bounded between two values mand M, El Karoui et al (1998) show that the price of an option is bounded between the Black- Scholes prices with volatilitiesmandM. In [17], Romano and Touzi show that the derivative of the value-function of an option with respect to the volatility under models such as Hull and White (1987) and Scott (1987) has a constant sign, and does not vanish before maturity. Henderson (2005) shows that convex option prices are decreasing in the market price of volatility risk. However, to our knowledge, the dependence of the European option price on the correlation parameter is not known in any stochastic volatility model.
In this article, we study the price monotonicity of European put options with respect to the parametersv,a,bandρ. We first show that the value function of put price is a classical solution of the Black-Scholes equation. Then using a Maximum principle we show that the price is increasing in the initial value of the variance process as well as in the constant term in the drift of the variance process and decreasing in the coefficient of the linear term in the drift of variance process. We also show that the price is increasing in the correlation for small values of the stock and decreasing for the large values.
This paper is organized as follows: In section 2 we recall some properties of the put price in the Heston model. In section 3 we study the monotonicity of the price with respect to the parameters of the drift of the variance process. The section 4 is devoted to the study of the monotonicity with respect to the correlation.
2. Preliminaries
Under a complete filtered probability space (Ω,F,(Ft)t≥0,P) satisfying the usual conditions, we consider the Heston stochastic volatility model for a price process St, defined by the following stochastic differential equations
dSt
St =√ VtdWt1,
dVt= (a−bVt)dt+σ√
VtdW2, dhW1, W2it=ρdt, (2.1) where a, b, σ > 0 and |ρ| < 1. Let (S, V) be the solution of this equation with initial value (S0=s, V0=v). One can writeSs as
St=s exp Z t
0
pVsdWs1−1 2
Z t 0
Vsds
(2.2) The process (St)t≥0is a local martingale; it is even a true martingale, by Mijatovi´c and Urusov [14]. Thereby, using the Call-Put parity, all the results of this paper hold for Call options.
We consider an European put option on S with strike K and maturity t. Its current price is given, for (s, v)∈R∗+×R+, by
P(t, s, v) =E[(K−St)+|S0=s;V0=v]. (2.3)
If we replace the put pay-off by a functiong∈ C2(R) such thatxg0andx2g00are bounded, then Ekstr¨om, Tysk 2010 (cf. [4] Theorem.2.3) show that the function
u(t, s, v) :=E[g(St)|S0=s, V0=v] (2.4) is a classical solution of the pricing equation. In particular, it satisfiesu∈ C R+3
∩ C1,2,2 R∗+
3
∩ C1,0,1 R∗+×R∗+×R+
. In addition, a probabilistic representation of the derivative ofuwith respect tov is given as
∂u
∂v(t, s, v) =E Z t
0
e−bτ( ˆSτs)2∂2u
∂s2(t−τ,Sˆτs,Vˆτv)dτ
, (2.5)
where ( ˆSs,Vˆv) is the unique solution starting from (s, v) to the stochastic differen- tial equation
dSˆst
Sˆts =ρσdt+
qVˆtvdWt1, dVˆtv=
a+σ22 −bVˆtv dt+σ
qVˆtvdWt2.
(2.6) Obviously, the European put pay-off does not satisfy the assumptions of this theorem. Nevertheless, Propositions 3.1 and 3.2 of [4] (which require only g to be continuous and bounded) ensure that P ∈ C(R3+) ∩ C1,2,2 R∗+
3
so that LP(t, s, v) = 0, ∀(t, s, v)∈R∗+×R∗+×R∗+,
P(0, s, v) = (K−s)+,∀(s, v)∈R∗+×R+, (2.7) where
Lϕ=−∂ϕ
∂t +
(a−bv) ∂
∂v+1 2s2v ∂2
∂s2+1 2σ2v ∂2
∂v2 +ρσsv ∂2
∂v∂s
ϕ. (2.8)
3. Monotonicity with respect to the parametersv, a and b
In this section we study the monotonicity properties of the put price with respect to the parameters v,a andb. We first give an extension of the result of [4] to the European put pay-off.
Theorem 3.1. In addition to (2.7), we haveP ∈ C1,0,1(R+×R∗+×R∗+). Further- more, the derivative ofP with respect tov is given by
∂P
∂v(t, s, v) =E Z t
0
e−bτh(t−τ,Sˆτs,Vˆτv)dτ
, (3.1)
where( ˆSs,Vˆv)is the solution starting from (s, v)to (2.6) andhis defined onR∗+ 3
by
h(τ, x, y) =Ey
K q
(1−ρ2)Rτ 0 Vudu
N0
−log(x/K)−ρRτ 0
√VudWu2+12Rτ 0 Vudu q
(1−ρ2)Rτ 0 Vudu
, (3.2) whereN is the cumulative distribution function of the standard normal law.
Remark 3.1. Note that the functionh(t, s, v) is simplys2∂ssP(t, s, v). As a direct consequence of this theorem, we have for anyt, s >0, the functionv7−→P(t, s, v) is increasing.
Proof. Writing St=sexp
ρ Z t
0
pVsdWs2+p 1−ρ2
Z t 0
pVsdWˆs2−1 2
Z t 0
Vsds
, (3.3) where the Brownian motion ˆW2 in independent fromW2, we have
P(t, s, v) =KE[N(d1)]−sE
eρR0t
√VudWu2−ρ22Rt
0VuduN(d2)
, (3.4) where
d1= −log(Ks)−ρRt 0
√VudWu2+12Rt 0Vudu q
(1−ρ2)Rt 0Vudu
(3.5)
and
d2=d1− s
(1−ρ2) Z t
0
Vudu. (3.6)
We can write∂ss2P(t, s, v), using this stochastic representation ofP, as
∂2P
∂s2 =Ev
K/s2 q
(1−ρ2)Rt 0Vudu
N0
−log(s/K)−ρRt 0
√VudWu2+12Rt 0Vudu q
(1−ρ2)Rt 0Vudu
. (3.7) Set
h(t, s, v) =s2∂2P
∂s2(t, s, v). (3.8)
The main purpose of the assumption (xg0 and x2g00 are bounded) is to give a stochastic representation of the second derivative of P with respect to s and to ensure that it is continuous and bounded. Here we see that we have a stochastic representation of∂ssP given by (3.7). Following the procedure of [4] (cf Proposition 4.1, 4.2), we only need to show that the function
(t, s, v)7−→H(t, s, v) :=E Z t
0
e−bτh(t−τ,Sˆτs,Vˆτv)dτ
(3.9) is continuous on R∗+×R+×R∗+ and bounded by an integrable random variable.
For this, we consider a sequence (tn, sn, vn)−→(t, s, v) and show thatH(tn, sn, vn) converges toH(t, s, v). As ( ˆSτsn,Vˆτvn) converges to ( ˆSsτ,Vˆτv) in probability, we only need to find an upper bound of H(t, s, v) by an integrable random variable and conclude by applying the dominated convergence theorem.
To obtain the desired upper bound, we first note that for anyx, y ∈ Rand 0≤τ ≤twe have
h(x, y, t−τ)≤Ey
K
√ 2π
q
(1−ρ2)Rt−τ 0 Vudu
=:M(t−τ, y). (3.10) We can easily see that for any 0≤y1≤y2, we have
M(t−τ, y1)≥M(t−τ, y2). (3.11) On the other hand, by the comparison theorem, we have
Vˆτv ≥ Vτv, a.s. (3.12)
It follows that
M(t−τ,Vˆτv) ≤ M(t−τ, Vτv), a.s. (3.13) Then,
E h
h( ˆSτs,Vˆτv, t−τ)i
≤E[M(t−τ, Vτv)] =E
EVτv
K
√2π q
(1−ρ2)Rt−τ 0 Vudu
=Ev
K
√2π q
(1−ρ2)Rt τVudu
. (3.14)
The last line follows from the Markov property of the processV. It follows that E
Z t 0
e−bτhv( ˆSτs,Vˆτv, t−τ)dτ
≤ Z t
0
Ev
K
√2π q
(1−ρ2)Rτ 0 Vudu
dτ . (3.15) We have, by Dufresne [3],
Ev
1 qRτ
0 Vudu
<+∞, ∀τ >0. (3.16) Moreover, for anyv≥0, we have
τ→0lim τ23Ev
1 qRτ
0 Vudu
= 0. (3.17)
It follows that for anyv≥0, Z t 0
E
1 qRτ
0 Vuvdu
dτ <+∞. (3.18)
The rest of the proof of the Theorem is identical to Proposition 3.1 of [4] by using this upper bound. Thus, the functionH is continuous onR+×R∗+×R∗+.
Monotonicity with respect to a and b
We now study the monotonicity properties of the put price with respect to the parameters a and b. Note that the paths of the variance process are increasing with respect to a and decreasing with respect tob. This means that increasinga generates higher volatility which will increase the Put price. To verify this claim, we will let the put price vary in terms ofaandb: We write
Pa,b(t, s, v) =E h
(K−Sta,b)+
S0a,b=s;V0a,b=vi
, (3.19)
where (Sa,b, Va,b) is the unique solution starting with (s, v) of the stochastic differ- ential equations
dSta,b Sa,bt =
q
Vta,bdWt1, dVta,b = (a−bVta,b)dt+σ
q
Vta,bdW2, dhW1, W2it=ρdt.
(3.20) The following maximum principle will be crucial for the proof of the main result of this section. The proof of this theorem can be found in the appendix.
Theorem 3.2 (Maximum Principle). Fort >0, let
µ∗t = sup{µ >0 : EStµ <∞}. (3.21) Let Lbe the operator defined by (2.8) and ϕ∈ C1,2,2(R∗+
3)∩ C(R3+)so that
∀t, M >0, ∃λ < µ∗t : sup
τ≤t, s≤M, v∈R
|ϕ(t, s, v)| ≤Mλ. (3.22) Supposeϕsatisfies
Lϕ(t, s, v) ≤ 0 (resp <0),∀(t, s, v)∈]0,+∞[×R∗+×R∗+,
ϕ(0, s, v) ≥ 0, ∀(s, v)∈R∗+×R∗+. (3.23) Thenϕ ≥ 0 (respϕ > 0) on R∗+
3.
We establish the monotonicity of P with respect to a and b in the following result
Proposition 3.1. Let a2> a1 andb1< b2. We have
Pa1,b(t, s, v)< Pa2,b(t, s, v), ∀b≥0, ∀(t, s, v)∈R∗+
3 (3.24)
and
Pa,b1(t, s, v)> Pa,b2(t, s, v), ∀a≥ σ2
2 , ∀(t, s, v)∈R∗+
3. (3.25) Proof. For anya, b≥0, let
La,bϕ=−rϕ−∂ϕ
∂t +
rs∂.
∂s+ (a−bv)∂.
∂v +1 2s2v∂2.
∂s2 +1 2σ2v∂2.
∂v2 +ρσsv ∂2.
∂v∂s
ϕ (3.26)
We can easily check that
La2,b(Pa2,b−Pa1,b)(t, s, v) =−(a2−a1)∂P∂va1,b,∀(t, s, v)∈R∗+ 3, (Pa2,b−Pa1,b)(0, s, v) = 0, ∀(s, v)∈R∗+
2. (3.27) and
La,b2(Pa,b1−Pa,b2)(t, s, v) =−(b2−b1)v ∂P∂va,b1,∀(t, s, v)∈R∗+ 3, (Pa,b1−Pa,b2)(0, s, v) = 0, ∀(s, v)∈R∗+
2. (3.28) We have, by Theorem 3.1, the function∂vPa1,band∂vPa,b1 are positive. Then, by Theorem 3.2, that (Pa2,b−Pa1,b)>0 and (Pa,b1−Pa,b2)>0.
4. Monotonicity with respect to the correlation
This section focuses on the monotonicity properties of the price of the European put with respect to the correlation. Note that the method we used in the previous section to establish the monotonicity with respect tov,aandbcan not be applied here. Indeed, the idea of this method was to differentiate (2.7) with respect to the parameter considered and obtain a differential system as (Lu <0 on Cand u≥0 on∂C), which gives the sign ofuby applying the maximum principle; while if we differentiate (2.7) with respect toρ, we obtain the system
L˜∂P∂ρ(t, s, v) =−σsv∂s∂v∂P (t, s, v),∀(t, s, v)∈]0, T]×R∗+×R∗+,
∂P
∂ρ(0, s, v) = 0, ∀(s, v)∈R∗+×R∗+. (4.1) As the sign of ∂svP is not necessarily constant, this does not allow us to deduce the sign of the derivative of P with respect toρusing the maximum principle. To analyze the impact ofρ in the priceP, we will study the sign of the derivative of P with respect to ρ. This derivative can be obtained by differentiating (3.4) with respect to ρ:
∂P
∂ρ =E
Z log(Ks)
−∞
exN0
x−ρRt 0
√VudWu2+12It
q
(1−ρ2)Rt 0Iudu
ρx−Rt 0
√VudWu2+ρ2It
p(1−ρ2)3It
! dx
, (4.2) whereIt:=Rt
0Vudu. The sign ∂P∂ρ is not obvious, however the following figure shows that there is a change of monotonicity depending on the value of the strike price.
We see that ∂P∂ρ is positive fors < K= 1 and negative for fors >1.
In order to determine if this change of monotonicity is unique, we will study in details the sign of the derivative of P with respect to ρ fors very large and very small. For this we define the quantities
sρ0(t, v) = inf
s >0 : ∂P
∂ρ(t, s, v)≤0
(4.3) and
sρ∞(t, v) = supn
s >0 : P˙ρ(t, s, v)≥0o
. (4.4)
Fig. 1. ˙Pρfors∈[0.4,2.5] (K= 1,v0= 0.1,b= 3,σ= 0.2 andt= 0.5).
Having sρ0 > 0 (resp sρ∞ < +∞) means that ∂P∂ρ is positive (resp negative) for s small (respslarge). We next present the main result of this section.
Theorem 4.1. For anyt, v >0 andρ∈]−1,1[, we have
0< sρ0(t, v)≤sρ∞(t, v)<+∞ (4.5) Proof. We use the results obtained in [16], where it is shown that forRsufficiently large, we have
lnP
−1 2It+
Z t 0
pVudWu1> R
∼ −µ+R (4.6)
and
lnP
−1 2It+
Z t 0
pVudWu1<−R
∼ −µ−R, (4.7)
withµ+= inf{p >0, T∗(p) =t}(>1),µ−= inf{p >0, T∗(−p) =t}and T∗(p) = sup
t >0, EQ exp
p2−p 2
Z t 0
Vudu
<+∞
, (4.8)
where underQthe processV satisfies the stochastic differential equation dVt= (a−(b−ρσp)Vt)dt+σp
VtdWtQ. (4.9)
We can easily see that, forksufficiently large, we have lnP(t, ek, v)∼ −µ−k, ln P(t, e−k, v)−1−e−k
∼ −µ+k (4.10) and
k→+∞lim
∂ρP(t, ek, v)
kP(t, ek, v) =−∂µ−
∂ρ , lim
k→+∞
∂ρP(t, e−k, v)
k(P(t, e−k, v)−1−e−k) =−∂µ+
∂ρ . (4.11)
By the comparison theorem, the processV is increasing with respect toρunderQ (see also [15]) forp >0 and decreasing forp <0. This means that forp >1,T∗(p) (as a function ofρ) is decreasing and for anyp0>0,ρ7−→T∗(−p0) is increasing. On the other hand,p7−→T∗(p) is decreasing nearµ+ andp7−→T∗(−p) is increasing nearµ−. It follows thatµ+is decreasing with respect toρandµ−is increasing with respect to ρ. This means that, for ksufficiently large, we have
∂ρP(t, e−k, v)>0 (4.12) and
∂ρP(t, ek, v)<0. (4.13) Thus (4.5).
So far we confirmed that 0< sρ0 ≤sρ∞<+∞, which means that the Put price is increasing in the correlation for small values of the stock price and decreasing for large values. The question is whethersρ0=sρ∞, which means that there is only one pointsρ(t, v) so that the derivative ofP with respect toρis positive fors≤sρand negative for s > sρ. All numerical experiments seem to confirm this intuition. In the next sections, we will show thatsρ0=sρ∞ for short and long maturities.
4.1. Small-Time Asymptotic Behavior
We study here the monotonicity with respect to the correlation for short maturities.
The main result of this section is the following Proposition Proposition 4.1. For any ρ∈]−1,1[and any v∈R+∗, we have
t→0lim sign ∂P
∂ρ(t, e−x, v) =sign(x). (4.14) Consequently,
t→0limsρ0(t, v) = lim
t→0sρ∞(t, v) = 1. (4.15) Proof. Let (S, V) be the unique solution of (2.6) starting with (s, v). By Forde and Jacquier (cf [6]), we have
t→∞lim tlogE(K−St)+=−Λ∗(log(K
s)), fors > K (4.16) and
t→∞lim tlogE(St−K)+=−Λ∗(log(K
s)), fors < K, (4.17) where Λ∗ is the Fenchel-Legendre transform of the function Λ defined by
Λ(p) = vp
σ(√
1−ρ2cot(12σp√
1−ρ2)−ρ), for p∈]p−, p+[,
Λ(p) =∞, for p∈R\]p−, p+[,
(4.18)
withp− and p+ are given by
p−= arctan
√
1−ρ2 ρ
1 2σp
1−ρ2 1ρ<0−π σ1ρ=0+
−π+ arctan √
1−ρ2 ρ
1 2σp
1−ρ2 1ρ>0, (4.19)
p+=
π+ arctan
√1−ρ2 ρ
1 2σp
1−ρ2 1ρ<0+π σ1ρ=0+
arctan √
1−ρ2 ρ
1 2σp
1−ρ2 1ρ>0. (4.20) The function Λ∗ is given by
Λ∗(x) =xp∗(x)−Λ(p∗(x)), (4.21) wherep∗(x) is the unique solution of
x= Λ0(p∗(x)) (4.22)
and Λ0 is given by
Λ0(p) = v
σ(p
1−ρ2cot(12σpp
1−ρ2)−ρ)+ σvp(1−ρ2) csc2(12σpp 1−ρ2) 2σ(p
1−ρ2cot(12σpp
1−ρ2)−ρ)2. (4.23) Let Σt(x) be the Black-Scholes implied volatility, defined as the unique solution of
P(t, K e−x, v) =PBS(t, K e−x, K; Σt(x)), (4.24) where
PBS(t, s, k,Σ) =KN(−log(s/k) +tΣ/2
√tΣ )−sN(−log(s/k)−tΣ/2
√tΣ ). (4.25)
By Theorem 2.4 of [6], we have
limt→0Σt(x) = |x|
p2Λ∗(x). (4.26)
WritingP(t, s, v) in terms of the Black-Schole implied volatility as in (4.24) and noting that the dependence of the right term of (4.24) with respect to ρ is only through Σ and using the fact that the Black-Scholes put price is is increasing with respect to the implied volatility, we see thatp(t, s, v) and Σt(log(K/s))) have the same monotonicity with respect to the correlation. Therefore
sign ˙Pρ(t, s, v) = sign ∂Σt(log(Ks)))
∂ρ . (4.27)
The implied volatility is differentiable with respect to the correlation. Moreover, using Lemma 4.1 below, we have
t→0lim
∂Σt(x)
∂ρ = −|x|
2Λ∗(x)p 2Λ∗(x)
∂Λ∗(x)
∂ρ . (4.28)
Let’s consider the derivative of Λ∗(x) with respect toρ, forx∈R. This derivative is given by
∂Λ∗(x)
∂ρ = ∂p∗(x)
∂ρ (x−Λ0(p∗(x)))−∂Λ
∂ρ(p∗(x)) =−∂Λ
∂ρ(p∗(x)) (as Λ0(p∗(x)) =x)
=
−vp∗(x)
√2ρ 1−ρ2
cot(θ∗(x))−(1−ρ2)θ∗(x) csc2(θ∗(x)) + 1
2σ(p
1−ρ2cot(θ∗(x))−ρ)2 , (4.29) where
θ∗(x) := 1
2σp∗(x)p
1−ρ2. (4.30)
Using Lemma 4.2 below, which ensures that, for anyx∈R, 2ρ
p1−ρ2 cot(θ∗(x))−(1−ρ2)θ∗(x) csc2(θ∗(x))
+ 1>0, (4.31) we have
sign ∂Λ∗(x)
∂ρ = sign (−vp∗(x)). (4.32)
On the other hand, asp∗(x) has the same sign asx, we deduce that fortsufficiently small, we have
sign ˙Pρ= sign ∂Σt(x))
∂ρ = sign ∂Λ∗(x)
∂ρ = sign (log(K s)).
Lemma 4.1. For any x6= 0, we have
t→0lim
∂Σt(x)
∂ρ = −|x|
2Λ∗(x)p 2Λ∗(x)
∂Λ∗(x)
∂ρ . (4.33)
Lemma 4.2. For any ρ∈[−1,1]andx∈R, we have 2ρ
p1−ρ2 cot(θ∗(x))−(1−ρ2)θ∗(x) csc2(θ∗(x))
+ 1>0. (4.34)
4.2. Large-Time Asymptotic Behavior
It is known that for long maturities the implied volatility curve in a stochastic volatility model flattens, so it does not depend on the strike. Under Heston model, Forde et al [7] showed that (under the assumptionb−ρσ >0) the implied volatility can be written as
Σ2t(x) = 8V∗(0) +a1(x)/t+o(t), (4.35) where V∗ and a1 are given below. The main result of this section is the following result
Proposition 4.2. For anyρ∈]−1,1[such thatb−ρσ >0 and for anyv >0, we have
t→+∞lim sρ0(t, v) = lim
t→+∞sρ∞(t, v) = +∞. (4.36) Proof. We will use the notations of [7]. Under the assumptionb−ρσ >0, we have, for anyp∈]p−, p+[,
V(p) = lim
t→∞t−1logE[exp (p(Xt−x0))] = a
σ2(b−ρσp−d(−ip)), (4.37) where
d(−ip) =p
(b−ρσp)2+σ2(p−p2) (4.38) and
p± :=
−2bρ+σ±p
σ2+ 4b2−4bσρ
. (4.39)
Let’s consider the functionp∗:R−→]p−, p+[ defined by
p∗(x) :=
σ−2bρ+ (aρ+xσ)
σ2+4b2−4bρσ x2σ2+2xaρσ+a2
1/2
2σ(1−ρ2) , for x∈R. (4.40)
Fortsufficiently large andx∈R, we have (cf. [7]) 1
S0E(St−S0e−x)+= 1 + A(0)
√2πtexp (−(1−p∗(0))x−V∗(0)t) (1 +O(1/t)), (4.41) whereV∗ is the Fenchel-Legendre transform ofV defined by
V∗(x) := sup{px−V(p), p∈]p−, p+[} (4.42) andAis the function defined in a neighborhood of 0 by
A(x) = −1
pV00(p∗(x))
U(p∗(x))
p∗(x)(1−p∗(x)), (4.43) where
U(p) :=
2d(−ip)
b−ρσp+d(−ip) σ2a2
expv aV(p)
. (4.44)
Similarly, the Black-Scholes implied volatility can be written as (cf. [7], Theorem 3.2)
Σ2t(x) = 8V∗(0) +a1(x)/t+o(t), (4.45) where
a1(x) =−8 log
−A(0)p
2V∗(0)
+ 4(2p∗(0)−1)x. (4.46) In particular, for anyx∈R, we have
t→+∞lim Σ2t(x) = 8V∗(0). (4.47)
Now using (4.45) we show, in a similar way as Lemma 4.1, that
t→+∞lim
∂Σt(x)
∂ρ = 8∂V∗(0)
∂ρ . (4.48)
We have
∂V∗(0)
∂ρ =−∂V
∂ρ(p∗(0)) + (x−V0(p∗(0)))∂p∗(0)
∂ρ =−∂V
∂ρ(p∗(0))
=−a
σ2 −σp∗(0) + σp∗(0)(b−ρσp∗(0))
p(b−ρσp∗(0))2+σ2(p∗(0)−p∗(0)2)
!
= −2ρap∗(0)(1−2p∗(0))
p(b−ρσp∗(0))2+σ2(p∗(0)−p∗(0)2)1ρ6=0. (4.49) The first two lines follow from the fact thatV0(p∗(0)) = 0. Forρ= 0, we have
∂V∗(0)
∂ρ
ρ=0=−a
2σ −1 + b
pb2+σ2/4
!
( > 0). (4.50) Lemma 4.3 below ensures that the functionϕ defined in (4.55) is increasing.
Note that, for anyρ∈[−1,1], we have
ϕ(ρ) =p∗(0). (4.51)
Asϕ(0) = 12, we deduce that for any ρ6= 0, (ϕ(ρ)−1/2) has the same sign as ρ. This means that
ϕ(ρ)−1/2
ρ >0. (4.52)
Therefore, we have
∂V∗(0)
∂ρ > 0, ∀ρ∈[−1,1]. (4.53)
It follows that for anyx∈R,
t→+∞lim
∂Σt(x)
∂ρ >0. (4.54)
It follows that fortsufficiently large, the put price is increasing with respect to the correlation. Thus (4.36).
Lemma 4.3. The functionϕ defined by
ρ∈]−1,1[7−→ ϕ(ρ) := σ−2bρ+ρp
σ2+ 4b2−4bρσ
2σ(1−ρ2) (4.55)
is increasing.
5. Appendix
Appendix A. Proof of Theorem 3.2 Suppose∃(t, s, v)∈R∗+
3so thatϕ(t, s, v)<0. Consider (Ss, Vv) the unique solution of the stochastic differential equations
dSts Sts =p
VtvdWt1,
dVtv= (a−bVtv)dt+σp
VtvdW2, dhW1, W2it=ρdt, S0s=s, V0v=v.
(A.1) Let’s define theF-stopping times
τ= inf
u∈[0, t] :ϕ(t−u, Sus, Vuv)≥ ϕ(t, s, v) 2
(A.2) and
¯ τn= inf
u∈[0, t] :Sus∧Vuv∈ 1
n, n c
∧t. (A.3)
We haveP(τ < t) = 1. Applying the Itˆo formula to the process (ϕ(t−u, Sus, Vuv))u≤t between 0 andτ∧τ¯n, we have
ϕ(t−τ∧τ¯n, Ssτ∧¯τn, Vτ∧¯vτn) =ϕ(t, s, v) + Z τ∧¯τn
0
Susp
Vuv∂sϕ(t−u, Sus, Vuv)dWu1+ σ
Z τ∧¯τn
0
pVuv∂vϕ(t−u, Sus, Vuv)dWτ2 +
Z τ∧¯τn 0
Lϕ(t−u, Sus, Vuv)du. (A.4) AsS andV are in ]0, n], we have
ϕ(t, s, v) =−E Z τ∧¯τn
0
Lϕ(t−u, Sus, Vuv)du+E
ϕ(t−τ∧τ¯n, Sτ∧¯s τn, Vτ∧¯v τn)
≥E
ϕ(t−τ∧¯τn, Sτ∧¯s τ
n, Vτ∧¯vτ
n)
≥ ϕ(t, s, v)
2 P(τ≤τ¯n) +E
ϕ(t−τ¯n, Sτs¯n, Vτ¯vn)1τ >¯τn
. (A.5)
Writing
{τ >τ¯n}={sup
u≤τ
Vu≥n} ∪ {sup
u≤τ
Su≥n}, (A.6)
we have E
ϕ(t−τ¯n, Sτs¯n, Vτ¯vn)1τ >¯τn
≤nλ
P
sup
u≤t
Vu≥n
+P
sup
u≤t
Su≥n
. (A.7) Now using Doob’s martingale inequality, we have
P
sup
u≤t
Su≥n
≤ ES
µ∗ t+λ
2
t
nµ
∗t+λ 2
=⇒nλP
sup
u≤t
Su≥n
≤ ES
µ∗ t+λ
2
t
nµ
∗t−λ 2
−→n→∞0. (A.8)
Similarly, applying Doob’s martingale inequality to the martingaleebt(Vt−ab) and taking into account the fact that
EVp <∞, ∀p >0, (A.9) we obtain
n→∞lim nλP
sup
u≤t
Vu≥n
= 0. (A.10)
Therefore
n→∞lim E
ϕ(t−τ¯n, Sτs¯
n, Vτ¯v
n)1τ >¯τn
= 0. (A.11)
This means that
ϕ(t, s, v)≥ ϕ(t, s, v)
2 . (A.12)
Hence the contradiction (ϕ(t, s, v) is supposed to be negative). Thusϕ≥0.
Now assumeLϕ <0. Let’s take (t, s, v) witht >0. Applying the Itˆo formula to the process (ϕ(t−u, Sus, Vuv))u≤tbetween 0 andt∧τ¯n, we have
ϕ(t− t
2∧τ¯n, Sst 2∧¯τn, Vtv
2∧¯τn) =ϕ(t, s, v) + Z 2t∧¯τn
0
Susp
Vuv∂sϕ(t−u, Sus, Vuv)dWu1+ σ
Z t2∧¯τn 0
pVuv∂vϕ(t−u, Sus, Vuv)dWτ2 +
Z 2t∧¯τn 0
Lϕ(t−u, Sus, Vuv)du. (A.13) We get, by the same way as before,
ϕ(t, s, v)≥ −E Z t2
0
Lϕ(t−u, Sus, Vuv)du >0. (A.14)
Thusϕ(t, s, v)>0.
Appendix B. Proof of Lemma 4.1:
The put priceP is given, in terms of the Black-Scholes implied volatility, by P(t, s, v) =KN
−log(s/K) +2tΣ2t
√tΣt
−sN
−log(s/K)−t2Σ2t
√tΣt
. (B.1) Differentiating this expression on both sides with respect toρ, we can write ˙Pρ as
P˙ρ(t, s, v) =KN0
−log(s/K) +2tΣ2t
√tΣt
√ t∂Σt
∂ρ (−log(s/K)). (B.2) On the other hand, by (4.26), we know that
limt→0Σt(x) = |x|
p2Λ∗(x). (B.3)
Moreover the function ρ 7−→ |x|/p
2Λ∗(x)(ρ) is C1 on ]−1,1[. We claim that t7−→ ∂Σ∂ρt is bounded near 0. This is equivalent to say that
P˙ρ(t, s, v) K√
t N0 (−log(s/K) +2tΣ2t)/(√
tΣt) is bounded. (B.4) Writing
P =E
K−s exp
ρ Z t
0
pVudWu1+p 1−ρ2
Z t 0
pVudWu2−It/2
+
, (B.5) we can write ˙Pρ as
P˙ρ(t, s, v) =E
"
− Z t
0
pVudWu1+ ρ p1−ρ2
Z t 0
pVudWu2
!
St1K≥St
#
. (B.6) Applying the H¨older inequality, withp >1, we have
P˙ρ(t, s, v)≤
"
E
−St Z t
0
pVudWu1+ Stρ p1−ρ2
Z t 0
pVudWu2
p#1/p
×[P(K≥St)]p−1p , (B.7) whereP(K≥St) can be written as
P(K≥St) = ∂P
∂K =N
−log(s/K) +2tΣ2t
√tΣt
+KN0
−log(s/K) +2tΣ2t
√tΣt
√ t∂Σt
∂K. (B.8) On the other hand, for anyy >0, we have
N(−y)≤ 1 y
exp −y2/2
√2π . (B.9)
It follows that for anys > Kandt sufficiently small, we have N
−log(s/K) +t2Σ2t
√tΣt
≤
√tΣt
log(s/K)−2tΣ2tN0
−log(s/K) +2tΣ2t
√tΣt
. (B.10) Then, for s > K, there exists a constantM >0 such that, fort sufficiently small, we have
P(K≥St)≤M √ t N0
−log(s/K) +2tΣ2t
√tΣt
. (B.11)
It follows that P˙ρ(t, s, v)
√tN0−log(s/K)+t 2Σ2t
√tΣt
≤M2 [E|Yt|p]1/p× √
tN0
−log(s/K) +2tΣ2t
√tΣt
−1 p
, (B.12)
where
Yt= − Z t
0
pVudWu1+ ρ p1−ρ2
Z t 0
pVudWu2
!
× exp
ρ Z t
0
pVudWu1+p 1−ρ2
Z t 0
pVudWu2−It/2
. (B.13) Setx= log(s/K). Fort small,
−x+2tΣt
√tΣt
∼ −x
√tΣt
. (B.14)
We choosepso that
p=p(t) =c
t. (B.15)
For this particularp, we have √
tN0
−log(s/K) +t2Σ2t
√tΣt
−1 p
∼M3 exp
−t c
log√
t− x2 2tΣt
≤M4. (B.16) We next show that [E|Yt|p(t)]1/p(t) is bounded fort close to 0. For this, we use the usual inequality
|y| ≤ey+e−y, ∀y∈R. (B.17) We get
|Y| ≤ Y1(t) +Y2(t), (B.18)
where
Y1(t) = exp (ρ−1) Z t
0
pVudWu1+ρ+ 1−ρ2 p1−ρ2
Z t 0
pVudWu2−It/2
!
(B.19) and
Y2(t) = exp (1 +ρ) Z t
0
pVudWu1+1−ρ−ρ2 p1−ρ2
Z t 0
pVudWu2−It/2
!
. (B.20) BothY1 andY2 can be written as
Yi(t) = exp
αi Z t
0
pVudWu1+βi Z t
0
pVudWu2−1 2It
, i= 1,2. (B.21) In particular, we have
EYip =E exp
αip Z t
0
pVudWu1+βip Z t
0
pVudWu2−p 2It
=E exp
αip Z t
0
pVudWu1+βi2p2−p 2 It
. (B.22)
By [16] and the fact that Z t
0
pVudWu1= (Vt−v−at+bIt)/σ, (B.23) we have, forpsufficiently large,
EYip= exp (−αip(v+at)/σ+aϕ(t) +vψ(t)), (B.24) where
ψ(t) = b σ2 +
p2λi2(p)σ2−b2
σ2 tan (g(t, p)), ϕ(t) = b
σ2t+ 2
σ2(log cosg(0, p)−log cosg(t, p)) (B.25) and
g(t, p) =
p2λi2(p)σ2−b2
2 t+ arctan( λi1(p)σ2−b
p2λi2(p)σ2−b2), (B.26) with
λi1(p) =αip/σ and λi2(p) =β2ip2−p
2 +αibp/σ. (B.27)
It follows that
[EYip]1/p= exp
−αi(v+at)/σ+aϕ(t)
p +vψ(t) p
. (B.28)
In particular, forp=p(t) =c/t, we have, fortsufficiently small, g(t, p(t)) ∼ cβiσ
2 + arctan(αi
βi). (B.29)
Similarly, we have aϕ(t)
p(t) +vψ(t)
p(t) ∼ vβic σ tan
cβiσ
2 + arctan(αi βi
)
. (B.30)
Note that the coefficientc in (B.15) was chosen so that
−π/2<cβiσ
2 + arctan(αi
βi)< π/2, for i= 1,2. (B.31) We finally have, for i= 1,2,
t→0lim[EYip]1/p= exp vβic
σ tan cβiσ
2 + arctan(αi
βi)
<+∞. (B.32) It follows that, using (B.12) and (B.16), the claim (B.4) is verified. We proceed similarly fors < K, by using the call price instead of the put price.
Appendix C. Proof of Lemma 4.2:
Let’s set
η(x) = 2ρ
p1−ρ2 cot(θ∗(x))−(1−ρ2)θ∗(x) csc2(θ∗(x))
+ 1 (C.1)
andη(x) =ϕ(θ∗(x)), whereϕis defined by ϕ(θ) = 2ρ
p1−ρ2
cos(θ)
sin(θ) −(1−ρ2) θ sin2(θ)
+ 1. (C.2)
For anyx∈R,p∗(x)∈[p−, p+], we have θ(ρ) :=1
2σp
1−ρ2 p− ≤θ∗(x)≤ 1 2σp
1−ρ2 p+ =: ¯θ(ρ), ∀x∈R. (C.3) So we only need to show thatϕis positive on [θ,θ].¯
We can easily see thatϕisC1 on θ,θ¯
\{0}, its derivative is given by ϕ0(θ) = 2ρ
p1−ρ2
(ρ2−2) sin(θ) + 2(1−ρ2)θcos(θ)
sin3(θ) . (C.4)
A simple study of the sign of the function
θ7−→(ρ2−2) + 2(1−ρ2)cos(θ)
sin(θ), (C.5)
shows that it reaches its maximum on θ,θ¯
at 0 and this maximum is equal to (−ρ2)<0. We deduce that
ρϕ0(θ)≤0, ∀θ∈ θ,θ¯
. (C.6)
We only have two possible situations:
Case ρ >0 : In this case, we have θ=−π+ arctan
p1−ρ2 ρ
!
et ¯θ= arctan
p1−ρ2 ρ
!
. (C.7) On the other hand, the functionϕis decreasing on
θ,θ¯
. In particular, we have, for anyθ∈
θ,θ¯ , ϕ(θ)≥ϕ(¯θ) = 2ρ
p1−ρ2
ρ
p1−ρ2 −(1−ρ2) arctan(
p1−ρ2
ρ )(−1− ρ2 1−ρ2)
! + 1
= 1 +ρ2
1−ρ2 − 2ρ
p1−ρ2arctan
p1−ρ2 ρ
! . We do the following change of variables
0≤y=
p1−ρ2
ρ ⇐⇒ ρ= 1
p1 +y2. (C.8)
We obtain
ϕ(θ)≥ϕ(¯θ) =1 y
2 +y2
y −2 arctany
>0, (C.9) as the minimum of the function (y7−→ 2+yy2 −2 arctany) is reached at the pointy0=
q
3+√ 17
2 and is≈0.78.
Case ρ <0 : In this case, we have θ= arctan
p1−ρ2 ρ
!
et ¯θ=π+ arctan
p1−ρ2 ρ
!
. (C.10) The functionϕis increasing on
θ,θ¯
. In particular, we have, for any θ∈ θ,θ¯
,
ϕ(θ)≥ϕ(θ) = 2ρ p1−ρ2
ρ
p1−ρ2 −(1−ρ2) arctan(
p1−ρ2
ρ )(−1− ρ2 1−ρ2)
! + 1
= 1 +ρ2
1−ρ2 − 2ρ
p1−ρ2arctan
p1−ρ2 ρ
!
. (C.11)
We do the following change of variables 0≤z=−
p1−ρ2
ρ ⇐⇒ ρ= −1
√1 +z2. (C.12) Thusϕ(θ)≥ϕ(θ) = 1z
2+z2
z −2 arctanz
>0.
Appendix D. Proof of Lemma 4.3 The functionϕis defined on [−1,1], by
ϕ(ρ) =σ−2bρ+ρp
σ2+ 4b2−4bρσ
2σ(1−ρ2) , (D.1)
forρ∈]−1,1[, and
ϕ(−1) = σ+ 4b
4(σ+ 2b), ϕ(1) = 4b−σ
4(2b−σ). (D.2)
The functionϕisC1on [−1,1], and its derivative is given, by ϕ0(ρ) = −2b+p
σ2+ 4b2−4bρσ
2σ(1−ρ2) − 2bρσ
2σ(1−ρ2)p
σ2+ 4b2−4bρσ+ 2ρ
1−ρ2p∗(0), (D.3) forρ∈]−1,1[, and
ϕ0(−1) = 2b2σ2
(σ+ 2b)3 + σ2
2(σ+ 2b), ϕ0(1) =
2b2σ2
(2b−σ)3 + σ2 2(2b−σ)
. (D.4)