on the link between the malliavin derivative operator and the...
TRANSCRIPT
On the link between the Malliavin derivativeoperator and the implied volatility behaviour .
Can we expect the Malliavin calculus to be useful in applications?
Elisa Alòs
Universitat Pompeu FabraBarcelona
Mathematics of Quantitative Finance, Oberwolfach 2017
Jointly with Jorge León and Josep Vives
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 1 / 58
Contents
1 Malliavin calculus : a useful tool ?
2 A modeling problem
3 Some extensions and applications
4 Conclusions
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 2 / 58
Contents
1 Malliavin calculus : a useful tool ?
2 A modeling problem
3 Some extensions and applications
4 Conclusions
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 3 / 58
Malliavin calculus : a useful tool ?
After a short search in the net, we can easily find sentences like :
Not to be a party-pooper, but Malliavin calculus is essentially uselessin finance. Any practical result ever obtained with Malliavin calculuscan be obtained by much simpler methods by eg differentiating thedensity of the underlying process.
So, is Malliavin calculus just a ’language’ used bymathematicians ?
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 4 / 58
Malliavin calculus : a useful tool ?
In this talk, we will take a walk through some recent applications ofMalliavin calculus.We will see examples where the Malliavin calculus techniques havebeen used to solve real problems.We will discuss about how, when and why Malliavin calculus canbecome a useful tool.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 5 / 58
Contents
1 Malliavin calculus : a useful tool ?
2 A modeling problem
3 Some extensions and applications
4 Conclusions
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 6 / 58
A modeling problemWe want to construct models for stock prices that reproduce theempirical properties of the implied volatility surface
Stock : Apple, Expiration : 16/4/2010. Data : courtesy of Rafael DeSantiago (IESE, Barcelona)
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 7 / 58
A modeling problem
In particular, a popular rule-of-thumb states (see Lee 2002) that skewslopes decay with maturity approximately as 1√
T .
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 8 / 58
A modeling problem
Is it possible to reproduce this short-time behaviour by usingstochastic volatility models ?
The literature, around 2005, was no so clear : It was commonlyassumed that this empirical fact could be explained by introducingjumps in the asset model. But not by introducing stochasticvolatilities. But at the same time :
Fouque, Papanicolaou, Sircar and Solna : "Maturity cycles in impliedvolatility." Finance and Stochastics 8.4 (2004) : 451-477.
Time-periodic volatility coefficients are able to reproduce theempirical skew.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 9 / 58
A modeling problem
Can we determine the class of stochastic volatility modelsthat can reproduce this skew behaviour ?
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 10 / 58
A modeling problem
Let us try to determine this class of stochastic volatilitymodels
Our objective will be to compute the at-the-money-skew, as thederivative of the implied volatility as a function of the log-stock price.Then we will compute its short-time limit.We will do this computations not for a concrete model, but for a bigclass of stochastic volatility models that will include both the classicaldiffusion volatilities and the time-periodic volatilities introduced byFouque, Papanicolaou, Sircar and Solna.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 11 / 58
A modeling problem : Why Malliavin calculus ?
Because of the Clark-Ocone formula, under some general hypotheses
F =∫ t
0Er (DrF )dWr ,
where D denotes the Malliavin derivative operator. Then, all therelevant information is included in the Malliavin derivative ofF .Because the future average volatility
1T − t
∫ T
tσ2
s ds
is a non-adapted process. And the anticipating Itô’s formula is anatural tool to work with anticipating processes
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 12 / 58
A modeling problem : Notations
In this talk we will consider the following model for the log-price of astock under a risk-neutral probability measure Q :
Xt = x + rt − 12
∫ t
0σ2
s ds
+∫ t
0σs(ρdWs +
√1− ρ2dBs), t ∈ [0,T ].
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 13 / 58
A modeling problem : Notations
In this talk we will consider the following model for the log-price of astock under a risk-neutral probability measure Q :
Xt = x + rt − 12
∫ t
0σ2
s ds
+∫ t
0σs(ρdWs +
√1− ρ2dBs), t ∈ [0,T ].
Here, x is the current log-price.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 14 / 58
A modeling problem : Notations
In this talk we will consider the following model for the log-price of astock under a risk-neutral probability measure Q :
Xt = x + r t − 12
∫ t
0σ2
s ds
+∫ t
0σs(ρdWs +
√1− ρ2dBs), t ∈ [0,T ].
Here, x is the current log-price, r is the instantaneous interest rate.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 15 / 58
A modeling problem : Notations
In this talk we will consider the following model for the log-price of astock under a risk-neutral probability measure Q :
Xt = x + r t − 12
∫ t
0σ2
s ds
+∫ t
0σs(ρdWs +
√1− ρ2dBs), t ∈ [0,T ].
Here, x is the current log-price, r is the instantaneous interest rate,W and B are independent standard Brownian motions.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 16 / 58
A modeling problem : Notations
In this talk we will consider the following model for the log-price of astock under a risk-neutral probability measure Q :
Xt = x + r t − 12
∫ t
0σ2
s ds
+∫ t
0σs(ρdWs +
√1− ρ2dBs), t ∈ [0,T ].
Here, x is the current log-price, r is the instantaneous interest rate,W and B are independent standard Brownian motions, ρ ∈ (−1, 1).
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 17 / 58
A modeling problem : Notations
In this talk we will consider the following model for the log-price of astock under a risk-neutral probability measure Q :
Xt = x + r t − 12
∫ t
0σ2
s ds
+∫ t
0σs(ρdWs +
√1− ρ2dBs), t ∈ [0,T ].
Here, x is the current log-price, r is the instantaneous interest rate,W and B are independent standard Brownian motions, ρ ∈ (−1, 1).Here, the volatility process σ is a square-integrable stochastic processwith right-continuous trajectories, adapted to the filtration generatedby W .
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 18 / 58
A modeling problem : Notations
It is well-known the price of an European call with strike price K isgiven by
Vt = e−r(T−t)E [(eXT − K )+|Ft ],
where E is the expectation with respect to Q and F := FW ∨ FB.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 19 / 58
A modeling problem : Notations
BS(t, x , σ) will denote the price of an European call option under theclassical Black-Scholes model with constant volatility σ, current logstock price x , time to maturity T − t, strike price K and interest rater :
BS(t, x , σ) = exN(d+)− Ke−r(T−t)N(d−),
where N denotes the cumulative probability function of the standardnormal law and
d± := x − x∗tσ√T − t
± σ
2√T − t,
with x∗t := lnK − r(T − t).
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 20 / 58
A modeling problem : Notations
LBS (σ) will denote the Black-Scholes differential operator, in the logvariable, with volatility σ :
LBS(σ) = ∂t + 12σ
2∂2xx + (r − 1
2σ2)∂x − r ·
It is well known that LBS(σ)BS(·, ·, σ) = 0. Moreover, we define
vt :=√
1T − t
∫ T
tσ2
s ds.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 21 / 58
A modeling problem : Implied volatility
Let It(Xt) denote the implied volatility process, which is anF -adapted process such that
Vt = BS(t,Xt , It(Xt)).
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 22 / 58
A modeling problem : Objective
We want to computelimT→t
∂It∂Xt
(x∗t ).
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 23 / 58
A modeling problem : Malliavin calculusLet us consider a standard Browian motion W = Wt ,t ∈ [0,T ]defined in a complete probability space (Ω,F ,P) .SetH = L2 ([0,T ]) , and denote W (h) =
∫ T0 h (s) dWs the Wiener
integral of a deterministic function h ∈ H .
Let S be the set of random variables of the form
F = f (W (h1) , ...,W (hn)) ,
where n ≥ 1, f ∈ C∞b (Rn) (f and all its derivatives are bounded), andh1, .., hn ∈ H . Given a random variable F of this form, we define itsderivative as the stochastic process
DW
t F , t ∈ [0,T ]given by
DWt F =
n∑i=1
∂f∂xi
(W (h1) , ...,W (hn)) hi (t) , t ∈ [0,T ] .
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 24 / 58
A modeling problem : The Malliavin derivative
Example : If F = WT , DWt F = 1[0,T ](t).
Example : If F = W HT =
∫ T0 K (T , s)dWs , DW
t F = K (T , s)1[0,T ](t).
We see that the Malliavin derivatives of both processes arevery different
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 25 / 58
A modeling problem : Some useful properties ofthe Malliavin derivative
Notice that, for an adapted process u and for s > t
DWs ut = 0
The Malliavin derivative operator satisfies the chain rule : for anyF ∈ D1,2
W and any function f satisfying some regularity conditions
Dt f (F ) = f ′(F )DWt F
Example : If σt = σ0E(νW Ht ) (see Bayer, Friz and Gatheral (2016)),
DWr σt = σtνK (t, r).
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 26 / 58
A modeling problem : Some useful properties ofthe Malliavin derivative
In particular, the two above properties give us that, for s > t
Ds(BS(t,Xt ,K , vt
))= ∂BS
∂σ
(t,Xt ,K , vt
)DW
s vt
= ∂BS∂σ
(t,Xt ,K , vt
) 12(T − t)vt
( ∫ T
sDW
s σ2r dr
)
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 27 / 58
A modeling problem : The Skorohod integral
We denote by δW the adjoint of the derivative operator DW :
E (δW (u)F ) = E∫ T
0(DW
s F )usds
The operator δW is an extension of the Itô integral in the sense thatthe set L2
a ([0,T ]× Ω) of square integrable and adapted processes isincluded in Domδ and the operator δ restricted to L2
a ([0,T ]× Ω)coincides with the Itô stochastic integral defined in [16]). We willmake use of the notation δ (u) =
∫ T0 utdWt .
The expectation of the Skorohod integral is zero.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 28 / 58
A modeling problem : The Skorohod integral
limn→∞
n−1∑i=1
g(iTn ,XiT/n,YiT/n
)(W(i+1)T/n −WiT/n)
=∫ T
0g(s,Xs ,Ys)dWs +
∫ T
0(DW
s )− (g(s,Xs ,Ys)) ds
=∫ T
0g(s,Xs ,Ys)dWs +
∫ T
0∂y(s,Xs ,Ys)
(∫ T
sDW
s σ2r dr
)ds,
where
(DWs )− (g(s,Xs ,Ys)) := limr→s(DW
s )− (g(r ,Xr ,Yr ))
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 29 / 58
The anticipating Itô’s formulaWe define Yt =
∫ Tt σ2
r dr . Then
f (T ,XT ,YT )= f (t,Xt ,Yt)
+∫ T
t
∂f∂s (s,Xs ,Ys)ds
+∫ T
t
∂f∂x (s,Xs ,Ys)
((r − σ2
s2
)ds + σs
(ρdWs +
√1− ρ2dBs
))
+12
∫ T
t
∂2f∂x2 (s,Xs ,Ys)σ2
s ds
+∫ T
t
∂f∂y (s,Xs ,Ys)dYs
+ρ∫ T
0(DW
s )−(∂f∂x (s,Xs ,Ys)
)σsds
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 30 / 58
A modeling problem : The anticipating Itô’sformulaThat is
f (T ,XT ,YT )
= f (t,Xt ,Yt) +∫ T
t
∂f∂s (s,Xs ,Ys)ds
+∫ T
t
∂f∂x (s,Xs ,Ys)((r − σ2
s )ds + σs(ρdWs +√1− ρ2dBs))
+12
∫ T
t
∂2f∂x2 (s,Xs ,Ys)σ2
s ds
−∫ T
t
∂f∂y (s,Xs ,Ys)σ2
s
+ρ∫ T
0
( ∂2f∂x∂y (s,Xs ,Ys)
)Λsσsds,
where Λs :=∫ T
s DWs σ2
r dr .Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 31 / 58
A modeling problem : computing the derivative ofthe implied volatility
How can we compute the derivative of the implied volatility ?
Take into account that
∂Vt
∂Xt= ∂xBS(t,Xt , It(Xt)) + ∂σBS(t,Xt , It(Xt)) ∂It
∂Xt(Xt).
Then, if we find a way to compute the derivative of Vt , it seems wecan compute the derivative of the implied volatility.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 32 / 58
A modeling problem : computing the derivative ofthe implied volatility
Previous results (see for example Renault and Touzi (1996)) give usthat, when ρ = 0
∂It∂Xt
(x∗t ) = 0.
So the skew effect is given by the correlation. So we need tounderstand the effect of the correlation on option prices, and then onthe implied volatility.To this end, we will prove a formula that decomposes option prices asthe sum of the price when ρ = 0, plus a term due by correlation.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 33 / 58
A modeling problem : An extension of the Hull andWhite formula
TheoremUnder some regularity conditions in the sense of Malliavin calculus,we have the following decomposition formula
Vt = E (BS(t,Xt , vt)|Ft)
+ ρ
2E(∫ T
te−r(s−t)∂xG(s,Xs , vs)σsΛsds|Ft
)
where G = (∂2xx − ∂x )BS
Notice that this expression is an exact decomposition.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 34 / 58
A modeling problem : proof of the decompositionformula
Applying Itô’s formula to the process
t → e−rtBS(t,Xt , vt)
and taking conditional expectations we get
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 35 / 58
A modeling problem : proof of the decompositionformula
E(e−rTBS(T ,XT , vT )
∣∣∣Ft)
= E(e−rtBS(t,Xt , , vt)
∣∣∣Ft)
+∫ T
tLBS(vs)BS(s,Xs , vs)ds
+12E
(∫ T
te−rs∂σBS(s,Xs , vs) v 2
s − σ2s
vs(T − s)ds∣∣∣Ft
)
+E(∫ T
te−rs∂2
xσBS(s,Xs , vs) σsρΛWs
2vs(T − s)ds∣∣∣Ft
)
+12E
(∫ T
te−rs
(∂2
xx − ∂x)BS(s,Xs , vs)
(σ2
s − v 2s
)ds∣∣∣Ft
)
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 36 / 58
A modeling problem : proof of the decompositionformula
Now, taking into account that
LBS(vs)BS(s,Xs , vs) = 0
and
∂2xσBS(s,Xs , vs) 1
vs(T − s)=(∂3
xxx − ∂2xx
)BS(s,Xs , vs)
the result follows.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 37 / 58
A modeling problem : An expression for thederivative of the implied volatility
Let It(Xt) denote the implied volatility process, which is anF -adapted process such that
Vt = BS(t,Xt , It(Xt)).
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 38 / 58
A modeling problem : An expression for thederivative of the implied volatilityPropositionUnder some general assumptions
∂It∂Xt
(x∗t ) =E (∫ T
t (∂xF (s,Xs , vs)− 12F (s,Xs , vs))ds|Ft)
∂σBS(t, x∗t , It(x∗t ))
∣∣∣∣∣Xt=x∗
t
,
whereF (s,Xs , vs) = ρ
2e−r(s−t)∂xG(s,Xs , vs)σsΛs
This is the expression from where we will compute theshort-time limit of the at-the-money derivative
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 39 / 58
A modeling problem : An expression for thederivative of the implied volatility
To prove this result, notice that
Vt = BS(t,Xt , It(Xt))
and
Vt = E (BS(t,Xt , vt)|Ft) + E(∫ T
tF (s,Xs , vs)ds|Ft
),
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 40 / 58
A modeling problem : An expression for thederivative of the implied volatility
Then we have
∂Vt
∂Xt= ∂xBS(t,Xt , It(Xt)) + ∂σBS(t,Xt , It(Xt)) ∂It
∂Xt(Xt).
and
∂Vt
∂Xt= E (∂xBS(t,Xt , vt)|Ft) + E
( ∫ T
t∂xF (s,Xs , vs)ds|Ft
).
Then the result follows after some algebra.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 41 / 58
A modeling problem : Short-time limit behavior
We consider that the volatility is a lower-bounded process, regularenough (in the Malliavin calculus sense) and that There exists aconstant δ > −1
2 such that, for all 0 < t < s < r < T ,
E((
DWs σr
)2∣∣∣∣Ft
)≤ C (r − s)2δ ,
E((
DWθ DW
s σr)2∣∣∣∣Ft
)≤ C (r − s)2δ (r − θ)2δ .
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 42 / 58
A modeling problem : Short-time limit behaviorPropositionAssume that the previous hypotheses hold. Then :
∂σBS(t, x∗t , It(x∗t )) ∂It∂Xt
(x∗t )
= ρ
2E (L(t, x∗t , vt)∫ T
tΛsds|Ft) + O(T − t)(1+2δ)∧1
as T → t, where
L(t, x∗t , vt) = (∂2xx −
12∂x )G(t, x∗t , vt)
Now, we only need to take limits !
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 43 / 58
A modeling problem : Short-time limit behavior
TheoremSuppose that the previous hypotheses (jointly with some continuityproperties on the volatility process) holld. Then1) Assume that δ is nonnegative and that there exists aFt-measurable random variable D+
t σt such that, for every t > 0,
sups,r∈[t,T ]∣∣∣E ((DW
s σr − D+t σt
)∣∣∣Ft)∣∣∣→ 0,
a.s. as T → t. Then
limT→t
∂It∂Xt
(x∗t ) = − ρ
2σtD+
t σt .
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 44 / 58
A modeling problem : Short-time limit behavior
Theorem2) Assume that δ is negative and that there exists a Ft-measurable
random variable Lδ,+t σt such that, for every t > 0,
1(T − t)2+δ
∫ T
t
∫ T
sE(DW
s σr∣∣∣Ft
)drds − Lδ,+t σt → 0,
a.s. as T → t. Then
limT→t
(T − t)−δ ∂It∂Xt
(x∗t ) = − ρ
σtLδ,+t σt .
We have solved the puzzle ! ! !
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 45 / 58
A modeling problem : Examples
ExampleAssume σ = f (Y ), where f ∈ C1
b(R) and Y is the solution of astochastic differential equation :
dYr = a (r ,Yr ) dr + b (r ,Yr ) dWr ,
for some real functions a, b ∈ C1b(R).
limT→t
∂It∂Xt
(x∗t ) = − ρ
2σt(f ′(Yt)b (t,Yt)),
that agrees with previous limit results.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 46 / 58
ExampleAssume σ = f (Y ), where f ∈ C1
b(R) and Y is a process of the form
Yr = m + (Yt −m) e−α(r−t) + c√2α∫ r
te−α(r−s)dW H
s ,
where W Hs :=
∫ s0 (s − u)H− 1
2dWu.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 47 / 58
ExampleAssume σ = f (Y ), where f ∈ C1
b(R) and Y is a process of the form
Yr = m + (Yt −m) e−α(r−t) + c√2α∫ r
te−α(r−s)dW H
s ,
where W Hs :=
∫ s0 (s − u)H− 1
2dWu.
Case H > 1/2. In this case∫ r
te−α(r−s)dW H
s
=(H − 1
2
) ∫ r
0
(∫ r
s11[t,r ](u)e−α(r−u)(u − s)H− 3
2du)dWs .
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 48 / 58
ExampleAssume σ = f (Y ), where f ∈ C1
b(R) and Y is a process of the form
Yr = m + (Yt −m) e−α(r−t) + c√2α∫ r
te−α(r−s)dW H
s ,
where W Hs :=
∫ s0 (s − u)H− 1
2dWu.
Case H > 1/2. In this case∫ r
te−α(r−s)dW H
s
=(H − 1
2
) ∫ r
0
(∫ r
s11[t,r ](u)e−α(r−u)(u − s)H− 3
2du)dWs .
andlimT→t
∂It∂Xt
(x∗t ) = 0.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 49 / 58
ExampleAssume σ = f (Y ), where f ∈ C1
b(R) and Y is a process of the form
Yr = m + (Yt −m) e−α(r−t) + c√2α∫ r
te−α(r−s)dW H
s ,
where W Hs :=
∫ s0 (s − u)H− 1
2dWu.
Case H < 1/2. In this case∫ r
te−α(r−s)dW H
s
=(12 − H
) ∫ r
0
(∫ r
s
[11[t,r ](u)e−α(r−u) − 11[t,r ](s)e−α(r−s)
]×(u − s)H− 3
2du)dWs
+∫ r
te−α(r−s)(r − s)H− 1
2dWs .
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 50 / 58
ExampleAssume σ = f (Y ), where f ∈ C1
b(R) and Y is a process of the form
Yr = m + (Yt −m) e−α(r−t) + c√2α∫ r
te−α(r−s)dW H
s ,
where W Hs :=
∫ s0 (s − u)H− 1
2dWu.
Case H < 1/2. In this case∫ r
te−α(r−s)dW H
s
=(12 − H
) ∫ r
0
(∫ r
s
[11[t,r ](u)e−α(r−u) − 11[t,r ](s)e−α(r−s)
]×(u − s)H− 3
2du)dWs +
∫ r
te−α(r−s)(r − s)H− 1
2dWs .
andlimT→t
(T − t) 12−H ∂It
∂Xt(x∗t ) = −c
√2α ρσt
f ′ (Yt) .Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 51 / 58
ExampleIf σt = σ0E(νW H
t ) (see Bayer, Friz and Gatheral (2016)) we get
DWr σt = νσtK (t, r)
and then, if K (t, r) ∼ (t − r)H−1/2, (H) holds with δ = H − 1/2.Moreover,
limT→t
(T − t) 12−H ∂It
∂Xt(x∗t ) = − ρν
(H + 1/2)(H + 3/2)
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 52 / 58
Contents
1 Malliavin calculus : a useful tool ?
2 A modeling problem
3 Some extensions and applications
4 Conclusions
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 53 / 58
Some extensions and applications : The secondderivative
We can study also the at-the-money short-time limit of the secondderivative. The first step is to study this derivative in the uncorrelatedcase. Then by using similar techniques, we can prove that, undersome regularity conditions
limT→t
∂2I∂k2 (t, k∗t ) =
( 112−
724ρ
2) (D+
t σ2t )2
σ5t
+ ρ2
6σ3t
((D+
t
)2σ2
t
),
which also agrees with previous results. Moreover, in the case ofnegative δ,
∂2It∂k2 (k∗t ) ∼ (T − t)2δ
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 54 / 58
Contents
1 Malliavin calculus : a useful tool ?
2 A modeling problem
3 Some extensions and applications
4 Conclusions
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 55 / 58
Conclusions
We have seen that (in the context of stochastic volatility models) theproperties of the implied volatility (and VIX, VVIX, etc.) can be’translated’ in terms of the Malliavin derivative of the volatilityprocess. Then, if we want to construct a model satisfying someproperties :
a) We will be able to see easily if there exist a model satisfying thisset or properties. Moreover, if the answer is yes :
b) We will be able to construct this model.
We have a powerful tool in modeling problems !
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 56 / 58
Open problems
Out-of-the-money implied volatilities, long-time behaviour, etc. etc.
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 57 / 58
Many thanks !
Elisa Alòs (Universitat Pompeu Fabra) Implied Volatility and Malliavin derivatives Oberwolfach 2017 58 / 58