Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Pathwise analysis and robustnessof hedging strategies forpath-dependent options
Candia Riga(joint work with Rama Cont)
University of Zurich
7th General Advanced Mathematical Methods in Financeand Swissquote Conference 2015
Lausanne, September 7-10
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Outline
1 Motivation
2 Pathwise Calculus for Non-AnticipativeFunctionals
3 A Pathwise Approach to Continuous-TimeTrading
4 A Pathwise Analysis of Hedging Strategies:Hedging Error and Robustness
5 Examples
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Model Ambiguity and Hedging Issues
Classical Framework: Traded assets X “ pX ptqqtPr0,T s modeled as
Rd`-valued semimartingale on pΩ,F , pFtqtPr0,T s,Pq
The choice of P may be challenged ‘a la De Finetti’(Knightian uncertainty)Ñ Our approach: we set up a probability-free financial model
The gain process is a stochastic integral, thus
it is not necessarily defined for a given path/price scenarioscenario analysis and stress tests cannot be performed
Ñ In our setting: for a certain class of trading strategies, thegain process is well-defined path-by-path (as a limit ofRiemann sums)
Robustness analyses are based on the existence of a ‘truemodel’, and study the performance of a ‘mis-specified model’Ñ Our analysis: we study the performance and robustness ofhedging strategies in given sets of scenarios.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Robustness of Hedging Strategies
Consider a market participant who sells an (exotic) option withpayoff H and maturity T on some underlying asset, at a modelprice given by
V ptq “ EQrH|Fts
and hedges the resulting Profit/Loss using the hedging strategyderived from the same model (say, Black-Scholes delta hedge).The actual dynamics of the underlying asset may, of course, bedifferent from the assumed dynamics.
How good is the result of the hedging strategy?
How ‘robust’ is it to model mis-specification?
How does the hedging error relate to model parameters andoption characteristics?
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Robustness of Hedging Strategies
El Karoui, Jeanblanc & Shreve (1998) provided an answer to theseimportant questions, for non-path-dependent options, when theunderlying dynamics is
dSptq “ Sptqrptqdt ` SptqσptqdW ptq under Qsuch that S is square-integrable. Then a hedging strategy,computed in a (mis-specified) Markovian model
dSptq “ Sptqrptqdt ` Sptqσ0pt, SptqqdW ptq
with local volatility σ0 leads to a profit
ż T
0
σ20pt,Sptqq ´ σ
2ptq
2Sptq2e
şTt rpsqds
Γptqhkkkkkkikkkkkkj
B2xx f pt, Sptqqdt
where f is the unique solution of the PDE
Bt f ` rptqxB2x f ` σ2
0pt, xqx2B2
xx f 2 “ rptqf f pT , xq “ Hpxq
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Notation: Non-Anticipative Functionals
Given x P Dpr0,T s,Rdq, for all t P r0,T s we denote:
xptq P Rd the value of x at t
xt “ xpt ^ ¨q P Dpr0,T s,Rdq the path stopped at t
xt´ “ x1r0,tq ` xpt´q1rt,T s P Dpr0,T s,Rdq
for δ P Rd , xδt “ xt ` δ1rt,T s P Dpr0,T s,Rdq
We define the space of stopped paths:
ΛT :“´
r0,T s ˆ Dpr0,T s,Rdq
¯
„,
where pt, xq „ pt 1, x 1q ðñ t “ t 1 and xt “ x 1t , and the metric
d8ppt, xq, pt1, x 1qq “ sup
uPr0,T s
ˇ
ˇxpu ^ tq ´ x 1pu ^ t 1qˇ
ˇ` |t ´ t 1|
Definition (Non-anticipative functional)
A non-anticipative functional on Dpr0,T s,Rdq is a mapF : pΛT , d8q Ñ R.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Notation: Regularity of F : Λ Ñ R
A non-anticipative functional F : ΛT Ñ R is said to be:
continuous at fixed times if for all t P r0,T s
F pt, ¨q :´´
ttu ˆ Dpr0,T s,Rdq
¯
„, || ¨ ||8
¯
ÞÑ R
is continuous
left-continuous, F P C0,0l pΛT q, if
@pt, xq P ΛT , @ε ą 0, Dη ą 0 : @h P r0, ts, @pt ´ h, x 1q P ΛT ,d8ppt, xq, pt ´ h, x 1qq ă η ñ |Ftpxq ´ Ft´hpx
1q| ă ε
boundedness-preserving, F P BpΛT q, if, @K Ă Rd compact,@t0 P r0,T s, DCK ,t0 ą 0 s.t. @t P r0, t0s, @pt, xq P ΛT
xpr0, tsq Ă K ñ |Ftpxq| ă CK ,t0
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Notation: Differentiability of F : Λ Ñ R
Definition (Horizontal derivative)
A non-anticipative functional F is horizontally differentiable atpt, xq P ΛT if t ÞÑ F pt, xtq is right-differentiable, with derivativedenoted by DFpt, xq; if it holds @pt, xq P ΛT , then we denoteDF :“ pDF pt, ¨qqtPr0,T q
Definition (Vertical derivative)
A non-anticipative functional F is vertically differentiable atpt, xq P ΛT if the map Rd Q e ÞÑ F pt, xe
t q is differentiable at 0; inthis case we denote ∇ωFpt, xq :“ pBiF pt, xqqi“1,...,d , where
BiF pt, xq :“ limhÑ0`
F pt, xheit q ´ F pt, xtq
h.
If this holds for all pt, xq P ΛT , then ∇ωF :“ p∇ωF pt, ¨qqtPr0,T s
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Sets of Smooth Non-Anticipative Functionals
Definition (C1,2b pΛq and C1,2
locpΛq)
Denote by C1,2b pΛTq the set of non-anticipative functionals
F P C0,0l pΛT q such that:
DDF continuous at fixed times,
D∇jωF P C0,0
l pΛT q j “ 1, 2,
DF ,∇ωF ,∇2ωF P BpΛT q
(
.
Denote by C1,2locpΛTq the set of non-anticipative functionals
F P C0,0l pΛT q such that there exists a sequence of stopping times
pτkqkě1 going to 8 and a sequence pF k P C1,2b pΛT qqkě1,
F pt, xtq “ÿ
kě1
F kpt, xtq1rτk pxq,τk`1pxqqptq @pt, xq P ΛT
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Quadratic Variation for Cadlag Paths
Fixed sequence of time partitions: Π “ tπnuně1,πn “ pt
ni qi“0,...,mpnq, 0 “ tn0 ă . . . ă tnmpnq “ T , |πn| ÝÝÝÑ
nÑ80
Definition (Paths of finite quadratic variation)
x P Dpr0,T s,Rq is of finite quadratic variation along Π if
@t P r0,T s, rxsptq :“ limnÑ8
ÿ
tni ďt
`
xptni`1q ´ xptni q˘2ă 8
x P Dpr0,T s,Rdq is of finite quadratic variation along Π if,@1 ď i , j ď d, x i , x i ` x j are so. In this case:
rxsi ,jptq ” rxi , x j sptq “
1
2
`
rx i ` x j sptq ´ rx i sptq ´ rx j sptq˘
Denote QpU,Πq :“tx P U Ă Dpr0,T s,Rdq, x is of f.q.v. along Πu
For every cadlag path ω P QpDpr0,T s,Rdq,Πq, we can alwaysassume that suptPr0,T s´πn |∆ωptq| ÝÝÝÑ
nÑ80.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Change of Variable Formula for Functionals
The piecewise constant approximation
ωn :“
mpnq´1ÿ
i“0
ωptni`1´q1rtni ,tni`1q
` ωpT q1tTu,
converges uniformly to ω
Theorem (Cont, Fournie (2010))
If F P C1,2locpΛT q and ω P QpDpr0,T s,Rdq,Πq, then the limit
ż T
0∇ωFtpωt´qd
Πω :“ limnÑ8
mpnq´1ÿ
i“0
∇ωFtnipω
n,∆ωptni qtni ´
q¨pωptni`1q´ωptni qq
exists and
FT pωT q “ F0pω0q `
ż T
0∇ωFtpωt´qd
Πω
`
ż T
0DtF pωt´qdt `
1
2
ż
p0,T str`
∇2ωFtpωt´qdrωs
cptq˘
`ÿ
tPp0,T s
pFtpωtq ´ Ftpωt´q ´∇ωFtpωt´q ¨∆ωptqq
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Functional Ito Formula for Functionals ofSemimartingales
Theorem (Cont, Fournie (2013))
If F P C1,2locpΛT q and X is an Rd -valued semimartingale, then a.s.
F pT ,XT q “ F p0,X0q `
ż T
0∇ωF pt,Xt´q ¨ dX
`
ż T
0DF pt,Xt´qdt `
1
2
ż
p0,T str`
∇2ωF pt,Xt´qdrX s
cptq˘
`ÿ
tPp0,T s
pF pt,Xtq ´ F pt,Xt´q ´∇ωF pt,Xt´q ¨∆X ptqq
In particular, Y defined by Y ptq “ F pt,Xtq for all t P r0,T s, is asemimartingale
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
A Pathwise Setting for Continuous-TimeTrading
Financial market: pΩ, || ¨ ||8q metric space, Ω :“ Dpr0,T s,Rd`q,
ω P Ω is a possible trajectory of the (forward) asset prices,F is the Borel σ-field and F “ pFtqtPr0,T s the canonical filtration
Trading strategies: pV0, φ, ψq, where
V0 Ω ÞÑ R, F0-measurable (initial investment)φ Rd -valued F-adapted caglad process (asset position)ψ R-valued F-adapted caglad process (bond position)
Portfolio value at time t P r0,T s along the price pathω P Dpr0,T s,Rd
`q:
V pt, ω;φ, ψq “ φpt, ωq ¨ ωptq ` ψpt, ωq
When is a strategy self-financing?When can we explicitly define its gain?
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Simple Trading Strategies
Definition (Simple self-financing trading strategies)
pV0, φ, ψq is a simple trading strategy if φ P ΣpRd ,Πq andψ P ΣpR,Πq, where ΣpU,Πq :“ Y
ně1ΣpU, πnq,
ΣpU, πnq :“
"
φ : @i “ 0, . . . ,mpnq ´ 1, Dλi : Ω Ñ U
Ftni-measurable, φpt, ωq “
mpnq´1ÿ
i“0
λi pωq1ptni ,tni`1s
*
A simple trading strategy pV0, φ, ψq is a self-financing if thereexists n ě 1 such that φ, ψ P Σpπnq and
ψpt, ωq :“V0´φp0`, ωq¨ωp0q´
mpnqÿ
i“1
ωptni ^tq¨pφptni`1^t, ωq´φptni ^t, ωqq
Equivalently: V pt, ω;φq “ V0 ` G pt, ω;φq, where
G pt, ω;φq :“
mpnqÿ
i“1
φptni ^ t, ωq ¨ pωptni ^ tq ´ ωptni´1 ^ tqq
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Self-financing Trading Strategies
Definition (Self-financing Trading Strategies on U)
pV0, φq is a self-financing trading strategy on U Ă Dpr0,T s,Rd`q if
there exists a sequence tpV0, φn, ψnq, n P Nu of self-financing
simple trading strategies, such that
@ω P U, @t P r0,T s, φnpt, ωq ÝÝÝÑnÑ8
φpt, ωq,
and any of the following equivalent conditions is satisfied:
(i) DF-adapted cadlag process G p¨, ¨;φq, @t P r0,T s, ω P U
G pt, ω;φnq ÝÝÝÑnÑ8
G pt, ω;φq, ∆G pt, ω;φq “ φpt, ωq∆ωptq;
(ii) DF-adapted cadlag process ψp¨, ¨;φq, such that@t P r0,T s, ω P U ψnpt, ωq ÝÝÝÑ
nÑ8ψpt, ω;φq and
ψpt`, ω;φq ´ ψpt, ω;φq “ ´ωptq pφpt`, ωq ´ φpt, ωqq ;
(iii) DF-adapted cadlag process V p¨, ¨;φq, @t P r0,T s, ω P U
V pt, ω;φnq ÝÝÝÑnÑ8
V pt, ω;φq, ∆V pt, ω;φq “ φpt, ωq∆ωptq.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Gain: Pathwise Construction
Proposition
If there exists F P C1,2locpΛT q X C0,0
r pΛT q, ∇ωF P C0,0pΛT q,
φpt, ωq “ ∇ωF pt, ωt´q @ω P QpΩ,Πq, t P r0,T s,
Then, φ is the asset position of a self-financing trading strategy onQpΩ,Πq with gain
G pt, ω;φq “
ż t
0φpu, ωuqd
Πω
“ limnÑ8
ÿ
tni ăt
∇ωF ptni , ωntni ´q ¨ pωptni`1q ´ ωpt
ni qq
and bond position
ψpt, ω;φq :“V0 ´ φ`p0, ωq`
´ limnÑ8
mnÿ
i“1
ωptni ^ tq ¨ pφnptni`1 ^ t, ωq ´ φnptni ^ t, ωqq
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Hedging Error and Super-Strategies
Definition (Hedging error in specific scenarios)
The hedging error of a self-financing trading strategy pV0, φq onU Ă Dpr0,T s,Rd
`q for a path-dependent derivative with payoff Hin a scenario ω P U is the value
V pT , ω;φq ´ Hpωq “ V0pωq ` G pT , ω;φq ´ Hpωq.
pV0, φq is called a super-strategy for H on U if its hedging error forH is non-negative on U, i.e.
V0pωq ` G pT , ω;φq ě HpωT q @ω P U.
Given A P Dpr0,T s,Sq, S :“ tM P Rdˆd ,M ą 0 symmetricu,
QApΠq :“
"
ω P QpΩ,Πq such that rωsptq “
ż t
0Apsqds @t P r0,T s
*
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Pathwise Replication of (Exotic) Derivatives
Proposition (Pathwise replication of exotic derivatives)
If H : pΩ, ‖¨‖8q ÞÑ R is continuous and F P C1,2locpWT q XC0,0pWT q
solves"
DF pt, ωtq `12tr
`
∇2ωF pt, ωtq ¨ Aptq
˘
“ 0, t P r0,T qF pT , ωq “ Hpωq, @ω P QApΠq
Then, the hedging error of the trading strategy pF0pω0q,∇ωF q,self-financing on QpC pr0,T s,Rdq, in any scenario ω P QApΠq is
1
2
ż T
0tr´
∇2ωF pt, ωtq ¨ pAptq ´ Aptqq
¯
dt
In particular, in all scenarios ω P QApΠq, pF p0, ω0q,∇ωF q replicatesH and its portfolio value at any time t P r0,T s is given by F pt, ωtq.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
A Hedging Formula for Path-DependentOptions
Let Ω :“ C pr0,T s,R` and S denote the coordinate process on thecanonical space pΩ,F ,Fq, i.e. Spt, ωq “ ωptq, @ω P Ω, t P r0,T s.
Assumption (Hedger’s model assumption)
The market participant assumes that the underlying asset price Sevolves according to dSptq “ σptqSptqdW ptq, i.e.
Sptq “ Sp0qeşt
0 σpuqdW puq´ 12
şt0 σpuq
2du, t P r0,T s, (1)
where W is a standard Brownian motion on pΩ,F ,F,Pq and thevolatility σ is a non-negative F-adapted process such that σ ‰ 0dt ˆ dP-almost surely and S is a square-integrable P-martingale.
The discounted price at time t of a path-dependent derivative withpayoff HpST q P L1pΩ,F ,Pq is given by
Y ptq “ EPrHpST q|Fts P-a.s.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
A Universal Hedging Equation
Theorem (Universal hedging formula (Cont-Fournie, 2013))
If EPr|HpST q|2s ă 8, then Y PM2pFq and the following
martingale representation formula holds:
HpST q “ Y p0q `
ż T
0∇SY puq ¨ dSpuq P-a.s.
Moreover, if Y P C1,2b pSq, where
C1,2b pSq :“tY : DF P C1,2
b pWT q, Y ptq“F pt,Stq P-a.s.@t P r0,T su,
then the hedging strategy for H is pathwise defined by∇SY “ ∇ωF p¨, S¨q dt ˆ dP´ a.s.
In case Y PM2pFq but Y R C1,2b pSq, the hedging strategy ∇SY
for H is a ‘weak vertical derivative’, yet it can be uniformlyapproximated by regular functionals (i.e. Lu-Cont 2015)
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
A Universal Pricing Equation
Theorem (Pricing equation for path-dependent derivatives)
Consider a path-dependent derivative with payoff HpST q. If
DF P C1,2b , F pt, Stq “ EPrHpST q|Fts
such that DF P C0,0l pWT q, then F is the unique solution of the
pricing equation
DF pt, ωtq `1
2tr`
∇2ωF pt, ωtq ¨ σ
2ptqω2ptq˘
“ 0, (2)
with the terminal condition F pT , ωq “ Hpωq, on the topologicalsupport of pS ,Pq in pC pr0,T s,R`q, ‖¨‖8q, i.e.
@ω P supppSq :“
ω P Ω, such that PpST P V q ą 0
@V neighborhood of ω in pΩ, ‖¨‖8q(
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Pathwise Analysis of Hedging Performance
Problem:
The hedger sells a path-dependent option with maturity Tand payoff HpST q such that EPr|HpST q|
2s ă 8
He computes the price and hedging strategy according to PHe trades according to that strategy, but taking in input therealized market prices ñ in a scenario ω, the final value ofthe hedging portfolio will differ from HpωT q
What is the performance of the hedging strategypY p0q,∇SY q with respect to H?
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Robustness of Delta-Hedging Strategies
Definition (Robust delta hedge)
The delta-hedging strategy pY p0q,∇SY q is said to be robust for Hon U Ă Ω if it is a super-strategy for H on U.
Notation: For paths of absolutely continuous finite quadraticvariation along Π, we define the local realized volatility as
σreal : r0,T s ˆA Ñ R,pt, ωq ÞÑ σrealpt, ωq “ 1
ωptq
b
ddt rωsptq,
where
A :“ tω P QpΩ,Πq, t ÞÑ rωsptq is absolutely continuousu.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
The Hedging Error Formula and Robustness
Proposition (The hedging error formula and robustness ofdelta-hedging)
If there exists F P C1,2b pWT q X C0,0pWT q such that
DF P C0,0l pWT q, and
F pt, Stq “ EPrHpST q|Fts dt ˆ dP-a.s.
Then, the hedging error of pF p0, ¨q,∇ωF q for H is explicitly givenby 1
2
ż T
0
´
σpt, ωq2 ´ σrealpt, ωq2¯
ω2ptq∇2ωF pt, ωqdt
In particular, if for all ω P U and Lebesgue-a.e. t P r0,T q
∇2ωF pt, ωq ě 0 and σpt, ωq ě σrealpt, ωq
then the delta hedge for H is robust on U.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
The Impact of Jumps
Proposition (Impact of jumps on delta hedging)
If there exists a non-anticipative functional F : ΛT Ñ R such that
F P C1,2b pΛT q X C0,0pΛT q, ∇ωF P C0,0pΛT q, DF P C0,0
l pWT q
F pt,Stq “ EPrHpST q|Fts dt ˆ dP-a.s.
Then, for any ω P QpDpr0,T s,R`q,Πq such that rωsc is absolutelycontinuous, the hedging error of the delta hedge pF p0, ¨q,∇ωF q forH is explicitly given by
1
2
ż T
0
´
σpt, ωq2 ´ σrealpt, ωq2¯
ω2ptq∇2ωF pt, ωqdt
´ÿ
tPp0,T s
pF pt, ωtq ´ F pt, ωt´q ´∇ωF pt, ωt´q ¨∆ωptqq .
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Vertical Smoothness
The previous results extend El Karoui, Jeanblanc & Shreve (1998):- to path-dependent options- to a pathwise setting: it gives the pathwise P&L of the strategyand removes unnecessary probabilistic assumptions
Definition (Vertical smoothness)
A functional h : Dpr0,T s,Rq ÞÑ R is vertically smooth onU Ă Dpr0,T s,Rq if, @pt,wq P r0,T s ˆ U, the real map
ghpt, ω; ¨q : RÑ R, e ÞÑ hpω ` e1rt,T sq
is of class C 2 on a neighborhood V of 0 and there existK , c , β ą 0 such that, for all ω, ω1 P U, t, t 1 P r0,T s,∣∣∣Beghpe; t, ωq
∣∣∣` ∣∣∣B2eeghpe; t, ωq
∣∣∣ ď K , e P V ,∣∣Beghp0; t, ωq ´ Beghp0; t 1, ω1q∣∣` ∣∣B2
eeghp0; t, ωq ´ B2eeghp0; t 1, ω1q
∣∣ď cp‖ω ´ ω1‖8 ` |t ´ t 1|βq.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Pricing Functional: Existence andRegularity
Proposition (Pricing functional: existence and regularity)
Let H : pDpr0,T s,Rq, ‖¨‖8q Ñ R a locally-Lipschitz payofffunctional such that EPr|HpST q|s ă 8 and define
h : pDpr0,T s,Rq Ñ R, hpωT q “ HpexpωT q,
where expωT ptq :“ eωptq for all t P r0,T s.
If h is vertically smooth on Cpr0,T s,R`q, then
DF P C0,2b pWT q X C0,0pWT q
such that
F pt,Stq “ EPrHpST q|Fts dt ˆ dP-a.s.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Propagation of ‘Vertical Convexity’
Definition (Vertical convexity of non-anticipative functionals)
A non-anticipative functional G : ΛT Ñ R is called verticallyconvex on U Ă ΛT if, for all pt, ωq P U, there exists aneighborhood V Ă R of 0 such that the map
V Ñ R, e ÞÑ G`
t, ω ` e1rt,T s˘
is convex.
Proposition (Propagation of vertical convexity)
Assume that, for all pt, ωq P Tˆ supppS ,Pq, there exists aninterval I Ă R, 0 P I, such that the map
vHp¨; t, ωq : I Ñ R, e ÞÑ vHpe; t, ωq “ H`
ωp1` e1rt,T sq˘
is convex. If
DF P C0,2b pWT q, F pt,Stq “ EPrHpST q|Fts dt ˆ dP-a.s.,
then F is vertically convex on Tˆ supppS ,Pq. In particular:
∇2ωF pt, ωq ě 0, @pt, ωq P Tˆ supppS ,Pq.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Example: Discretely-Monitored ExoticDerivatives with Black-Scholes
Lemma (Discretely-monitored path-dependent derivativeswith Black-Scholes)
Let σ : r0,T s Ñ R` such thatşT
0 σ2ptqdt ă 8.
Assume that H : Dpr0,T s,R`q and there exist a partition0 “ t0 ă t1 ă . . . ă tn ď T and a function h P C 2
b pRn;R`q suchthat
@ω P Dpr0,T s,R`q, HpωT q “ hpωpt1q, ωpt2q, . . . , ωptnqq.
Then,
DF P C1,2locpWT q, F pt,Stq “ EPrHpST q|Fts dt ˆ dP-a.s.,
and the horizontal and vertical derivatives of F are given in aclosed form.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Example: Hedging Asian Options withBlack-Scholes
Consider an arithmetic Asian call option
HpST q “
ˆ
1
T
ż T
0Spuqdu ´ K
˙`
and assume σ : r0,T s Ñ R` such thatşT
0 σ2ptqdt ă 8.
The pricing functional F is given by: @pt, ωq PWT ,
F pt, ωtq “ f pt, aptq, ωptqq, aptq “
ż t
0ωpsqds,
where f P C1,1,2pr0.T q ˆ R` ˆ R`q X Cpr0.T s ˆ R` ˆ R`q solves
#
σ2ptqx2
2 B2xx f pt, a, xq ` xBaf pt, a, xq ` Bt f pt, a, xq “ 0,
f pT , a, xq “ g`
aT
˘ (3)
Note that (3) turns out to be a particular case of the universalpricing equation.
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Example: Robustness of Black-ScholesDelta-Hedging for Asian Options
Corollary (Robustness of delta-hedging for Asian options)
If the Black-Scholes volatility term structure over-estimates thelocal realized volatility on AX supppS ,Pq, i.e.
σptq ě σrealpt, ωq @ω P AX supppS ,Pq,
Then the Black-Scholes-delta hedge for the arithmetic Asian calloption is robust on AX supppS ,Pq. Moreover, the hedging erroris given by
1
2
ż T
t
´
σpuq2 ´ σrealpt, ωq2¯
ω2puqB2xx f pu, apuq, ωpuqqdu
where f is the solution to the Cauchy problem (3).
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Example: Hedging Asian Options withHobson-Rogers
dSptq “ SptqσnpOptqqdW ptq,where
dOptq “ σnpOptqqdW ptq ´1
2pσnpOptqq2 ` λOptqqdt.
Consider a geometric Asian call option
HpST q “
´
eMpT q ´ K¯`
, MpT q “1
T
ż T
0log Spuqdu
The pricing functional F is given by: @pt, ωq PWT ,
F pt, ωq “ upT ´ t, logωptq, logωptq ´ opt, ωq, gpt, ωqq,
where u is the classical solution of the following Cauchy problemon r0.T q ˆ Rˆ Rˆ R#
12σ
npx1 ´ x2q2pB2
x1x1u ´ Bx1uq ` λpx1 ´ x2qBx2u ` x1Bx3u ´ Btu “ 0,
up0, x1, x2, x3q “ ΨG pex1 , x3T q.
(4)(4) is another particular case of the universal pricing equation
Motivation Pathwise Functional Calculus Pathwise Continuous-Time Trading Pathwise analysis of Hedging Examples
Hedging Asian Options with Path-DependentModels
Corollary
If σnpopt, ωqq ě σrealpt, ωq @ω P AX supppS ,Pq, then theHobson-Rogers delta hedge for H is robust on AX supppS ,Pq.Moreover, the hedging error at maturity is given by
12
şT0
`
σnpopt, ωqq2 ´ σmktpt, ωq2˘
ω2ptqB2xupT ´ t, logωptq, logωptq ´ opt, ωq, gpt, ωqqdt,
where u is the solution of the Cauchy problem (4).
Other models that generalize Hobson-Rogers and allow to derivefinite-dimensional Markovian representation for S and itsarithmetic mean are given by
Pascucci,Foschi 2006,Salvatore, Tankov 2014.
They thus guarantee the existence of a smooth pricing functionalfor arithmetic Asian options, then robustness of the delta hedgecan be proved the same way as we showed in the Black-Scholesand Hobson-Rogers cases.