• Nie Znaleziono Wyników

Ald´ericJoulinUniversit´edeLaRochelle(France)Workshop“Stochasticandharmonicanalysisofprocesseswithjumps”2-9May2006-Angers Onmaximalinequalitiesforstablestochasticintegrals

N/A
N/A
Protected

Academic year: 2022

Share "Ald´ericJoulinUniversit´edeLaRochelle(France)Workshop“Stochasticandharmonicanalysisofprocesseswithjumps”2-9May2006-Angers Onmaximalinequalitiesforstablestochasticintegrals"

Copied!
22
0
0

Pełen tekst

(1)

On maximal inequalities for stable stochastic integrals

Ald´eric Joulin

Universit´e de La Rochelle (France)

Workshop

“Stochastic and harmonic analysis of processes with jumps”

2-9 May 2006 - Angers

(2)

Content

1 Motivations

2 Pruitt’s truncation method

3 Large range estimates

4 Small range estimates

5 Application to first crossing problems

(3)

Motivations

Z α-stable process, H predictable integrable.

Our aim: to understand the behavior of H · Z as α is “close to 2”, i.e. to give upper bounds on:

- Large range: P sup

t∈[0,1]

Z t 0

HsdZs

≥ x

!

, x large.

- Small range: P sup

t∈[0,1]

Z t 0

HsdZs ≥ x

!

, x small.

Main difficulties:

- Absence of close formulae for stable densities.

- Infinite variance ⇒ classical maximal inequalities fail.

(4)

Motivations

Pruitt ’81. Regularity of paths of L´evy processes. In the SαS case:

P sup

t∈[0,1]

|Zt| ≥ x

!

≤ D1

α(2 − α)xα, xα> D1 α(2 − α). Optimal in x since behavior similar to ν ({|x |, ∞}).

Linear explosion as α → 2.

Proof: Truncation of the jumps + Doob’s maximal inequalities.

(5)

Motivations

Gin´e-Marcus ’83. To derive a CLT for stable stochastic integrals. Generalization of Pruitt’s result: for any

xα > D2kHkαα α(2 − α)2,

P sup

t∈[0,1]

Z t 0

HsdZs

≥ x

!

≤ D2

α(2 − α)2xα Z 1

0

E|Ht|αdt.

Optimal in x .

H ∈ Lα(Ω × [0, 1]) is sufficient.

Explosion of quadratic order as α → 2.

(6)

Motivations

Hult-Lindskog ’06. Extremal behavior of L´evy-type stochastic integrals:

P sup

t∈[0,1]

Z t

0

HsdZs

≥ x

!

x →∞ Dα

αxα Z 1

0

E|Ht|αdt.

Uniform integrability assumption E supt∈[0,1]|Ht|α+p < ∞.

Dα bounded in α ⇒ No explosion phenomenon !

⇒ loss of information from asymptotic to non-asymptotic estimates.

(7)

Motivations

Natural question: is it possible to avoid the explosion of the upper bound as α → 2 and to involve at the same time optimal bounds in terms of the deviation level x and the Lα-norm of H ? I do not know.

Our purposes: to conserve the optimal rate x−α and:

- to weaken the speed of explosion in Gin´e-Marcus’ estimate, with the same optimal Lα-norm.

- to avoid the explosion as α is close to 2, at the price of stronger integrability assumptions on H.

(8)

Pruitt’s truncation method

Z α-stable with characteristics (b, 0, ν), where ν stable L´evy measure:

ν(dy ) = c1{y <0}+ c+1{y >0} dy

|y |α+1. L´evy-Itˆo decomposition with truncation level R:

Zt = bRt + Z t

0

Z

|y |≤R

y ˜µ(dy , dt) + Z t

0

Z

|y |>R

y µ(dy , dt)

= drift + L2-martingale Zt(R−) + compound Poisson Zt(R+).

(9)

Pruitt’s truncation method

Construction of stable stochastic integrals:

H ∈ L2(Ω × [0, 1]).

Xt := A(R)t + Xt(R−)+ Xt(R+), with A(R)t := bR

Z t 0

Hsds, Xt(R−):= H·Zt(R−), Xt(R+) := H·Zt(R+).

(10)

Pruitt’s truncation method

Crucial point: the truncation level R is chosen at the end of the proofs, depending on:

the deviation level x . a suitable Lp-norm kHkp.

(11)

Large range estimates (1)

Proposition

Underlying driving Z SαS. Then

P sup

t∈[0,1]

|Xt| ≥ x

!

≤ D

(2 − α)xα Z 1

0

E|Ht|αdt, xα ≥ DkHkαα 2 − α .

(12)

Large range estimates (1)

Remark

Optimal in the deviation level x and the Lα-norm of H.

Linear explosion (instead of quadratic) as α → 2:

⇒ generalization of Gin´e-Marcus’ estimate.

Proof: adaptation to stable stochastic integrals of

Bass-Levin-Vondracek’s method + choice of truncation level R = x .

(13)

Large range estimates (2)

For non-explosion as α is close to 2, we need the tricky inequality:

P sup

t∈[0,1]

|Xt| ≥ x

!

≤ P sup

t∈[0,1]

|A(R)t | ≥ x/2

! (1)

+P sup

t∈[0,1]

|Xt(R−)| ≥ x/2

! (2)

+P sup

t∈[0,1]

|Xt(R+)| > 0

!

≤ (1) + (2) + K αRα.

⇒ Reduced to bound suitably (2), avoidingR

|y |≤Ry2ν(dy )

(14)

Large range estimates (2)

Theorem

H ∈ Lα+p(Ω × [0, 1]), p ∼ 2 − α. Then for xα ≥ K kHkαα+p

(2 − α)β , β := βα+p > 2 :

P sup

t∈[0,1]

|Xt| ≥ x

!

≤ M xα

Z 1 0

E|Ht|α+pdt

α/(α+p)

.

(15)

Large range estimates (2)

Remark

Optimal in x and no longer explosion as α → 2.

Price to pay compared to Gin´e-Marcus’ estimate:

- H ∈ Lα+p and not only in Lα.

- Restriction on the range of validity of x (β > 2).

Proof: moment estimates + choice of truncation level R = x /kHkα+p.

(16)

Small range estimates

Theorem

Underlying driving Z SαS, H bounded. Then for any  > 0 and any x ∈ (0, xα,),

P sup

t∈[0,1]

Xt ≥ x

!

≤ (1 + ) exp −Kα

 x

kHk∞,α

α/(α−1)! .

(17)

Small range estimates

Remark

Different behaviors as x is small or large. What about intermediate range ?

As α → 2, recover the Gaussian case ?

(18)

Small range estimates

Corollary

Yes: choosing a suitable weight in the stable L´evy measure and taking limit as α → 2 yields

P sup

t∈[0,1]

Bt≥ x

!

≤ exp −x2/2 , x > 0.

(19)

Application to first crossing problems

Z SαS. Stable (recurrent) O.U. process of parameter λ > 0:

Xt(λ):=

Z t 0

e−λ (t−s)dZs, t ≥ 0, X0(λ )= 0.

First passage times: if x > 0, define

Tx(λ ):= inf{t ≥ 0 : |Xt(λ)| ≥ x}, Tcurv := inf{t ≥ 0 : |Zt| ≥ x(1 + λαt)1/α}.

By Alili-Patie’s transform: Tx(λ )

(d )= λ α1 log (1 + λαTcurv).

Maximal inequalities for X(λ ) ⇒ Estimates on Tx(λ )

⇒ Estimates on Tcurv.

(20)

Application to first crossing problems

Theorem

Estimates on first crossing times:

P (Tcurv > r ) ≤ exp



− K1

λxα log (1 + λαr )



, r > 0, x > 0,

P (Tcurv≤ r ) ≤ K2log (1 + λαr )

λxα , 0 < r < r0(α, x , λ).

(21)

L. Alili and P. Patie.

On the first crossing times of a Brownian motion and a family of continuous curves.

C. R. Math. Acad. Sci. Paris, 340(3):225–228, 2005.

E. Gin´e and M.B. Marcus.

The central limit theorem for stochastic integrals with respect to L´evy processes.

Ann. Probab., 11(1):58–77, 1983.

H. Hult and F. Lindskog.

Extremal behavior of stochastic integrals driven by regularly varying L´evy processes.

To appear in Ann. Probab., 2006.

W.E. Pruitt.

The growth of random walks and L´evy processes.

Ann. Probab., 9(6):948–956, 1981.

(22)

T H E E N D ...

Cytaty

Powiązane dokumenty

[r]

In Reinach’s opinion, legal entities are granted the same sort of independent existence as “numbers, houses and trees”; and principles that regulate the order and interdependencies

Find possibly greatest number ro e(O,l) such that independently of the choice of functions f and F, the following implication is satisfied:. (f,

W pracy xnaietiono dokładne ostacowanie od dołu funkcjonału Re prawie wypukłych i wyznaczono promień a-gwiazdristcśei tej klasy. /(-’) w klasie

In the Jasło poviat, the respondents considered the factors supporting innovation to be the most important group of factors influencing the creation of an appropriate

The objective of the research study was to analyze the chemical composition, in- cluding amino acid composition, of the rapeseed protein-fibre concentrate (RPFC) as well as to

and one can obtain the estimate (H) for an arbitrary plurisubharmonic func- tion ϕ in Ω, where instead of |α| 2 i∂∂ϕ we take a function H satisfying (2) (see [3] for

Na rycinach 1 i 2 widać, że kaplica usytuowana w południowo-wschodnim narożu piętra pałacu spełniała wymóg wyodrębnienia od części mieszkalnej. Boczne drzwi