P. 1
lundberg claims

lundberg claims

|Views: 30|Likes:
Published by postscript

More info:

Published by: postscript on Apr 23, 2008
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PS, PDF, TXT or read online from Scribd
See more
See less

06/16/2009

pdf

text

original

J. Appl. Prob.

37, 914–917 (2000) Printed in Israel © Applied Probability Trust 2000

AN INSENSITIVITY PROPERTY OF LUNDBERG’S ESTIMATE FOR DELAYED CLAIMS
´ ´ PIERRE BREMAUD,∗ CNRS and Ecole Polytechnique F´ d´ rale de Lausanne e e

Abstract This short note shows that the Lundberg exponential upper bound in the ruin problem of non-life insurance with compound Poisson claims is also valid for the Poisson shot noise delayed-claims model, and that the optimal exponent depends only on the distribution of the total claim per accident, not on the time it takes to honour the claim. This result holds under Cramer’s condition. Keywords: Poisson shot noise; ruin problem; large deviations AMS 2000 Subject Classification: Primary 62P05

1. Poisson shot noise A Poisson shot noise (PSN) is a process of the form X(t) =
n

h(t − Tn , Xn )1(0,t] (Tn ),

where {Tn }n∈Z is the sequence of times of a homogeneous Poisson process of rate λ, {Xn }n∈Z is an i.i.d. sequence of random elements of a measurable space (E, E ), independent of the Poisson process, and h(t, x) = 0 for negative times, and is otherwise non-negative. In the sequel, the time t is always positive. Campbell’s formula gives
t

E[X(t)] = λ
0

E[h(s, X1 )] ds,

and, therefore, the PSN, which is always well defined since the function h is non-negative, is finite if ρ = λ
0 def ∞

E[h(t, X1 )] dt < ∞.

A PSN is also called a (randomly) filtered Poisson process, with the interpretation that h(t − Tn , Xn ) is the random impulse response associated with the spike, or impulse, at Tn . t The total (or integrated) input in the time interval (0, t] is A((0, t]) = 0 X(s) ds, and it takes the form A((0, t]) =
n

H (t − Tn , Xn )1(0,t] (Tn ) +
n

(H (t − Tn , Xn ) − H (−Tn , Xn ))1(−∞,0] (Tn ), (1.1)

Received 4 October 1999; revision received 31 January 2000. ∗ Postal address: Laboratoire des Signaux et Syst` mes, CNRS-ESE, Plateau de Moulon, 91192 Gif-sur-Yvette, France. e Email address: bremaud@lss.supelec.fr

914

Lundberg’s estimate for delayed claims

915

where H (t, x) =
0

t

h(s, x) ds.

The interpretation of this model in terms of non-life insurance is the following. A claim occurs at time Tn and the insurance company honours this claim at the rate h(t − Tn , Xn ). The ∞ total claim is therefore σn = 0 h(t, Xn ) dt. This is a fluid model of claims. To account for claims which are not necessarily fluid (as for the compound Poisson model of the classical theory), we adopt the integrated Poisson shot noise (IPSN) model (1.1) where H (t, x) = µ(x, [0, t]) (1.2)

and µ(x, ·) is for each x a measure on the non-negative half-line. If µ(x, ·) = xδ(·) where δ(·) is the Dirac unit mass at 0, we have the input of the classical compound Poisson model. In this case we denote Xn by σn , and therefore A((0, t]) =
n

σn 1(0,t] (Tn ).

In the general model
t ∞

E[A([0, t))] = λ
0

E[H (s, X1 )] ds + λ
0

E[H (t + s, X1 ) − H (s, X1 )] ds.

In particular if this quantity is finite for some t > 0, then it is finite for all t > 0, and proportional to t. The traffic intensity is then ρ =
def

1 λ t

t

E[H (s, X1 )] ds +
0 0

E[H (t + s, X1 ) − H (s, X1 )] ds .

(1.3)

The total claim for accident 1 is σ1 = H (∞, X1 ) = µ(X1 , R+ ). For background in the classical compound Poisson model of claims, see the first pages of [3] or [4], and for the delayed-claims problem, see [5] where the shot noise model is studied. 2. Lundberg’s estimate The insurance company starts with an initial fortune u. The total amount paid by the insurance company in the interval (0, t] is A0 ((0, t]) =
n

H (t − Tn , Xn )1(0,t] (Tn ).

(2.1)

The ruin probability is, therefore, when the gross premium rate is c,
SN

(u) = P (sup{A0 ((0, t]) − ct ≥ u},
t≥0

where the upper index refers to shot noise. In the classical ruin problem, with compound Poisson claim process, we omit the upper index. We assume the usual condition ρ < c. (2.2)

916

´ P. BREMAUD

We also assume that the claim process is light-tailed, that is E[eθH (∞,X1 ) ] < ∞, for all θ in a neighbourhood of 0. We have the following result: Theorem 2.1. Under Cramer’s condition, that is if there exists R > 0 which is a solution of

(2.3)

λE[σ1 ]
0 t

eRz dFI (z) = c,

(2.4)

where FI (t) = (E[σ1 ])−1 0 (1 − F (z)) dz is the stationary residual claim distribution, then we have Lundberg’s inequality SN (u) ≤ e−Ru . (2.5) Moreover R is the best exponent in Lundberg’s inequality for the delayed-claims process. Proof. If in the classical (non-delayed) model we take for claim size σ1 = H (∞, X1 ), we have the inequality SN (u) ≤ (u), (2.6) which results from the observation that the classical ruin process is dominated by the corresponding delayed-claims ruin process. To prove that R is the best exponent in Lundberg’s inequality for the delayed-claims process we begin by showing that 1 SN lim (u) = −R ′ , u↑∞ u where R ′ is the solution of λE[eR σ1 − 1] = cR ′ . To show this, we use the general results of [2], which give R ′ = sup{θ; λ(θ ) ≤ 0}, where 1 ln E[eθ(A0 ([0,t))−ct) ]. t↑∞ t Straightforward computations (see Section 2.1 of [5]) show that λ(θ ) = lim 1 1 ln E[eθA0 ([0,t)) ] = λ t t
t 0

(2.7)

(2.8)

(2.9)

E[eθ H (s,X1 ) − 1] ds.

(2.10)

Under the light-tail assumption (2.3), the first term on the left-hand side converges to λE[eθH (∞,X1 ) − 1]. It now remains to show that R = R ′ . But this follows from the observation that

λ
0

eθz (1 − F (z)) dz =

λ E[eθσ1 − 1]. θ

In conclusion, we mention that an analogous result of insensitivity holds in a queueing context; see [1].

Lundberg’s estimate for delayed claims

917

References
[1] Br´ maud, P. (2000). An insensitivity property for light-tail shot noise traffic overflow asymptotics. Res. Rept e DSC/2000/05, Dept. of Communication Systems, EPFL, Switzerland. [2] Duffield, N. G. and O’Connell, N. (1995). Large deviations and overflow probabilities for the general singleserver queue, with applications. Math. Proc. Camb. Phil. Soc. 118, 363–374. ¨ [3] Embrechts, P., Kluppelberg, C. and Mikosch, T. (1997). Modelling Extremal Events. Springer, New York. [4] Grandell, J. (1992). Aspects of Risk Theory. Springer, New York. ¨ [5] Kluppelberg, C. and Mikosch, T. (1995). Explosive Poisson shot noise processes with applications to risk reserves. Bernoulli 1, 125–147.

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->