28.05.2013 Views

Etude des marchés d'assurance non-vie à l'aide d'équilibres de ...

Etude des marchés d'assurance non-vie à l'aide d'équilibres de ...

Etude des marchés d'assurance non-vie à l'aide d'équilibres de ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

tel-00703797, version 2 - 7 Jun 2012<br />

4.3. Asymptotics – the A + B/u rule<br />

Using an integration by part theorem requires that the following limit to exist<br />

fΘ(− log(1 − x))<br />

lim<br />

= lim<br />

x→1 1 − x<br />

t→+∞ fΘ(t)e t .<br />

This requirement is strong and will be satisfied only for light-tailed distributions and a certain<br />

range of parameter values. For example, when Θ is exponentially distributed E(λ), the previous<br />

constraint imposes λ > 1.<br />

Another way to <strong>de</strong>al with such integral asymptotic is to apply the Pascal formula, assuming<br />

u is an integer. We get<br />

P (X > u) =<br />

u<br />

k=0<br />

<br />

u<br />

(1 − q)(−1)<br />

k<br />

k<br />

+∞<br />

0<br />

e −kθ dFΘ(θ).<br />

The integral is (once again) the Laplace transform LΘ of the random variable Θ at k. This<br />

binomial alternating sum requires a special treatment because finding an asymptotic for LΘ(k)<br />

will not help us to <strong>de</strong>rive an asymptotic of the sum. This issue is studied in the next subsection.<br />

4.3.5 Binomial alternating sum for claim tails<br />

In the discrete time framework, the survival function can be expressed as<br />

P (X > u) =<br />

u<br />

k=0<br />

<br />

u<br />

(1 − q)(−1)<br />

k<br />

k LΘ(k),<br />

where LΘ <strong>de</strong>notes the Laplace transform of the random variable Θ.<br />

This integral falls within the framework of the alternating binomial sum <strong>de</strong>fined as<br />

Sn(φ) =<br />

n<br />

k=n0<br />

<br />

n<br />

(−1)<br />

k<br />

k φ(k), (4.7)<br />

where 0 ≤ n0 ≤ n, φ is a real function and n ∈ N is large. n0 can be used to exclu<strong>de</strong> first few<br />

points of the sum that would not be <strong>de</strong>fined.<br />

Letting<br />

λ<br />

φ(k) = (1 − q)<br />

α<br />

,<br />

(λ + j) α<br />

in Equation (4.7), we get the distribution function of X when Θ is gamma distributed of<br />

Subsection 4.2.2. Note that, having started with the integral representation +∞<br />

0 P (X =<br />

k|Θ = θ)dFΘ(θ) of P (X = k), we know that the alternating sum is valued on [0, 1]. This is<br />

not immediate without that integral representation.<br />

Let us point out that the probability P (X = k) is a <strong>de</strong>creasing function of k. This is<br />

not easy to see by using the alternating binomial sum representation (4.7). Here is a simle<br />

proof. To indicate the <strong>de</strong>pen<strong>de</strong>nce on the parameter λ, <strong>de</strong>note by P (X = k)λ. From algebraic<br />

manipulation and using the binomial recurrence equation n n−1 n−1<br />

k = k + k−1 , we get<br />

λ<br />

P (X = k + 1)λ = P (X = k)λ − P (X = k)λ+1<br />

α<br />

.<br />

(λ + 1) α<br />

191

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!