01.06.2015 Views

Actuarial Modelling of Claim Counts Risk Classification, Credibility ...

Actuarial Modelling of Claim Counts Risk Classification, Credibility ...

Actuarial Modelling of Claim Counts Risk Classification, Credibility ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

330 <strong>Actuarial</strong> <strong>Modelling</strong> <strong>of</strong> <strong>Claim</strong> <strong>Counts</strong><br />

Some algebra immediately leads to the following system <strong>of</strong> equations to solve:<br />

∑<br />

a t t<br />

t=1<br />

∫ <br />

∑<br />

= 1 + a t t<br />

0<br />

t=1<br />

2 (e − e − ) t−1e f d<br />

∫ <br />

0<br />

(<br />

t−1e<br />

e 2+2 + e − 2 2+<br />

− 2) 2 f d<br />

and<br />

∑<br />

a t t<br />

t=1<br />

∫ <br />

∑<br />

= 1 − a t t<br />

0<br />

t=1<br />

( ) t−1e<br />

e − e − − f d<br />

∫ <br />

0<br />

(<br />

e 2+2 + e − 2 − 2) t−1e − f d<br />

Again, this system does not admit any closed-form solution, but can be solved numerically<br />

(using an appropriate SAS/IML optimization algorithm).<br />

Remark 9.1 Note that here, we have made an averaging with respect to the age structure<br />

<strong>of</strong> the portfolio. In the case where the portfolio is partitioned into a series <strong>of</strong> risk classes, an<br />

average with respect to the composition <strong>of</strong> the portfolio (in terms <strong>of</strong> classification variables)<br />

could also be performed. If some explanatory variables are correlated with A, care must be<br />

taken in the second averaging.<br />

9.2.6 Multivariate Panjer and De Pril Recursive Formulas<br />

Notations<br />

In Sections 9.2.7 and 9.3.3, we will need the bivariate and trivariate extensions <strong>of</strong> the<br />

Panjer algorithm that was described in Section 7.2. The present section is devoted to the<br />

presentation <strong>of</strong> this method as well as a particular case for the sum <strong>of</strong> independent and<br />

identically distributed random vectors, known as multivariate De Pril’s recursive formula.<br />

Assume independent and identically distributed realizations <strong>of</strong> possibly dependent losses<br />

X i = X i1 X ik T affected by a common event, denoted by the counting variable N . For<br />

example N may count the number <strong>of</strong> hurricanes hitting the United States and the X j s may<br />

represent the cost <strong>of</strong> the hurricane in state numbered j, j = 1k. It is natural to try to<br />

obtain the distribution <strong>of</strong> the aggregate claim:<br />

( ) T<br />

∑ N N∑<br />

S = S 1 S k T = X i1 X ik (9.3)<br />

Even when the components <strong>of</strong> X are independent, the components <strong>of</strong> S will have some<br />

positive dependence due to the common counter N .<br />

When N belongs to the Panjer family <strong>of</strong> counting random variables, let us show that a<br />

multivariate version <strong>of</strong> Panjer’s recursive formula emerges. To this end, we will use the<br />

following notations:<br />

i=1<br />

i=1

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

Saved successfully!

Ooh no, something went wrong!