The Panjer recursion is an algorithm to compute the probability distribution approximation of a compound random variable ${\displaystyle S=\sum _{i=1}^{N}X_{i}\,}$ where both ${\displaystyle N\,}$ and ${\displaystyle X_{i}\,}$ are random variables and of special types. In more general cases the distribution of S is a compound distribution. The recursion for the special cases considered was introduced in a paper ^[1] by Harry Panjer (Distinguished Emeritus Professor, University of Waterloo^[2]). It is heavily used in actuarial science (see also systemic risk).

Preliminaries

We are interested in the compound random variable ${\displaystyle S=\sum _{i=1}^{N}X_{i}\,}$ where ${\displaystyle N\,}$ and ${\displaystyle X_{i}\,}$ fulfill the following preconditions.

Claim size distribution

We assume the ${\displaystyle X_{i}\,}$ to be i.i.d. and independent of ${\displaystyle N\,}$. Furthermore the ${\displaystyle X_{i}\,}$ have to be distributed on a lattice ${\displaystyle h\mathbb {N} _{0}\,}$ with latticewidth ${\displaystyle h>0\,}$.

${\displaystyle f_{k}=P[X_{i}=hk].\,}$

In actuarial practice, ${\displaystyle X_{i}\,}$ is obtained by discretisation of the claim density function (upper, lower...).

Claim number distribution

The number of claims N is a random variable, which is said to have a "claim number distribution", and which can take values 0, 1, 2, .... etc.. For the "Panjer recursion", the probability distribution of N has to be a member of the Panjer class, otherwise known as the (a,b,0) class of distributions. This class consists of all counting random variables which fulfill the following relation:

${\displaystyle P[N=k]=p_{k}=\left(a+{\frac {b}{k))\right)\cdot p_{k-1},~~k\geq 1.\,}$

for some ${\displaystyle a}$ and ${\displaystyle b}$ which fulfill ${\displaystyle a+b\geq 0\,}$. The initial value ${\displaystyle p_{0}\,}$ is determined such that ${\displaystyle \sum _{k=0}^{\infty }p_{k}=1.\,}$

The Panjer recursion makes use of this iterative relationship to specify a recursive way of constructing the probability distribution of S. In the following ${\displaystyle W_{N}(x)\,}$ denotes the probability generating function of N: for this see the table in (a,b,0) class of distributions.

In the case of claim number is known, please note the De Pril algorithm.^[3] This algorithm is suitable to compute the sum distribution of ${\displaystyle n}$ discrete random variables.^[4]

Recursion

The algorithm now gives a recursion to compute the ${\displaystyle g_{k}=P[S=hk]\,}$.

The starting value is ${\displaystyle g_{0}=W_{N}(f_{0})\,}$ with the special cases

${\displaystyle g_{0}=p_{0}\cdot \exp(f_{0}b)\quad {\text{ if ))\quad a=0,\,}$

and

${\displaystyle g_{0}={\frac {p_{0)){(1-f_{0}a)^{1+b/a))}\quad {\text{ for ))\quad a\neq 0,\,}$

and proceed with

${\displaystyle g_{k}={\frac {1}{1-f_{0}a))\sum _{j=1}^{k}\left(a+{\frac {b\cdot j}{k))\right)\cdot f_{j}\cdot g_{k-j}.\,}$

Example

The following example shows the approximated density of ${\displaystyle \scriptstyle S\,=\,\sum _{i=1}^{N}X_{i))$ where ${\displaystyle \scriptstyle N\,\sim \,{\text{NegBin))(3.5,0.3)\,}$ and ${\displaystyle \scriptstyle X\,\sim \,{\text{Frechet))(1.7,1)}$ with lattice width h = 0.04. (See Fréchet distribution.)

As observed, an issue may arise at the initialization of the recursion. Guégan and Hassani (2009) have proposed a solution to deal with that issue .^[5]