Collective Risk Models

for Short-Term Periods

Tuhinshubhra Bhattacharya

In collective risk model we assume a random process that generates claims for
a portfolio of policies. This process is characterized in terms of portfolio as a
whole rather than individual policies comprising the portfolio. In individual
risk model, we had assumed that the number of times the claim will be made
within the insured period is 0 or 1 and number of policies is fixed as n. But
at the begining of a period of insurance cover, the company does not know
that how many claims will occur and what the amounts of the claims will be.
The collective risk models takes into account these two sources of variability.

Suppose N be the random variable that denotes the number of claims pro-
duced by a portfolio of policies in a given period. Then the possible values
of N are 0, 1, 2, . . .. If Xi denotes the amount of the ith claim in the given
period, then the aggregate of claims S generated by the portfolio for a given
period is modelled as
N
X
S= Xi with S = 0 when N = 0.
i=1

Assumptions: (i) X1 , X2 , . . . are independent and identically distributed ran-

dom variables .
(ii)N is independent of (Xi )∞
i=1 .

Let P (x) denote the common d.f. of the independent and identically dis-
tributed Xi0 s. Let X be a random variable with this d.f. Let

pk = E(X k )

1
, then expectation of S is

E(S) = E(E(S|N )) = E(N · p1 ) = p1 E(N )

and variance is S, and

V (S) = V (E(S|N ))+E(V (S|N )) = V (p1 N )+E(N (p2 −p21 )) = p21 V (N )+(p2 −p21 )E(N )

So, expected aggregate claim is product of expected number of claims and

expected individual claim amount and variance of S has two parts, first due
to variance of number of claims and second due variance of individual claim
amount.

Let MX (t) = E(etX ) is MGF of X and MN (t) = E(etN ) is MGF of N , then

moment generating function of S is

MS (t) = E(etS )
  N

P
t Xi
= E E e i=1 |N 

= E[(MX (t))N |N ]
= E[exp(N ln MX (t)|N )]
= MN [ln MX (t)]

Example 1:Let N is a geometric random variable with p.m.f. as

p(x) = pq n n = 0, 1, 2, . . .
N
P
where 0 < q < 1. Obtain the moment generating function of S = Xi in
i=1
terms of MGF of X, MX (t) when N is assumed to be independent of i.i.d.
random variables X1 , X2 , . . ..

Now, the MGF of N is

∞
X p
MN (t) = etn pq n = .
1 − qet
n=0

2
 N
P
As, S = Xi where N is assumed to be independent of i.i.d. random
i=1
variables X1 , X2 , . . ., the MGF of S can be written as
p p
MS (t) = MN [ln MX (t)] = =
1 − qeln MX (t) 1 − qMX (t)

The distribution of S

Let G(x) = P (S ≤ x) be distribution function of X and F (x) = P (Xi ≤ x)

is the common distribution function of Xi , i = 1, 2, . . .. Let pn = P (N = n) is
the probability mass function of N where n = 0, 1, 2, . . .. Now,
∞
[
{S ≤ x} = {S ≤ x, N = n}
n=1

So, for x ≥ 0,
∞
X
P (S ≤ x) = P (S ≤ x, N = n)
n=0
∞ n
!
X X
= P Xi ≤ x|N = n P (N = n)
n=0 i=1
∞ n
!
X X
= P Xi ≤ x P (N = n)
n=0 i=1
n∗
= F (x)pn

where F n∗ is the distribution function of X1 + X2 + . . . + Xn and F 0∗ (x) = 1

if x ≥ 0 and = 0 if x < 0.

Example 2: As in Example 1, let the numner of claims N is a geometric

random variable with

p(x) = pq n n = 0, 1, 2, . . .

and commnon ditribution unction of X is

F (x) = 1 − e−x , x>0

that is individual claim amount is exponential random variable with mean 1.

Then show that
p
MS (t) = p + q .
p−t

3
Solution: The MGF of X is
Z∞
MX (t) = etx e−x dx = (1 − t)−1 .
0

Then from the result in Example 1,

p
MS (t) = MN [ln MX (t)] =
1 − qMX (t)
p
=
1 − q(1 − t)−1
p(1 − t)
=
p−t
p
= p + (1 − p)
p−t

As 1 is the MGF of the random variable which is degenerated at 0 and p/(p−t)

is the MGF of exponential random variable with cdf H(x) = 1 − e−px . Hence
we S has a mixed distribution with weight p at 0 and weight q for x > 0. The
cdf of S is

FS (x) = p · 1 + q · (1 − e−px ) = 1 − qe−px , x≥0

Example 3 An insurance portfolio produces N claims where N is random

variable defined as



0 with probability 0.5

N= 1 with probability 0.4 .



3 with probability 0.1


The individual claim X(in lakhs) follows the probability distribution give as

1 with probability 0.9
X= .
10 with probability 0.1

It is assumed that the individual claim amount and N are independent ran-
dom variables. Calculate the exact probability that aggregate claims exceed
3 times the expected claims.

4
Solution E(N ) = 0.7 and E(X) = 1.9. Hence E(S) = E(N )E(X) = 1.33.
Hence, required probability is

P (S > 3(1.33)) = P (S > 3.99) = 1 − P (S < 3.99)

= 1 − P (S = 0|N = 0)P (N = 0)
− P (X1 = 1|N = 1)P (N = 1)
− P (X1 = 1, X2 = 1, X3 = 1|N = 3)P (N = 3)
= 1 − 1(0.5) − 0.9(0.4) − (0.9)3 · (0.1)
= 0.0671

Distribution of N

1. Let the distribution of N is Poisson with parameter λ. Then E(N ) =

V ar(N ) = λ. With this choice of distribution of N , the distribution of
S is called compound Poisson distribution. As before, we assume
pk = E(Xik ), then expectation is

E(S) = λp1

and variance is

V (S) = V (E(S|N )) + E(V (S|N )) = p21 λ + (p2 − p21 )λ = λp2 .

The moment generating function of N is

MN (t) = exp{λ(et − 1)}.

Hence, moment generating function of S is

MS (t) = MN [ln MX (t)] = exp{λ(MX (t) − 1)}

2. When the variance number of claims exceeds its mean, then Poisson
distribution is not appropriate. Suppose the distribution of N is negative
binomial with p.m.f.
 
r+n−1 r n
P (N = n) = p q n = 0, 1, 2, . . . .
n

5
 This distribution has two parameters r and 0 < p < 1. For this distri-
bution, we have  r
p
MN (t) =
1 − qet
rq rq
and E(X) = p and V (X) = p2
. When negative binomial is chosen as
the distribution of N , then distribution of S is called compound negative
N
Xi and E(Xik ) = pk we have
P
binomial distribution. As, S =
i=1

rq
E(S) = p1
p
and variance is
rq rq rq rq 2
V (S) = V (E(S|N )) + E(V (S|N )) = p21 + (p2 − p21 ) 2 = p2 + p21 2 .
p p p p
Also, the MGF of S is
 r
p
MS (t) = MN [ln MX (t)] =
1 − qMX (t)

Exercises:

1. Suppose that S has a compound Poisson distribution with λ = 3 and

the common distribution of X1 , X2 , . . . is charaterrized by the p.m.f.
p(x) = 0.1x where x = 1, 2, 3, 4.Calculate probabilities that aggregate
claims equal to 0, 1, 2, 3. Also, find the probability that aggregate claims
exceed 3 units.

2. In a collective risk model, suppose the frequency N of claims is a ran-

dom variable with possible values 0, 1, 2 with probabilities 0.4, 0.2, 0.4
respectively. Amunt X of claims is a random variable with possible val-
ues 1000, 3000, 7000 with probabilities 0.3, 0.4, 0.3 respectively. Find the
distribution of total risk to the company for a short period.

3. For a collective risk model, the number of claims has a negative bino-
mial distribution with parameters r = 2 and p = 1/3. Claim amounts
are mutually independent with two possible values 1 and 2 with equal

6
 chance. Find moment generating function of total claims.

4. Aggregate claim S in a collective risk model has a compound negative

binomial distribution, with parameters k = 100 and p = 0.3 for the
negative binomial distribution and individual claim distribution is ex-
ponential with mean 100 units. Find mean and variance of S. Using
normal approximation, find P [S > 1.3E(S)].

Send the solutions of the exercises at tuhinsubhra.bh@gmail.com. Do

the exercises on your workbook clearly and scan those to send to
my mail