c20 Notes

om
i.c
CM2-20: Ruin theory Page 1
ng
si
om
o m
as
.m
w
w
w
Ruin theory
Syllabus objectives
5.1 Ruin theory
5.1.1 Explain what is meant by the aggregate claim process and the cash-flow
process for a risk.
5.1.2 Use the Poisson process and the distribution of inter-event times to
calculate probabilities of the number of events in a given time interval and
waiting times.
5.1.3 Define a compound Poisson process and calculate probabilities using
simulation.
5.1.4 Define the probability of ruin in infinite/finite and continuous/discrete time
and state and explain relationships between the different probabilities of
ruin.
5.1.5 Describe the effect on the probability of ruin, in both finite and infinite
time, of changing parameter values by reasoning or simulation.
5.1.6 Calculate probabilities of ruin by simulation.
The Actuarial Education Company  IFE: 2019 Examinations

om
i.c
Page 2 CM2-20: Ruin theory
ng
si
om
o m
0 Introduction
as
.m
w
In this chapter we consider the aggregate claims S(t ) arising up to time t. If N(t ) is the number of
w
w
claims arising by time t, and Xi is the amount of the i-th claim, then S(t)  X1  X2   XN(t ) .
N(t ) is called a Poisson process and S(t ) is called a compound Poisson process. We can use S(t )
to model claims received by an insurance company and hence consider the probability that this
insurance company is ruined.
We start with the notation and the other basic concepts before giving formal definitions of both
the Poisson process and the compound Poisson process. We will also introduce the concept of a
premium security loading. Briefly, this is an additional amount charged on an insurance premium
to reduce the likelihood of an insurance company becoming ruined.
Later we will introduce the adjustment coefficient, a parameter associated with risk, and
Lundberg’s inequality.
We will consider the effect of changing parameter values on the probability of ruin for an
insurance company before finally considering the impact of introducing reinsurance.
 IFE: 2019 Examinations The Actuarial Education Company

om
i.c
ng
si
om
o m
1 Basic concepts
as
.m
w
1.1 Notation
w
w
One technical point needed later in this chapter is that a function f ( x ) is described as being
o( x ) as x goes to zero, if:
f (x )
lim 0
x 0 x
You can use this notation to simplify your working. For example, the function
0.5x 2  0.004 x 3
g(x)  3x  0.5x 2  0.004 x 3 can be rewritten as g(x)  3x  o(x) , since  0 as
x
x  0 . Note that o(x) does not represent an actual number so that c  o(x) ( c is a constant),
o(x) and o(x) are all equivalent.
Question
Which of the following functions are o(x) as x  0 ?
(i) x2 (ii) ex (iii) e x  1  x
Solution
(i) Yes
(ii) No
(iii) Yes, because if we expand e  x as a power series and simplify, we get:
x2 x 3
e x  1  x   
2! 3!
e x  1  x 1
This gives  x  terms in x 2 and higher powers.
x 2
In the actuarial literature, the word ‘risk’ is often used instead of the phrase ‘portfolio of
policies’. In this chapter both terms will be used, so that by a ‘risk’ will be meant either a
single policy or a collection of policies. This chapter will focus on claims generated by a
portfolio over successive time periods. Some notation is needed.

om
i.c
ng
si
om
m
 N (t ) the number of claims generated by the portfolio in the time interval [0, t], for
o
as
all t  0
.m
w
 X i the amount of the i-th claim, i = 1, 2, 3, ...
w
w
 S (t ) the aggregate claims in the time interval [0, t], for all t  0.
 X i i 1 is a sequence of random variables. N(t )t 0 and S(t )t 0 are both families of
random variables, one for each time t  0 ; in other words N (t )t 0 and S(t )t 0 are
stochastic processes.
You can think of a stochastic process as being a whole family of different random variables.
Consider a time line. On the line there are an infinite number of different time intervals. For each
interval of time, there is a random variable that corresponds to the aggregate claim amount
arising in that time interval. This is what we mean here by a stochastic process.
It can be seen that:
N (t )
S(t )   Xi
i 1
with the understanding that S (t ) is zero if N (t ) is zero.
The stochastic process S(t )t 0 as defined above is known as the aggregate claims
process for the risk. The random variables N (1) and S (1) represent the number of claims
and the aggregate claims respectively from the portfolio in the first unit of time.
The insurer of this portfolio will receive premiums from the policyholders. It is convenient
at this stage to assume, as will be assumed throughout this chapter, that the premium
income is received continuously and at a constant rate. Here is some more notation:
c = the rate of premium income per unit time
so that the total premium income received in the time interval  0, t  is ct . It will also be
assumed that c is strictly positive.
1.2 The surplus process

Suppose that at time 0 the insurer has an amount of money set aside for this portfolio. This
amount of money is called the initial surplus and is denoted by U . It will always be
assumed that U  0 . The insurer needs this initial surplus because the future premium
income on its own may not be sufficient to cover the future claims. Here we are ignoring
expenses. The insurer’s surplus at any future time t   0  is a random variable since its
value depends on the claims experience up to time t. The insurer’s surplus at time t is
denoted by U (t ) .

om
i.c
ng
si
om
m
The following formula for U (t ) can be written:
o
as
.m
U (t )  U  ct  S (t )
w
w
w
In words this formula says that the insurer’s surplus at time t is the initial surplus plus the
premium income up to time t minus the aggregate claims up to time t. Notice that the initial
surplus and the premium income are not random variables since they are determined before
the risk process starts. The above formula is valid for t  0 with the understanding that U(0)
is equal to U. For a given value of t, U(t) is a random variable because S(t) is a random
variable. Hence U (t )t 0 is a stochastic process, which is known as the cash flow process
or surplus process.
Figure 1
Figure 1 shows one possible outcome of the surplus process. Claims occur at times
T1,T2 ,T3 ,T4 and T5 and at these times the surplus immediately falls by the amount of the
claim. Between claims the surplus increases at constant rate c per unit time. The model
being used for the insurer’s surplus incorporates many simplifications, as will any model of
a complex real-life operation. Some important simplifications are that it is assumed that
claims are settled as soon as they occur and that no interest is earned on the insurer’s
surplus. Despite its simplicity this model can give an interesting insight into the
mathematics of an insurance operation.
We are also assuming that there are no expenses associated with the process (or, equivalently,
that S(t ) makes allowance for expense amounts as well as claim amounts), and that the insurer
cannot vary the premium rate c .
We are also ignoring the possibility of reinsurance. Simple forms of reinsurance will be
incorporated into the model later in this chapter.

om
i.c
ng
si
om
o m
1.3 The probability of ruin in continuous time
as
.m
It can be seen from Figure 1 that the insurer’s surplus falls below zero as a result of the
w
w
claim at time T3 . Speaking loosely for the moment, when the surplus falls below zero the
w
insurer has run out of money and it is said that ruin has occurred. In this simplified model,
the insurer will want to keep the probability of this event, that is, the probability of ruin, as
small as possible, or at least below a predetermined bound. Still speaking loosely, ruin can
be thought of as meaning insolvency, although determining whether or not an insurance
company is insolvent is, in practice, a very complex problem. Another way of looking at the
probability of ruin is to think of it as the probability that, at some future time, the insurance
company will need to provide more capital to finance this particular portfolio.
Now to be more precise. The following two probabilities are defined:
 (U )  P [U (t )  0, for some t , 0  t   ]
 (U , t )  P [U ( )  0, for some  , 0    t ]
 (U ) is the probability of ultimate ruin (given initial surplus U) and  (U , t ) is the probability
of ruin within time t (given initial surplus U). These probabilities are sometimes referred to
as the probability of ruin in infinite time and the probability of ruin in finite time. Here are
some important logical relationships between these probabilities for 0 < t1  t2 <  and
for 0  U1  U2:
 U2 , t    U1, t  (1.1)
 U2    U1  (1.2)
 U , t1    U , t2    U  (1.3)
lim  U , t    U  (1.4)
t 
The intuitive explanations for these relationships are as follows:
The larger the initial surplus, the less likely it is that ruin will occur either in a finite time
period, hence (1.1), or an unlimited time period, hence (1.2).
For a given initial surplus U, the longer the period considered when checking for ruin, the
more likely it is that ruin will occur, hence (1.3).
Finally, the probability of ultimate ruin can be approximated by the probability of ruin within
finite time t provided t is sufficiently large, hence (1.4).
Question
What is lim   u , t  ?
u

om
i.c
ng
si
om
o m
Solution
as
.m
w
As the amount of initial surplus increases, ruin will become less and less likely. So the limit is zero.
w
w
You may be wondering whether it is possible to find numerical values for these ruin probabilities.
In some very simple cases it is. However, for most practical situations, finding an exact value for
the probability of ruin is impossible. In some cases there are useful approximations to  (u) , even
if calculation of an exact value is not possible.
1.4 The probability of ruin in discrete time

The two probabilities of ruin considered so far have been continuous time probabilities of
ruin, so-called because they check for ruin in continuous time. In practice it may be
possible (or even desirable) to check for ruin only at discrete intervals of time.
For a given interval of time, denoted h, the following two discrete time probabilities of ruin
are defined:
 h (U )  P U  t  < 0, for some t , t = h, 2h, 3h,
 h (U , t )  P U   < 0, for some  ,  = h, 2h, ,t  h, t 
Note that it is assumed for convenience in the definition of  h (U , t ) that t is an integer

multiple of h. Figure 2 shows the same realisation of the surplus process as given in
Figure 1 but assuming now that the process is checked only at discrete time intervals. The
black markers show the values of the surplus process at integer time intervals (ie h = 1); the
black markers together with the white ones show the values of the surplus process at time
intervals of length ½.
Figure 2
It can be seen from Figure 2 that in discrete time with h = 1, ruin does not occur for this
realisation of the surplus process before time 5, but ruin does occur (at time 2½) in discrete
time with h = ½.

om
i.c
ng
si
om
m
Listed below are five relationships between different discrete time probabilities of ruin for
o
as
0  U1  U2 and for 0  t1  t2   . Formulae (1.5), (1.6), (1.7) and (1.8) are the discrete time
.m
versions of formulae (1.1), (1.2), (1.3) and (1.4) above and their intuitive explanations are
w
w
similar. The intuitive explanation of (1.9) comes from Figure 2.
w
 h (U2 , t )   h (U1, t ) (1.5)
 h (U2 )   h (U1) (1.6)
 h (U , t1)   h (U , t2 )   h (U ) (1.7)
lim  h (U , t )   h (U ) (1.8)
t 
  h (U , t )   (U , t ) (1.9)
Question
Explain why Equation 1.9 is true.
Solution
 (U , t ) involves checking for ruin at all possible times. Since the more often we check for ruin,
the more likely we are to find it, we would expect that  (U , t ) would be greater than  h (U, t) .
Intuitively, it is expected that the following two relationships are true since the probability of
ruin in continuous time could be approximated by the probability of ruin in discrete time,
with the same initial surplus, U, and time horizon, t, provided ruin is checked for sufficiently
often, ie provided h is sufficiently small.
lim  h (U , t )   (U , t ) (1.10)
h 0
lim  h (U )   (U ) (1.11)
h 0
Formulae (1.10) and (1.11) are true but the proofs are rather messy and will not be given
here.

om
i.c
ng
si
om
o m
2 The Poisson and compound Poisson processes
as
.m
w
2.1 Introduction
w
w
In this section some assumptions will be made about the claim number process, N (t )t 0 ,

and the claim amounts,  X i i 1 . The claim number process will be assumed to be a
Poisson process, leading to a compound Poisson process S(t )t 0 for aggregate claims.
The assumptions made in this section will hold for the remainder of this chapter.
2.2 The Poisson process

We use the term “Poisson process” to describe the number of claims arising from a time period of
length t.
If the number of claims N arising from a single time period has a Poisson distribution with
parameter  then the number of claims N(t ) which arise over a time period of length t is a
Poisson process, ie N(t ) has a Poisson distribution with parameter t .
The Poisson process is an example of a counting process. Here the number of claims
arising from a risk is of interest. Since the number of claims is being counted over time, the
claim number process N (t )t 0 must satisfy the following conditions:
(i) N (0)  0 , ie there are no claims at time 0
(ii) for any t  0 , N (t ) must be integer valued
(iii) when s  t , N (s )  N (t ) , ie the number of claims over time is non-decreasing
(iv) when s  t , N (t )  N (s ) represents the number of claims occurring in the time

interval  s , t  .
The claim number process N (t )t 0 is defined to be a Poisson process with parameter  if
the following conditions are satisfied:
(i) N (0)  0 , and N (s )  N (t ) when s  t
(ii) P  N (t  h )  r | N (t )  r   1   h  o(h)
P  N (t  h )  r  1| N (t )  r    h  o(h ) (2.1)
P  N (t  h )  r  1| N (t )  r   o(h )
(iii) when s  t , the number of claims in the time interval (s , t ] is independent of the
number of claims up to time s . (2.2)
Condition (ii) implies that there can be a maximum of one claim in a very short time interval h. It
also implies that the number of claims in a time interval of length h does not depend on when
that time interval starts.

om
i.c
ng
si
om
o m
Question
as
.m
w
Explain how motor insurance claims could be represented by a Poisson process.
w
w
Solution
The events in this case are occurrences of claim events (ie accidents, fires, thefts etc) or claims
reported to the insurer. The parameter  represents the average rate of occurrence of claims
(eg 50 per day), which we are assuming remains constant throughout the year and at different
times of day. The assumption that, in a sufficiently short time interval, there can be at most one
claim is satisfied if we assume that claim events cannot lead to multiple claims (ie no motorway
pile-ups etc).
When studying a Poisson process the distribution of the time to the first claim and the times
between claims is often of particular interest.
Time to the first claim

This section will show that the time to the first claim has an exponential distribution with
parameter  .
Let the random variable T1 denote the time of the first claim. Then, for a fixed value of t , if
no claims have occurred by time t , T1  t . Hence:
P T1  t   P  N (t )  0   exp{  t }
This last step follows from the formula for the probability function of a Poisson(t ) distribution
with x  0 .
And:
P (T1  t )  1  exp{   t }
so that T1 has an exponential distribution with parameter 
This is because the RHS matches the formula for the distribution function of an exponential
distribution.
The time to the first claim in a Poisson process has an exponential distribution with parameter  .
Time between claims

This section will show that the time between claims has an exponential distribution with
parameter  .

om
i.c
ng
si
om
m
For i  2,3, , let the random variable Ti denote the time between the  i  1 th and the i th
o
as
claims. Then:
.m
w
w
 n   n 1 n 
 Ti  r   P   Ti  t  r  Ti
w
P  Tn 1  t r
 
 i 1   i 1 i 1 

 P N t  r   n N  r   n 
 P  N t  r   N  r   0 N  r   n 
Then we use the independence of claim numbers in different time periods to remove the
conditioning.
By condition (2.2):
P (N (t  r )  N (r )  0 N (r )  n)  P (N (t  r )  N (r )  0)
Finally:
P (N (t  r )  N (r )  0)  P (N (t )  0)  exp{   t }
since the number of claims in a time interval of length r does not depend on when that time
interval starts (condition (2.1)). Thus inter-event times also have an exponential distribution
with parameter  .
The time between claims in a Poisson process has an exponential distribution with parameter  .
Note that the inter-event time is independent of the absolute time. In other words the time until
the next event has the same distribution, irrespective of the time since the last event or the
number of events that have already occurred. This is referred to as the memoryless property of
the exponential distribution.
Question
If reported claims follow a Poisson process with rate 5 per day (and the insurer has a 24 hour
hotline), calculate:
(i) the probability that there will be fewer than 2 claims reported on a given day
(ii) the probability that another claim will be reported during the next hour.
Solution
(i) The expected number of claims reported on a given day is 5. So the number of claims
reported on a given day has a Poisson(5) distribution and the probability that there will
be fewer than 2 claims is:
P(N  2)  P(N  0)  P(N  1)  e5  5e5  0.040
Here we have used the formula for the Poisson probability, but alternatively, using page
176 of the Tables, P(N  2)  P(N  1)  0.04043 .

om
i.c
ng
si
om
m
(ii) The waiting time until the next event has an Exp(5) distribution. We need to find a
o
as
probability using the exponential distribution. To do this, we can use the cumulative
.m
w
distribution function:
w
w
P(T  t )  1  et
1 of
So the probability that there will be a claim (or several claims) during the next hour ( 24
a day) is:
 24
5
P(T  24
1 )1e  0.1881
Note that it is unlikely that the rate would be constant over time in reality.
2.3 The compound Poisson process

In this section the Poisson process for the number of claims will be combined with a claim
amount distribution to give a compound Poisson process for the aggregate claims.
The following three important assumptions are made:

 the random variables  X i i 1 are independent and identically distributed

 the random variables  X i i 1 are independent of N (t ) for all t  0
 the stochastic process N (t )t 0 is a Poisson process whose parameter is

denoted  .
This last assumption means that for any t  0 , the random variable N (t ) has a Poisson
distribution with parameter  t , so that:
( t )k
P [N (t )  k ]  exp{   t } for k  0,1,2,
k!
With these assumptions the aggregate claims process, S(t )t 0 , is called a compound
Poisson process with Poisson parameter  . By comparing the assumptions above with the
assumptions in Section 0, it can be seen that the connection between the two is that if
S(t )t 0 is a compound Poisson process with Poisson parameter  , then, for a fixed value
of t ( 0) , S (t ) has a compound Poisson distribution with Poisson parameter  t .
Note the slight change in terminology here: ‘Poisson parameter  ’ becomes ‘Poisson
parameter  t ’ when a change is made from the process to the distribution.
The common distribution function of the X i s will be denoted F ( x ) and it will be assumed
for the remainder of this chapter that F (0)  0 so that all claims are for positive amounts.

om
i.c
ng
si
om
m
F (x) is defined to be P( X  x) . In the continuous case we would find F (x) by integrating the
o
as
probability density function (PDF):
.m
w
w
x
F x 
w
 f (t ) dt

The probability density function of the X i s, if it exists, will be denoted f ( x ) and the kth
moment about zero of the X i s, if it exists, will be denoted mk , so that:
mk  E  X ik  for k  1, 2, 3,
 
Whenever the common moment generating function of the X i s exists, its value at the point
r will be denoted by M X (r ) .
The definition of a moment generating function is:
M X (r )  E[e rX ]
Note that we are using r for the dummy variable to avoid confusion with time t .
Since, for a fixed value of t , S (t ) has a compound Poisson distribution, it follows from an
earlier subject that the process {S (t )}t  0 has mean  tm1 , variance  tm2 , and moment
generating function MS (r ) , where:
MS (r )  exp { t (M X (r )  1)}
Subject CS2 shows that if:
S  X1  X2    XN
where N  Poisson    , then:
E  S   E  X   m1 Var (S)  E  X 2   m2

 
MS  r   MN ln M X  r  
These formulae can be found on page 16 of the Tables.
Here N  t   Poisson  t  , therefore E  S  t    tm1 and Var  S(t)  tm2 . Also
 
MN t   r   exp t er  1 , so: 
 
MS(t ) (r )  exp t (eln Mx (r )  1)  exp  t  MX (r )  1  

om
i.c
ng
si
om
m
For the remainder of this chapter the following (intuitively reasonable) assumption will be
o
as
made concerning the rate of premium income:
.m
w
c >  m1 (2.3)
w
w
so that the insurer’s premium income (per unit time) is greater than the expected claims
outgo (per unit time).
Question
Why is this intuitively reasonable?
Solution
Otherwise the insurer would be charging premiums that were less than the amount it expected to
pay out in claims.
In the real world this assumption may not always be true, especially during periods of competitive
pressure when premium rates are soft.
2.4 Probability of ruin in the short term

If we know the distribution of the aggregate claims S(t ) , we can often determine the probability of
ruin for the discrete model over a finite time horizon directly (without reference to the models), by
looking at the cashflows involved.
Question
The claims arising during each year from a particular type of annual insurance policy are assumed
to follow a normal distribution with mean 0.7P and standard deviation 2.0P , where P is the
annual premium. Claims are assumed to arise independently. Insurers assess their solvency
position at the end of each year.
A small insurer with an initial surplus of £0.1m expects to sell 100 policies at the beginning of the
coming year in respect of identical risks for an annual premium of £5,000. The insurer incurs
expenses of 0.2P at the time of writing each policy. Calculate the probability that the insurer will
prove to be insolvent at the end of the coming year. Ignore interest.
Solution
Using the information given, the insurer’s surplus at the end of the coming year will be:
U(1)  initial surplus + premiums  expenses  claims
 0.1m  100  5,000  100  0.2  5,000  S(1)
 0.5m  S(1)

om
i.c
ng
si
om
m
The distribution of S(1) is:
o
as
.m
S(1)  N 100  0.7  5,000, 100   2.0  5,000    N  0.35m,  0.1m 
2 2
w
   
w
w
So the probability that the surplus will be negative is:
P U(1)  0   P  S(1)  0.5m
 P N 0.35m,  0.1m    0.5m

2
   
 0.5m  0.35m 
1   1  0.93319  0.067
 0.1m 
Question
If the insurer expects to sell 200 policies during the second year for the same premium and
expects to incur expenses at the same rate, calculate the probability that the insurer will prove to
be insolvent at the end of the second year.
Solution
The insurer’s surplus at the end of the second year will be:
U(2)  initial surplus + premiums  expenses  claims

 0.1m + (100  200)  5,000  (100  200)  0.2  5,000  S(2)
 1.3m  S(2)
The distribution of S(2) is:
S(2)  N(300  0.7  5,000, 300  (2.0  5,000)2 )  N[1.05m,(0.173m)2 ]
The probability that the surplus will be negative at the end of the second year is:
P [U(2)  0]  P [ S(2)  1.3m]
 P (N[1.05m,(0.173m)2 ]  1.3m)
 1.3m  1.05m 
1 
 0.173m 
 1  (1.443)  0.074
The normal distribution is probably not a very realistic distribution to use for the claim amount
distribution in most portfolios, as it is symmetrical, whereas many claim amount portfolios will
have skewed underlying distributions. However, it is commonly used in exam questions.

om
i.c
ng
si
om
o m
Question
as
.m
w
The number of claims from a portfolio of policies has a Poisson distribution with parameter 30 per
w
w
year. The individual claim amount distribution is lognormal with parameters   3 and  2  1.1 .
The rate of premium income from the portfolio is 1,200 per year.
If the insurer has an initial surplus of 1,000, estimate the probability that the insurer’s surplus at
time 2 will be negative, by assuming that the aggregate claims distribution is approximately
normal.
Solution
First we need the mean and variance of the aggregate claims in a two-year period. The expected
number of claims will be 60. So the mean and variance are (using the formulae for the first two
moments of a lognormal distribution):
E  S(2)  60e3 0.55  2,088.80
and Var  S(2)   60e6 2.2  218,457
Ruin will occur if S(2) is greater than the initial surplus plus premiums received. So we want:
 3,400  2,088.80 
P  S(2)  1,000  2  1,200  P N(0,1)    1  (2.8053)  0.0025
 218,457 
The probability of ruin is approximately 0.25%.
2.5 Premium security loadings

So far, we have used c to denote the rate of premium income per unit time, independent of the
claims outgo. In some circumstances it is more useful to think of the rate of premium income as
being related to the rate of claims outgo.
For the insurer to survive, the rate at which premium income comes in needs to be greater than
the rate at which claims are paid out. If this is not true, the insurer is certain to be ruined at some
point.
Sometimes c will be written as:
c   1     m1
where    0  is the premium loading factor.

om
i.c
ng
si
om
m
The security loading is the percentage by which the rate of premium income exceeds the rate of
o
as
claims outgo. So, for the Poisson process outlined above, we have:
.m
w
c  1    E  S   1    m1
w
w
where  is the security loading.  is also sometimes called the ‘relative security loading’. It
might typically be a figure such as 0.2, ie 20%.
The insurer will need to adopt a positive security loading when pricing policies, in order to cover
expenses, profit, contingency margins and so on.
Note that this does not mean that ruin is impossible. It is quite possible for the actual claims
outgo to exceed substantially its expected value. So even in this situation the insurer’s probability
of ruin is non-zero.
Mean, variance and MGF of the total claim amount

For a compound Poisson process S(t ) , the mean and variance of the total claim amount are given
by:
E  S  t     t E  X  Var  S(t )   t E  X 2 
 
The moment generating function of the process is given by:
MS t   r   exp  t (M X  r   1 
2.6 A technicality
In the next section a technical result will be needed concerning M X  r  (the moment
generating function of the individual claim amount distribution), which, for convenience, will
be presented here. It will be assumed throughout the remainder of this chapter that there is
some number  (0     ) such that M X  r  is finite for all r   and:
lim M X  r    (2.4)
r  
(For example, if the X i s have a range bounded above by some finite number, then  will
be  ; if the X i s have an exponential distribution with parameter  , then  will be equal
to  .)
Suppose for example that claim amounts have a continuous uniform distribution on the
interval (0,10) , so that they are bounded above by 10. Then the moment generating function of
the claim distribution is (from the Tables):
e10r  1
MX  r  
10r

om
i.c
ng
si
om
m
This is defined for all positive values of r , and so in this case    . We can see that as r   ,
o
as
the limit of the MGF is infinite. If the claim distribution is Exp( ) , the MGF (as stated in the
.m
w
Tables) is:
w
w

MX (r )  (1  r /  )1 
 r
This tends to infinity as r tends to  from below.
In the next section the following result will be needed:
lim   M X  r   cr    (2.5)
r  
If  is finite, (2.5) follows immediately from (2.4).
This is because  , c and r would all have finite values in the limit.
Now it will be shown that (2.5) holds when  is infinite. This requires a little more care.
First note that there is a positive number,  say, such that:
P  X i     0
The reason for this is that all claim amounts are positive.
So, if we pick a small enough number (   0.01 maybe), we’re bound to get a proportion of claims
whose amount exceeds this.
This probability will be denoted by  . Then:
M X (r )  e r  
This follows by considering the claims below and above  :
M X (r )  E  erX   E  e rX | X    P( X   )  E  e rX | X    P( X   )
     
 0  e r 
Hence:
lim ( M X (r )  cr )  lim ( e r    cr )  
r  r 
Here the e r term is tending to  , while the cr term is tending to  . Remember that in
such cases the exponential term always ‘wins’. So the limit is  .

om
i.c
ng
si
om
o m
3 The adjustment coefficient and Lundberg’s inequality
as
.m
w
This section will look at the probability of ruin and introduce the adjustment coefficient, a
w
w
parameter associated with risk. The letters R and r will be used interchangeably for the
adjustment coefficient.
3.1 Lundberg’s inequality

Lundberg’s inequality states that:
 (U )  exp {  RU }
where U is the insurer’s initial surplus and  (U ) is the probability of ultimate ruin. R is a
parameter associated with a surplus process known as the adjustment coefficient. Its value
depends upon the distribution of aggregate claims and on the rate of premium income.
Before defining R the importance of the result and some features of the adjustment
coefficient will be illustrated.
Don’t worry at this stage about what R actually represents. It will be defined shortly. Until then
just think of it as a parameter associated with the surplus process.
Note that if we can find a value for R , then Lundberg’s inequality tells us that we can find an
upper bound for the probability of ruin. This is a very useful result.
Figure 3 shows a graph of both exp{  RU } and  (U ) against U when claim amounts are
exponentially distributed with mean 1, and when the premium loading factor is 10%. (The
solution for R will be found in Section 3.2. The formula for  (U ) is given in Section 4.2.)
10
In fact it can be shown that the value of R in this case is .
11
Figure 3 – Exponentially distributed claims with mean 1

and premium loading factor 10%

om
i.c
ng
si
om
m
It can be seen that, for large values of U ,  (U ) is very close to the upper bound, so that
o
as
 (U )  exp {  RU } .
.m
w
w
In the actuarial literature, exp{  RU } is often used as an approximation to  (U ) .
w
R can be interpreted as measuring risk. The larger the value of R , the smaller the upper
bound for  (U ) will be. Hence,  (U ) would be expected to decrease as R increases. R is
a function of the parameters that affect the probability of ruin, and R ’s behaviour as a
function of these parameters can be observed.
Note that R is an inverse measure of risk. Larger values of R imply smaller ruin probabilities, and
vice versa.
Figure 4 shows a graph of R as a function of the loading factor,  , when:
(i) the claim amount distribution is exponential with mean 10, and
(ii) all claims are of amount 10.
Figure 4 – R by loading
Note that in both cases, R is an increasing function of  . This is not surprising as  (U )

would be expected to be a decreasing function of  , and since  (U )  exp  RU  , any
factor causing a decrease in  (U ) would cause R to increase.
Question
Why would  (U) be expected to be a decreasing function of  ?
Solution
 is the security loading, and  (U) is the probability of ruin for a fixed level of surplus U . If 
increases the premiums we charge will increase, and we should become more secure, ie the
probability of ruin should fall.

om
i.c
ng
si
om
m
Note also that the value of R when claim amounts are exponentially distributed is less than
o
as
the value of R when all claim amounts are 10. Again, this result is not surprising. Both
.m
claim amount distributions have the same mean, but the exponential distribution has
w
greater variability. Greater variability is associated with greater risk, and hence a larger
w
w
value of  (U ) would be expected for the exponential distribution, and a lower value of R .
This example illustrates that R is affected by the premium loading factor and by the
characteristics of the individual claim amount distribution. R is now defined and shown, in
general, to encapsulate all the factors affecting a surplus process.
3.2 The adjustment coefficient – compound Poisson processes

The surplus process depends on the initial surplus, on the aggregate claims process and on
the rate of premium income. The adjustment coefficient is a parameter associated with a
surplus process which takes account of two of these factors: aggregate claims and
premium income. The adjustment coefficient gives a measure of risk for a surplus process.
When aggregate claims are a compound Poisson process, the adjustment coefficient is
defined in terms of the Poisson parameter, the moment generating function of individual
claim amounts and the premium income per unit time.
The adjustment coefficient, denoted R , is defined to be the unique positive root of:
 M X (r )    cr  0 (3.1)
So, R is given by:
  M X (R )    cR (3.2)
Note that, although R relates to the aggregate claims, the MGF used in the definition is for the
individual claim amount.
It is probably not at all obvious to you at this stage why R is defined in this way. The reason is
bound up with the proof of Lundberg’s inequality, which you are not required to know. Please
accept the definition, so that you can find the value of R in simple cases.
Note that equation (3.1) implies that the value of the adjustment coefficient depends on the
Poisson parameter, the individual claim amount distribution and the rate of premium
income. However, writing c  (1   ) m1 gives:
M X (r )  1  (1   )m1r
so that R is independent of the Poisson parameter and simply depends on the loading
factor,  , and the individual claim amount distribution.
You can see from this equation that all the  ’s have cancelled.
Question
An insurer knows from past experience that the number of claims received per month has a
Poisson distribution with mean 15, and that claim amounts have an exponential distribution with
mean 500. The insurer uses a security loading of 30%. Calculate the insurer’s adjustment
coefficient and give an upper bound for the insurer’s probability of ruin, if the insurer sets aside
an initial surplus of 1,000.

om
i.c
ng
si
om
o m
Solution
as
.m
w
The equation for the adjustment coefficient is:
w
w
MX (r )  1  (1   )m1r
 1  1
We have X  Exp   , so that M X (r )  (1  500r ) (this comes from the Tables),   0.3 , and
 500 
m1  E[ X ]  500 . Substituting these into the equation:
(1  500r )1  1  1.3  500r  1  650r
Rearranging:
1  (1  500r )(1  650r )
1  1  500r  650r  325,000r 2

0  150  325,000r
 r  0.000462
From Lundberg’s inequality,  (U)  e rU , so here:
 (U)  e 0.0004621,000  0.630
Note that we didn’t use the Poisson parameter in our solution.
We will see another example of finding the adjustment coefficient shortly.
Question
In the previous question we ignored the fact that r could be 0. Why?
Solution
The first reason is that Core Reading defines R to be the unique positive root, so R cannot be
zero.
The second reason is that R  0 should always be a solution to the equation. Why? Consider the
LHS. MX (0)  E[e0 ]  1 . Consider the RHS. 1  (1   )m1  0  1 . So R  0 is always a solution, but
why do we ignore it?
Consider Lundberg’s inequality,  (U)  e RU . If R  0 , we have an upper bound of 1 for the
probability of ruin. That should have been obvious!
In practice, ignore R  0 , just as we have done here.

om
i.c
ng
si
om
m
It may not be obvious to you why R does not depend on the Poisson parameter. The basic
o
as
reason is that increasing the Poisson parameter speeds up the whole process, so that claims arise
.m
more quickly. This means that ruin, if it is going to happen, will happen sooner, rather than later.
w
w
However it does not affect the probability that ruin does actually occur, when we are considering
w
ruin at any time in the future.
It can be shown that there is indeed only one positive root of (3.1) as follows.
Define g (r )   M X (r )    cr and consider the graph of g (r ) over the interval [0,  ] . Note
first that g (0)  0 .
This is what we discovered in the previous example.
Further, g (r ) is a decreasing function at r  0 since:
d d
g (r )   M X (r )  c
dr dr
so that the derivative of g (r ) at r  0 is  m1  c which is less than zero by assumption

(2.3).
MX (0) gives the mean of X , which is E[ X ] or m1 .
It can also be shown that if the function g (r ) has a turning point, it must be at the minimum
of the function. The second derivative is:
d2 d2
g (r )   M X (r )
dr 2 dr 2
which is always strictly positive.
The second derivative can be written:
M X (r )  E  X 2erX 
 
The function in this expectation is made up of two positive factors, and hence the expectation
must have a positive value.
Hence, there can only be one turning point, since any turning point is a minimum. To show
that there is a turning point, note from (2.5) that lim g (r )   . Since g (r ) is a decreasing
r  
function at r  0 , it must have a minimum turning point and so the graph of g (r ) is as
shown in Figure 5.

om
i.c
ng
si
om
o m
as
.m
w
w
w
Figure 5
Thus there is a unique positive number R satisfying equation (3.1).
Equation (3.2) is an implicit equation for R . For some forms of F ( x ) it is possible to solve
explicitly for R ; otherwise the equation has to be solved numerically.
Consider the exponential distribution where F ( x )  1  e  x .
This is in the Tables, using  as the parameter for the exponential distribution, which avoids
confusion with the Poisson parameter.

For this distribution, M X (r )  , so:
 r

  cR 
 R
    R  cR  cR 2  
  
 R2      R  0  R    (3.3)
 c c
since R is the positive root of (3.1).
(1   ) 
If c  , then R  .
 (1   )
Question
Write down the equation for the adjustment coefficient for personal accident claims if 90% of claims
are for £10,000 and 10% of claims are for £25,000, assuming a proportional security loading of 20%.
Show that this equation has a solution in the range 0.00002599  R  0.00002601 .

om
i.c
ng
si
om
o m
Solution
as
.m
w
The adjustment coefficient satisfies:
w
w
1  (1   ) m1R  MX (R)
The distribution of the individual claim sizes X is:
10,000 with probability 0.9

X 
25,000 with probability 0.1
So E[ X )   xP( X  x)  0.9  10,000  0.1  25,000  11,500
and M X (R)  E eRX    eRx P( X  x)  0.9 e10,000R  0.1 e25,000R

 
The security loading is   0.2 .
So the equation for the adjustment coefficient is:
1  1.2  11,500R  0.9e10,000R  0.1e25,000R
ie 1  13,800R  0.9e10,000R  0.1e25,000R
We can show that there is a solution in the range stated by looking at the values of LHS  RHS :
At R  0.00002599 : 1  13,800R  (0.9e10,000R  0.1e25,000R )  0.000035
At R  0.00002601 : 1  13,800R  (0.9e10,000R  0.1e25,000R )  0.000018
Since there is a reversal of signs (and we are dealing with a continuous function), the difference
must be zero at some point between these two values, ie there is a solution of the equation in the
range 0.00002599  R  0.00002601 .
If the equation for R has to be solved numerically, it is useful to have a rough idea of R ’s
value. Equation (3.2) can be used to find a simple upper bound for R as follows:
  cR   M X (R )

   e Rx f ( x )dx
0

   (1  Rx  21 R 2 x 2 )f ( x )dx
0
  (1  Rm1  21 R 2 m2 )
The inequality is true because all the terms in the series for e Rx are positive. So e Rx must always
be greater than the total of the first few terms.

om
i.c
ng
si
om
m
Alternatively, we could present this proof as:
o
as
.m
  cR  E[eRX ]
w
w
w
 1 
 E 1  RX  R2 X 2  
 2 
 1  
  E[1]  E[RX ]  E  R2 X 2   
 2  
 1 
  1  RE[ X ]  R2E[ X 2 ]
 2 
So that (c   m1 )R  1  R2m , giving:

2 2
R  2(c   m1) /  m2 (3.4)
so that R  2 m1 / m2 when c  (1   ) m1 . Notice that if the value of R is small, then it

should be very close to this upper bound since the approximation to e Rx should be good.
Question
Find an upper limit for the adjustment coefficient in the previous question, and comment on your
answer.
Solution
Here:
E[ X ]  11,500
Also:
E[ X 2 ]   x 2P( X  x)  0.9  10,0002  0.1  25,0002  152,500,000
So:
2 m1 2  0.2  11,500
R   0.0000302
m2 152,500,000
So this is a reasonable initial estimate, compared with the correct value of approximately
0.000026.
1 1 1
R log(c / m1 )  log(1   )  log1.2  0.00000729
M M 25,000
The steps used in the proof are equally valid for a discrete claims distribution. However, note that
the lower bound obtained here is not very close to the accurate value for R .

om
i.c
ng
si
om
o m
3.3 A lower bound for R
as
.m
A lower bound for R can be derived when there is an upper limit, say M , to the amount of
w
w
an individual claim. For example, if individual claim amounts are uniformly distributed on
w
(0,100) , then M  100 . The result is proved in a similar fashion to Result (3.4). The lower
bound is found by applying the inequality:
x x
exp(Rx )  exp(RM )  1  for 0  x  M (3.5)
M M
The inequality is proved through the series expansion of exp(RM ) :

x x x (RM ) j x
M
exp(RM )  1  
M M
 j !
 1
M
j 0

R j M j 1x
 1  j!
j 1

(Rx ) j
 1  j!
for 0  x  M
j 1
 exp(Rx )
since x j  M j 1 x if 0  x  M .
Inequality (3.5) can be used to show that:
1
R log(c /  m1)
M
when individual claim amounts have a continuous distribution on (0, M ) .
This is the lower bound for R that we are trying to find.
The starting point is the equation defining R,
ie Equation 3.2:
M
  cR    exp(Rx )f ( x )dx
0
M
 x x
    exp(RM )  1   f ( x )dx
0
M M
 
 exp(RM )m1    m1
M M
Hence, rearranging:
c 1 RM (RM )2
 (exp(RM )  1)  1   
 m1 RM 2 3!

om
i.c
ng
si
om
m
RM  RM 
o
2
as
1  
.m
1! 2!
w
w
 exp(RM )
w
1  c 
This gives R  log   as required.
M   m1 
1
If c  (1   )m1 , this is just R  log(1   ) .
M
Other approximations for R can be found, particularly when R is small, by truncating the
series expansion of exp(Rx ) .
3.4 The adjustment coefficient – general aggregate claims processes

In Section 3.2 the existence of the adjustment coefficient, R, was proved for a compound
Poisson aggregate claims process. In this section the existence of the adjustment
coefficient for a general aggregate claims process is proved.
Let {Si }
i 1 be a sequence of independent identically distributed random variables:
Si  aggregate claims from a risk in time period i .
c is the constant premium charged to insure this risk
The following assumptions are made:
c  E [Si ] (3.6)
There is some number   0 such that:
lim E e  i      
r S c
  (3.7)
r    
Si has density function h( x ) ,   x   (3.8)
In the general situation the adjustment coefficient is the positive number R that can be
shown to satisfy the following:
E [e R (Si c ) ]  1
The proof that there is one, and only one, positive number R to satisfy this is as follows.
Let f (r )  E [e r (Si c ) ] for   r   .
Then f (0)  1 , f (0)  E [Si  c ]  0 , f ( x )  0 , and:
lim f (r )  
r  

om
i.c
ng
si
om
m
We then use the same argument as for the compound Poisson case, based on the graph below.
o
as
This graph forms part of the Core Reading.
.m
w
w
f(r)
w
1
0 R  r
Figure 6 – f(r) by r
Suppose Si has a compound Poisson distribution with Poisson parameter  and claim size
random variable X .
Then:
E e R (Si c )   1
 
 E e RSi   e Rc
 
 MS (R)  eRc
 e  (M X (R )1)  e Rc
  M X (R )  Rc  
which is the same as (3.2).

om
i.c
ng
si
om
o m
4 The effect of changing parameter values on ruin probabilities
as
.m
w
4.1 Introduction
w
w
Recall that  (U) was defined to be P U( )  0,  0  , and  (U , t ) was defined to be
 (U, t )  P U( )  0, 0    t  .
In this section the effect of changing parameter values on  (U , t ) and  (U ) will be

discussed.
No new theory will be introduced and the method for obtaining numerical values for  (U , t )
will not be discussed. Features of  (U , t ) , and in some cases of  (U ) , will be illustrated by
a series of numerical examples. In these examples the same basic assumptions will be
made as in previous sections. In particular, it will be assumed that the aggregate claims
process is a compound Poisson process. In addition it will be assumed throughout Section
4.3, Section 4.4 and Section 4.5 that:
 the Poisson parameter for the number of claims is 1 (4.1)
 the expected value of an individual claim is 1 (4.2)
 individual claims have an exponential distribution. (4.3)

In Section 4.6, Assumptions (4.2) and (4.3) will be made, but the Poisson parameter will be
allowed to vary.
The implication of Assumption (4.1) is that the unit of time has been chosen to be such that
the expected number of claims in a unit of time is 1. Hence  (U ,500) is the probability of
ruin (given initial surplus U ) over a time period in which 500 claims are expected. The
actual number of claims over this time period has a Poisson distribution (with parameter
500) and could take any non-negative integer value.
The implication of Assumption (4.2) is that the monetary unit has been chosen to be equal
to the expected amount of a single claim. Hence  (20,500) is the probability of ruin (over a
time period in which 500 claims are expected) given an initial surplus equal to 20 times the
expected amount of a single claim.
The advantage of using an exponential distribution for individual claims (Assumption (4.3))
is that both exp( RU ) and  (U ) can be calculated for these examples. See Section 3.2 and
Section 4.2.
4.2 A formula for  (U ) when F(x) is the exponential distribution

The formula for  (U ) when individual claims amounts are exponentially distributed with
mean 1, and when the premium loading factor is  , is given by the following result.
When F ( x )  1  exp(  x ) :
1  U 
 (U )  exp    (4.4)
1   1  
The syllabus does not require this result to be derived or memorised.

om
i.c
ng
si
om
m
This result has been stated in order to illustrate how, for this particular distribution, the
o
as
ultimate ruin probability is affected by changes in parameter values.
.m
w
4.3  (U , t ) as a function of t
w
w
Question
Is  (U , t ) an increasing or decreasing function of t ?
Solution
 (U , t ) is the probability of ruin at some point between times 0 and t . This should increase with
time since the longer the time period, the more chance there is of ruin. It should be intuitively
obvious that  (U , t1 )  (U , t2 ) for t1  t2 , since if a scenario produces ruin before time t1 , then
ruin has also occurred before time t2 .
Figure 7 shows a graph of  (15, t ) for 0  t  500 . The premium loading factor,  , is 0.1 so
that the premium income per unit time is 1.1. Also shown in Figure 7 are  (15) (dotted line)
and   0.1 (solid line) for this portfolio. These last two values are shown as lines parallel
to the time axis since their values are independent of time.
Here:
  (15, t ) has been worked out using a numerical method not described here
  (15) has been calculated using equation 4.4
 exp15R is the upper bound given by Lundberg’s inequality.
Figure 7 – Values over time

om
i.c
ng
si
om
o m
Question
as
.m
Calculate the value of  (15) .
w
w
w
Solution
15
1 0.11.1
 (15)  e  0.23248
1.1
Question
Calculate the value of e 15R .
Solution
 1  0.1
R is worked out from R   .
1  1.1
15 1.1
0.1
So e 15R  e  0.2557
The features of note in Figure 7 are:
(i)  (15, t ) is an increasing function of t
(ii) for small values of t,  (15, t ) increases very quickly (its value doubles as t increases
from 25 to 50 and doubles again as t increases from 50 to 100)
(iii) for larger values of t,  (15, t ) increases less quickly and approaches asymptotically
the value of  (15) .
General reasoning should help you to understand (ii) and (iii). You would expect a much higher
probability of ruin before time 50 than before time 25 since the overall performance of the fund
could easily change in such a short time period. However, if premium rates are expected to be
profitable in the long term, then at time 400, say, a significant surplus will have built up in most
cases and so the probability of ruin at time 425 won’t be that much higher than at time 400. We
are assuming here that accumulated surpluses stay in the fund and are not, for example,
distributed to shareholders.
4.4 Ruin probability as a function of initial surplus
Question
Is  (U , t ) an increasing or decreasing function of U ?

om
i.c
ng
si
om
o m
Solution
as
.m
w
Decreasing. The bigger the initial surplus, the less chance there should be of ruin.
w
w
Figure 8 shows values of  (U , t ) for  and for three values of the initial surplus, U  15, 20,
and 25. The premium loading factor is 0.1 as in Figure 7. For U  15 the graph of  is as in
Figure 7.
Figure 8 –  for different values of U
Question
Calculate the value of  (20) .
Solution
1 0.1 1.1
20
 (20)  e  0.14756
1.1
This is the limit to which the middle of the three lines is tending to as t tends to  .
The features of note in Figure 8 are:
(i) the graphs all have the same general shape
(ii) increasing the value of U decreases the value of  (U , t ) for any value of t
(iii) each of the three graphs approaches an asymptotic limit as t increases (as has
already been noted for U equal to 15 in the discussion of Figure 4). Note that
 (20)  0.1476 and  (25)  0.0937 .
 (U ) is a non-increasing function of U .

om
i.c
ng
si
om
m
In the case of exponentially distributed individual claim amounts, the derivative with respect
o
as
to U of  (U ) is:
.m
w
d 
w
 (U )   (U )
w
dU 1 
which is negative since   0 . Hence  (U ) is a decreasing function of U .
It is intuitively clear that  (U , t ) (of which  (U ) is a special case) should be a decreasing

function of U . An increase in U represents an increase in the insurer’s surplus without any
corresponding increase in claim amounts. Thus, an increase in U represents an increase
in the insurer’s security and hence will reduce the probability of ruin.
4.5 Ruin probability as a function of premium loading
Question
Is  (U , t ) an increasing or decreasing function of  ?
Solution
Decreasing. If everything else remains unchanged, then increasing the premium income will
reduce the probability of ruin.
Figure 9 shows values of  (15, t ) for 0  t  500 and for three values of the premium
loading factor,   0.1, 0.2 and 0.3. The graph of  (15, t ) for   0.1 is the same as the
graph in Figure 7 and the same as one of the graphs in Figure 8. Figure 9 is, in many
respects, similar to Figure 8. The features of note in Figure 9 are:
(i) the graphs of  (15, t ) all have the same general shape,
(ii) increasing the value of  decreases the value of  (15, t ) for any given value of t ;
this is in fact true for any value of U , and is an obvious result since an increase in
 is equivalent to an increase in the rate of premium income with no change in the
aggregate claims process,
(iii) it can be seen that when   0.1, 0.2 and 0.3,  (15, t ) is more or less constant for t
greater than about 150. For t1  t 2 , the difference  (15, t 2 )   (15, t1) represents the
probability that ruin occurs between times t1 and t 2 . Hence for these values of  ,
0.2 and 0.3, (and for this value of the initial surplus, 15, and for this aggregate claims
process) ruin, if it occurs at all, is far more likely to occur before time 150, ie within
the time period for 150 claims to be expected, than after time 150. This point will be
discussed further in Section 4.6.

om
i.c
ng
si
om
o m
as
.m
w
w
w
Figure 9 –  (15, t ) for different values of 
It is clear by general reasoning that  (U ) must be a non-increasing function of  . In the

case of exponential individual claim amounts,  (U ) is a decreasing function of  .
d
 (U )  (1   )1 (U )  U (1   )2 (U )
d
Question
d
Verify this expression for  (U) .
d
Solution
This is probably easiest if we start by writing:
 (U) 
1
1 

exp U(1   )1  U 
We need to differentiate this using the product rule:
d dv du
(uv)  u v
d d d
Differentiating  (U) with respect to  :
d
d
  
 (U)  (1   )1  U(1   )2 exp U(1   )1  U   (1   )2 exp U(1   )1  U
  
Now substituting back in each term for  (U) :
d
 (U)   U(1   )2  (U)  (1   )1 (U)
d  

om
i.c
ng
si
om
m
This is clearly negative since  , U and  (U ) are all positive quantities. Since the
o
as
derivative is less than zero for all values of  ,  (U ) is a decreasing function of  .
.m
w
Figure 10 shows  (10) as a function of  .
w
w
Figure 10
4.6 Ruin probability as a function of the Poisson parameter

Figure 11 shows  (15,10) as a function of  for three values of the premium loading factor,
  0.1, 0.2 and 0.3. This graph is identical to Figure 9 apart from the labelling of the x-axis.
This can be explained by considering the following two risks.
Figure 11
Risk 1: aggregate claims are a compound Poisson process with Poisson parameter 1 and
F ( x )  1  e  x . The premium income per unit time to cover this risk is (1   ) .
Risk 2: aggregate claims are a compound Poisson process with Poisson parameter 0.5 and
F ( x )  1  e  x . The premium income per unit time to cover this risk is 0.5(1   ) .

om
i.c
ng
si
om
m
The unit of time is taken to be one year. The only difference between these risks is that
o
as
twice as many claims are expected each year under Risk 1. This is reflected in the two
.m
premiums.
w
w
Consider Risk 2 over a new time unit equivalent to two years. Then the distribution of
w
aggregate claims and the premium income per unit time are now identical to the
corresponding quantities for Risk 1. Hence, the probability of ruin over an infinite time span
is the same for both risks.
The solid line in Figure 12 shows an outcome of the surplus process for Risk 1 when
  0.1. The dotted line shows the same surplus process when the unit of time is two years.
This illustrates that any outcome of the surplus process that causes ultimate ruin for Risk 1
will also cause ultimate ruin for Risk 2. There is thus no difference in the probability of
ultimate ruin for these two risks. It is only the time (in years) until ruin that will differ.
Measuring times in years, the probability of ruin by time 1 for Risk 1 is the same as the
probability of ruin by time 2 for Risk 2. This explains why Figures 9 and 11 show the same
functions. For example, the value of  (15,10) when   50 (Figure 11) is the same as the
value of  (15,500) when   1 (Figure 9).
Figure 12
Point (iii) in Section 4.5 will now be investigated, where it was noted that values of  (15, t )
were more or less constant for values of t greater than 150 when   0.2 and 0.3. In
particular, the situation will be considered when the premium loading factor is 0.2.
Consider a second aggregate claims process, which is the same as the process considered
throughout this section except that its Poisson parameter is 150 and not 1. (This second
process is really identical to the original one; all that has happened is that the time unit has
been changed.) Use  * to denote ruin probabilities for the second process and  to
denote, as before, ruin probabilities for the original process. The change of time unit means
that for any t  0 :
 * (U , t )   (U ,150t )
but it has no effect on the probability of ultimate ruin (put t   in the relationship above)
so that:
 * (U )   (U )

om
i.c
ng
si
om
m
The point made in (iii) above was that:
o
as
.m
 (15,150)   (15)
w
w
From this and the previous two relations it can be seen that:
w
 * (15,1)   * (15)
In words this relation says that for the second process, starting with initial surplus 15, the
probability of ruin within one time period is almost equal to (actually a little less than) the
probability of ultimate ruin. This conclusion depends crucially on the fact that  * (15,1) is a
continuous time probability of ruin. To see this, consider  * (15,1) , which is just the
probability that for the second process the surplus at the end of one time period is negative.
 * (15,1) can be calculated approximately by assuming that the aggregate claims in one
time period, which will be denoted S*(1), have a normal distribution. Recall that individual
claims have an exponential distribution with mean 1 and that the number of claims in one
time period has a Poisson distribution with mean 150. From this:
E S   1   150
 
and  
Var S   1  300
These are calculated as m1  150  1  150 and m2  150  2  300 , where E  X 2   2 for an
 
Exp 1 distribution.
Now, using tables of the normal distribution:
 * (15,1)  P S * (1)  15  1.2  150 
 P  (S * (1)  150) / 17.32  45 / 17.32
 0.005
X 
Recall that if X  N( , 2 ) , then Z  has a standard normal distribution, ie Z  N(0,1) .

Probabilities for this distribution can be looked up in the Tables.
From Figure 9 it can be seen that the value of  (15,150) , and hence of  * (15,1) , is about
0.07 which is very different from the (approximate) value of the discrete time probability of
ruin  * (15,1) calculated above.
4.7 Concluding remarks

When individual claim amounts are exponentially distributed with mean 1, first note that if
  0 , then  (U )  1 irrespective of the value of U .
We’re thinking here of substituting   0 into equation 4.4.
This result is in fact true for any form of F ( x ) (it trivially follows that if   0 , then  (U )  1 ).
In other words a positive premium loading is essential if ultimate ruin is not to be certain.

om
i.c
ng
si
om
m
Also note that throughout this section it has been assumed that individual claim amounts
o
as
are exponentially distributed with mean 1. This mean could be measured in units of £100,
.m
£1,000 or perhaps £1,000,000. The parameter of the exponential distribution can be set to 1
w
without loss of generality provided that the monetary unit is correctly specified. In simple
w
w
terms, the probability of ruin when U is £1 is the same as the probability of ruin when U is
100 pence. It can be said that:
 (U ) when F ( x )  1  e  x
is the same as:
 ( U ) when F ( x )  1  e  x
In other words, if the expected claims per unit time increase by a factor  , so too must the
initial surplus if the probability of ultimate ruin is to be unchanged.

om
i.c
ng
si
om
o m
5 Reinsurance and ruin
as
.m
w
5.1 Introduction
w
w
One of the options open to an insurer who wishes to reduce the variability of aggregate
claims from a risk is to effect reinsurance. This is a form of insurance in which an
insurance company obtains insurance cover from other insurance companies (reinsurers)
against the risk of losses.
A reduction in variability would be expected to increase an insurer’s security, and hence

reduce the probability of ruin. A reinsurance arrangement could be considered optimal if it
minimises the probability of ruin. As it is difficult to find explicit solutions for the
probability of ruin, the effect of reinsurance on the adjustment coefficient will be considered
instead.
If a reinsurance arrangement can be found that maximises the value of the adjustment
coefficient, the upper bound for the probability of ultimate ruin will be minimised. As the
adjustment coefficient is a measure of risk, it seems a reasonable objective to maximise its
value. In the following, the effect on the adjustment coefficient of proportional and of
excess of loss reinsurance arrangements will be considered.
Throughout this section we will use the notation X  individual claim amount, Y  amount paid
by the direct insurer and Z  amount paid by the reinsurer.
5.2 Proportional reinsurance

Under proportional reinsurance, the reinsurer covers an agreed proportion of each risk and
the reinsurance premium is in proportion to this risk ceded.
For example, the reinsurer might agree to pay 20% of each claim. The insurer would then pay
80% of the claim amount.
If we have a retained proportion  then:
Y X Z  (1   ) X
 E[Y ]  E[ X ]   E[ X ] E[ Z ]  E[(1   ) X ]  (1   )E[ X ]
Question
Write down expressions for Var (Y ) and Var (Z ) .
Solution
Var (Y )  Var ( X )   2Var ( X )
Var (Z )  Var ((1   ) X )  (1   )2 Var ( X )

om
i.c
ng
si
om
m
Let us consider the idea of a proportional reinsurance approach by way of a question:
o
as
.m
Question
w
w
w
Consider the insurer in the third question of Section 2.4. The number of claims has a Poisson
distribution with parameter 30 per year. The individual claim amount distribution is lognormal
with parameters   3 and  2  1.1 . The rate of premium income from the portfolio is 1,200 per
year. The insurer has an initial surplus of 1,000
This insurer is investigating the possibility of using proportional reinsurance. It has approached a
reinsurer, who uses a security loading of 50% to calculate its reinsurance premiums. If the insurer
decides to reinsure 20% of each risk in the portfolio, estimate the effect the reinsurance will have
on its probability of ruin at Time 2. Again you can assume that the aggregate claim distribution is
approximately normal.
Solution
We first need to calculate the reinsurance premium. Since the reinsurer takes responsibility for
20% of each risk, and uses a loading factor of 50%, the reinsurance premium (per annum) will be:
RP  1  R   m1  1.5  30  0.2  e3.55  313.32
So over a two year period, the insurer will pay 626.64 for the reinsurance.
We now use Snet (2) for the insurers aggregate payments (net of reinsurance). We need the
mean and variance of Snet (2) , which are, using the formulae for a compound Poisson distribution
as before:
E[ Snet (2)]  60  0.8  e3.55  1,671.04
and: Var  Snet (2)   60  0.82  e8.2  139,812.49
So ruin will occur if:
Snet (2)  1,000  2,400  626.64  2,773.36
Using a normal distribution approach, we have:
 2,773.36  1,671.04 
P  Snet (2)  2,773.36   P N(0,1)  
 139,812.49 
 1  (2.9481)  0.0016
Question
Comment on the usefulness of reinsurance in this context.

om
i.c
ng
si
om
o m
Solution
as
.m
w
The reinsurance has reduced the probability of ruin to some extent, ie from about 0.25% (which
w
was calculated in Section 2.4) to about 0.16%. However, this result is probably quite sensitive to
w
the assumptions made (we are near the tail of the normal distribution), and slightly different
assumptions might give us very different results.
We will also want to look at the effect of reinsurance on profitability. As we are paying a
reinsurance premium, it is likely that the overall effect on profitability is negative (although the
effect on security is positive, as we have seen). There is likely to be a trade-off between security
and profitability here.
5.3 Excess of loss reinsurance

Under excess of loss reinsurance, the cost to an insurer of a large claim, a cluster of claims
arising from a single event (such as an explosion) or claims over a given period (such as a
catastrophe) is capped with the liability above a certain level being passed to a reinsurer. In
most cases the reinsurer’s maximum liability is limited, but we will assume there is no such
limit within this chapter.
We refer to the capped amount as the retention of the insurer.
For example, the insurer might retain the first ten million of claims arising during the year, with
the reinsurer paying any excess above this amount. If the claim amount turns out to be less than
ten million, then the insurer will pay this amount in full and the reinsurer will pay nothing.
However, if the claim amount is twelve million, then the insurer will pay ten million and the
reinsurer will pay the excess of two million.
Subject SP7 discusses the different forms of reinsurance contracts and their relative merits
in greater detail.
We can apply the same type of logic as used in the previous section if the insurer decides to buy
excess of loss reinsurance. You might like to think about the effect on the probability of ruin if the
insurer in the previous example purchases excess of loss reinsurance with an individual retention
of 2,000, say, and a security loading of 50% as before.
If we have a retention limit M , and no upper limit, then:
X X M  0 X M
Y  Z 
M X  M X  M X  M
Y can be also be written as min( X , M) , and Z can also be written as max(0, X  M) .
Question
Calculate E Y  if X has an exponential distribution with parameter 0.01, and the insurer has an
excess of loss reinsurance arrangement with retention limit M .

om
i.c
ng
si
om
o m
Solution
as
.m
The formula for an expectation is  xf (x) dx . We have to calculate E[Y ] by carrying out two
w
w
w
x
integrals, to allow for the two different ranges of X :
M 
E[Y ]   xf (x) dx   Mf (x) dx
0 M
M 
  0.01xe 0.01 x dx   0.01Me 0.01 x dx
0 M
Using integration by parts:
M M 
 0.01x 0.01 x  0.01 0.01 x  0.01 0.01 x 
  e   e dx   M e 
 0.01  0 0 0.01  0.01 M
M  1 M 

   xe 0.01 x    e 0.01 x    Me 0.01 x 
  0  0.01 0  M
1 0.01M 1
 Me 0.01M  e   Me 0.01M
0.01 0.01

1
0.01

(1  e 0.01M )  100 1  e 0.01M 
Question
Calculate Var (Z ) (in terms of M ) if X  U(0,100) , where the insurer has an excess of loss
reinsurance arrangement with retention limit M , 0  M  100 .
Solution
To find Var (Z ) , we need to find E[ Z 2 ] , since Var (Z )  E[ Z 2 ]  E 2[ Z ] .
M 100
E[ Z 2 ]   0 f (x) dx  
2
(x  M)2 f (x) dx
0 M
1
The PDF of the U(0,100) distribution is (see page 13 of the Tables), so:
100
100
100
(x  M)2  (x  M)3  (100  M)3

2
E[ Z ]  dx    
M
100  300  M
300

om
i.c
ng
si
om
m
We now need E[ Z ] :
o
as
.m
100
100  (x  M)2  (100  M)2
w
( x  M)
E[ Z ]   dx   
w

w
M
100  200  M
200
2
(100  M)3  (100  M)2 
So Var (Z )   
300  200 
 
(100  M)3 (100  M)4

 
300 40,000
5.4 Maximising the adjustment coefficient under proportional reinsurance

First consider the effect of proportional reinsurance with retention  on the insurer’s
adjustment coefficient. Throughout Section 5.4 the insurer’s premium income per unit time,
before payment of the reinsurance premium, will be written as (1   ) m1 , which represents
the expected aggregate claims per unit time for the compound Poisson process with a
loading factor  . It will also be assumed that the reinsurance premium is calculated as
(1   )(1   ) m1 . Since the reinsurer pays proportion 1   of each claim, (1   ) m1
represents the reinsurer’s expected claims per unit time.
Thus,  is the premium loading factor used by the reinsurer. Hence, the insurer’s premium
income, net of reinsurance, is:
 1      1    1      m1 (5.1)
Question
Explain this formula.
Solution
The insurer will charge a premium of:
1    E  X   1    m1
The reinsurer will charge a premium of:
(1   )E[ Z ]
But E[ Z ]  (1   )E[ X ]  (1   )m1 , so the reinsurer’s premium is:
(1   ) (1   )m1

om
i.c
ng
si
om
m
So the net premium received by the insurer is the difference:
o
as
.m
(1   )m1  (1   ) (1   )m1  (1   )  (1   )(1   ) m1
w
w
w
It will also be assumed that    . If this were not true, it would be possible for the insurer
to pass the entire risk on to the reinsurer and to make a certain profit.
This of course ignores commission, expenses and other adjustments to the theoretical risk
premium.
For the insurer’s premium income, net of reinsurance, to be positive:
1    (1   )(1   )
ie   (   ) / (1   )
Question
What range of values is possible for  if   0.2 and   0.4 ?
Solution
0.2
  0.1429
1.4
But since  cannot exceed 1, the possible range of values is 0.1429    1 .
There is, however, a more important constraint on the insurer. The insurer’s net of
reinsurance premium income per unit time must exceed the expected aggregate claims per
unit time. Otherwise ultimate ruin is certain (as noted in Section 0). Net of reinsurance, the
insurer’s expected aggregate claims per unit time are  m1 .
Thus:
(1   )  (1   )(1   )  

or   1 (5.2)

Question
So what is the range of possible values of  now, given the figures in the previous question?
Solution
0.2
 1  0.5 . So 0.5    1 .
0.4

om
i.c
ng
si
om
m
Equation (5.2) specifies the insurer’s minimum retention level since:
o
as
.m
1   /   (   ) / (1   ) when   
w
w
the only case of interest. If the premium loading factors are equal, then inequality (5.2)
w
becomes   0 . In this case there exists a risk sharing arrangement and any retention level
is possible. If, however,    then the insurer has to retain part of the risk.
Same loadings
First consider the case where both the insurer and the reinsurer use  as the premium
loading factor. The adjustment coefficient will be found as a function of the retention level
 , when F ( x )  1  e 0.1x .
The distribution of the insurer’s individual claims net of reinsurance is exponential with
parameter 0.1 /  . This can be seen by noting that if Y   X , then:
P [Y  y ]  P [ X  y /  ]  1  exp{ 0.1y /  }
Note that the assumptions for the claims process and the adjustment coefficient equation apply
equally well in the presence of reinsurance, provided that we use the net premium and the net
claim amounts in the adjustment coefficient equation.
Question
What will be the general equation for R , the direct insurer’s adjustment coefficient, when there
is reinsurance?
Solution
From the previous work, we know that the equation for R is:
  cr  MX (r )
With reinsurance this will become:
  cnet r  MY (r )
for the direct insurer. But we know that cnet  (1   )E[ X ]  (1   )E[ Z ] , so the equation for R
becomes:
   (1   )E[ X ]  (1   )E[ Z ] r  MY (r )
or 1   (1   )E[ X ]  (1   )E[ Z ] r  MY (r )
In the case of proportional reinsurance this is:
1   (1   )E[ X ]  (1   )(1   )E[ X ] r  MY (r )

om
i.c
ng
si
om
m
Hence, the equation defining R (see formula (3.2)) is:
o
as
.m

  (1   ) 10 R    e Rx (0.1/  )e 0.1x / dx
w
0
w
w
1
  1  (1   )10 R   (5.3)
1  10 R

R for 0    1 (5.4)
(1   )10
It can be seen that R is a decreasing function of  . This is sensible as the larger the
retention  , the larger the risk for the insurer and so  (U ) would be expected to increase,
and R to decrease, with  .
Different loadings
Now consider what happens when    . For the rest of Section 5.4 assume that:
 the individual claim amount distribution is F ( x )  1  e 0.1x
 the insurer’s premium loading factor is   0.1 .
Case A:   0.2
Suppose first that the reinsurer’s premium loading factor is   0.2 , so that the insurer’s
(net) premium income per unit time is (12  1) .
This comes from Equation 5.1.
Equation (5.2) shows that the insurer must retain at least 50% of each claim. Hence, a value
of  will be sought in the interval [0.5,1] that maximises the value of R . The equation
defining R is:

  (12  1) R 
1  10 R
Question
Derive this formula for R .
Solution
Using the equation from the previous question where   0.1 , E[ X ]  1 / 0.1  10 and   0.2 , we
get:
1  1.1  10  1.2(1   )  10  r  MY (r )
 1  11  12(1   ) r  MY (r )
 1  (12  1)r  MY (r )

om
i.c
ng
si
om
o m
But what about MY (r ) ?
as
.m
1
 r 
w
1
MY (r )  E[erY ]  E[er X ]  MX ( r )   1  
w
 1  10r
 0.1 
w
Therefore the equation is:
1
1  (12  1)r 
1  10r
This follows from (5.3) as only the premium is different  which leads to:
2  1
R for 0.5    1 (5.5)
10(12 2   )
The right hand side is based on the MGF of the net claim amounts which have an   
distribution.
Question
Derive this formula for R .
Solution
Multiplying the equation in the previous question through by 1  10 R gives:
(1  12 R  R)(1  10 R)  1
Multiplying out the left hand side and subtracting 1 from both sides gives:
(10  120 2 )R2  (2  1)R  0
Dividing through by R and rearranging gives the required expression.
As when the loading factors were equal, the adjustment coefficient depends on the
retention level.
The value of  that maximises R in (5.5) is sought.
Differentiate R with respect to  (using the quotient rule for differentiation) to give:
dR 20(12 2   )  (2  1)10(24  1)


d 100(12 2   )2
du dv
v u
d u d d .
The quotient rule is  
d  v  v 2

om
i.c
ng
si
om
o m
1 1
Alternatively the algebra is a little easier if you write R  
as
.
10 12  1
.m
w
Now the denominator is always positive for values of  in [0.5,1] , so there will be a turning
w
w
point of the function when:
20(12 2   )  (2  1)10(24  1)
ie when 24 2  24  1  0
The roots of this quadratic are 0.9564 and 0.0436, and so the turning point which is of
interest is 0.9564.
Remember that  must lie in the range (0.5,1).
Consider the following values:
 R
0.5 0
0.9564 0.00911
1.0 0.00909
This shows that R has a maximum in [0.5,1] at 0.9564.
Alternatively you can show that the second derivative is negative.
Figure 13 shows R as a function of  (as given by (5.5)) for values of  greater than 0.85.
This range of  values has been chosen to highlight the important features of the graph.
The dotted line shows the value of R when   1 (ie no reinsurance).
Figure 13 – R as a function of 

om
i.c
ng
si
om
m
It can be seen from Figure 13 that there is a range of values for  ,     1 , such that if the
o
as
retention level is in this range, the value of the adjustment coefficient exceeds the value
.m
when   1 . The value of  can be calculated from (5.5) by setting the value of R at 
w
w
equal to the value of R at 1, giving   0.9167 . The arrow in Figure 13 indicates the value
w
of .
You can check for yourself that   11 / 12 .
In terms of maximising the adjustment coefficient, the optimal retention level is   0.9564 .
It should be noted, however, that optimality in one sense does not imply optimality in
another. For example, if the insurer does not effect reinsurance, then the expected profit
per unit time is  m1 (ie  , since   0.1 and m1  10 ).
m1 is the ‘loading’ bit.
If the insurer effects reinsurance with retention level 0.9564, then the expected profit per
unit time is 0.9128 (ie premium income, from (5.1), less expected claims).
The expected profit per unit time is now found in terms of  and  .
It has already been calculated from (5.1), that with   0.1 ,   0.2 and m1  10 , the
insurer’s net premium income is (12  1) . The insurer’s expected claims per unit time are
10 . Hence, the expected profit per unit time is (2  1) .
If we put   0.9564 , this gives 0.9128 , as stated above.
This shows that expected profit per unit time is an increasing function of  , and if the
insurer were to choose  to maximise the expected profit per unit time, the choice would
be   1 . This example illustrates a general point – the level of reinsurance is a trade-off
between security and profit.
Case B:   0.3
The value of  is now found that maximises R when the reinsurer’s premium loading
factor is 0.3.
The calculations are very similar to the previous case.
From (5.1), the insurer’s net premium income is (13  2) , so that the equation defining R
is:

  (13  2) R 
1  10 R
which leads to:
3  2
R for 0.67    1
10(13 2  2 )
1 1
Or R   , adopting the same approach as before.
10 13  2

om
i.c
ng
si
om
m
By (5.2), the insurer must retain at least 2/3 of each claim so the value of  in the range
o
as
 2 3,1 that maximises R is sought. Differentiation gives:
.m
w
w
dR 30(13 2  2 )  (3  2)10(26  2)
w

d 100(13 2  2 )2
and the function has a turning point when:
30(13 2  2 )=(3  2)10(26  2)
ie when:
39 2  52  4  0
The roots of this quadratic are 0.0820 and 1.2514, so there are no turning points in the
interval [2 3 ,1] and R as a function of  in this interval increases from 0 at   2 / 3 to
0.00909 at   1 . Thus, the value of  which maximises the adjustment coefficient is 1.
It is not always possible to increase the value of the adjustment coefficient by effecting
reinsurance. Note that when an insurer effects reinsurance, this reduces the variability of
the insurer’s aggregate claims. A reduction in variability is associated with an increase in
the value of the adjustment coefficient. However, when    , the insurer’s premium
loading factor, net of reinsurance, decreases, and the value of the adjustment coefficient is
expected to decrease with the loading factor. When the reinsurer’s premium loading factor
was 0.3, the reduction in the insurer’s security caused by the reduction in the loading factor
has a greater effect on the adjustment coefficient than the increase resulting from
reinsurance for all values of  .
Loading factor (net of reinsurance)

The insurer’s premium loading factor, net of reinsurance, implied by (5.1) is now found, and
shown to be an increasing function of  .
The loading factor is found by dividing the expected profit per unit time by the expected
claims per unit time. The expected profit per unit time is:
[(1   )  (1   )(1   )] m1   m1
This is just an algebraic expression for net premiums less expected net claims.
So the loading factor is:
   [(1   )  (1   )(1   )   ] / 
   (   ) / 
d 
Now  (   ) /  2 which is positive since    , so that   is an increasing function
d
of  . Thus, the net loading factor increases as the retention level increases.

om
i.c
ng
si
om
o m
5.5 Maximising the adjustment coefficient under excess of loss reinsurance
as
.m
In this section the effect of excess of loss reinsurance on the adjustment coefficient will be
w
w
considered. The following assumptions will be made for Section 0:
w
 the insurer’s premium income (before reinsurance) per unit time is (1   ) m1
 the reinsurance premium per unit time is (1   ) E (Z ) , where  (  ) is the

reinsurer’s premium loading factor, and Z  max(0, X  M ) .
The insurer’s individual net claim payments are distributed as Y  min( X , M ) , and the
insurer’s premium income, net of reinsurance, is:
c *  (1   ) m1  (1   ) E (Z )
which gives the equation defining R as:
M 
  c *R     e Rx f ( x )dx  e RM [1  F (M )]
0 
 
You may see c* referred to as cnet .
Question
Explain where the right hand side of this equation comes from.
Solution
X X M
We need MY (r ) , where Y   and X has PDF f (x) . By definition, MY (r )  E[erY ] , but
M X  M
we need to express this as two separate integrals to take into account the different ranges of X :
M 
e f (x) dx   e rM f (x) dx
rY rx
E [e ] 
0 M
M 
 e f (x) dx  e  f (x) dx
rx rM

0 M
But the second integral is just integrating the PDF from M to  . This is the same as P( X  M) ,
which can be written as 1  F (M) . So the right hand side of the equation is:
M 
   e rx f (x) dx  e rM (1  F (M))
 0 
This is formula (3.2) with a truncated claim amount distribution as a result of the excess of
loss reinsurance.

om
i.c
ng
si
om
m
To illustrate ideas, look at the situation when X  U (0,20) , so that f ( x )  0.05 for 0  x  20.
o
as
Then for 0  M  20 :
.m
w
20
w
 ( x  M ) 0.05 dx  10  M  0.025M
2
w
E[Z ] 
M
e
Rx
and MY (R )  0.05dx  e RM (1  0.05M )
0
0.05 RM
 (e  1)  e RM (1  0.05M )
R
The equation for R must be solved numerically for given values of  and  . Figure 14
shows R as a function of M when     0.1 . As in Section 5.4, any retention level is
possible when the premium loading factors are equal. R is a decreasing function of M .
Figure 14 – R as a function of M
In Figure 14, R goes to  as M goes to 0.
When    , there is a minimum retention level for the same reason as in the previous
section.
Recall that the lower limit for  given by Equation 5.2 applied when we were considering
proportional reinsurance.
For example, when   0.1 and   0.2 the insurer’s net premium income, c , is
11  1.2 (10  M  0.025M 2 ) and this must exceed the insurer’s expected claims, net of
reinsurance. The insurer’s expected net claims equal  E [ X ]   E [ Z ] , which gives
 (M  0.025M 2 ) .

om
i.c
ng
si
om
m
Thus:
o
as
.m
1  1.2M  0.03M 2  M  0.025M 2
w
w
w
 M 2  40M  200  0
 5.8579  M  34.1421
Hence, the minimum retention level is 5.8579. Similarly, when   0.4 , the minimum
retention level is 10.
Figure 15 shows R as a function of M for the following combinations of  and  :
(a)   0.1 and   0.2 (solid line)
(b)   0.1 and   0.4 (dotted line).
Without reinsurance, ie for M  20 , the insurer’s adjustment coefficient is 0.014

(irrespective of the reinsurer’s loading factor).
From Figure 15, it can be seen that, for   0.2 :
R (M )  R (20) for 9.6  M  20
R (M )  R (20) for M  9.6
and for   0.4 :
R (M )  R (20) for R  20
Figure 15 – R as a function of M
Hence, for   0.2 it is possible for the insurer to increase the value of the adjustment
coefficient by effecting reinsurance, provided that the retention level is above 9.6. However,
when   0.4 , the insurer should retain the entire risk in order to maximise the value of the
adjustment coefficient. As in the case of proportional reinsurance, the insurer’s expected
profit per unit time is reduced if reinsurance is effected.

om
i.c
ng
si
om
o m
Question
as
.m
Claims occur as a Poisson process with rate  and individual claim sizes X follow an Exp( )
w
w
distribution. The office premium includes a security loading 1 . An individual excess of loss
w
arrangement operates under which the reinsurer pays the excess of individual claims above an
amount M in return for a premium equal to the reinsurer’s risk premium increased by a
proportionate security loading 2 . Derive and simplify as far as possible an equation satisfied by
the adjustment coefficient for the direct insurer.
Solution
The adjustment coefficient equation is   cR   MX (R) . The net rate of premium income for the
direct insurer equals the rate of premiums charged to the policyholder minus the rate of premiums
paid to the reinsurer:

1
c  (1  1 )  (1  2 )  (x  M) e   x dx
 M
The second term can be integrated using the substitution y  x  M , and identifying the integral as
the mean of an Exp( ) distribution. This gives:
1
c [(1  1 )  (1  2 )e   M ]

The individual net claims are the claims paid to policyholders minus the recoveries from the
reinsurer. So the MGF (which is valid for all values of R   ) is:
M 
 x  x 1
M X (R)   e  e dx   e  e dx 
Rx RM
[   R e ( R)M ]
0 M
 R
So the equation for the adjustment coefficient is:
1 1
  [(1  1 )  (1  2 )e   M ]R   [   R e ( R)M ]
  R
Cancelling  ’s and multiplying through by  (  R) gives:
 (  R)  (  R) (1  1 )  (1  2 )e   M  R      Re ( R)M 

   
Cancelling the  2 ’s from both sides gives:
  R  (  R)[(1  1 )  (1  2 )e   M ]R    R e ( R)M
Cancelling R’s to exclude the trivial solution gives:
   (  R)[(1  1 )  (1  2 )e   M ]    e ( R)M

om
i.c
ng
si
om
m
The adjustment coefficient R is the smallest positive solution of this equation.
o
as
.m
w
w
Question
w
Use the approximation e x  1  x  x2 / 2 to find an approximate numerical value for the
adjustment coefficient for the previous example in the case where   0.05 , 1  0.3 , 2  0.4
and M  10 .
Solution
Using the values given, the adjustment coefficient equation becomes:
0.05  (0.05  R)(1.3  1.4e0.5 )  0.05e0.510R
Multiplying by 20e0.5 to clear some of the fractions gives:
e0.5  (1  20R)(1.3e0.5  1.4)  e10R
Expanding the LHS and applying the approximation to the RHS:
0.90538  14.8668R  1  10R  50R2
ie 0.09462  4.8668R  50R2  0
Solving this using the quadratic formula (taking the smallest positive root) gives:
4.8668  4.86682  4(50)(0.09462)

R  0.0268
2(50)

om
i.c
ng
si
om
o m
as
Chapter 20 Summary
.m
w
w
Poisson process – claim numbers
w
Claim numbers N(t )t 0 can be modelled using a Poisson process with parameter  so that
N(t)  Poisson(t ) . A Poisson process is an example of a counting process.
Time until first claim and inter claim time

If N(t )t 0 is a Poisson process with parameter  , then the time until the first claim and
the inter claim time follow exponential distributions with parameter  :
fT (t )  e t (t  0 )
Compound Poisson process - aggregate claim amounts

S(t )  X1  X2    XN(t )
E  S(t )  tE( X ) Var  S(t )   tE  X 2  MS(t ) (u)  e t[MX (u)1]

 
Surplus process
U(t )  U  ct  S(t) , t0
U is the initial surplus and c is the premium income per unit time
c  1    E[S] where  is the insurer’s premium loading
Ruin probabilities
 (u)  P[U(t )  0 for some t ]
 (u, t0 )  P[U(t )  0 for some t  t0 ]
 h (u)  P[U(t )  0 for some t  h,2h,3h,]
 h (u, t0 )  P[U(t )  0 for some t  h,2h,3h, and t  t0 ]

om
i.c
ng
si
om
o m
as
Lundberg’s inequality and the adjustment coefficient
.m
w
For the continuous time model with an infinite time horizon, Lundberg’s inequality, which
w
w
uses a parameter R called the adjustment coefficient, provides an upper bound for the
probability of ultimate ruin. The adjustment coefficient R is an inverse measure of risk, ie the
higher the value of R, the lower the upper bound on the probability of ultimate ruin.
For a compound Poisson process with parameter  , the adjustment coefficient R is the
unique positive root of the equation:
  cr   MX (r )
where  is the Poisson parameter, c is the premium rate per unit of time and MX (r ) is the
MGF of the individual claim amounts at point r .
It is possible to derive upper and lower bounds for R.
The adjustment coefficient in the presence of reinsurance

In the presence of reinsurance, for a compound Poisson process with parameter  , the
adjustment coefficient R is the unique positive root r of the equation:
  cnet r   MY (r )
where  is the Poisson parameter, cnet is the premium rate per unit of time net of the rate
paid to the reinsurer and MY (r ) is the MGF of the individual claim amounts paid by the insurer
(net of reinsurance) at point r .
In order to maximise security, the insurer will want to find a reinsurance arrangement that
maximises the adjustment coefficient R. However, this will not necessarily be the
arrangement that maximises expected profits. There is a trade-off between security and
profit.
Effect of changes in parameter values on the probability of ruin

The probability of ultimate ruin decreases if the insurer’s premium loading  is increased or if
the insurer’s initial surplus U is increased. This is because the insurer has more of a buffer
against claims.
An increase in the value of the Poisson parameter  will not affect the probability of ultimate
ruin since the expected aggregate claims E[S]  E[ X ] , the variance of aggregate claims
Var (S)  E  X 2  and the premium rate 1    E[ X ] all increase proportionately in line with
 
 . However, it will reduce the time it takes for ruin to occur.

om
i.c
ng
si
om
o m
An increase in the variance Var ( X ) of the individual claim amounts will increase the probability
as
.m
of ruin as it will increase the uncertainty associated with the aggregate claims process without
w
any corresponding increase in premium.
w
w
An increase in the expected individual claim amount E[ X ] will increase the probability of ruin.
The expected aggregate claims and the premium rate both increase proportionately in line with
E[ X ] , however the variance of the aggregate claims amount increases disproportionately since

Var (S)  E  X 2    Var  X    E[ X ]
 
2
 . The variance of the aggregate claim amount increases
in line with  E[ X ] .
2

c20 Notes

Uploaded by

Copyright:

Available Formats

c20 Notes

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

c20 Notes

Uploaded by

Copyright:

Available Formats

om

The Actuarial Education Company  IFE: 2019 Examinations

 IFE: 2019 Examinations The Actuarial Education Company

Which of the following functions are o(x) as x  0 ?

(i) x2 (ii) ex (iii) e x  1  x

(iii) Yes, because if we expand e  x as a power series and simplify, we get:

The Actuarial Education Company  IFE: 2019 Examinations

It can be seen that:

with the understanding that S (t ) is zero if N (t ) is zero.

c = the rate of premium income per unit time

1.2 The surplus process

 IFE: 2019 Examinations The Actuarial Education Company

The Actuarial Education Company  IFE: 2019 Examinations

Now to be more precise. The following two probabilities are defined:

 U2 , t    U1, t  (1.1)

 U2    U1  (1.2)

The intuitive explanations for these relationships are as follows:

 IFE: 2019 Examinations The Actuarial Education Company

1.4 The probability of ruin in discrete time

 h (U )  P U  t  < 0, for some t , t = h, 2h, 3h,

 h (U , t )  P U   < 0, for some  ,  = h, 2h, ,t  h, t 

Note that it is assumed for convenience in the definition of  h (U , t ) that t is an integer

The Actuarial Education Company  IFE: 2019 Examinations

 h (U2 )   h (U1) (1.6)

Explain why Equation 1.9 is true.

 IFE: 2019 Examinations The Actuarial Education Company

2.2 The Poisson process

(i) N (0)  0 , ie there are no claims at time 0

(ii) for any t  0 , N (t ) must be integer valued

(iii) when s  t , N (s )  N (t ) , ie the number of claims over time is non-decreasing

(iv) when s  t , N (t )  N (s ) represents the number of claims occurring in the time

(i) N (0)  0 , and N (s )  N (t ) when s  t

The Actuarial Education Company  IFE: 2019 Examinations

Time to the first claim

so that T1 has an exponential distribution with parameter 

Time between claims

 IFE: 2019 Examinations The Actuarial Education Company

P(N  2)  P(N  0)  P(N  1)  e5  5e5  0.040

The Actuarial Education Company  IFE: 2019 Examinations

2.3 The compound Poisson process

The following three important assumptions are made:

 the stochastic process N (t )t 0 is a Poisson process whose parameter is

 IFE: 2019 Examinations The Actuarial Education Company

The definition of a moment generating function is:

Subject CS2 shows that if:

where N  Poisson    , then:

E  S   E  X   m1 Var (S)  E  X 2   m2

These formulae can be found on page 16 of the Tables.

Here N  t   Poisson  t  , therefore E  S  t    tm1 and Var  S(t)  tm2 . Also

The Actuarial Education Company  IFE: 2019 Examinations

Why is this intuitively reasonable?

2.4 Probability of ruin in the short term

U(1)  initial surplus + premiums  expenses  claims

 0.1m  100  5,000  100  0.2  5,000  S(1)

 IFE: 2019 Examinations The Actuarial Education Company

P U(1)  0   P  S(1)  0.5m

 P N 0.35m,  0.1m    0.5m

U(2)  initial surplus + premiums  expenses  claims

The distribution of S(2) is: