4

Ruin theory

4.1 INTRODUCTION

In this chapter we focus again on collective risk models, but now in the long
term. We consider the development in time of the capital of an insurer.
This is a stochastic process which increases continuously because of the earned
premiums, and decreases stepwise because of the payment of claims. When the
capital becomes negative, we say that ruin occurs. Let denote the probability
that this ever happens, provided that the annual premium and the claims process
remain unchanged. This probability is a useful tool for the management since it
serves as an indication of the soundness of the insurer’s combined premiums and
claims process, given the available initial capital A high probability of
ruin indicates instability: measures such as reinsurance or raising some premiums
should be considered, or the insurer should attract extra working capital.
The probability of ruin enables one to compare portfolios with each other, but
we cannot attach any absolute meaning to the probability of ruin, as it doesn’t
actually represent the probability that the insurer will go bankrupt in the near
future. First of all, it might take centuries for ruin to actually happen. Moreover,
potential interventions in the process, for instance paying out dividends or raising
the premium for risks with an unfavorable claims performance, are ruled out in the
81
82 RUIN THEORY

determination of the probability of ruin. Furthermore, the effects of inflation on

the one hand and the return on the capital on the other hand are supposed to cancel
each other out exactly. The ruin probability only accounts for the insurance risk,
not the managerial blunders that might occur. Finally, the state of ruin is nothing
but a mathematical abstraction: with a capital of –1 Euro, the insurer isn’t broke
in practice, and with a capital of +1 Euro, the insurer can hardly be called solvent.
The calculation of the probability of ruin is one of the classical problems in actu-
arial science. Although it is possible to determine the moment generating function
with the probability of not getting ruined (the non-ruin probability), only
two types of claim distributions are known for which the probability of ruin can
easily be calculated. These are the exponential distributions and sums, mixtures
and combinations of these distributions, as well as the distributions with only a
finite number of values. For other distributions, however, an elegant and usually
sufficiently tight upper bound can be found. The real number R
in this expression is called the adjustment coefficient. This so-called Lundberg
upper bound can often be used instead of the actual ruin probability: the higher
R, the lower the upper bound for the ruin probability and, hence, the safer the
situation. The adjustment coefficient R can be calculated by solving an equation
which contains the mgf of the claims, their expectation and the ratio of premium
and expected claims.
Multiplying both the premium rate and the expected claim frequency by the
same factor does not change the probability of eventual ruin: it doesn’t matter
if we make the clock run faster. There have been attempts to replace the ruin
probability by a more ‘realistic’ quantity, for instance the finite ruin probability,
which is the probability of ruin before time But this quantity behaves somewhat
less orderly and introduces an extra problem, namely the choice of the length of
the time interval. Another alternative arises if we consider the capital in discrete
time points 0, 1, 2, … only, for instance at the time of the closing of the books.
For this discrete time model, we will derive some results.
First, we will discuss the Poisson process as a model to describe the development
in time of the number of claims. A characteristic feature of the Poisson process is
that it is memoryless: the occurrence of a claim in the next second is independent
of the history of the process. The advantage of a process being memoryless is the
mathematical simplicity; the disadvantage is that it is often not realistic. The total
of the claims paid in a Poisson process constitutes a compound Poisson process.
In the second part of this chapter, we will derive the mgf of the non-ruin prob-
ability by studying the maximal aggregate loss, which represents the maximum
 THE RISK PROCESS 83

difference between the earned premiums and the total payments up to any mo-
ment. Using this mgf, we will determine the value of the ruin probability in case
the claims are distributed according to variants of the exponential distribution.
Next, we will consider some approximations for the ruin probability.

4.2 THE RISK PROCESS

A stochastic process consists of related random variables, indexed by the time

We define the surplus process or risk process as follows:

where

with

A typical realization of the risk process is depicted in Figure 4.1. The random
variables denote the time points at which a claim occurs. The slope of
the process is if there are no claims; if, however, for some then the
capital drops by which is the size of the claim. Since in Figure 4.1, at time
the total of the incurred claims is larger than the initial
capital plus the earned premium the remaining surplus is less than
0. This state of the process is called ruin and the point in time at which this occurs
for the first time is denoted by T. So,

if for all
The random variable T is defective, as the probability of is positive. The
probability that ruin ever occurs, i.e., the probability that T is finite, is called the
84 RUIN THEORY

ruin probability. It is written as follows:

Before we turn to the claim process i.e., the total claims up to time we
first look at the process of the number of claims up to We will assume that
is a so-called Poisson process:
Definition 4.2.1 (Poisson process)
The process is a Poisson process if for some intensity the increments
of the process have the following property:

for all and each history

As a result, a Poisson process has the following properties:
the increments are independent: if the intervals
are disjoint, then the increments are independent;
the increments are stationary: is Poisson distributed
for every value of

Next to this global definition of the claim number process, we can also consider
infinitesimal increments where the infinitesimal ‘number’
 EXPONENTIAL UPPER BOUND 85

again is positive, but smaller than any real number larger than 0. For the Poisson
process we have:

Actually, these equalities are not really quite equalities: they are only valid if we
ignore terms of order
A third way to define such a process is by considering the waiting times

Because Poisson processes are memoryless, these waiting times are independent
exponential random variables, and they are also independent of the history of
the process. This can be shown as follows: if the history H represents an arbitrary
realization of the process up to time with the property that then

If is a Poisson process, then is a compound Poisson process; for a fixed

the aggregate claims have a compound Poisson distribution with
parameter
Some more notation: the cdf and the moments of the individual claims are

The loading factor or safety loading is defined by hence

4.3 EXPONENTIAL UPPER BOUND

In this section we give a short and elegant proof of F. Lundberg’s exponential

upper bound. Later on, we will derive more accurate results. First we introduce
the adjustment coefficient.
86 RUIN THEORY

Definition 4.3.1 (Adjustment coefficient)

The adjustment coefficient R for claims with is the positive
solution of the following equation in

See also Figure 4.2.

In general, the adjustment coefficient equation (4.10) has one positive solution:
is strictly convex since
and, almost without exception, continuously. Note that for the
limit of R is 0, while for we see that R tends to the asymptote of
or to
Remark 4.3.2 (Equivalent equations for the adjustment coefficient)
The adjustment coefficient can also be found as the positive solution of any of the
following equivalent equations, see Exercise 4.3.1:

where S denotes the total claims in an interval of length 1 and consequently – S

is the profit in that interval. Note that S is compound Poisson distributed with
 EXPONENTIAL UPPER BOUND 87

parameter and hence From the last equation we

see that the adjustment coefficient R corresponds to the risk aversion in case of
an exponential utility function which leads to an annual premium see (1.20). In
the second equation of (4.11), which can be proven by partial integration, R = 0
is no longer a root. The other equations still admit the solution R = 0.

Example 4.3.3 (Adjustment coefficient for an exponential distribution)

Assume that X is exponentially distributed with parameter The corre-
sponding adjustment coefficient is the positive solution of

The solutions of this equation are the trivial solution and

This situation admits an explicit expression for the adjustment coefficient.

For most distributions, there is no explicit expression for the adjustment coeffi-
cient. To facilitate solving (4.10) by a spreadsheet or a computer program, one can
use the fact that see Exercise 4.3.2.

In the next theorem, we prove F. Lundberg’s famous exponential inequality for the
ruin probability. Surprisingly, the proof involves mathematical induction.

Theorem 4.3.4 (Lundberg’s exponential bound for the ruin probability)

For a compound Poisson risk process with an initial capital a premium per unit
of time claims with cdf and mgf and an adjustment coefficient R
that satisfies (4.10), we have the following inequality for the ruin probability:

Proof. Define and as the probability that

ruin occurs at or before the claim. Since the limit of for equals
for all it suffices to prove that for each For the
inequality holds, since if and if Assume
that the first claim occurs at time This event has a ‘probability’ Also
88 RUIN THEORY

assume it has a size which has a probability Then the capital at that
moment equals Integrating over and yields

Now assume that the induction hypothesis holds for i.e.,

for all real Then, (4.15) leads to

where the last equality follows from (4.11).

Remark 4.3.5 (Interpretation of the adjustment coefficient; martingale)

The adjustment coefficient R has the property that is constant in
In other words, is a martingale: it can be interpreted as the fortune of a
gambler who is involved in a sequence of fair games. This can be shown as follows:
since and ~ compound Poisson with parameter we
have, using again (4.11):

Note that if R is replaced by any other real number, the expression in square
brackets in (4.17) is unequal to 1, so in fact the adjustment coefficient is the unique
positive number R with the property that is a martingale.

4.4 RUIN PROBABILITY AND EXPONENTIAL CLAIMS

In this section we give an expression for the ruin probability which involves the
mgf of U (T), i.e., the capital at the moment of ruin, conditionally given the event
 RUIN PROBABILITY AND EXPONENTIAL CLAIMS 89

that ruin occurs in a finite time period. This expression enables us to give an exact
expression for the ruin probability in case of an exponential distribution.
Theorem 4.4.1 (Ruin probability)
The ruin probability for satisfies

Proof. Let R > 0 and Then,

From Remark 4.3.5, we know that the left-hand side equals For the first
conditional expectation in (4.19) we take and write, using
see also (4.17):

The total claims between and has again a compound Poisson

distribution. What happens after is independent of what happened before
so and are independent. The term in curly brackets equals 1.
Equality (4.20) holds for all so
also holds.
Since for it suffices to show that the last
term in (4.19) vanishes for For that purpose, we split the event
according to the size of More precisely, we consider the cases
and for some function Notice that implies that we are
not in ruin at time i.e., so We have
90 RUIN THEORY

The second term vanishes if For the first term, note that
has an expected value and a variance
Because of Chebyshev’s inequality, it suffices to choose the function such
that We can for instance take

Corollary 4.4.2 (Some consequences of Theorem 4.4.1)

1. If then the chord in Figure 4.2 tends to a tangent line and, because of
Theorem 4.4.1, if then the ruin probability equals 1, see
Exercise 4.4.1.

2. If then Hence, the denominator in (4.18) is larger than

or equal to 1, so this is yet another proof of Theorem 4.3.4.

3. If the claims cannot be larger than then from which we

can deduce an exponential lower bound for the ruin probability:

4. It is quite plausible that the denominator of (4.18) has a finite limit for
say Then, of course, This yields the following asymptotic
approximation for for large we have

5. If R > 0, then for all As a consequence, if

for some then R = 0 and for all

Example 4.4.3 (Expression for the ruin probability, exponential claims)

From (4.18), we can derive an exact expression for the ruin probability if the claims
have an exponential distribution. For this purpose, assume that ruin occurs at a
finite time and that the capital U (T – 0) just before ruin equals Then, for
each value of and if H represents an arbitrary history of the process with
and we have:

Apparently, the deficit –U (T) at ruin also has an exponential distribution, so

the denominator of (4.18) equals ). With and
see (4.13), and thus we have the following exact expression
 DISCRETE TIME MODEL 91

for the ruin probability in case of exponential claims:

Notice that Lundberg’s exponential upper bound boils down to an equality here
except for the constant In this case, the denominator of (4.18) does not
depend on In general, however, it will depend on

4.5 DISCRETE TIME MODEL

In the discrete time model, we consider more general risk processes than the
compound Poisson process from the previous sections, but now only on the time
points 0,1,2,… Instead of we write Let denote the
profit between the time points and therefore

Later on, we will discuss what happens if we assume that is a compound

Poisson process, but for the moment we only assume that the profits
are independent and identically distributed, with but
We define a discrete time version of the ruin time the ruin probability
and the adjustment coefficient as follows:

The last equation has a unique solution. This can be seen as follows: since E [G] > 0
and Pr[G < 0] > 0, we have and for while
so is a convex function.
Example 4.5.1 (Compound Poisson distributed annual claims)
In the special case that is a compound Poisson process, we have
where denotes the compound Poisson distributed total claims in year From
(4.11), we know that R satisfies the equation Hence,
Example 4.5.2 (Normally distributed annual claims)
If with then follows from:
92 RUIN THEORY

Combining this result with the previous example, we observe the following. If we
consider a compound Poisson process with a large Poisson parameter, i.e., with
many claims between the time points 0 , 1 , 2 , . . . , then will approximately follow
a normal distribution. Consequently, the adjustment coefficients will be close to
each other, so On the other hand, if we take
and in Exercise 4.3.2, then it turns out that is an upper bound for
R.
Analogously to Theorem 4.4.1, one can prove the following equality:

So in the discrete time model one can give an exponential upper bound for the ruin
probability, too, which is

4.6 REINSURANCE AND RUIN PROBABILITIES

In the economic environment we postulated, reinsurance contracts should be com-

pared by their expected utility. In practice, however, this method is not applicable.
As an alternative, one could compare the ruin probabilities after a reinsurance
policy. This too is quite difficult. Therefore we will concentrate on the adjustment
coefficient and try to obtain a more favorable one by reinsurance. It is exactly from
this possibility that the adjustment coefficient takes its name.
In reinsurance we transfer some of our expected profit to the reinsurer, in
exchange for more stability in our position. These two conflicting criteria cannot
be optimized at the same time. A similar problem arises in statistics where one
finds a trade-off between the power and the size of a test. In our situation, we can
follow the same procedure as it is used in statistics, i.e., maximizing one criterion
while restricting the other. We could, for instance, maximize the expected profit
subject to the condition that the adjustment coefficient R is larger than some
We will consider two situations. First, we use the discrete time ruin model,
take out a reinsurance policy on the total claims in one year and then examine the
discrete adjustment coefficient In the continuous time model, we compare R
for two types of reinsurance, namely proportional reinsurance and excess of loss
reinsurance, with a retention for each claim.
 REINSURANCE AND RUIN PROBABILITIES 93

Example 4.6.1 (Discretisized compound Poisson claim process)

Again consider the compound Poisson distribution with and
from Example 3.5.4. What is the discrete adjustment coefficient for the total
claims S in one year, if the loading factor equals 0.2, i.e., the annual premium
equals 1.8?
The adjustment coefficient is calculated as follows:

Now assume that we take out a stop-loss reinsurance with For a reinsurance
with payment Y, the reinsurer asks a premium with If
the reinsurance premium amounts to
To determine the adjustment coefficient, we calculate the distribution of the profit
in one year which consists of the premium income minus the reinsurance
premium minus the retained loss. Hence,

The corresponding discrete adjustment coefficient which is the solution of

(4.25), is approximately 0.199.
Because of the reinsurance, our expected annual profit is reduced. It is equal to
our original expected profit minus the one of the reinsurer. For instance, for
it equals In the following table, we show the results
for different values of the retention

We see that the decision is not rational: it is dominated by

as well as i.e., no reinsurance, since they all yield both a higher expected
profit and more stability in the sense of a larger adjustment coefficient.
Example 4.6.2 (Reinsurance, individual claims)
Reinsurance may also affect each individual claim, instead of only the total claims
in a period. Assume that the reinsurer pays an amount if the claim amount
94 RUIN THEORY

is In other words, the retained loss equals We consider two special

cases:

Obviously, proportional reinsurance can be considered as a reinsurance on the total

claims just as well. We will examine the usual adjustment coefficient which
is the root of

where denotes the reinsurance premium. The reinsurer uses a loading factor
on the net premium. Assume that and for and
Furthermore, let and consider two values and
In case of proportional reinsurance the premium equals

so, because of (4.32) leads to the equation

For we have and we have

Next, we consider the excess of loss reinsurance with
The reinsurance premium equals

while and therefore is the root of

 BEEKMAN'S CONVOLUTION FORMULA 95

In the table below, we give the results for different values of compared
with the same results in case of proportional reinsurance with the same expected
payment by the reinsurer:
For the loading factors of the reinsurer and the insurer are equal,
and the more reinsurance we take, the larger the adjustment coefficient is. If
the reinsurer’s loading factor equals then for the expected retained loss
is not less than the retained premium
Consequently, the resulting retained loading factor is not positive, and eventual ruin
is a certainty. The same phenomenon occurs in case of excess of loss reinsurance
with In the table below, this situation is denoted by the symbol *.

From the table we see that all adjustment coefficients for excess of loss coverage
(XL) are at least as large as those for proportional reinsurance (Prop.) with the
same expected payment. This is not a coincidence: by using the theory on ordering
of risks, it can be shown that XL coverage always yields the best R-value as well
as the smallest ruin probability among all reinsurance contracts with the same
expected value of the payment, see Example 10.4.4.

4.7 BEEKMAN’S CONVOLUTION FORMULA

In this section we show that the non-ruin probability can be written as a com-
pound geometric distribution function. For this purpose, we consider the maximal
aggregate loss, i.e., the maximal difference between the payments and the earned
premium up to time

Since S(0) = 0, we have The event occurs if, and only if, a finite
point in time exists for which In other words, the inequalities
96 RUIN THEORY

and are equivalent and consequently

Next, we consider the points where the surplus process reaches a new record low.
This happens necessarily at points in time when a claim is paid. Let the random
variables denote the amounts by which the record low is
less than the – 1st one, see Figure 4.3 where there are three new record lows,
assuming that the process drifts away to in the time period not shown. Let M
be the random number of new records. We have

From the fact that a Poisson process is memoryless, it follows that the probability
that a particular record low is the last one is the same every time. Hence, M follows
a geometric distribution. For the same reason, the amounts of the improvements
are independent and identically distributed. The parameter of M, i.e.,
the probability that the previous record is the last one, equals the probability to
avoid ruin starting with initial capital 0, hence it equals
So L has a compound geometric distribution. Both the value of the geometric
parameter and the distribution of conditionally given follow
from the following theorem:
 BEEKMAN'S CONVOLUTION FORMULA 97

Theorem 4.7.1 (Distribution of the capital at time of ruin)

If the initial capital equals 0, then for all we have:

Proof. In a compound Poisson process, the probability of having a claim in the

interval equals which is independent of and of the history of
the process up to that time. So, between 0 and there is either no claim (with
probability ), and the capital increases from to or one claim with
size X. In the latter case, there are two possibilities. If the claim size is less than
then the process continues with capital Otherwise ruin occurs, but
the capital at ruin is only larger than if Defining

we can write

If denotes the partial derivative of G with respect to then

Substitute (4.43) into (4.42), subtract from both sides and divide by
Then we get

Integrating this over yields

98 RUIN THEORY

The double integrals in (4.45) can be reduced to single integrals as follows. For
the first double integral, exchange the order of integration, substitute
and again exchange the integration order. This leads to

In the second double integral in (4.45), we substitute Then,

Hence,

For the first term on both sides of (4.48) vanishes, leaving

which completes the proof.

This theorem has many important consequences.

Corollary 4.7.2 (Consequences of Theorem 4.7.1)

1. The ruin probability at 0 depends on the safety loading only. Integrating

(4.40) for yields so regardless of we have

2. Assuming that there is at least one new record low, has the same distri-
bution as the amount with which ruin occurs starting from (if ruin
 BEEKMAN’S CONVOLUTION FORMULA 99

occurs). So we have the following expression for the density function of the
record improvements:

3. Let H denote the cdf of and the parameter of M. Then, since L

has a compound geometric distribution, the non-ruin probability of a risk
process is given by Beekman’s convolution formula:

where

and

4. The mgf of the maximal aggregate loss L which because of (4.38) is also
the mgf of the non-ruin probability is given by

Proof. Only the last assertion requires a proof. Since with

M ~ for we have

The mgf of follows from its density (4.51):

100 RUIN THEORY

since the integrated term disappears at because for

Substituting (4.56) into (4.55) then yields (4.54).

Remark 4.7.3 (Recursive formula for ruin probabilities)
The ruin probability in can be expressed in the ruin probabilities at smaller initial
capitals, as follows:

To prove this, note that implies that the surplus eventually will drop below
the initial level, so

where we have substituted

4.8 EXPLICIT EXPRESSIONS FOR RUIN PROBABILITIES

Two situations exist for which we can give expressions for the ruin probabilities.
In case of exponential distributions, and mixtures or combinations of these, an
analytical expression arises. For discrete distributions, we can derive an algorithm.
In the previous section, we derived the mgf with the non-ruin probability
In some cases, it is possible to identify this mgf, and thus give an
expression for the ruin probability. We will describe how this works for mix-
tures and combinations of two exponential distributions, see Section 3.7. Since
and
 EXPLICIT EXPRESSIONS FOR RUIN PROBABILITIES 101

it follows from (4.54) that the ‘mgf’ of the function equals

Note that, except for a constant, is a density function, see Exercise 4.8.1.
Now, if X is a combination or a mixture of two exponential distributions as in
(3.27), i.e., for some and it has density function

then the right-hand side of (4.61), after multiplying both the numerator and the
denominator by can be written as the ratio of two polynomials
in By using partial fractions, this can be written as a sum of terms of the form
corresponding to times an exponential distribution. We give two
examples to clarify this method.

Example 4.8.1 (Ruin probability for exponential distributions)

In (4.62), let and hence the claims distribution is exponential(1).
Then, for and the right-hand side of (4.61) leads to

Except for the constant this is the mgf of an exponential distribution. We con-
clude from (4.61) that – is equal to the density function of this distribution.
By using the boundary condition we see that for the exponential(1)
distribution

which corresponds to (4.23) in Section 4.4 for

Example 4.8.2 (Ruin probability, mixtures of exponential distributions)

Choose and Then
102 RUIN THEORY

So, after some calculations, the right-hand side of (4.61) leads to

for

The ruin probability for this situation is given by

Notice that indeed holds.

This method works fine for combinations of exponential distributions, too, and
also for the limiting case gamma see Exercises 4.8.5–7. It is possible to
generalize the method to mixtures/combinations of more than two exponential
distributions, but then roots of polynomials of order three and higher have to be
determined.
To find the coefficients in the exponents of expressions like (4.67) for the ruin
probability, i.e., the asymptotes of (4.66), we need the roots of the denominator
of the right-hand side of (4.61). Assume that, in the density (4.62), and
We have to solve the following equation:

Notice that the right-hand side of this equation corresponds to the mgf of the
claims only if is to the left of the asymptotes, i.e., if If is larger, then
this mgf is hence we write instead of for these branches in
Figure 4.4. From this figure, one sees immediately that the positive roots and
are real numbers that satisfy

Remark 4.8.3 (Ruin probability for discrete distributions)

If the claims X can have only a finite number of positive values
with probabilities the ruin probability equals
 APPROXIMATION OF RUIN PROBABILITIES 103

where The summation extends over all values

of leading to and hence is finite. For a proof of
(4.70), see Gerber (1989).

4.9 APPROXIMATION OF RUIN PROBABILITIES

For other distributions than the ones above, it is difficult to calculate the exact
value of the ruin probability Furthermore, one may argue that this exact
value is not very important, since in case of doubt, other factors will be decisive.
So there is a need for good and simple approximations for the ruin probability.
First of all, we give some global properties of the ruin probability that should
preferably be satisfied by the approximations. Equation (4.50) yields

Next, we know that and thus, with partial integration,

These moments of the maximal aggregate loss L follow easily since

has a compound geometric distribution, with the distribution of M and
104 RUIN THEORY

given in Section 4.7. The required moments of are

Since we have

It can also be shown that

hence

After this groundwork, we are ready to introduce a number of possible approxi-

mations.

1. Replacing the claims distribution by an exponential distribution with the

same expected value, we get, see (4.23):

For the approximation is correct, but in general, the integrals over

the left-hand side and the right-hand side are different.

2. Approximating by with chosen such that (4.74) holds

yields as an approximation

Note that if the claims are exponential distributed, then so

not only (4.77) but also (4.78) gives the correct ruin probability.
 APPROXIMATION OF RUIN PROBABILITIES 105

3. We can approximate the ruin probability by a gamma distribution:

To fit the first two moments, the parameters and of the gamma cdf
must meet the following conditions:

4. Just as in the first approximation, one can replace the claims distribution
by another with a few moments in common, for which the corresponding
ruin probability can be easily calculated. A suitable candidate for such a
replacement is a mixture or combination of exponential distributions.
5. Another possible replacement is a discrete distribution. The ruin probabili-
ties can easily be computed from (4.70). For each claims distribution, one
can find a two-point distribution with the same first three moments. This
is not always possible in case of a mixture/combination of two exponential
distributions. Both methods yield good approximations.
6. From the theory of ordering of risks, it follows that one gets a lower bound for
the ruin probability if one replaces the claims distribution with expectation
by a one-point distribution on A simple upper bound can be obtained
if one knows the maximum value of the claims. If one takes a claims
distribution with probability for and probability for 0, then a
Poisson process arises which is equivalent to a Poisson process with claims
always equal to and claim number parameter instead of So, both
the lower bound and the upper bound can be calculated by using (4.70) with

7. The geometric distribution allows the use of Panjer’s recursion, provided

the individual terms are integer-valued. This is not the case for the terms
of L, see (4.51). But we can easily derive lower and upper bounds this
way, by simply rounding the down to an integer multiple of to get a
random variable which is suitable for Panjer’s recursion, and gives an
upper bound for since Rounding up leads
to a lower bound for By taking small, we get quite good upper and
lower bounds with little computational effort.
106 RUIN THEORY

4.10 EXERCISES

Section 4.2

1. Assume that the waiting times are independent and identically distributed random
variables with cdf and density function Given and for
some what is the conditional probability of a claim occurring between points in time
and (This generalization of a Poisson process is called a renewal process.)
2. Let be a Poisson process with parameter and let and
Show that and interpret these
formulas by comparing with

Section 4.3

1. Prove that the expressions in (4.11) are indeed equivalent to (4.10).

2. Use for and to prove that
3. For and determine the values of for which is
finite, and also determine R.
4. If the claims distribution is discrete with then determine if it is given that
R = log 3.
5. Which premium yields
6. If then determine R by using a spreadsheet, for and
7. Assume that the claims X in a ruin process with arise as follows: first, a value Y is drawn
from two possible values 3 and 7, each with probability Next, conditionally on the
claim X is drawn from an exponential distribution. Determine the adjustment coefficient R.
If R = 2 for the same distribution, is larger or smaller than
8. In some ruin process, the individual claims have a gamma(2,1) distribution. Determine the
loading factor as a function of the adjustment coefficient R. Also, determine If the
adjustment coefficient equals does hold? Using a sketch of the graph of the mgf of the
claims, discuss the behavior of R as a function of
9. Discuss the determination of the adjustment coefficient R if in a ruin process the claims are
lognormally distributed. Also, if the claims are inverse Gaussian.
10. Argue that Use the relation where S denotes the total claim
in some period of length 1, to derive that an exponential premium increases with the parameter
(risk aversion)

Section 4.4

1. From Theorem 4.4.1, we know that if Why does this imply that if

2. Which compound Poisson processes have a ruin probability

 EXERCISES 107

3. For a compound Poisson process, it is known that the continuous ruin probability depends on
the initial capital in the following way: Determine the adjustment
coefficient for this process. Can anything be said about the Poisson parameter in this risk process?
What is
4. Assume that By looking at the event "non-ruin & no claim before with
denoting the premium income per unit of time, show that must hold.
5. For a certain risk process, it is given that and Which of the
numbers 0, 1 and 6 are roots of the adjustment coefficient equation
Which one is the real adjustment coefficient?
One of the four expressions below is the ruin probability for this process; determine which
expression is the correct one, and argue why the other expressions can’t be the ruin probability.

6. The ruin probability for some ruin process equals By using

the fact that for ruin processes, in general, for some
determine the adjustment coefficient R and the appropriate constant in this case.

Section 4.5

1. Assume that the distribution of satisfies and

Further, and is an integer. Determine if Express in terms of
and both and in terms of and
2. Assume that an insurer uses an exponential utility function with risk aversion Prove
that if and only if and interpret this result.
3. Show that with probability 1, as well as for all if both are determined
for a compound Poisson risk process.
4. Assume that the continuous infinite ruin probability for a compound Poisson process equals
in case of an initial capital for some constant Furthermore, the claims follow an
exponential distribution with parameter 2 and the expected number of claims a year is 50.
Determine the safety loading for this process. Also determine an upper bound for the discrete
infinite ruin probability.

Section 4.6

1. The claim process on some insurance portfolio is compound Poisson with and
The loading factor is Calculate the adjustment coefficient in case one takes out a
proportional reinsurance with a loading factor Calculate the relative loading
factor after this reinsurance. Which restrictions apply to
108 RUIN THEORY

2. For the same situation as in the previous exercise, but now with excess of loss coverage
write down the adjustment coefficient equation, and determine the loading factor after
reinsurance.
3. Assume that the claims per year are R(5,1) distributed and that A
reinsurer covers a fraction of each risk, applying a premium loading factor Give the
adjustment coefficient for the reinsured portfolio, as a function of Which value optimizes
the security of the insurer?
4. A total claims process is compound Poisson with and The relative
loading factor is One takes out a proportional reinsurance The relative
loading factor of the reinsurer equals 1. Determine the adjustment coefficient For which
values of is ruin not a certainty?

Section 4.7

1. What is the mgf of if the claims (a) are equal to with probability 1, and (b) have an
exponential distribution?
2. Prove that
3. In Exercises 4.4.3 and 4.4.6, what is

Section 4.8

1. For which constant is a density?

2. Make sketches like in Figure 4.4 to determine the asymptotes of (4.61), for a proper combination
of exponential distributions and for a gamma distribution.
3. Calculate and R if Which values of are
possible taking into account that decreases in and that the safety loading is positive?
4. If and then determine (4.61), and an
explicit expression for
5. Determine ] in the previous exercise, with the help of (4.17). Determine independent
random variables X, Y and I such that IX + (1 – I)Y has density
6. Just as Exercise 4.8.4, but now is a gamma density, and
7. Determine if and the claims are equal to with and
exponential( 1) and independent.
8. Sketch the density of in case of a discrete claims distribution.
9. Prove (4.70) in case of and
10. Assume that the individual claims in a ruin process are equal to the maximum of two independent
exponential(1) random variables, i.e., with exponential(l). Deter-
mine the cdf of and use this to prove that the corresponding density is a combination
of exponential distributions. Determine the loading factor in the cases that for the adjustment
coefficient, we have R = 0.5 and R = 2.5.
 EXERCISES 109

11. The ruin processes of company 1 and 2 are both compound Poisson with intensities
and claims distributions exponential(3) and exponential(6), and loading factors
and The claims process of company 1 is independent of the one of company 2.
These companies decide to merge, without changing their premiums. Determine the intensity,
claims distribution and loading factor of the ruin process for the merged company. Assume
that both company 1 and 2 have an initial capital equal to 0, then obviously so does the
merged company. Compare the probabilities of the following events (continuous infinite ruin
probabilities): “both companies never go bankrupt” with “the merged company never goes
bankrupt”. Argue that, regardless of the values of the initial capitals and for the separate
companies, and consequently of for the merged company, the following holds: the event
“both companies never go bankrupt” has a smaller probability than “the merged company never
goes bankrupt”.

Section 4.9

1. Verify (4.72), (4.73), (4.75)/(4.76), and (4.80). Solve and from (4.80).
2. Work out the details of the final approximation.