UNIT 6 STOCHASTIC MODELS IN

INSURANCE
Structure
6.0 Objectives
6.1 Introduction
6.2 Insurance Process
6.3 Survival Function
6.4 Life Insurance Contracts
6.4.1 Simple Life Insurance Contracts

6.5 Non-life Insurance and Ruin Probability

6.6 Stochastic Processes in Insurance
6.6.1 Generalisation of the Claim Number Process
6.6.1.1 Mixed Poisson Processes
6.6.1.2 Ordinary ~ e n e w i Processes
l

6.7 Let Us Sum Up

6.8 Key Words
6.9 Some Useful Books
6.10 Answer or Hints to Check Your Progress

6.0 OBJECTIVES
After going through this unit you will be able to:
understand the use of basic probability based models in insurance that take
into account the time factor; and
evaluate insurance claims and premiums using stochastic techniciues.

6.1 INTRODUCTION
Insurance operates from the platform of probability theory. Therefore, it is
necessary to understand the formulation of claim and premium processes in the
framework of probability analysis. Particularly, probability distributions of
these variables have to be studied.

6.2 INSURANCE PROCESS

In order to model an insurer's portfolio of risk over time we consider variables
such as claim number, claim size and premium rate. We will see below that the
.' nature of these variables differs. For example, in non-life insurance not only Stochastic Models
in lnsurance
the claims arrival times are random but also the claim severities are random.
Needless to point out that these are different in case of life insurance.
Moreover, the task of modelling risk involves various stochastic processes.
Consequently, we have at hand a difficult job.

To consider a portfolio for risk evaluation, we often take into account the
sequence of claim arrival times and the sequence of claim severities, the claim
arrival process and the claim size process. To develop the analytical
framework it is often required to assume that variables considered under a
portfolio have independent identical distribution (iid).

6.3 SURVIVAL FUNCTION - - - - -

Let us see the preliminary idea of survival function given in the Wakipedia.

Survival analysis in insurance is used to deal with death of a life. It attempts to

answer questions such as: what is the fraction of a population which will
survive past a certain time? If it survives, at what rate will it die? Can multiple
causes of death or failure be taken into account?

Take survival function as a property of any random variable that maps a set of
events associated with mortality onto time. Through it we get the probability
that the life will survive beyond a specified time. To see the functional form
consider the following (as given in the above source):

Let X be a continuous random variable describing life time. It has a

cumulative distribution F ( r ) on the interval [O,oo). Then the survival h c t i o n
is given as

The survival function R(t) is monotone decreasing, i.e.,

R ( u ) < R ( t ) for u > t

The time, t = 0 represents starting of the insurace. We will assume that

lim R ( t ) is zero.
1-m

Take a failure rate A ( t ) as the probability that a failure occurs in a specified

interval, given no failure before timet. It cah be defined with the aid of the
survival~nctionR ( t ) ,the probability of no failure before time t , as:

where tl (or t ) and t2 are respective@ the beginning and ending of a specified
interval-of time spanning At.
Actuarial Techniques I In the above equation as lim ~ ( t+ )O . Thus the failure rate becomes
A1 +o

infinitesimally small. Then we have the hazard function, which gives the
'

instantaneous failure rate at any point in time:

~ ( t ) ~- (+ t~ t )
h(t) = lim
A + ~t.~(t)
Continuous failure rate depends on a failure distribution, F(t), which is a
cumulative distribution function that describes the probability of failure prior
to time t,
~ ( t ~ t ) = ~ ( t ) = i - ~ t (2t0) ,
The failure distribution function is the integral of the failure density,f(x),

Then the hazard function can be defined as

A common failure distribution is the exponential failure distribution,

which is based on the exponential density function. Consequently, we get a

constant hazard rate.

6.4 LIFE INSURANCE CONTRACTS

In life insurance the company has to pay predetermined amount of insurance
claims. Such claims are the sums insured by policyholders. The amount of
claim is defined in the contract to cover the events such as death, survival,
disability etc. Claims could occur either in policy period or after the expiry
date of the insurance. Since claims are payment by the company it depends
premiums charged from them to meet the costs. The premiums are determined
through equivalent principle at the time of issuing the policy. Essentially, the
principle strikes the balance between the expected value of discounted future
benefits payable by the insurer plus the administrative costs with that of the
expected value of discounted future premiums payable by the policyholder.
The factors taken into account for premium calculation are interest rate,
probability distributions and cost shares. In the following discussion, we will
ignore the costs for simplicity of analysis.

We will use the following notations:

r interest rate,
1+ r interest factor,

v=- 1 discount factor,

l+r '
 Stochastic Models
n integer duration of years of a contract, in Insurance
-
L, the residual life time of a person aged x ,

L,, the number of completed years of a person aged x , i.e.,

L, = [ I , ] the integer part of I , .

An annuity is a sequence of payments of limited duration n (we will discuss

details in Unit 7). The present value of an annuity with n annual payments of
1 starting at time 0 is

Let us denote the lifetime of a newborn person as I , . It is a random variable

with distribution function F with range [ 0 , o ] , where o is the end of the
predetermined value assigned to a life. We assume F to be a smooth function,
that is, without jumps. In this set up the future lifetime of a person aged x is a
random variable I, with probability distribution function F,, where

We use the L, = [ I x ] for integer value of x so that L, is the number of

completed years lived by the life after the xlhbirthday. Then the one-year
survival probability of the life is given as

and a similar duration death probability is

Setting p, = 0 and q, = 1 we have

The values of q, can be found in mortality tables issued by statistical bureaus.

If one assumes that these values do not change over time, then the distribution
of L, can be calculated. Thus,

which denotes the probability that the person aged x will survive k years but
not k + l years, for O ~ k s o - x .
We will introduce a refinement to model and have &timations of cohort
dependent mortalities in Unit 8.
Actuarial Techniques 1 6.4.1 Simple Life Insurance Contracts
To understand the life insurance contracts, we include the basic tools in the
following discussion. The notations that will be used are:

zh = discounted benefits
Z" = discounted contributions.

Taking ,Zh and Z' as random variables, we get the equivalence principle:

E (z') = E (zh)
)+ costs.

We follow the simplification norm as mentioned above and ignore the costs.
Moreover, we assume that all contracts start at the xth birthday of the insured.

Usually, benefits in life insurance contracts consist of at most one payment of

the sum insured. To see the underlying formulation we consider two basic
types of life insurance viz., (a) term insurance and (b) pure endowment.

Term Insurance

A term insurance of duration n consists of payment of 1 unit at the end of the

year of death, if it falls within the contract period. The sum insured is fixed at
the time of contract. However, the time crf payment L, is random. The net
present value (NPV) of the contract is

The probability distribution of zbis given by

'(zb = d + ' ) = I P ' { ~ , = k } , r,~p X= . q , + ,
for k I n - 1. Otherwise, the value pf zh is zero.
Therefore, the expected netpresent value (ENPV) is

The ENPV is called net single premium.

In the above equation when n m we get a "whole life insurance". In that

case the net single premium A, is given by

When benefits c, vary from year to year in a life insurance, we have

28
 Stochastic Models
and the ENPV equals to in Insurance

Pure Endowment

Let us consider a life insurance of duration n that consists of payment of 1 unit

only if the insured is alive at the end the nih year. Then

We call it pure endowment. Its ENPV denoted by Ex is given as

Endowments

Consider a payment of I'unit at the end of the year of death, if this occurs
within the first n years, otherwise at the end of the nib year. We call it
endowment life insurance. Under this form the discounted benefit is given as

This has the features of a term insurance and a pure endowment. As a result we
can derive it by adding ENPV of these two forms.

Premiums can be paid periodically leading to as a series of payments which

have to be paid while the insured person is alive. For such payments we have
to include the duration and the frequency of premiums and the additional
amount to be paid. The regular premiums can by interpreted as life annuities
payable in advance. If there is yearly payment of 1, then the discounted ENPV
is given by
n-l

zc =~ v ~ l ( L s > k - l l '
k 00

Check Your Progress 1

1) Define the term survival hnction.

Actuarial Techniques I 2) Write the meaning of term life insurance.

3) How endowment life insurance is different from pure endowment?

6.5 NON-LIFE INSURANCE AND RUIN

PROBABILITY
Lundberg (1903) provided the mathematical foundation of non-life insurance.
He developed a model considering premiums and claims. Subsequently, he
gave another model by considering a homogenous Poisson process, which we
presenftbelow.

The following discussion is devdoped taking some select concepts given in

Embrechts et a1 (1999). We have taken the definitions as given there.

Definition: The stochastic counting process N = ( N (t))/ is a homogeneous

Poisson process with rate the intensity rate of A > 0 if
(a) N(0) = 0 ass.,
(b) N has stationary, independent increments, and
(c) f o r a l l O ~ s < t < m : N ( t ) - N ( s ) - P O I S ( A ( ~ - s ) )i.e.,
,

In insuranceapplications, N(t) is used for the number of claims in the interval

(O,t] in a well defined portfolio. If we denote the claim arrival of the nthclaim
by S,, then
 Stochastic Models
The inter-arrival times T,, T, = Sk - Sk-,, k = 2,3 ... are independent, in Insurance
identically exponentially distributed ( e x p ( A ) ) with finite mean EI; = 112.
The latter property also characterises the homogeneous Poisson process. The
main size process X k is at first a ~ u m e dto be iid with distribution
function (df) F (F(0) = 0) and finite mean p = EX, . The rv Xk denotes the
claim size occurring at time Sk. As a resuit, the total claim amount up to time t
is given by S ( t ) = z Nt (=' ~) x k. The latter rv is referred to as compound Poisson
rv. Its df can be written as

!
I
where df Fn*in (5.2) denotes the n -fold convolution of F and FO*is the
Dirac measure in 0 (see Unit 9 for meaning) .
C
In addition to the liability process ( ~ ( t ) )an~ insurance
~ ~ , company depends
on premiums in order to meet the losses. In the above standard the premium
process ( ~ ( t ) )is assumed to be linear (deterministic), i.e., P ( t ) = u + c t ,
where u 2 0 is the initial capital and c > 0. That is, we have a constant
premium rate chosen in such a way that the company (or portfolio) has a fair
chance of "survival". With this set up we present the CramCr-Lundberg model
in the following where random variable (rv) plays a crucial role. Let r be the
ruin time of the risk process and

i.e., r = inf { t 2 0 :u ( t ) < 0) (5.4)

We assume that inf # = a.The ruin probabilities in this formulation are

are definedas Y ( u , T ) = ~ (S Tr ) , T <a.

When Y (u,m) ,the infinite horizon ruin probability, we write Y ( u ) . With the
net-profit condition

lim,,, Y ( u ) = 0 . It is implied by (5.5) that on the average we obtain a I- lgher

premium income than a claim loss. The basic risk process (5.3) can now be
rewritten as

&here Apt = ES ( t ) and S = c 1 (Ap) - 1 > 0 is the safety loading which

guarantees "survival".
Actuarial Techniques 1 Definition. The stochastic process ( ~ ( t ) defined
) in (5.3) with the net-profit
condition (5.5) is called the Cramer-Lundberg risk process.

Let us denote R = 1- H for any df H concentrated on [0,a).Then

where p = A p l c < 1 and the integrated tail df FI is defined as

a

The function Y ( u ) in (5.6) also allows a compound df expression.

6.6 STOCHASTIC PROCESSES IN INSURANCE

Some Basic Results

The Craper-Lundberg model ~ ( t= )u + c t - S ( t ) , t 2 0; (see (5.3)) prompts .

us consider two related results, viz., Lundberg coefficient and CramCr-
Lundberg approximation which,we give below:

Lundberg Coefficient
Definiti~n.Let the claim df F allows for a constant R > 0 to exist for which
B ( R ) = 0 , then R is called the Lundberg coefficient (or Lundberg adjustment
coefficimt) of the risk process (U ( t )), .

Examples where R exists are the exponential and gamma distributions.

However, R does not exist for Pareto or lognormal distributions.

The CramCr-Lundberg Approximatiob

Theorem. Assume that the Lundberg coefficient R exists and that

Then

c-np
limy ( u ) e h =
u-car A~;(R)-C'

The moment condition (6.1) is satisfied in all examples where kuxists. The
asymptotic estimate (6.2) is called the Cram&-Lundberg approximation.

The imprtant assumption in the Cram&-Lundberg approximation is that the

exponential moments of the claim size distribution exist for some r > 0 . This
 Stochastic Models
means that the right tail of F decreases at least exponentially fast. However, in Insurance
analysis of insurance and financial data typically indicates the presence of
heavy tails.

6.6.1 Generalisation of the Claim Number Process

The classical risk process has been generalized to obtain more realistic
description of the insurance closer to reality. The homogeneous Poisson
process, which is a stationary implying that the size of the portfolio cannot
change (increase or decrease) has been extended to consider inhomogeneous
process. Moreover, since some of the insurance products such as fire and
automobile insurances, which require consideration for inclusion of risk
fluctuations in the modelling have been complied with. As already mentioned
in Section 6.5 the simplest way to take size fluctuations into account is to
consider in homogenous Poisson processes with intensity measure A ( t ). The
purpose of this section is mainly to discuss the choice of point processes
describing such risk fluctuations.

6.6.1.1 Mixed Poisson Processes

Definition

Let 8 be a homogeneous Poisson process with intensity 1 and A a random

variable with P ( A > 0) = 1, independent of fi . Then the process

is called a mixed Poisson Process. The random variable A is called structure

variable.

A mixed Poisson process has stationary increments. However, the independent

increments condition can no longer be retained. The stochastic variation of the
claim number intensity can be interpreted as random changes of the Poisson
parameter from its expected value A. To model this type of distribution of the
structure gamma distribution is used. The density function of the gamma
distribution is given by

-
The notation A r ( y , S ) indicates that the random variable A has a gamma
distribution with density hnction given in (61).

Definition

A mixed Poisson process N is called a negative binomial process or Polya

process if A - r ( y , 6)

We then have for a Polya process N

.%ctuorialTechniques I

i.e., N(t) has a negative binomial distribution. If we compare the total claim
amount up to time t for the Poisson model keeping means equal, then 'the
variance of the Polya model is found to be bigger than in the Poisson model.
This is called an over-dispersipn and is which often found in real insurance
data (see Embrechts et a1 (1999)).

6.6.1.2 Ordinary Renewal Processes

Definition

Let the occurrence of claims be described by N . Take T, to be the interarrival

times between two successive claims. A point process on R+is called a
renewal process if the variables (T),rl are independent and if 7;, T,, T,... have
the same df G . N is called ordinary renewal process if also has df G .

1
We call N a stationary renewal process if G has definite mean- and if Go of
A
N , satisfies Go( x ) = A [ C ( s )ds.

Ordinary Renewal Process

Take N to be an ordinary renewal process and assume that Tk has finite mean
1/ A . In this formulation N is not stationary, and E N ( t ) # At unless Tk has an
exponential distribution. We consider the associated random walk

where Y, = -cT, + X,. We assume that EY, = -c 1 A + u < 0 , implying that the
rand'om walk S,, drifts to -a. The safety loading 9 is defined as
9 = c 1 ( A p ) - 1. Since ruin can occur only at renewal epochs, we have that
 Stochastic Models
Cheek Your Progress 2
1-1 in Insurance

I) Why Lund berg model considered Poisson process?

........................................................................................
2) How do you define ruin probabilities?

........................................................................................
3) Explain Lundberg coefficient.

4) Why do you use renewat process in insurance modelling?

Actuarial Techniques I LET US SUM UP
In this unit we have discussed the modelling techniques of insurance in the
stochastic framework. While claim arrival time and claim size are pointed out
to be random, the premium determination is seen as in the context of
equivalent principle which states that the expected value of discounted future
benefits payable by the insurer is equal to the expected value of discounted
future premiums payable by the policyholder. Survival function has been
defined to get the probability that the life will survive beyond a specified time
and expected net present value of (or net single premium) in life insurace
contracts of term insurance and pure endowment have been studied. The
classical risk model formulated by Lundberg for non-life insurance considering
the homogenous Poisson process have been examined and ruin probabilities
have been derived. The classical risk model has been seen by relaxing the
assumption of homogenous statonary process.

6.8 KEY WORDS

Compound Poisson Process: A stochastic process with rate h > 0 and jump
size distribution G in a continuous-time. If ~ ( t:t )2 0 then
N(t)

Y ( t )= D, where, N ( t ) : t 2 0 is a Poisson process with rate A , and

I

{D, : i 2 1} are independent and identically distributed random variables, with

distribution function G, which are also independent of N ( t ) :t 2 0 .

Convolution: A mathematical operator which takes two functions (f and g)

and produces a third function that in a sense represents the amount of overlap
between f and a reversed and translated version of g. A convolution is a kind of
very general moving average.

Dirac Measure: A measure 6, on a set X (with any a - algebra of subsets of X )

that gives the singleton set {x) the measure 1, for a chosen element
x€x:sx({x})=1.

Endowments: An insurance contract that requires payment of 1 unit at the end

of the year of death, if this occurs within the first n years, otherwise at the end
of'the nth year.

Equivalence Principle: A premium fixation principle in insurance where the

expected value of discounted future benefits payablz by the insurer is equal to
the expected value of discounted future premiums payable by the policyholder.

Erlang Distribution: A probability distribution describing waiting times in

queuina systems.

Erlang Unit: A unit to measure telecommunications (or other) traffic.

Erlang Unit: a unit to measure telecommunications (or other) traffic

 Stochastic Models
Law of Large Numbers: In repeated, independent trials with the same in Insurance
probability of success in each trial, the percentage of successes is increasingly
likely to be close to the chance of success as the number of trials increases.

Markov Process: A stochastic process where the h r e expected value of a

value (such as an asset price) is dependent only on the current value.

Net Present Value: Present value of cash flow minus initial investment.
Pure Endowment: An insurance contract of duration nthat consists of
payment of 1 unit only if the insured is alive at the end the nth year.

Renewal Process : A generalisation of the Poisson prcess. The Poisson

process is a continuos-time Makov process on the positive integers (usually
starting at zero) which has iid holding times at each integer i exponentially
distributed before advancing (with probability 1) to the next integer:i + 1. We
may define a renewal process to be similar except that the holding times take
on a more general distribution.

Renewal Theory: A theory used for the purpose of comparing the long term
benefits of different insurance policies.

Survival Function: A relation between mortality and time. Through it we get

the probability that the life will survive beyond a specified time.

Term Insurance: An insurance contract of duration n that consists of payment

of 1 unit at the end of the year of death, if it falls within the contract period.

6.9 SOME USEFUL BOOKS

Butcher. M.V., Nesbitt. C.J. (197 I), Mathematics of Compound Interest.
CTlrich's Books. Ann. Arbor. Michigan
Embrechts, Paul, RiidigerFrey, Hansjorg Furrer(1999), Stochastic Process in
Insurance and Finance, ETH Ziirich, Switzerland (see internet) .
Gerbar H. U. (1997), Life Insurance, Third Edition. Springer

t
Salih N. Neftci.(1996), An Introduction to the Mathematics of Financial
Derivatives, Academic Press. San Diego-California, USA

h
6.10 ANSWER OR HINTS TO CHECK YOUR
mOGRESS
Check Your Progress 1
1) See Section 6.3 and answer.
2) See Sub-section 6.4.1 and answer.
3) See Sub-section 6.4.1 and answer.
Actuarial Techniques I Check Your Progress 2
1) Read Section 6.5 and answer.
2) Read Section 6.5 and answer.
3) Read Sub-section 6.6 and answer.
4) Read Sub-section 6.6.1.2 and answer.