Faculty and Institute of Actuaries

Claims Reserving Manual v.1

(09/1997)
Section 2

Section 2
INTRODUCTION TO RESERVING

[2A]
DESCRIPTION OF CLAIM PROCESS

The insurance background and the influences that need to be taken into account in
the reserving process are described in some detail in later sections. In this
Introduction we shall give a brief overview of what the claims process in
insurance entails, and how the problem of reserving arises.
We may start by supposing that we have a "risk" situation, associated with an
insurance "cover". The essential features are that there is a person or corporate
body, whose financial condition is directly affected by the occurrence of certain
"events" occurring over a defined period of time. An obvious example would be
an individual with Private Car Damage insurance who would suffer repair costs
following an accident causing damage to their vehicle.
If an insurance cover exists, an event occurring under the cover will give rise
to an insured loss, which subsequently becomes a claim on the insurer.
Typically there will be some delay between the event giving rise to the claim,
and the ultimate settlement of the claim with the insured. In the case of a Motor
claim, this delay from event to settlement may stretch from a number of days to
several years, depending upon factors such as complexity or severity of the claim.
Other significant dates are involved in this process. Following the occurrence
of the event, a significant date would be the date upon which that event becomes
known to the insurer. Whilst notification would normally occur quite soon after
the event, there can be circumstances where a considerable period of time may
elapse between the occurrence of an event and the notification of a claim to an
insurer (for example, when a ship is damaged in harbour, but the damage
becomes evident only when she is dry-docked at some later date).

Claim Reserves
The delay between event and settlement dates means that the insurer must set up
"reserves" in respect of those claims still to be settled. The reserves required at
any time are the resources needed to meet the costs, as they arise, of all claims not
finally settled at that time. The insurer must be able to quantify this liability if it is
to assess its financial position correctly, both for statutory and for internal
purposes.
Throughout this Manual we are dealing with claims reserves in respect of
events which have already occurred. This is distinct from future claims, arising

09/97 2A.1
 INTRODUCTON TO RESERVING

from risks covered by the insurer over the remaining period of the policy, where
the insured event has not yet occurred. Such liabilities are covered through the
mechanism of an Unearned Premium or Unexpired Risk reserve, which are
outside the scope of this Manual.
We are concerned then with reserves in respect of claims which have
occurred as at a particular date — which we shall call the "valuation date". We
can distinguish two categories of such claims:

- claims for which the event has occurred, and which are already known and
reported to the insurer;

- claims for which the event has occurred, but which have not yet been
reported to the insurer.

Reserves relating to the former category are normally referred to as "Known (or
Reported) Claims" reserves; the latter as "Incurred But Not Reported (IBNR)"
reserves.
This definition of IBNR applies usually in relation to Direct Insurance: in the
context of the London Market, the usage of "IBNR" will normally include, in
addition to reserves for unreported claims, allowance for any future deterioration
in amounts outstanding on claims already reported. The latter element is normally
called "IBNER" (incurred but not enough reserved).
A number of variations on the type of claims reserve are needed because of
various features including:

- a claim may involve a number of partial payments separated over an

extended period of time, culminating in its final settlement;

- claims may be settled prematurely by an insurer, and then require to be

reopened for further payments or recoveries that subsequently come to light
(this may occur on more than one occasion for a particular claim);

- the insurer may share the liability for the claim with other insurers, either by
reinsurance or by co-insurance.

Thus the need may arise for reserves for re-opened claims, and for reserves net of
reinsurance or co-insurance.
From this description, it is clear that reserves represent an attempt, at a point
in time, to attribute a financial value to those payments still to be made in respect
of a set of incurred losses, as yet unsettled. This cannot therefore be quantified
with precision, but must be the subject of estimation. Varying assumptions about
future influences on the outcome of those losses will lead to greater or smaller
estimates in a given context, leading to the idea of strength of reserves. In
consequence, information will often be needed as to the likely adequacy of a
given estimate of reserves. This may in turn involve careful examination of the
methods by which estimates are reached, and the assumptions on which they are
based.

09/97 2A.2
 DESCRIPTION OF CLAIM PROCESS

The description of claim events given above is somewhat idealised. In many

instances, it will be possible to identify a finite event or occurrence which gave
rise to a claim in the way described. However, in others, the question of defining
the relevant event can be one of considerable difficulty and argument. We need
only consider the situation with claims for compensation for employment-related
disease, where the dates of the relevant event, or indeed its definition, can be
difficult to determine. The need for modification and extension of the picture
included here will be addressed in certain specific areas outlined below.

Need for Reserves

The need to calculate reserves arises in a number of different circumstances, of
which the following are examples:

- assessing the financial condition of an insurer, since movements in reserves

over a period are key to assessing its progress;

- pricing insurance business in the sense of estimating the future cost of claims
on risks yet to be taken on (by extrapolation of past paid and reserved claim
cost);

- assessing the solvency of an insurer, in terms of its ability to meet its

liabilities (requiring assessment of likely upper limits of outstanding claim
cost);

- putting a value on the net worth of an insurer, particularly for purposes of

sales or acquisitions;

- commutations and reinsurance to close: that is, putting a financial value on

the run-off of a portfolio of insurance business.

The strength that is appropriate for the reserves may vary from one of these
circumstances to another.

Reserving Methods
Until the early 1970s, the approach to reserves commonly related solely to the
area of reserves for known claims. The practice was for each claim to be
individually assessed by a claims official at an early stage of its existence, and
possibly at subsequent stages. These individual "case" estimates would then be
aggregated to form a total reserve for outstanding claims. With the passage of
time, and the increasing ability to subject the results of this process to statistical
scrutiny, it was found that other methods or approaches to reserving might be
more appropriate. Nevertheless, it is still the case that, in some areas of business,
particularly where the numbers of claims are relatively small, or where they are
particularly complex, case estimation is employed, possibly in conjunction with
other methods
However, in many instances the volume of claims is such that it would be
impractical or too expensive to assess each claim individually and, in such cases,

09/97 2A.3
 INTRODUCTON TO RESERVING

an alternative approach is required. Furthermore, in the case of IBNR reserves,

specific claim files are not available for examination. In such cases, an alternative
method of reserving is required.
The case estimator — claims official — requires access to details of policy
cover, claim event and subsequent history before making their assessment of each
claim. Similarly, in using statistical estimation methods, the reserver requires
access to relevant data relating to the group of claims that are required to be
reserved.
The more obvious of such data would normally relate to claim cost levels
and perhaps numbers of claims settled in the recent past — together with some
information such as the numbers of claims in the group unsettled at the present
time. It might however be that, in some instances, the claims reserving specialist
has available not only what might be called "hard fact" relating to actual past
settlement experience, but additional "softer" information relating to the claims
concerned. This may include more generalised market and economic information.
As will be discussed in Section 2B below, reserving methods would
normally involve the application of a series of assumptions to underlying data
from which a reserve would be evaluated. It is crucial, if use is to be made of soft
data, that this aspect is fully reflected in the methodology employed and reports
written. Indeed, in some methodologies, it might be felt that the nature of factual
data employed (for example, assuming ratios of claims paid and outstanding to
premiums for past claim years), is too questionable to lead to valid results. This is
when the "softer" data may be of some value, even if they relate to more general
market conditions.
In reviewing methods of reserving to be employed, an important initial point
of consideration will be the nature and quality of data upon which the method
will be based, together with the extent to which the use of a particular model is
likely to be valid for those data.
For this reason, it is crucial that the individual carrying out the reserving
exercise should be as familiar as possible with the underlying business concerned.
If that person is situated within an insurance operation, then they should maintain
close contact with underwriters and claims management. If not, then it is still
important to understand as closely as possible the origins of whatever data
sources are to be used.

09/97 2A.4
 [2B]
RESERVING METHODOLOGY — GENERAL

Introduction
This section gives a general overview of the methodology used in reserving. Most
of the comments that follow could equally well apply to any situation where one
is constructing a model, fitting it to past observations, and using it to infer results
about future statistics of interest.
Whatever the purpose of reserving, in essence it involves the following
steps:

1. Construct a model of the process, setting out the assumptions made.

2. Fit the model, using past observations.

3. Test the fit of the model and the assumptions, rejecting or adjusting it.

4. Use the model to make predictions about future statistics of interest.

By "model" we refer to an artificial creation whose function is to represent what

is important in the process under consideration, but to omit aspects not
considered relevant to the particular area of understanding. The design of an
appropriate model thus involves a process of selection from among many possible
models. Normally, as an aid to understanding, we would try to select as simple a
model as possible — indeed, the model can be regarded as a deliberate
simplification of the underlying process itself.
There is a very close analogy with the underlying scientific method, which
involves the successive refinement and replacement of models to improve their
validity and accuracy for the purpose in hand.
Unfortunately, in the real world of claims reserving, restricted availability of
data often places severe limitations on the models that can be applied in practice.
It is clearly inappropriate to use a model which cannot be supported by the data
available.
We are thus led to a situation where, in some practical instances, although a
particular model may be felt to have a number of disadvantages, the available
data and knowledge of the claims environment do not support any further
refinement of this model.
The alternative is to revert to an earlier stage of model development, and to
produce an alternative approach based on somewhat differing assumptions as to
the underlying nature of the claim process. This might be used to create an
alternative version of the original reserving exercise. This process may in turn be
repeated, each such development producing alternative estimates of reserves.

09/97 2B.1
 INTRODUCTON TO RESERVING

In practice, a consequence of the limited availability of information may be

that we have no alternative but to introduce a fifth step to those described above,
as follows:

5. Apply professional judgement and experience to choose a number.

In other contexts, particularly those where direct access to the underlying claims
data files can be made, it may be possible to develop apparently differing
approaches to a point of broadly similar estimate values. The more elaborate
models to which this development gives rise are among those contained in
Volume 2 of the Manual. Even here, however, judgement as to the choice of
model and the value to be chosen to represent the reserve is still largely at the
reserver's discretion.
Ultimately, our objective is to formalise, as far as possible, the use of
numeric or other available information, together with the judgement of the
reserver, through the use of appropriate models.
Section 2E discusses at some length the elements that make up
professionalism and judgement, the need for which cannot be emphasised too
strongly.
The following sections describe some of the general considerations that
affect the steps outlined above, and some of the more specific points that arise
when performing such a process in the context of reserving.

Constructing a Model
The type of model constructed will depend on the purpose of the exercise.
Estimates of future claims for pricing purposes may be required at a very detailed
level (by type of car or geographical location, for example), and to be produced
quarterly or monthly. Reserves for high level management information or
statutory purposes may be required only by broad classification or on an annual
basis.
Similarly, the model must reflect the data that are available and their
limitation. Some types of information may not be adequately recorded (details of
claims settled at no cost, for example). In other instances, past data may not be
available for a sufficient period of time historically, or be of too small a volume,
to form a credible basis for fitting the model. When data of an adequate quantity
or quality are not available, it may be possible to supplement the available
company-specific data by reference to industry-wide data. Section E13 refers to
some of the possible sources of these additional external data. This section also
sets out techniques for projecting data beyond the available experience, by using
regression or graphical techniques. Data availability and limitations are also
discussed further in Volume 1 Section B.
The foundations of any model are the assumptions that underlie it. Ideally,
any modelling process should start with a clear statement of the assumptions
being made. The model can then be fitted to the data, and the assumptions tested.
Some simple reserving methods do not set out explicitly the assumptions
underlying the method. This may make it hard or impossible to gauge the
appropriateness of that particular model.

09/97 2B.2
 RESERVING METHODOLOGY — GENERAL

There is no easy classification of the types of model that can be used for
reserving, but there are various distinctions that can be made. In the first instance,
one can make the distinction between models that do not appear to start from a
formal set of assumptions, and those which rigorously set out the foundations on
which the model is built.
Many of the methods in Volume 1 fall into the former category, and those in
Volume 2 the latter. To produce valid reserve estimates using methods that fall
into the first category, it is essential that the user should consider the implicit and
explicit assumptions being made, and endeavour to test their validity or
otherwise. Just because a method jumps straight to step 4, does not mean that
steps 1, 2 and 3 can be forgotten! The apparent simplicity of the arithmetic
involved in performing some of the simpler methods in Volume 1 should not
distract the user from the complexity of the underlying assumptions being made
when applying these techniques.
A second general distinction can be made in the approach taken in
constructing a model. Many of the simpler models do not, on the surface in any
event, start from a base point that considers the underlying process influencing
the future claims payments (cars crashing, houses burning down, people receiving
compensation for injury, and so on). Instead, at a crude level, they are simply
taking one aggregate set of values (the past claims experience), and making some
estimate from this as to how the progression of values will continue (the future
claims experience). Step 3 of the reserving process (testing and adjusting the
model), will, of course, bring information about the nature of the claims into the
process, as will step 5.
By contrast with this initially arithmetic approach, other models begin with a
recognition of the underlying nature of the risks, perhaps starting with
assumptions about the frequency and severity of claims, and modelling how these
may change over time.
A final distinction that can be made is between deterministic models and
stochastic models. The future claims payments predicted by a reserving model are
random events. We can never know with certainty what the future payments will
be. The best that any model can do is to produce an estimated value of those
payments. Deterministic models just make assumptions about the expected value
of future payments. Stochastic models also model the variation of the future
payments. By making assumptions about the random component of a model,
stochastic models allow the validity of the assumptions to be tested statistically.
They therefore provide estimates not just of the expected value of the future
payments, but also of the size of the possible variation about that expected value.
All the models in Volume 1 are deterministic. Several of the models in
Volume 2 are stochastic. The introduction to Volume 2 describes the general
nature of stochastic models, and sets out some of their strengths and weaknesses.

Fitting and Testing the Model

Fitting a model may very often be a simple arithmetic process. Testing a model is
not so straightforward. A model can be broken down into a number of
parameters. For example, the basic chain-ladder model for an n × n triangle of
data may be summarised as n different "levels" for each year of origin, and n
different development factors for the n years of claims development, making 2n

09/97 2B.3
 INTRODUCTON TO RESERVING

parameters in all. Other models may have more or fewer parameters for the same
sized triangle of data.
The more parameters a model has, the better it may appear to "fit" the
observed data, but the less stable it will become for predicting future values. As
an extreme example, if one had, say, twenty data points, one could fit a model
with twenty parameters that would "fit" the data perfectly. The model would,
however, be completely unstable. A small change in any of the data points may
result in a large change in the parameters fitted and the future values predicted.
There is always a compromise between having enough parameters to fit the data
adequately, but few enough to produce a model with a certain amount of stability
and predictive power.
The amount of data available for reserving is often very limited. It is
important therefore to appreciate the limitations of an over-parameterised model.
This may be done, for example, by examining the effect on the model, and the
predicted results, of small changes in the data to which the model is being fitted.
We know the data are just one realisation of some random process, so a given
data point could reasonably be a little larger or smaller than in fact was the case.
For a model to be acceptable, ideally, the predicted future payments should not be
greatly affected by small changes in the observed data. The user should therefore
check carefully whether the results are heavily influenced by a few data points.
Testing the assumptions of a model can take many forms. Whilst there are a
variety of statistical techniques that can be used for this purpose, the approach in
practice may often be more pragmatic. This is particularly so for the methods in
Volume 1, where "soft" information may be used as a guide in choosing
development factors, for example. This is not easily amenable to explicit testing.
Many of the simpler, chain-ladder based, models are based on broad assumptions
about the stability of the types of business, and the speed of settlement of claims,
for example. The user of a reserving model should attempt to validate that these
assumptions apply to the particular class of business being considered. Ideally,
this should be done in a quantitative fashion but, as a practical compromise, it
may have to take the form of a qualitative assessment. For example, this may
involve a review of the types of policy written, or a visual examination of whether
settlement rates appear to have remained stable.
Throughout the text of Volume 1, the reader's attention is drawn to some of
the assumptions underlying the different methods. An awareness and questioning
of these and other implicit assumptions should be kept in mind whenever one
applies the methods in practice. Section B of Volume 2 discusses in a little more
detail the requirements of a "good" model, and some of the more general
techniques for testing models in general and stochastic models in particular.

09/97 2B.4
 [2C]
RESERVING IN CONTEXT

So far, reserving has been presented in terms of points of principle, removed from
the particular features of specific classes of general insurance business. As
reserving must in the end be a practical application, it is worth setting out the
main characteristics of some specific reserving situations, as they in turn will
affect the choice of appropriate reserving method. The examples given are not
intended to be exhaustive, but indicate the range of characteristics found in
general insurance.
Reference should be made to a suitable general insurance primer for basic
details of policy coverage found in different classes of business.

Private Car
The main classifications for analysing data are:

cover type
- third party
- comprehensive
peril
- fire
- theft
- windscreen
- third party bodily injury
- third party physical damage
- accidental damage

In general this business is amenable to statistical treatment, as there will usually

be a large number of similar policies, generating a large volume of claims.
Claims are mainly in respect of property damage, with a smaller number of
liability claims. Third Party bodily injury claims are subject to the greatest
uncertainty, since

- the size of claims can vary enormously, and for seemingly similar situations

- claims can remain outstanding for 7-10 years in some cases

- claims inflation is difficult to extrapolate, as judicial inflation in particular

tends to be fairly erratic in its development.

09/97 2C.1
 INTRODUCTON TO RESERVING

Creditor
The claims should be split between the main perils (accident, sickness and
unemployment), as the frequency and duration of claims may differ markedly,
and vary according to the economic cycle.
If sufficient information is available, it is possible to use an annuity-type
approach to reserving. Each outstanding claim is reserved according to the
number of months the claim has already been running.
Experience can differ markedly by scheme, reflecting different selling
practices and customer bases.

Mortgage Indemnity Guarantee

Following a period of severe market losses, the standard insurance cover changed
during 1993/4. The new cover required the insured (the mortgage lender) to retain
part of the loss on a coinsurance basis, together with the imposition of a
maximum indemnity per property. Whilst this will serve to spread the loss more
evenly between the insured and the insurer, the basic claim characteristics remain
unaltered, namely:

_ claims arise from a combination of repayment default and loss of property

value. Whilst there is a steady underlying level of default due to marital
breakdown and individual financial problems, the main underlying cause is
economic. Sudden sharp rises in interest rates, and increasing
unemployment, are two such influences;

- the cost of claims is heavily geared to house price deflation;

- the delays in claim settlements reflect the degree of forbearance of the lender
following the borrower's first falling into arrears, the attitude of the courts to
granting repossession orders and, finally, the length of time it takes to sell
the property in a depressed market.
New forms of cover continue to evolve, and there is currently considerable
discussion about the possibility of substituting some form of aggregate excess of
loss contract for the cover described above. Under this, the insurer would be able
to cap its liability, both in terms of the aggregate losses retained by the insured
before the contract comes into play, and in terms of the upper limit of aggregate
losses then payable by the insurer.
Because, under existing arrangements, cover normally extends over the
whole life of a loan, the main reserving problem concerns the Unexpired Risk
Reserve — i.e. the reserve needed if the cost of claims yet to arise on existing
loans seems likely to exceed the Unearned Premium Reserve. The determination
of reserves for outstanding or IBNR claims for this class of business is relatively
straightforward, compared to that of Unexpired Risk.

09/97 2C.2
 RESERVING IN CONTEXT

Catastrophes
Most catastrophes arise from natural causes (windstorms, earthquakes, floods,
etc.). They will usually give rise to a large number of related individual losses
occurring within a short period of time defining that event.
Other catastrophes may be due to man-made causes (explosions, air crashes,
etc.). These may give rise to a relatively small number of claims, but ones of
exceptionally high cost, both in terms of material damage and of liability.
If we confine ourselves to property insurance, the main thing to note is their
atypical features:

- reporting patterns are a function of the exact nature of the catastrophe, as the
speed with which the insured reports the claim will depend on the
seriousness of the loss, whether it was possible or practical to report losses
initially, and any emergency procedures set up by the insurer;

- settlement patterns are also likely to be different for each catastrophe,

particularly in the first few days, being dependent on the insurer's ability to
cope with the problem.

In view of the singular behaviour of each natural catastrophe, it is preferable to

try to project each such event separately. This may take the form of a curve-fitting
exercise. Alternatively, it may be possible to refer to the daily development
pattern of previous catastrophes, subject to the problems in the immediate
aftermath as the insurer gears up to tackling the problem. An assessment of the
insurer's aggregate exposure may be utilised in this initial period. In the extreme
case, it may be necessary to rely on any external market comment or assessment
of total insured loss available at that time.

Reinsurance and the London Market

The London Market specialises in writing those risks that are too large for the
smaller direct insurers to handle. A significant part of this business relates to
overseas risks. In view of the size of the risks involved, much of the business is
written on a co-insurance basis, spreading the risk amongst several insurers. The
London Market is also a major centre for the acceptance of reinsurance business,
again on an international basis.
A given piece of London Market business could therefore range from a very
specific property insurance on one risk, to an excess of loss reinsurance contract
covering an insurance company's entire world-wide general liability business. In
practice, the common features of both Reinsurance and London Market business
are:

- the data available to the insurer of this business are limited particularly for
retrocession business;

- numbers of claims and individual claims information will often not be

available, particularly for proportional business. Hence, any reserving
methods requiring such data are not applicable;

09/97 2C.3
 INTRODUCTON TO RESERVING

- classification of the business is difficult, and the description and nature of the
cover may need careful interpretation;

- sub-dividing the data into too small groups, to improve homogeneity, can
give rise to excessive volatility as the statistical significance of the
development model becomes increasingly suspect;

- the development of claims data is generally medium/long-tailed. Hence,

IBNR often forms a significant element of the reserve amounts;

- the length of tail also means that projections are usually based on incurred
claims data, rather than on paid claims, since the latter may require
development factors to ultimate which are too large and sensitive to be
reliable. This is particularly so in the early periods of development, when
little or no claim payment may have been made for some classes;

- differing underlying currencies and inflation rates may distort aggregated

data;

- There may be a "spiral" effect for catastrophe losses, due to the number of
retrocession contracts written by all the companies making up the market.
This means that such losses can cycle round the market, particularly for those
years where there is an active retrocession market;

- Latent claims are even more difficult to deal with, as different cedants may
use different triggers to determine reinsurance coverage. This will have a
knock-on effect on the reinsurer's own reinsurance coverage.

Latent Claims
Most of the methods examined in the Claims Reserving Manual are dependent on
having historic data which can be modelled statistically, in order to project the
total liabilities still outstanding. However, there are circumstances where this is
not possible, latent claims being a prime example. These are losses which may
lie dormant for several years after the originating event before manifesting
themselves. Industrial diseases and environmental pollution are typical sources
of such claims.
Even when the latent hazard becomes a reality and claims are being settled
and paid, as is now the case with asbestosis, standard projection methods are still
not applicable. The normal course of events for latent claims is that, when the
nature of the hazard becomes apparent, claims start to be reported at roughly the
same time, irrespective of the underwriting year to which they relate. Amounts
then increase rapidly.
This is completely different from the normal course of events, where regular
development by "development year" might be expected. In the case of latent
claims, however, large movements occur across each of the affected diagonals of
the data when displayed in the familiar triangle format. Conventional
"triangulation" methods of projection will not work in this situation.

09/97 2C.4
 RESERVING IN CONTEXT

Another problem with latent claims is that the historic data may not exist.
Even when they do exist, they give little indication of what will happen in the
future and, hence, how claims will develop. This is because the claims are
dependent on the results of court actions, and these are highly unpredictable. The
court actions determine whether coverage applies, if it is deemed to apply then
who is liable, when coverage is deemed to apply, and the amount of liability
(which may include considerable amounts for punitive damages). Also, expenses
are very high covering, inter alia, lawyers' contingency fees, and are often
incurred even when coverage is found not to apply.
Although the previous development of claims data may not be of much use
in forecasting future development, there are alternative approaches that might be
tried. These might include the following:

Share of total market: tracking losses

This involves initially determining what the total loss will be world-wide and the
sources of the loss.
For example, for pollution in the USA, this may be done by site, first for
known sites and secondly for not-yet-known sites. Assessment of the size of loss
by site and the probability that insurers will have to pay ("win-factors") have to
be determined, followed by how the losses are aggregated and spread by year and
between insurers and reinsurers. This is dependent on court actions to date and
will vary state by state.
The individual insurer or reinsurer then has to follow through the alternative
insurance, reinsurance and retrocession linkages to obtain its own ultimate
forecast figures. This might be done via a decision tree, based on tracking the
total market loss through the consecutive layers of insurance, reinsurance and
retrocession. The total market loss will have been estimated by splitting the
exposure into four markets (US, Lloyd's, London Market and Others), considering
retained/insured percentages and share of the insured amounts by market at each
level.

Share of total market: relativities

Again, the total market loss is considered. However, for this simpler approach the
percentage of the total market is then estimated, using suitable market parameters
(for example, by considering the insurer's premium income for the relevant
classes of business relative to the total premium income for those same
classes/insurers over the whole market).

Exposure
The book of business written is examined to determine, for each contract written,
what the possible exposure is to the various latent hazards, and somehow
estimating numbers of losses to each contract. This must take into account
numbers of reinstatements, aggregation issues, occurrence and aggregate excesses
and limits, and so on.

09/97 2C.5
 INTRODUCTON TO RESERVING

Market Practice
This is based on the assumption that, as the results are so unpredictable, the best
that can be done is to be in line with the market. Therefore, if the comparable
market is in general using an IBNR of (say) 1.5 times reported outstanding claims
for asbestos-related claims, then this would be used as a benchmark. A higher
multiplier might be considered as placing an unnecessary strain on the resources
of the company; a lower multiplier might be questioned by the company's
auditors, unless it could be proved that the company's relative exposure differed
from that of the whole market.

Further Research
Much work is currently being undertaken worldwide in researching approaches to
reserving for latent claims, particularly in the USA. To date, however, there is no
universally accepted approach to these problems. Each case must therefore be
considered carefully on its own merits by the reserver, bearing the above points in
mind.

09/97 2C.6
 [2D]
COMMONLY USED SIMPLE RESERVING METHODS

Introduction
The following section lists all the reserving methods found in Volume 1. It briefly
describes the model underlying each method, together with its advantages and
disadvantages, and refers to where more detailed discussions and examples can be
found.
The main purposes of this section are:

a) while reading through the Manual, to act as a simple summary to help recap
and consolidate the knowledge gained to date,

b) having read the Manual, to help identify a shortlist of possible methods to

apply to a given reserving situation, and to act as a quick reference point for
those wanting to refresh their memory about a particular method.

Consequently some of the terminology used may not be familiar until subsequent
sections have been read.

Methods of estimating ultimate liability which do not explicitly allow for

inflation
In the following descriptions, unless otherwise stated, "ultimate liability" is taken
to mean the ultimate loss from a particular claim year, however defined. The total
incurred claims cost can then be found by adding across all claim years and
deducting the paid to date.

Method and Description of the Model General Comments

References

1. Grossing up Ultimate liability = Simple.

paid* at delay d Assumes a stable run-off pattern,
(general) / grossing up factor for delay d but is susceptible to error if this
is not the case (e.g. if inflation is
D6 varying rapidly), particularly for
[various ways in which the grossing up factor can
the most recent years.
El-4, 11-12 be derived]
F3,7

09/97 2D.1
 INTRODUCTON TO RESERVING

2. Grossing up Ultimate liability = Asl.

(case + ( case reserves at delay d Tests the adequacy and
reserves) / grossing up factor for delay d) consistency of case reserves if
paid at used with data grouped by report
delay d [various ways in which the grossing up factor can year.
be derived]
F5-6

3. Link ratio Ultimate liability = Asl.

(general) paid* at delay d
× link ratio at delay d
D6 × link ratio at delay d+ 1
E5-12 ×…
F4-7 × last link ratio

[various ways in which the link ratios can be

derived)

4. Link ratio As 3. with the link ratio at delay d derived from Asl.
(basic chain the run-off triangle as the sum of column d+1 Trends and anomalies in the data
ladder) divided by the sum of column d excluding the last are not allowed for if operated
entry. blindly.
E8 Produces a single rigid estimate,
without any indication of how to
look for possible variations.

5. Loss ratio Ultimate liability = premium Very simple.

(general) × ultimate loss ratio Requires little information.
May be the only method if
D6-7 [various ways in which the ultimate loss ratio can claims data are scanty or
G2, 10-12 be derived, and in the way that premium and loss unreliable.
ratio can be defined] Ignores the claim development
pattern to date.
Prejudges answer if used
naively.

6. Loss ratio Ultimate liability = Simple.

(step-by-step) paid* at delay d Insensitive to data in the most
+ premium recent years.
G9 × ( loss ratio for delay d+ 1 Loss ratios are based on the
+ loss ratio for delay d+2 observed claims development
+ ... pattern to date, rather than
chosen arbitrarily.
+ loss ratio for last delay)

[various ways in which the loss ratios can be

derived, and in the way that premium and loss
ratio can be defined]

09/97 2D.2
 COMMONLY USED SIMPLE RESERVING METHODS

7. Bornhuetter Ultimate liability = Credibility approach between

Ferguson paid1 at delay d statistical estimate and
+ ( premium predetermined figure.
D6 × ultimate loss ratio Insensitive to data in the most
G3-8 × proportion of the ultimate liability recent years.
which will emerge in future) Prejudges answer to some
extent.
[various ways in which the ultimate loss ratio can
be derived, and in the way that premium and loss
ratio can be defined]

8. Average cost Ultimate liability = Makes use of extra information.

per claim ultimate number of claims Easier to detect and allow for
× ultimate average cost per claim changes in the development
D8 pattern of numbers.
Hl-5 [various ways in which the number of claims and Doesn't allow for changes in the
average cost per claim can be derived] number of zero claims or in the
company's definition of settled.
Difficult to apply if claims are
relatively few in number.

1
incurred claims (paid plus case reserves) may be used as an alternative to paid claims only,
and can give a further perspective on the estimating of ultimate liability. However, whereas
methods based on paid data require only a stable settlement pattern, those based on incurred data
also require a stable reporting pattern and consistency in the setting of case reserves.

09/97 2D.3
 INTRODUCTON TO RESERVING

Methods of estimating ultimate liability which explicitly allow for inflation

These methods can be used when inflation is varying rapidly.

Method and Description of the Model General Comments

References

9. Inflation Most of the above methods can be based on Need to choose appropriate past
adjusted historical payments inflated to current money inflation rates.
methods terms. The resulting projected payments are
increased in accordance with expected future
J2 inflation.

10. Bennett & Ultimate (report year) liability = paid at delay d Need to choose appropriate past
Taylor + number of claims reported inflation rates.
Method A × ( inflation adjusted average payments
in delay d+1
J3 + inflation adjusted average payments
in delay d+2
+…
+ inflation adjusted average payments
in last delay)

11. Separation Ultimate liability = Past inflation rates derived from

method paid at delay d data — no need to choose
+ number of claims appropriate rates. Also derives
J4 × ( proportion of payments in delay d+1 any other calendar year effects
× index of average payments within the data.
Relatively complicated.
in payment year a + d + 1
Unstable.
+ proportion of payments in delay d + 2
× index of average payments in
payment year a + d + 2
+ ...
+ proportion of payments in last delay
× index of average payments in
payment year a + last delay)

09/97 2D.4
 COMMONLY USED SIMPLE RESERVING METHODS

Methods of estimating IBNR values directly

Method and Description of the Model General Comments

References

12. Simple ratios IBNR value = Very simple.

(general) previous period IBNR value Requires little information.
× current value of a quantity related to IBNR May be the only method if
13-4 / previous period value of the same quantity claims data are scanty or
unreliable.
[various quantities can be used] Insensitive to changes in the
relationship between IBNR and
the chosen quantity.
Extremely limited applicability.

13. Simple ratios As 12. with the quantity defined as the product of Very simple.
(Tarbell) the number of claims reported in the last 3 months Too crude for medium or long
and the average size of claims reported in the last 3 tailed business.
15 months.

14. Average cost IBNR value = Makes use of additional

per IBNR number of IBNR claims information.
claim × average cost per IBNR claim Difficult to apply if claims are
relatively few in number.
16
15. Loss ratio IBNR value = More suited to long tailed
(step-by-step) IBNR emerged at delay d business.
+ premium
17-8 × ( IBNR emergence/premium at delay d+1
+ IBNR emergence/premium at delay d+2
+…
+ IBNR emergence/premium at last delay)

[various ways in which IBNR emergence can be

defined]

Other methods
Volume 1 also includes methods

for tail fitting E13

for estimating reserves for reopened claims K2-3
for estimating reserves for claims expenses K4-5
for estimating reserves allowing for discounting L2-4

09/97 2D.5