Sie sind auf Seite 1von 14



In these notes we examine the basic principles that apply to the underwriting of general (non-life)
insurance risks and look briefly at some of the problems which underwriters face.

We can begin by attempting to define the activity of underwriting. The purpose of underwriting was
described by Bernard Benjamin in the following way:

‘The purpose of underwriting is to choose to cover, from all the risks which are proposed to
the insurer, those which collectively will be profitable.’

Benjamin notes that underwriting covers several practical aspects:

• accepting or rejecting risks offered and fixing premiums and retentions (the insurer’s retention
being the amount of the risk that it is willing to cover);

• classifying and rating each risk, establishing the standards of coverage to be offered to
acceptable risks;

• securing a safe and profitable distribution of risks and establishing standards, rules and
procedures to preserve those standards.

If a risk is regarded as generally acceptable the underwriter must decide what cover is to be offered
and fix a premium for it. The ‘pricing’ aspect of underwriting is often described as rating. There are
three basic approaches to rating:

• Judgement rating
• Experience rating
• Manual or group rating


Judgement rating means the fixing of a premium by means of personal intuition alone, without the use
of statistical data. Obviously, few insurances are now underwritten purely on the basis of personal
judgement except, perhaps, some of the unusual risks that are sometimes insured in the Lloyd’s
market. These have included policies on actresses’ legs, singers’ voices, a comedian’s moustache
and the prospect of a Martian invasion. Judgement rating is worth mentioning, however, if only to
emphasise that the underwriter will often have to exercise at least some degree of personal
judgement when interpreting and applying the statistical data that he has at his disposal. Such
judgement, however, is itself the product of the underwriter’s experience with other risks and thus
might be described as the ‘unconscious’ use of statistical data. However it is viewed, an underwriter’s
personal experience is a valuable resource, particularly when new or unusual risks are to be insured.


Experience rating is the pricing of a risk on the basis of the proposer’s own past claims experience, or
rating the risk on its own merits. For obvious reasons this method is inappropriate for most individual
risks where past claims experience (if any) will not provide a sufficiently reliable guide to future claims
trends. However, where a single insurance contract effectively covers a large number of individual
risks past claims experience may give a statistical base that is wide enough. Thus, experience rating
can be used successfully for motor fleet risks (where there is a sufficiently large number of vehicles in
one ownership), employers’ liability business (where there is a sufficiently large number of employees

‘Underwriting and the Selection of a Liability Portfolio’, in A Guide to Insurance Management edited by
Stephen Diacon.

in one employment) and non-proportional reinsurance risks (e.g. excess of loss treaties). The no-
claims discount (NCD) or Bonus-Malus, system, used by virtually all motor insurers provides a further
example. Here experience rating is effectively used to modify a premium that is initially set by the
group (class) method discussed below.


Group rating is the classic method, and is used for the vast majority of general insurance risks.

First, some preliminary points. An insurance premium is often (although not always) the product of
two factors:

1 a measure (or unit) of exposure

2 a premium rate.

The measure or unit of exposure reflects the extent to which the insured subject matter is exposed to
the risk of loss. In the case of property insurances the measure is usually value (either the total value
of the property or maximum possible loss). In liability insurance it is usually something that reflects
the amount of activity generated by the insured business, such as payroll or turnover.

The premium rate is generally expressed as a rate per unit of exposure, e.g. per £0.50% of payroll or
sum insured. The rate itself reflects the degree of risk and is based largely on the average claims
frequency per annum per unit of exposure and the present value of average claim size. However, it
will also depend on a number of other factors, such as the loading for expenses and contingencies.
These factors are considered later.

Because not all risks are the same, it is necessary to divide them into distinct groups, placing those
with similar characteristics in the same group and charging similar rates of premium. The factors that
are most likely to influence annual claims cost are used to distinguish one group from another. These
are known as underwriting factors.

In the early days of general insurance little attempt was made to group risks in this way. Not only
were the ‘unlucky’ policyholders who suffered losses subsidised by ‘lucky’ policyholders who did not
(this being the basic principle on which insurance has always been based – the losses of the few
being borne by the many) but the ‘good’ risks also effectively subsidised the ‘bad’ ones, since all paid
the same price for their cover. However, it was quickly realised that such an arrangement was
inequitable and, moreover, that an underwriter could gain an advantage over his competitors by
identifying good risks and attracting them through lower premiums, whilst penalising bad risks. The
business of an underwriter who ignored distinctions between risks would tend to be unprofitable,
because he would lose good business to his competitors and collect insufficient premium to cover
losses on the bad business that he would inevitably tend to retain. This is the phenomenon of
‘adverse selection’, discussed below.

The history of fire insurance in the U.K. provides a good example of the way in which rating systems
develop. Rating was very simple at first, with a standard rate for brick or stone buildings which was
doubled for timber properties. There were thus only two basic rating groups. Further rating factors
came to be added over time, producing higher premiums for properties outside London and for
‘hazardous trades’ requiring the storage of flammable materials. The Industrial Revolution brought
many new trades and processes, each with its own hazards, ultimately resulting in highly complicated
fire rating structures and ‘tariff’ agreements.

Since 1932 in the U.K.
The effectiveness of the NCD system in this regard, and the justification for using it, has been weakened
in the U.K. by the increasing availability of a ‘protected’ no claim discount by virtue of which a certain
number of claims are allowed without any penalty in terms of loss of premium discount. About 60% of
U.K. motor policyholders with maximum NCD entitlement have such a ‘protected’ discount. Again, a
competitive motor market has lead many insurers to offer ‘introductory’ discounts of up to 60% in a wide
variety of circumstances, without reference to previous entitlement.
i.e. agreements amongst insurance offices to provide standard forms of policy and charge standard
rates for particular risks. Most classes of general insurance business have been subject to a tariff at

This historical process, of increasingly sophisticated discrimination between risks, is seen also in the
development of motor insurance rating. Early motor underwriters used a very simple classification,
but modern rating systems divide motor risks into an astronomical number of possible sub-groups. In
the UK, for instance, insurers currently employ 20 basic vehicle groups, up to 50 age bands for
drivers, a minimum of 20 rating zones, offer three levels of cover, and at least four classes of use.
When all the other underwriting factors are taken into account there are about 550 million possible
rating combinations. Since there are only about 25 million vehicles on the road in the U.K. adequate
claims experience for each individual sub-group will, of course, not be available. Sophisticated
actuarial techniques must therefore be used to analyse claims data and produce appropriate
premiums for a multitude of different risk combinations (see below).

Given the high level of competition between insurers this process of increasing sophistication in risk
discrimination is likely to continue. The most successful underwriters are likely to be those who can
identify groups of policyholders who are effectively being overcharged by the rest of the market and
gain their business by offering them lower premiums, whilst identifying those who are being
undercharged and take corrective action.


We will now examine all the factors that effect the premium for a general insurance risk. First, we
must distinguish between the ‘risk’ premium and the ‘office’ premium. The risk (or pure) premium is
the amount which the underwriter must collect in order to cover the cost of claims only. The amount
finally charged to the insured is the ‘office’ premium. This will be a greater amount because it will
include loadings for expenses and be affected by a number of other matters that are mentioned
below. The factors governing the ‘office’ premium for a general insurance risk are as follows:

Premium components

A the risk premium: This is determined by:

• claim frequency
• claim size
• claims-handling expenses

B modifying factors

• effect of a deductible (excess)

• effect of excess of loss reinsurance
• loading for contingences
• adjustments for inflation
• adjustments for investment income
• expenses: these include: - expenses proportional to premium (e.g. commission)
- expenses per policy (e.g. renewal costs)
- general overheads
• profit loading and cost of capital
• market forces

We have already discussed claim frequency and claim size in relation to the risk premium. The
inclusion of an amount to cover claims handling expenses is self-explanatory.

one time or another in the U.K., although there have always been ‘non-tariff’ insurers competing
alongside them. Competition from non-tariff offices (including overseas insurers) ultimately lead to the
dissolution of these agreements, none of which have survived in the UK. Such tariffs are now illegal
under European law.


A deductible (or excess) is an agreed amount that the insured must bear in the event of a loss, thus
reducing the insurer’s liability for particular losses and its overall claims cost. This will obviously allow
a corresponding reduction in the amount of premium that needs to be collected. Deductibles are also
a useful way of combating moral hazard (see below).

Excess of loss reinsurance

In very simple terms, under an excess of loss reinsurance the insurer agrees with a reinsurer that the
latter will pay the balance of claims above a certain level in exchange for an appropriate portion of the
premium. Just as a deductible reduces the insurer’s liability by removing the ‘bottom slice’ of a loss,
so excess of loss reinsurance reduces liability by removing the ‘top slice’ of larger individual losses or
large accumulations of loss. However, the effect of reinsurance arrangements on the premium
charged to the insured is less straightforward, because the insurer must, of course, pay the reinsurer
for the services that he provides and pass on these costs to the insured.

Loading for contingencies

Unusual and unforeseen events may produce a level of claims or expenses which is higher than that
suggested by the available statistical data. In an effort to guard against the unknown and build a
margin of safety underwriters therefore include an extra amount in their calculations for contingencies.
Where there is little loss data to go on the risk will be very uncertain and a high ‘safety factor’ will have
to be built in.


Inflation must be taken into account when using existing claims data to fix future premium levels.
Thus, it will usually be necessary to increase the value of past losses to reflect the effects of inflation
between the date of the loss and the present day. Again, for lines of business where claims take a
long time to settle (such as liability insurance) an allowance must be made for inflation between the
date of the policy and the date when claims on it are likely to be paid. In some cases (such as
personal injury claims) the cost of claims may be rising at a level that is much higher than the general
level of inflation. Inflation must also be taken into account when deductibles and retention levels for
reinsurance are fixed.

Investment income

General insurers once regarded investment income as an extra ‘windfall’ on top of the underwriting
profit that they expected to make. However, most insurers have seen little in the way of underwriting
profit in recent years. Any trading profit they have made has been as a result of investment income,
which is now taken into account by general insurers when setting premium rates. The life offices have
always done this, because life insurance contracts are long term and investment income is likely to be
substantial. Investment earnings are likely to be modest for most lines of general insurance business,
but more significant for some ‘long-tail’ business (such as employers’ liability insurance). Claims
reserves are one of the main sources of investment income and, in the case of liability insurance,
these usually stand at a level that is about 400% of premium income. Thus, for every £1 of new
premium collected by liability insurers each year there is about £4 set aside in a reserve for
outstanding claims. Unfortunately, long-tail liability business brings with it uncertainties that are likely

Equally, inflation eats away at the insurance cover itself, since sums insured become inadequate as
property values and rebuilding costs rise with the effects of inflation, resulting in under insurance. It is
important to provide insurance contracts that are flexible enough to accommodate the effects of the fall
in the value of money. The problem became particularly acute in the U.K. during the 1970s, when
inflation levels were relatively high. A number of techniques (such as index-linking of sums insured in
personal lines business and various special schemes for commercial risks) were widely adopted at this
I.e. insurance business characterised by a long time lag between the collection of the premium at the
beginning of the insurance year and the settlement of claims arising from that policy year.
Investment income is also derived from unexpired premium reserves and shareholders’ funds – the
capital base of the business.

to outweigh the benefits of this enhanced investment income (this problem is discussed later).
Furthermore, the falling interest rates and poor returns on equity holdings that have been experienced
in Europe recently have dramatically reduced insurers’ investment income.


Expenses include all the general costs of running an insurance business, including commission
payable to intermediaries. For some lines the latter may account for 10-20% of the premium. As
regards expenses, it is easy to say that they should be as low as possible: insurers everywhere try to
keep expenses down to gain a competitive edge. This is seen most dramatically in ‘personal lines’
business where there has been a move towards direct marketing (to reduce acquisition and
distribution costs), simplified underwriting (which can be carried out by less well-qualified and less
highly paid staff) and streamlined policy processing and claims handling techniques.

It is worth noting, however, that the optimal level of expenses is not necessarily the lowest that can be
achieved. For example, expenses can be lowered by a reduction in the number and/or quality of risk
surveys, simplified underwriting that disregards some of the finer distinctions between risks, and less
thorough claims investigation. However, the result may well be an influx of lower quality business,
more claims and more fraud. As in all things, a balance must be preserved between containing costs
and maintaining quality.

Profit loading and the cost of capital

Capital is the raw material from which insurance is ‘manufactured’. Regulators require that insurers
(like banks and other financial institutions) should hold sufficient capital to ensure that they are able to
meet their obligations (i.e. pay claims). This ‘cushion’ must be deep enough to deal with possible
fluctuations in an insurer’s business results and the impact of unexpected events. The tendency on
the part of regulators now is to insist that capital standards should be ‘risk based’ – i.e. that the
amount of capital to be held should depend not just on the amount of insurance business written but
also on the relative ‘riskiness’ of that business, which is liable to vary according to which lines of
insurance are written and various other features of the insurer’s business, such as the extent to which
it is diversified. Insurers must obviously give investors a rate of return which is generous enough to
attract the amount of capital which they need. Obviously, the higher the cost of that capital (i.e. the
higher the return demanded by investors) the greater the loading on premiums must be.

Market forces and the underwriting cycle

From a purely actuarial viewpoint the underwriting process is one of observation and statistical
modeling, with the establishing of distinct classes of risk as its object (see below). However, market
forces also play a powerful role in fixing insurance premiums and levels of cover. Insurance is bought
and sold in a market where competition is fierce, and sometimes destructive. Underwriters know that
if they cannot match their competitors in terms of price and cover customers will simply take their
business elsewhere: contracts are short-tem (normally one year) and brokers make it their business to
ensure that their clients get the best deal. In recent years underwriters have often accepted business
– unwisely – at rates of premium that they know to be inadequate in order to retain or increase their
market share. Apparently, they have done so in the hope that rates will subsequently harden or
claims experience will improve.

Competition of this sort is one factor (but certainly not the only factor) in the ‘underwriting cycle’ – the
phenomenon whereby insurance markets swing between ‘hard’ and ‘soft’ markets, with periods of
(relative) profitability and (relative) unprofitability alternating over a cycle of 6-9 years. This is
discussed below.

Defining the underwriting cycle

A key feature of general insurance markets is that prices, and hence underwriting profitability, are
subject to sharp swings. This is known as the ‘underwriting cycle’ or ‘insurance cycle’. Since
underwriting profitability is the difference between insurance premiums received (and earned) and
claims and expenses that are incurred, the frequency and severity of claims during a period will be a
key determinant of underwriting profitability. But the term ‘underwriting cycle’ is usually defined in

terms of a pattern of upward and downward movements in insurance prices, broadly cyclical in nature,
and their subsequent impact on underwriting profitability, which is similarly cyclical.

Causes of the underwriting cycle

Cyclical patterns are common in most industries, and in the wider economy as a whole, where we
speak of the ‘business cycle’. However, cycles in non-life insurance markets tend to be more
pronounced than in industry generally. What are the reasons for this high degree of cyclicality?
There have been many studies of the insurance cycle in the United States, and a few in other
countries. One common empirical finding is that while both the demand and supply for insurance
varies over time, it is variations in supply that are more important than variations in demand. This is
because financial capital is the main determinant of supply. New financial capital can come into a
market quickly to increase supply when premiums are high, but can withdraw quickly from the market
if it appears that the rate of return on capital is falling below a required rate. Again, new companies
can come into the insurance industry fairly quickly, provided there are no unusual delays in gaining
regulatory approval. This contrasts with the manufacturing and agricultural sectors, where supply is
less elastic. However, there are many factors that might cause financial capital to increase or
decrease, and the availability of capital does not mean that it will be used to provide underwriting

Three main theories of the insurance cycle have some degree of empirical support, although no single
theory is fully supported by the facts. One of these theories relates to expected investment income
(or more generally investment returns) that can be earned on insurance premiums. Since insurance
companies receive premiums at the outset of the insurance contract, they can invest these premiums
(less administration and marketing costs and reinsurance costs) until claims are eventually paid. The
longer the time lags between the receipt of premiums and the payment of claims, the greater the
potential to earn investment income. If interest rates (or rates of return) are expected to rise, then
some insurance companies will seek to reduce insurance prices in order to attract more premiums to
invest in expectation of these higher interest rates (rates of return). Other insurance companies, not
wishing to lose market share in a competitive market, will also cut insurance prices. Similarly, if
interest rates fall or are expected to fall, insurance rates will rise, after a time lag. See Figure 1 below:

Figure 1: cycle caused by interest rate movements ('cash flow' underwriting)

interest rates

x insurance rates or
underwriting profits
time-lag time-lag


Inverse relationship, with a time lag

A second theory is based on available equity capital and the cost of equity capital. Equity capital is
more relevant than debt capital because insurance regulators accept equity capital for solvency
purposes and restrict the use of debt capital (non-subordinated debt). Hence, when stock markets
rise above a normal level, there are two capital effects. The first is that the cost of capital falls for
existing and for new insurance companies. Second, the rise in share prices increases the value of the
financial asset holdings of insurance companies and hence brings about an even greater increase in

A useful discussion of the underwriting cycle can be found in ‘Profitability of the non-Life insurance
industry: its back-to-basics time’, Sigma No. 5, Swiss Re, 2001.

their capital and reserves. This increase in available capital, and a reduction in the cost of capital, will
tend to increase supply and hence impart a downward pressure on insurance prices. Similarly, if
stock markets fall and remain depressed, this will tend to cause insurance prices to rise. See Figures
2 and 3 below.

Figure 2: Impact of a rise in investment values on the capital of an Insurance Company

Balance Sheet (before) Balance Sheet (after)

Assets Liabilities Assets Liabilities

Policyholder Policyholder
Investments liabilities liabilities
(Technical Investments (Technical
provisions) provisions)

Current Current
liabilities liabilities

Current assets
Capital Current assets
Offices & other & Reserves Capital
fixed assets & Reserves
Offices & other
fixed assets
rise in
of investments this rise in value of investments
e.g. buoyant stock is reflected in a greater proportionate
and/or bond market increase in the size of Capital &

Figure 3: cycle caused by impact of movements in the stock market:

(increasing capital & reserves and hence capacity)

capital market values

(stock market values)

x insurance rates or
underwriting profits
time-lag time-lag


Inverse relationship, with a time lag

A third theory supposes that it is claims experience and not capital markets effects that are most
important. This theory holds that insurance pricing tends to underestimate the potential for claims
during periods when there are no large individual claims or clusters of claims. However, when a very
large claim occurs, insurance prices rise sharply, especially if the large loss also depletes capital or
causes insurer insolvencies. Then, if there are no major claims in the ensuing period, insurance
prices tend to drift downwards until another large loss occurs. This theory is based on the concept of
an ‘economic shock’ and assumes that the insurance market has a short memory. There is a further
behavioural aspect of the ‘claims shock’ theory. It supposes that after a major loss, insurers will seek
to recover some the losses that they have paid out, especially if these payments were well in excess
of those anticipated when prices were set. However, after there has been a sharp rise in prices,
competitive forces start to cause prices to fall, even though insurers may seek to resist market
pressures for lower prices for a period. See Figure 4 below.

Figure4: ‘Claims shock’ theory

insurance rates or
underwriting profits
(remaining high
euro afterwards to allow
recouping of losses)



Empirical evidence suggests that all these theories apply at the same time but in differing degrees of
importance, varying with the type of insurance. However, this evidence is far from conclusive since
the data available for analysis is inadequate. Nevertheless, it would appear that cycles in the price of
insurance in respect of natural catastrophes are more likely to be caused by the ‘claims shock’
phenomenon, while less risky types of insurance are more likely to be influenced by capital market
effects. There also is evidence that capital market effects have become more important across all
classes of business than hitherto, because capital management within insurance companies has
become more important and top management are now more sensitive to stock market conditions.

Motor insurance, which is not as risky as some other types of insurance (such as commercial property
insurance and marine and aviation insurance) is most likely to be influenced by interest rate and
capital market factors. Liability insurance is also likely to be affected by interest and capital market
factors. Interest rate effects, in particular, will be especially important in the case of liability insurance.
This is so because of the high levels of investment income that liability insurers expect to earn on
premiums that are received, which arise from the long time lags between the collection of liability
premiums and the settlement of claims. Similarly, since supplying liability insurance ties up the capital
of insurance companies, in part because insurance companies cannot discount future claim payment
at realistic rates, a fall in capital and reserves is likely to reduce supply. The rapid rise in the price of
liability insurance which occurred a few years ago is consistent with the fall in interest rates and
depressed stock market conditions which obtained during the period in question.

Since many insurance companies underwrite a wide range of insurances, there is likely to be ‘knock-
on’ effect across their underwriting portfolios as a whole. For example, the fact that the 2001 World

Trade Centre disaster occurred at a time of falling interest rates and stock markets had a
compounding effect on the supply of insurance. Moreover, the events of September 11 probably
increased risk aversion among customers and hence increased their willingness to pay higher
insurance prices, at least to some degree.



The techniques that are used to analyse data will depend on the level of detail and reliability of the
data available, and also the extent to which the data are a good predictor for the future. In the case of
some lines of insurance (e.g. liability insurance) the data is usually rather limited in scope, but in other
lines, such as motor insurance, there are often large amounts of data at a very fine level of detail.
These can be used to construct premium rates based entirely on previous data, which can then be
compared with the rates in use. The results have to be treated with caution, since they are based on
one (or possibly a few) years’ experience, but they can give indications of issues to be considered.
For example, it is possible to detect which areas of the business are profitable and which are loss
making. Companies can also seek to exploit niche markets where they are able to make profits and
steer away from markets where they make losses. It should be noted that, while price is extremely
important, it is not the only consideration for the consumer. Other factors include the service that is
provided to customers after a claim, the convenience of using a particular insurer, and so on.

Motor claim types (e.g. third party property damage, third party bodily injury, theft, windscreen, own
damage, etc) are modelled separately. This is known as ‘component pricing’. The reasons for
modelling these separately are that different factors may affect each, and the factors may affect each
in different ways. Thus, it is expected that more accurate modelling can be achieved in this way.

For each claim type, separate models are formed for the claim frequency and severity, using past
data, to determine the effect of various factors. The expected values for frequency and severity are
multiplied together to get an expected claim cost for each claim type (per unit of exposure). These
are then added over claim types and adjustments are made to allow for expenses, contingencies, etc.
Finally, if a simple structure is required, it is common to fit an appropriate model to these to recover
the structure of the premiums. This gives a complete set of prices based on past experience, which
can be used in conjunction with the present rating structure, market knowledge, information on claim
trends, and so on, to make decisions about premiums to be charged in future. This summarises the
process, and more details are given below.

It is assumes that data is available at the individual policy level at a sufficient level of detail. In the
past this may not have been available, but it has become standard practice to record policy records at
a great level of detail, so that statistical investigations of the data have become possible. The
recorded information should include characteristics of both the policyholder and the risk being insured.
These include the age of driver, No Claims Discount (NCD) or Bonus-Malus level, occupation, claims
record, vehicle rating group, location, use, etc. The vehicle rating group is a grouping of car types,
whereby small family cars will be in a low group, and large, high-performance or expensive cars will
be in a high group. The intention with this, and indeed all the recorded information, is to use it to
predict the number of claims that will occur per annum, per policy, and also the size of each claim.
The data on the policies that the company has written over a specific period is one part of the
information needed for this analysis. This gives information on the exposure to risk that the company
has faced: roughly, how many policies there were of each level of the grouping factors. It is from these
that the claims will emerge, and it is necessary to know the exposure in order to quantify the risk. Ten
claims arising from one hundred policies is a much higher risk than ten claims arising from one
thousand policies. Thus, the insurer needs information on the number of policies as well as the
number of claims.

It is also assumed that the insurer has corresponding data on claims which links a claim record with a
policy. This means that the insurer can associate the risk factors with each claim, and match them up
with the exposure data. Claims are classified by type of claim: bodily injury, windscreen, etc. These
are analysed separately. Let us consider one claim type in more detail: call this claim type i . The
data available are the exposure in vehicle years or earned premium, number of claims and claims
cost. Both are cross-classified by the risk/rating factors. At this point, some further explanation is
needed of rating factors. It is assumed that the risk associated with each policy varies according to

risk factors. These may not be directly measurable: for example, an obvious risk factor may be how
good a driver is, or how willing the driver is to take risks. As these are difficult, or impossible to
measure, some proxies which are easily measurable are needed, which are called ‘rating (or
underwriting) factors’. For example, the age of the driver is important as a measure of the risk factor
of how good, or risk averse, the driver is. The aim is to find rating factors that are good predictors of
the risk, and it is likely that companies will continually search for refinements to try to do this more

An example of some data, looking at claim frequency rather than claim severity, is given below. This
is a small extract from a data set. Among the information shown is the year of account, and it is
sometimes the aim to examine trends in claim rates or claim sizes over time. Then come the rating
factors: policyholder age, marital status, NCD in years, ABI group (for the vehicle), policyholder sex.
Following that is given the total exposure, which has been calculated in policy years (after taking into
account new policies, withdrawals, policy alterations and so on), and finally the number of claims
(here for accidental damage). Sometimes claims are recorded even though they do not eventually
result in a payment, and these need to be excluded: hence the wording ‘non-nil AD claims’.

Rating Factors
Year PolicyholderAge MaritalStatus NCDYears ABI Group PolicyholderSex Total Exposure Total of Non-nil AD Claims
1997 22 S 4 6 M 199 1
1997 22 S 5 9 F 144 1
1997 25 M 6 9 M 23 1
1997 27 S 3 6 M 292 1
1997 30 S 1 12 M 258 1
1997 37 S 3 6 M 152 1
1997 45 M 0 3 M 66 1
1998 22 S 1 13 M 177 1
1998 22 S 5 9 F 221 1
1998 24 S 0 4 M 163 1
1998 25 M 6 9 M 315 1
1998 25 S 7 14 F 153 1
1998 26 S 2 6 M 192 1
1998 27 S 3 6 M 73 1
1998 28 S 0 7 M 648 3
1998 30 M 3 5 M 245 1
1998 30 S 1 12 M 107 1
1998 34 M 7 9 M 40 1
1998 45 M 0 3 M 299 1

Once this data has been set up, a model can be formulated. For frequency, a Poisson distribution is
used, with the exposure affecting the mean number of claims in each category. Thus, the number of
claims for a particular category (eg 1997, age 26, married, NCD 4, ABI group 3, female) has a
Poisson distribution with mean:

Exposure × Claim frequency per policy per annum

The aim of the exercise is to find a reliable estimate of the claim frequency per policy per annum for
each category. A modelling exercise is then followed to examine the effect of each of the rating
factors. This explores whether the claim frequency differs by policyholder sex, age, NCD, and so on.
If it does, then these effects have to be quantified, and this gives an estimate of the claim frequency
per annum, for this claim type.

Next the average claim size has to be investigated in a similar way. The total cost of claims in each
category is calculated from the database. This is divided by the number of claims in each category to

give an average claim size. These average claim sizes, cross-classified by the rating factors is
modelled using a gamma distribution, and the effects of all rating factors is explored. A final model is
decided upon, which gives the estimated average claim size for this type of claim in each category.

The estimates of claim frequency and claim severity from these models can then be combined, to
obtain the expected claim payment per policy per annum in each category, often known as the ‘risk
premium’. This is for one claim type, and the process has to be repeated for each claim type.
Typically, it is found that different factors are used for each claim type, and in the frequency and
severity models.

These risk premium, by claim type, can then be summed as appropriate to obtain the total risk

Total Risk Premium = ∑

fitted frequency × fitted severity

These figures will capture the relative risks of each category, known as the ‘relativities’. It is usual to
apply adjustments, to allow for changes in the risks over time and to project into the future, to allow for
possible distortions in the reserves, and so on. In addition, as mentioned earlier, allowance has to be
made for expenses (per policy and per claim), investment income, commission, profit loading and so

The relativities are usually multiplied by a ‘base premium’, which is a single figure and does not vary
by the rating categories. The effect of this is to shift all the premiums up or down, and this can be
done at any time.

While it is the case that most claim types can be analysed in this way, it is sometimes found that the
risk cannot be quantified at such a fine level of detail. This may be due to lack of data, or simply
because the risk does not vary by any significant amount. Typically, this would apply to larger claim
sizes for third party bodily injury. One way forward in this situation is to cap all such claims at a
reasonable amount for data analysis purposes, and work with the capped claims rather than the
actual claims. An adjustment would need to be made such that all risk premiums are increased by the
expected excess of the actual claims over the capped claims. This would be done as a constant
loading irrespective of the rating factors.

The final figures from the data modelling exercise can be compared with the present rating structure,
and used in a competitor analysis, to make decisions on new rates, to find markets where the
company can sell profitable policies at a cheaper rate than their competitors (niche marketing), and so



Moral hazard is a problem that affects insurance generally: it is the risk that, by giving insurance
cover, the insurer will bring about a change in human behaviour which makes the adverse and
economically undesirable insured event more likely to happen. For example, insured persons may
become less careful or even cause losses deliberately in order to get the insurance money.

To discourage this, insurers generally seek to restrict cover to losses which are ‘fortuitous’
(accidental) and restrict payments to an indemnity only – i.e. exact compensation for the loss and no
more. However, this is not always simple. For example, the dividing line between losses caused by
mere carelessness (which insurers must be prepared to cover) and losses caused deliberately (which
are uninsurable) is difficult to draw. Again, there is a commercial need for cover that gives more than
a full indemnity, e.g. ‘new for old’ cover on personal possessions.

Moral hazard can be reduced by a number of standard techniques. These include the exposing of the
insured to part of the risk by means of deductibles or coinsurance, the use of policy conditions to
restrict coverage for high risk insureds, either in advance of losses occurring or as a consequence of
claims experience, and the levying of accurate variable premiums according to risk. In this way the
incentive for taking care is provided by the insurer, rather than by the threat of suffering a loss, a risk
which is now (mainly) transferred to the insurer. Of course, the absence of complete information
about the risk, and the cost of control, means that insurers cannot hope to influence the insured’s
behaviour fully and decisively.

Moral hazard takes on some extra dimensions in the context of liability insurance. See the Parsons,
C. Moral Hazard in Liability Insurance, Geneva Papers on Risk and Insurance, Vol 28, No 3, 2003.


Insurers aim to achieve a balanced portfolio of risks for each class of business that they write.
Usually, the portfolio will include a spread of ‘good’, ‘bad’ and ‘average’ risks, although it will not
always be possible to identify in advance the category into which a given risk falls. Adverse selection
occurs when an insurer attracts an unduly high proportion of ‘bad’ risks and fails to attract enough
‘good’ risks to balance them out. This is usually a consequence of inaccurate pricing by the insurer
whereby ‘good’ risks are overpriced and ‘bad’ risks are underpriced, creating a perverse incentive
whereby the insurer concerned will attract mainly the latter rather than the former. This assumes, of
course, that the insured in question know that the risks they present are wrongly priced, but this
knowledge will be easily acquired in a competitive market.

A standard technique to limit adverse selection is accurate risk differentiation, as described above in
connection with moral hazard – which is closely related to adverse selection. In other words, low-risk
policyholders must be identified and rewarded with reduced premiums while high risk policyholders
must pay more for their cover. It may not be possible to identify high-risk policyholders in advance, in
which case premium loadings and/or restrictive policy terms may be applied as a corrective after
claims(s) have occurred.

To some extent, there is a tension between the need for accurate risk discrimination and the basic
risk-spreading and loss-sharing principles of insurance. For example, precise risk discrimination may
lead to insurance being unaffordable or even unobtainable for some very high risks. As a result,
insurers may be accused of ‘red-lining’ – denying cover to vulnerable people such as those with
homes and businesses in socially deprived areas where crime and vandalism is rife. Unfortunately,
the unavailability of insurance (and, often, of local banking services) may contribute to the further
decline of the area.

We should note that adverse selection is an especially prominent feature of some risks. For example,
people who live on mountain tops will not want to buy any flood insurance. However, the propensity
to insure will increase as one moves down the mountainside. At the valley floor, where the risk is
greatest, everyone will want insurance!

Another way of dealing with adverse selection, at least, in theory is through compulsory insurance.
For example, policyholders can effectively be compelled to buy flood insurance (which some, given
the choice would not buy at all) by writing this risk only as part of a package along with other risks that
nearly everyone will want to insure, such as fire and theft. In this way the risk can be spread, with an
element of cross-subsidy whereby some premium is (notionally) charged to policyholders who present
little or no flood risk in order to fund claims by higher risk policyholders. Of course, this will only work
if the whole insurance market adopts the same strategy. However, raises the possibility of insurers
infringing competition law.


Many of the problems that underwriters face are problems of uncertainty or, rather, of added
uncertainty. The law of large numbers and the underwriting principles outlined earlier give the
underwriter with a means of dealing with conventional risks and ‘normal’ levels of uncertainty that go
with them. However, an underwriter’s calculations can go astray if the circumstances are such that an
extra layer of risk or further dimension of uncertainty is added to the normal insurance risk. This ‘extra
layer’ of risk can arise in various circumstances, e.g.:

(a) where the subject matter of insurance is new, or of an unusual character (such as a new type
of vehicle, aircraft, or industrial product);

(b) where the subject matter is familiar, but the risk alters as a result of unforeseen changes in
factors external to it, or the emergence of some hidden facet of the risk;

(c) the subject matter is familiar, but the cover granted by the insurer is wider than, or different
from, that which normally granted.

(a) In the first case (new or unusual subject matter) little or no statistical data will be available and
successful underwriting will depend to a greater than normal extent on the personal skill, judgement
and experience of the underwriter. However, no risk is likely to be entirely novel since it will always
share some characteristics with risks that are familiar. For example, a new aircraft or road vehicle
should be quite easy to compare with existing aircraft or vehicles in terms of established risk factors,
such as method of construction and repair costs, type of motive power, performance figures and so
forth. In any case, risks of this sort are likely to account for only a very small proportion of a typical
insurer’s business. It would normally be appropriate to include substantial margins for contingencies
and profits and ignore investment income when fixing the premium for risks of this sort.

(b) Risks in the second category (those affected by unforeseen changes in external factors) are
far more of a problem. The changes in question may be physical in nature (e.g. climatic change),
economic, social, legal, or a combination of several of these. If the underwriter is caught unawares
the consequences can be disastrous, because a large proportion of the risks in his portfolio may be
affected. Some examples from the recent (unpleasant!) experience of U.K. and other insurers
illustrate the point:

1 The heavy losses suffered by U.K. general insurers on mortgage indemnity business as a
result of their failure to foresee the collapse of the housing market at the end of the 1980’s;

2 The losses suffered by American and European insurers on pollution and asbestos-related
risks through their failure to appreciate the physical nature of these exposures and the failure
to anticipate changes in the legal environment brought about by (especially U.S.) legislatures
and courts;

3 The sharp increase in employers’ liability claims arising from cancer, noise induced deafness,
RSI (repetitive strain injury) and other gradually developing diseases.

Many more examples could be quoted. Much can be seen with the aid of hindsight and it is not
suggested that any of the above could easily have been predicted by underwriters. When
professional economic forecasters, lawyers, physicians and various other experts have failed to
predict these trends we can hardly expect insurers and their own advisers to do much better.
However, examples such as these provide lessons for the future and highlight the need for insurers to
be aware of the latest developments in law, medicine, natural science and all the other fields of
knowledge that impinge on insurance. There are signs that insurers have learned from past mistakes
and are prepared to look much further ahead than they have done in the past. For example, property
insurers take a keen interest in global warming and its possible effects on weather patterns and

The ‘law of large numbers’ is a principle whereby probability density function of average loss tends to
become concentrated around the mean as the sample number increases – in other words, the more
risks there are in the insurance ‘pool’, the more stable and predictable will be the pattern of losses.

climatic change, and employers’ liability insurers are already thinking about the possibility of claims
from ‘new’ diseases, from sources such as work-related stress, passive smoking and electro-magnetic
fields (EMF).

(c) Risks in the third category (novel forms of insurance) are not unusual. Insurers naturally seek
to develop their products and underwriters may be asked to contribute by granting new and wider
forms of cover. Pressure to do so may arise directly, from competitors, customers or brokers, or
indirectly, through colleagues involved in the sales and marketing function. Where new forms of cover
are granted the potential consequences must be thought through carefully. Again, two examples from
the recent experience of U.K. insurers illustrate the hazards:

1 The unexpectedly heavy claims for subsidence paid by U.K. insurers after cover was
introduced for the first time following the collapse of the ‘tariff’ for home insurance business in
the 1960s (a problem which is only now being properly addressed by underwriters through the
scientific assessment of subsidence risks).

2 The problem of fraudulent or inflated claims for damage to personal property under household
policies written on a ‘new for old’ (i.e. replacement cost) basis when this form of cover
became widely available in the 1970s.

The lesson, again, is to exercise caution when contemplating the introduction of new cover. Most
underwriters are well aware of this, being conservative by nature. On the other hand, marketing
people, and those involved in direct sales or liaison with intermediaries (agency inspectors), like to
say "yes", to requests for wider cover, special terms or relaxed premium rates, perhaps because their
success is sometimes judged by business volumes rather than business quality. The job of the
insurance manager is to reconcile these conflicting interests and to strike a balance between the
conservatism of the underwriter and the zeal of colleagues in sales and marketing.


It is fair to say that the rating and underwriting of liability insurance is generally more problematic than
that of other lines. As we have seen, some forms of liability insurance are characterised by a ‘long
tail’ of claims, which extends many years beyond the periods of insurance where the claims have their
origin. The potentially long delay between underwriting a risk or group of risks and settling the last
claims that arise from the years of insurance in question creates considerable uncertainty as regards
the final cost of such claims and the level of premium that is necessary to cover them. The uncertain
effects of inflation, investment yields and potential increases in the size of court awards over long time
periods are likely to make pricing inherently difficult. Furthermore, the longer the time span of the
whole insurance transaction, the greater is the risk that changes in legislation, scientific knowledge or
accident victims’ general propensity to claim will make claims greater in amount or more frequent than
was anticipated. It is well known, for example, that a failure to predict accurately the level of future
claims for asbestos related illness and environmental damage has caused the collapse of many
insurers in the United States this century and contributed significantly to the problems of the Lloyd’s
insurance market in recent years.

It is also worth noting that problems of moral hazard are particularly complex and acute in the case of
liability insurance, adding further to the problems of accurate pricing.

Finally, we should emphasise a key difficulty – the fact that experience rating may be impractical even
for large liability risks if such risks are of the type that generate long-tail claims. The point here is that
when there are long time delays in the development and reporting of claims the current loss
experience may reflect, not the present state of the risk, but rather the state of the risk thirty or forty
years ago. This problem is particularly acute in respect of employers’ liability risks that generate a
high proportion of claims for latent or gradually-developing diseases.

There have been around 265,000 asbestos-related deaths in the USA. Despite the fact that the use of
use of asbestos in Western Europe had more or less ceased by the 1970s, it is anticipated that
asbestos-related claims will not peak until the year 2020.
11 106 American insurers became insolvent between 1988 and 1990, asbestos-related and pollution claims
being a major source of failure.
See Parsons, C. (2003) ‘Moral hazard in liability insurance’ The Geneva Papers on Risk and Insurance