CM2-21: Run-off triangles Page 37

4 Loss ratios
The ratio of incurred claims to earned premiums over a defined period is called the loss
ratio.

incurred claims
ie loss ratio =
earned premiums

Investigation of the loss ratios for each of several different origin years would normally
show some consistency, provided that there have not been any distortions, and in particular
no significant change in premium rates.

A good example of a ‘distortion’ would be the hurricane that affected the South of England in
1987. This changed the pattern of household insurance claims completely during that origin year.

An increase in premiums would affect the loss ratio directly.

The expected loss ratios will also have formed part of the derivation of the premium basis.

It is therefore logical that a loss ratio based on trends of past data, underwriters’ views, or
market data, could be used as a basis for an estimate of the eventual loss and hence the
outstanding claims. It is, however, on its own a very crude measure due to the fluctuations
that are inherent in any claims experience.

This is similar to the approach that we used earlier when we applied the chain ladder method to a
triangle of loss ratios when we used the ACPC method. However, the approach here is more
general, in that the ultimate loss ratios can be estimated using any method, including subjective
methods involving personal judgement. However, once the ultimate estimated loss ratios have
been found, they are applied to the premium figures in order to calculate outstanding claims
reserve figures, just as we did before.

Such estimated loss ratios may be useful as an input to Bornhuetter-Ferguson based

estimates. These are discussed in the next section.

The Actuarial Education Company IFE: 2019 Examinations

 Page 38 CM2-21: Run-off triangles

5 The Bornhuetter-Ferguson method

5.1 Concept of the Bornhuetter-Ferguson method

The Bornhuetter-Ferguson method combines the estimated loss ratio with a projection
method.

Here, ‘projection method’ refers to methods such as the basic chain ladder method which are
based on past claim amounts and/or numbers.

It therefore improves on the crude use of a loss ratio by taking account of the information
provided by the latest development pattern of the claims, whilst the addition of the loss ratio
to a projection method serves to add some stability against distortions in the development
pattern.

The concepts behind the method are:

That whatever claims have already developed in relation to a given origin year, the
future development pattern will follow that experienced for other origin years.

The past development for a given origin year does not necessarily provide a better clue
to future claims than the more general loss ratio.
In other words it is a compromise that combines the loss ratios with the development pattern.

5.2 Description of the method

In its simplest form the concept leads to the following approach to calculations:

1. Determine the initial estimate of the total ultimate claims from each origin year using
premiums and loss ratios.

2. Divide these estimates by projection factors (f) determined, in a normal manner,

from a claims development table. These are effectively estimates of the claims that
should have developed to date.

3. Subtract these amounts from the corresponding total ultimate claims figures to give
an estimate of the amount of claims that are yet to develop.

Clearly, the three stages could be combined and expressed as:

Future claims development = Premium Estimated Loss Ratio 1 1f

We can relate this formula to the explanation given above:

1 Step 1 gives you premium expected loss ratio

2 Step 2 gives you premium expected loss ratio 1/f

3 Step 3 gives you Step 1 minus Step 2, which is the formula given.

As the final estimate of the ultimate loss is based on observed data and an initial estimate
ignoring the observations, this method could be viewed as using a Bayesian approach.
Bayesian statistics is covered in Subject CS1.

IFE: 2019 Examinations The Actuarial Education Company

 CM2-21: Run-off triangles Page 39

In standard Bayesian work we combine data from two sources. We find an estimate of a
parameter value based on a prior distribution, and combine it with an estimate based on a
likelihood function. Our final estimate for the parameter value is usually a weighted average of
the two estimates from these two different sources.

We are doing the same thing here. We find an estimate of the outstanding losses based on loss
ratios (ie not from the data in the run off triangle), and use the data in the triangle as a second
source of information to refine our estimate.

The analysis is not precise here because there is no single parameter that we’re trying to
estimate.

5.3 Application of the method

In its original form, the Bornhuetter-Ferguson method was applied to the development of
incurred claims. However, it could equally be applied to the development of paid claims,
using either an accident year or policy year cohort.

Further, the original projection was done using a chain ladder approach, although
alternative development factors or grossing-up factors (g) could easily be applied instead
(ie g would replace 1/f in the above expression).

The original form also made no explicit adjustment for inflation, although the method could
be adjusted in a similar way to the other methods.

The example below is based on the original form of the method, but examiners would
expect students to also be able to apply the method to paid claims.

The first stage is to determine the development factors, using the same method as for the
chain ladder methods.

The Actuarial Education Company IFE: 2019 Examinations

Page 40 CM2-21: Run-off triangles

Cumulative incurred claims data, by years of accident and reporting development

DY

0 1 2 3 4 5 Ult

1 2,866 3,334 3,503 3,624 3,719 3,717 3,717

2 3,359 3,889 4,033 4,231 4,319

AY 3 3,848 4,503 4,779 4,946

4 4,673 5,422 5,676

5 5,369 6,142

6 5,818

TOTAL 25,933 23,290 17,991 12,801 8,038 3,717

TOTAL – last no: 20,115 17,148 12,315 7,855 3,719

RATIO (r) 1.158 1.049 1.039 1.023 0.999 1.000

DEVELOPMENT
1.290 1.114 1.062 1.022 0.999 1.000
FACTOR (f)

Don’t get confused by the numbers in this table. The numbers in the top left-hand corner of the
triangle are slightly different from the ones used in the ACPC example.

The ‘TOTAL’ figures have been calculated by summing the entries in each column, eg the second
TOTAL is:

3,334 3,889 4,503 5,422 6,142 23,290

The ‘TOTAL last no’ figures have been calculated by summing all but the last entry in each
column, eg the first TOTAL last no figure is:

25,933 5,818 20,115

The RATIO figures are what we previously referred to as the development factors. These are
calculated in the usual way. Note, however, that since we are given the TOTAL and TOTAL last
no for each development year, the calculation can be simplified. For example, the ratio of DY1 to
DY0 is given by:

23,290
1.158
20,115

IFE: 2019 Examinations The Actuarial Education Company

CM2-21: Run-off triangles Page 41

What the Core Reading refers to as DEVELOPMENT FACTORS are actually cumulative development
factors, ie they apply from the development year of the column they are in up to Development
Year 5. For example, the figure of 1.290 is calculated as:

1.158 1.049 1.039 1.023 0.999 1.000 1.290

So, if we were to use the Basic Chain Ladder method to estimate the cumulative claims incurred
by the end of Development Year 5 in respect of Accident Year 6, we would obtain:

5,818 1.290 7,505

Question

Confirm the figures of f 1.062 and r 1.039 for Development Year 3.

Solution

The figures are:

12,801
r 1.0394
12,315

and f 1.000 0.999 1.023 1.039 1.0618 1.062

Next, the expected Ultimate Loss Ratio, say 83%, is applied to the earned premium (EP) to
give the initial estimate of the total ultimate loss (UL) for each accident year.

This figure of 83% (or whatever) would normally be derived from a different source from the data
in the triangle.

Initial estimate of total ultimate losses, by accident year:

AY 1 2 3 4 5 6

EP 4,486 5,024 5,680 6,590 7,482 8,502

UL
3,723 4,170 4,714 5,470 6,210 7,057
(0.83 EP)

Again, the figures for earned premium are ‘new data’ that would be derived from a separate
source.

Note that in this example, the expected loss ratio has been taken as that experienced for the
fully developed first accident year. This has been done due to lack of other information.

In other words the figure of 83% we are assuming has been estimated by dividing the ultimate
claims incurred for Accident Year 1 (ie 3,717) by the earned premium for Accident Year 1 (ie
4,486). If we knew that the claims experience was likely to be different for the other accident
years, we would use different percentages for the other years.

The next stage is the application of the development factors to the estimated ultimate
losses and the addition of the incurred claims that have already been reported.

The Actuarial Education Company IFE: 2019 Examinations

Page 42 CM2-21: Run-off triangles

Note that the accident years are in reverse order in the following table.

Revised estimate of total ultimate losses, by accident year

AY 6 5 4 3 2 1

f 1.290 1.114 1.062 1.022 0.999 1.000

1 – 1/f 0.225 0.102 0.058 0.022 –0.001 0

Initial UL 7,057 6,210 5,470 4,714 4,170 3,723

Emerging liability 1,588 633 317 104 –4 0

Reported liability 5,818 6,142 5,676 4,946 4,319 3,717

Ultimate liability 7,406 6,775 5,993 5,050 4,315 3,717

The emerging liability is calculated by multiplying the initial UL by the corresponding value of
1 1 f . The reported liability for a particular accident year is the last known figure in the run-off
triangle for that accident year. The ultimate liability is the sum of the emerging liability and
reported liability.

The total ultimate liability relating to these six accident years is, therefore, 33,256.

If the claims paid to date amounted to 20,334 (the same figure that we used before), the total
reserve required would be 12,922.

If you are having some difficulty getting to terms with the method, it may be helpful to look at the
procedure from a slightly different point of view.

(i) The earned premium for Accident Year 2 is 5,024 (given).

(ii) The initial expected ultimate loss amount for Accident Year 2 (ie before combining with
the development information) is 83% of this, ie 4,170.

(iii) If the ultimate claim amount was 4,170, then we would expect to have paid out so far
1 1
4,170 4,174 (ie initial UL ).
0.999 f

(iv) So that would mean we would have 4,170 4,174 4 to pay out in the future (ie we
expect a rebate of 4 next year).

(v) We have actually incurred so far 4,319 (from the triangle).

(vi) So we would expect to pay out 4,319 4 4,315 in total.

This is our Bornhuetter-Ferguson estimate for the total claim payment on claims arising in
Accident Year 2.

IFE: 2019 Examinations The Actuarial Education Company

 CM2-21: Run-off triangles Page 43

Question

Apply the same logic to the amount needed for Accident Year 3.

Solution

(i) The earned premium for Accident Year 3 is 5,680.

(ii) The initial expected ultimate loss amount for accident Year 3 is 83% of this, ie 4,714.

(iii) If the ultimate claim amount was 4,714, then we would expect to have paid out so far
1
4,714 4,613 .
1.022

(iv) So that would mean we have 4,714 4,613 101 to pay in the future.

(v) We have actually incurred 4,946 so far.

(vi) So we would expect to pay out 4,946 101 5,047 in total.

This is our Bornhuetter-Ferguson estimate for the total claim payment on claims arising in
Accident Year 3.

Our figure differs slightly from the 5,050 shown in the table because of rounding.

The same approach is used to get the figures for Accident Years 4, 5 and 6.

Ultimately we add up the expected outgo for the whole period, and subtract what we have
already paid out.

5.4 Assumptions underlying the method

Again the assumptions depend on whether the original or an amended version of the
method is being used.

The ‘original’ method is the simplest form, which was outlined at the start of this section.

For the original method, the underlying assumptions are the same as for the basic chain
ladder method, together with the assumption that the estimated loss ratio is appropriate.

The Actuarial Education Company IFE: 2019 Examinations