R. E.

BEARD
(A paper discussed before the Society on 6 March 1959)
INTRODUCTION
JUST over four years ago I had the pleasure of presenting a paper
to the Students' Society on some aspects of non-life insurance
{J.S.S. 13, 139) and I would first like to thank the Society for again
giving me an opportunity of putting on paper some further ideas
which have largely flowed from my interests in the non-life field.
Frankly this paper is of an exploratory nature and suffers from all
the defects of reporting continuing experimental work. I hope
therefore that the many obvious loose ends will be viewed in this
light and that members will be able to sense the excitement of this
relatively undeveloped field.
The purpose of this paper is not to provide a guide through the
various papers that have appeared relating to the three related
topics of retention, reinsurance and risk but rather to provide an
elementary approach to these problems with the object of stimu-
lating interest in some aspects of actuarial practice which are
largely regarded as lying in the realm of 'experience judgment' but
which are capable of precise formulation against a background of
mathematical statistics.
At the outset I would mention that the problems discussed in
this paper are of far greater significance in non-life branches of
insurance than they are in the life field, but the differences are of
degree and not of principle. I have, however, approached the
subject from the life side so that the nature of the problems will
be more readily appreciated. Furthermore, the study of basic
principles will, it is hoped, be of some value in directing attention
to some of the more subtle aspects of life insurance finance referred
25-2
THREE R'S OF INSURANCERISK,
RETENTION AND REINSURANCE
by
JSS 15 (6) (1960) 399-421
400 R. E. BEARD
to by the President in his address at the commencement of this
session (J.I.A. 85, 1).
In broad terms it can be asserted that the technical processes of
actuarial calculations, e.g. premium calculations and valuations,
are based on mean values derived from a deterministic model. The
actuary approaches his problem from the basis that he is dealing
with large aggregates whose future behaviour follows the pattern
of a mortality table of his own selection. He recognizes that there
will be fluctuations, both random and systematic, of the actual
experience from that implied by his adoption of a particular set of
assumptions and he introduces margins, largely determined by
experience, to cover such fluctuations. He also relies on periodical
tests to ensure that his margins are adequate.
It would not be inappropriate at this point to digress into a dis-
cussion on expenses, but although the expense question is critical,
as will be apparent later, it would mean moving a long way from
the main issues of this paper. I would however draw attention to
the lack of precision in much of this area and the almost complete
absence of scientific procedures as compared with some other com-
mercial enterprises. Thus a cost analysis of life business by year of
entry would provide some shocks and bring out some features
which have been obscured by the relatively slow decline in the
value of money in this country but which have given rise to serious
problems in those countries where the decline has been much
more rapid.
It is not necessary to discuss expenses in detail and I shall in
general restrict myself solely to the risk components of the business,
in other words it will be assumed that actual expenses are just equal
to the expense loadings provided.
There is however one aspect of expenses which is pertinent.
It can be reasonably assumed that there is an element of costs which
is independent of the size of case, whether it is measured against
the sum insured or against the premium. There is therefore a good
reason for writing as large a sum assured as possible on the grounds
that the relative expense component is thereby minimized. The
upper limit is provided by the fluctuation in surplus likely to be
caused in the event of claim and this at once leads into the region
THREE R'S OF INSURANCE 401
of tonight's discussion. A failure to set objective standards for
maximum net retentions will result in an unnecessarily high
expense ratio.
For most established life insurance companies random fluctua-
tions in claims experience are likely to be small relative to the other
factors involved so that the problem, although studied in the past,
has not attracted much recent attention in this country. For non-
life companies, however, the random fluctuations are frequently of
much greater relative significance and thus the problems assume
a much greater importance. Unfortunately, suitable statistical
material for study is not available in the U.K. and is relatively
scarce elsewhere. It is in this field that ASTIN is concerned in the
hope of assisting in the scientific analysis of the problems involved.
NATURE OF THE PROBLEMS
To bring the nature of the problems into clearer relief it is con-
venient to study a very simple insurance process, namely, business
consisting wholly of one year temporary insurances of similar
amounts on lives of the same age. Actual expenses are assumed to
equal the loadings and interest will be ignored. It will be assumed
that a policy of 100 sum assured is issued on 1st January to each
of 1000 lives aged 50. It will also be assumed that the expected
mortality rate of this group is q
50
by the Hypothetical Select Table,
i.e. .00700.
The risk premium to be charged for each case is .7 giving rise
to a total premium income of 7oo, exactly sufficient to meet the
7 expected claims of 100 each. However, due to random fluctua-
tion it is unlikely that exactly 7 lives will die in the year. If less
than 7 die a mortality profit will be made and if more than 7, a loss
will emerge.
If there were no other resources than the risk premiums available
then in the latter case the 'company' would be insolventruined
in the language of risk theory. Hence there is a need to have
reserves available to meet adverse fluctuations in experience and
we may call such reserves' capital'. Clearly the capital required can
be accurately expressed in a probability form, i.e. given the prob-
ability distribution of the expected claims, the probability that the
402 R. E. BEARD
actual claims exceed the risk premiums plus capital can be stated.
In other words, to a given amount of capital can be attached a' ruin
probability'.
Now since capital has to be serviced it is clear that the risk
premium must be loaded with an amount sufficient to meet the
service charge. It also follows that the amount of this loading is
related to the underlying risk distribution and to the level of ruin
probability adopted. The lower the ruin probability adopted, the
greater will be the loading needed to service the larger capital.
We are now comfortably in the region of the first of the three
' R's', namely, risk theory. The other two R's, retention and re-
insurance, are different facets of the same problem. If it be con-
sidered that the risk distribution is such that the ruin probability
is too high there are two courses open. One is to increase the
capital (assuming the premium loadings are adequate) to such a
level that the ruin probability is reduced to an acceptable level. The
other is to modify the risk distribution so that the ruin probability
(measured against the capital) is appropriate. This modification of
the risk distribution is the essential function of reinsurance.
Some interesting points arise from considering the net retention
from a risk standpoint and whilst it may be noted that the
'reduction' of the risk distribution is achieved in surplus line
reinsurance by cutting off the tail of the distribution by setting
maximum retentions for individual cases, the quota share form
achieves it by a proportionate overall reduction of all cases. The
non-proportional forms, e.g. excess loss and stop loss covers
achieve the same object, but have considerable advantages
expense-wise.
EXAMPLE
The simple case set out at the beginning of this section will be
used as the basis of numerical work. In Table I is set out the
probability distribution of claims assuming the lives to be
independent.
From Table I we see that the probability that 7 or less claims
arise in the year is .59871. The ' ruin' probability is thus .40129,
i.e. the chance is about -4 that the claims will exceed the premiums.
If it was decided that a ruin probability of .00092 was appropriate,
THREE R'S OF INSURANCE
403
the table shows that it would be necessary to be able to meet up to
16 claims; since premiums of 700 are available, capital reserves of
900 would be needed. If it be assumed that the ' service charge'
for the capital is 6% p.a. (adopted in this example purely for
illustration purposes) then an amount of 54 is required for
service, so that the risk premiums require a loading of 7.7 %. The
corresponding loadings for other values of the ruin probability
have been interpolated from Table 1 and are given in Table 2.
Ruin proby.
Loading %
(w=1ooo) 2
1.4
1
26
Table
05
3.4
2
01
5.4
005
6.1
001
7-6
0005
8.2
0001
9-6
The situation arising when reinsurance is introduced is now
considered. Assume that 1000 risks are written at a premium of
742 each, i.e. a loading of 6 %, and that capital reserves of 700
are available. From Table 1 we see the ruin probability is "00555;
the loadings (42) are sufficient to service the capital. The insurer
now decides that the ruin probability is too high and decides to
seek reinsurance. If he seeks a 50 % quota share arrangement on
original terms he will retain 350 risk premium plus loadings of
j2i. Reference to Table 1 shows that his ruin probability would be
reduced to -ooooi (since his total resources of 1050 will suffice to
meet up to 21 claims) but he cannot properly service his capital
Table I. Probability of n claims = () oo7" (.993)
1000
-
n
No. of
claims
0
I
2
3
4
5
6
7
8
9
IO
II
Proby.
00089
00627
02209
05179
09099
'12778
14937
'14953
'13083
10167
07101
04506
Cum. proby.
00089
00716
02925
08104
17203
29981
44918
5987I
72954
83121
90222
94728
No. of
claims
I2
13
14
15
16
17
18
19
20
21
22
Proby.
02618
01402
00697
00323
00140
00058
'00022
00008
00002
00001
00001
Cum. proby.
'97346
98748
99445
99768
99908
99966
99988
99996
99998
99999
1'ooooo
404
R. E. BEARD
since he has only 21 loadings available against the 42 assumed
to be needed. To service his capital properly it would be necessary
for him to retain all the original loading; this could only be done
at the expense of the reinsurer.
Of course, it may happen that the reinsurer by virtue of a
greater spread of business would be prepared to accept a lower
margin in the premium he receives. But unless the reinsurer was
prepared to forego all loading, merely offloading a quota share will
not provide enough margin to service the capital in this example.
If, on the other hand, the reinsurer agreed to pass to the company
a further 1000 cases of 50 each in reciprocation for the business
ceded then the company would have enough loading to service the
capital and his ruin probability would be reduced as a calculation
on the basis of 2000 cases shows. The relevant figures are set out
in Tables 3 and 4.
Thus a 6 % loading on 2000 cases of 50 each will amount to
42, sufficient to service a capital of 700; premiums will amount
Table 3. Probability of n claims = (
2
) '007" (993)
2000
-n
No. of
claims
I
2
3
4
5
6
7
8
9
10
11
I2
13
14
I5
16
17
Proby.
00001
00008
00037
00130
00366
00856
01721
03023
04715
06618
08440
09862
10631
10636
09927
08682
07142
Cum. proby.
00001
00009
00046
00176
'00542
'01398
03119
06142
10857
I7475
25915
35777
46408
57O44
66971
75653
82795
No. of
claims
18
19
20
21
22
23
24
25
26
27
28
29
30
3I
32
33
Proby.
O5547
04079
02848
01893
01201
00727
00423
00235
00126
00066
00032
00015
00007
00004
00001
00001
Cum. proby.
88342
92421
95269
97162
98363
99090
99513
99748
99874
99940
99972
99987
99994
99998
99999
I'ooooo
Ruin proby. (n
Loading %
= 2000)
I
2
I
' I
19
5
2'5
0I
3'8
005
4'3
001
5'3
0005
5'7
0001
6'6
Table 4
THREE R'S OF INSURANCE
45
to 700 so that a total of 28 claims could be met. The probability of
this or a lower number is .99972 so that the ruin probability is
00028, only one-twentieth of the value when 1000 cases were
involved.
The position can be looked at from another angle. Assume that
the direct writing company has capital of 700 on which the
service charge is 42; assume that the expected writings are 1000
cases. It is decided that a ruin probability of .00012 is acceptable.
Table 1 shows that this probability is reached at 18 claims and that
capital reserves of 1100 are needed, calling for a loading of 9.4 %
or 66. If business is obtained at this rate and the full amount
retained the ruin probability would be .00555, rnuch higher than
desired, but the loading of 66 would be more than adequate to
meet the service charge of 42 on the actual capital of 700. If
4/11 of the business was reinsured on a quota share basis the
required conditions would just be met.
Instead, however, of reinsuring on a quota share basis, the direct
insurer could decide to reinsure on some kind of excess basis. To
maintain his desired ruin probability while retaining the full
amount of each case, would mean obtaining cover for the excess
claims over his resources up to 18 claims. The theoretical net risk
premium may be obtained by noting that if P is the premium, the
total risk premiums + capital resources are 1400 P, so that the
reinsurer has to provide P if there are 14 claims, 100 + P for 15
claims, 200 + P for 16, 300 + P for 17 and 400 + P for 18
claims, 500 for 19 claims, 600 for 20 claims and so on. (To
ensure consistency in the calculations it is necessary to adopt a
situation in which the cedant has just enough resources to meet the
net claims when the total number of claims lies between 14 and
18 inclusive.) Using the figures in Table 1 we find P = -944.
The remaining question is how much of the loading of 66 should
be passed on to the reinsurer? The probability that the total
number of claims is 19 or more is .00012, when he has to pay at
least 500. If the total claims are 18 or less he will have sufficient
to meet his commitments if his capital is 400, hence to meet the
same conditions as in the quota share example, the loading passed
on should be 24.
406 R. E. BEARD
This example is, of course, highly theoretical, but the important
thing to notice from it is that whilst in the quota share case the
'service charge' amounted to only 9.4% of the risk premium
ceded, in the excess case the 'service charge' amounted to over
25 times the risk premium ceded.
By way of a finishing comment on this section it is of interest to
note the different structure of the ruin position in the quota share
and excess cases. In Table 5 the balance available in the event of
various numbers of claims is set out for the two cases. In the excess
case there is a sequence of years in which the company is ' nearly
ruined', a reminder that the spread of the components of the ruin
probability may on occasion be of significance.
No. of
claims
o
i
2
3
4
S
6
7
8
9
Proby.
00089
00627
02209
05179
09099
12778
14937
14953
13083
10167
Table 5. Balance of resources
Balance
, " , No. of
Q.S. X.S. claims Proby.
1145 1399
1082 1299
1018 1199
955 i99
891 999
827 899
764 799
700 699
636 599
573 499
10
11
12
13
14
15
16
17
18
07101
04506
02618
01402
00697
00323
00140
00058
00022
Balance
Q.S. X.S.
59 399
445 299
382 199
318 99
255 0
191 0
127 0
64 0
0 0
FURTHER DEVELOPMENT
The previous part of this paper has dealt with the analysis of a
simple portfolio in essentially theoretical conditions. I t is hoped
that the critical underlying principles have emerged and that the
basis for further development is secure. The position may be
recapitulated in that we have shown that a relationship exists
between the net retention, the reserves and premium loadings, this
relationship being expressed in probability terms. Thus, if Z is the
total amount of claims, u the 'free' reserves, P the net risk pre
mium and a the ruin probability we have
Pr{Z > u + P} = .
THREE R'S OF INSURANCE 407
Clearly we can immediately extend the model to cover the case
when the sums at risk are no longer constant by calculating the
probability distribution of expected claims. Following the analysis
given in J.S.S. 13, 141, if there is an aggregate of N
1
identical risks
each subject to a probability of p
1
that a claim arises in a year and
such that the proportion of claims falling between a and a + da is

1
(a)da then the moment generating function of the total amount
of claims arising is
If we now consider a mixed portfolio consisting of N
1
N
2
, ..., risks,
subject to claim probabilities p
1
,p
2
, , and claim distributions

1
(a),
2
(a),..., the m.g.f. of the whole portfolio will be given by
the product being taken over all the different groups. By taking
the logarithm we get the cumulant generating function, namely
whence
etc.
If a Poisson distribution be used instead of a binomial the
moments of the distribution of total claims would reduce to
from which it will be seen that the moments are based on products
of expected numbers of claims (N
r
p
r
) and moments (about zero) of
the distributions. The expected claim rate (p
r
) enters only in
directly (unless it happens to be high), a feature which is of use in
sorting out questions such as the attitude to be adopted to, say,
408
R. E. BEARD
insurances on old lives. It is also useful to note that the usual
measure of skewness and kurtosis may be written in the form
where the m's lie between the extreme values taken by these
functions.
The purpose of this very general treatment is to sort out the
principles upon which a net retention policy should be based. I t
sets out the theory behind the combination of claims observations
to build up the total claim distribution for the portfolio being
considered. The mean is given by the average of the observations
and the variance by the sum of the variances of the respective
strata. The approach to normality, needed to decide whether a
normal curve can be safely used as an approximation, follows from
the and
1
and
2
estimates. Once the moments of the distribution of
total claims are known the calculation of the one year ruin
probability for the appropriate free reserves can be made. If this
is too high consideration can then be given to the effect of different
reinsurance methods until the ruin probability is reduced to the
desired level.
In calculating the ruin probability allowance will have been
made for the risk premium receivable and it is assumed that there
are sufficient loadings to provide the necessary service on the
reserves employed. Implicit in the assumption is that the aggre
gate of risk premiums is equal to the expected value of total claims.
Suppose now that it is found that the maximum claim on any one
case should be limited to say 10,000 to produce the desired ruin
probability how should detailed underwriting instructions be
built up?
On the basis that the distribution and level of business will be
the same as that on which the experience is based, and that risk
premiums are fair, then clearly the instruction should be to retain
the maximum amount on each case, subject to an upper limit of
10,000. By retention we refer to 'probable maximum claim' and
assume that a lower limit has been fixed on expense considerations,
involving special premium treatment. However, for many reasons
the
1
and
2
estimates. Once the moments of the distribution of
total claims are known the calculation of the one year ruin
probability for the appropriate free reserves can be made. If this
is too high consideration can then be given to the effect of different
reinsurance methods until the ruin probability is reduced to the
desired level.
THREE R'S OF INSURANCE
409
risk premiums may not be fair (competition, accommodation, etc.)
and it is important to see what variations should be applied. First,
it must be assumed that business is not accepted at 'under claim
cost', i.e. that the expected claims exceed the risk premiums. If
for special reasons this is done then (theoretically) the shortfall
in risk premium should be charged to an expense account appro-
priate to the special reason. On this basis, no change in retention
policy is called for.
Regard must however be given to the 'service loading'. If this
is too low then there will be insufficient money to service the
capital employed. If it is too high there is a danger that business
growth will be inhibited. It is, however, implicit that the service
loadings are in the aggregate sufficient for capital needs and it may
well be that the loading in different cases may be considered out of
line. Thus suppose that for the conditions of the company the
service loading required overall is 5 % of the expected risk pre-
miums and it is considered that for 20 % of the business the margin
will be zero, for another 20% the margin is 71/2% and for the
remaining 60 % the margin is 5 %. On this distribution the overall
margin will be 41/2% only, but if one-half of the first group is
reinsured the margin will rise to 5 %, but on a total of business
reduced by 10 %. If therefore the business budgeting is raised by
10% then the total premiums after reinsurance will provide the
required amount for service.
It is assumed that the pattern of business is largely dictated by
factors beyond the control of the underwriter so he has to devise
rules to balance the portfolio. In this very simple example he does
this by fixing his retentions for the substandard class at 50 % of the
standard so that the adjustment is secured automatically. Theo-
retically this gives rise to a change in his expected claim distribu-
tion by a slight shift towards the lower end and thus the conditions
would permit a slightly higher standard retention. However, since
the critical aspect, namely the judgment of margins in the pre-
miums, cannot be precisely defined, there is little justification for
fine adjustments.
The problem of net retentions then reduces to (1) determination
of the maximum net line, and (2) a pattern of reductions from this
410 R. E. BEARD
maximum to secure the necessary 'service' loading. Clearly for
a given portfolio there are an unlimited number of ways of adjusting
for (2) depending on the relative amounts of business in the various
loading groups. I t is therefore necessary to set up some criteria so
that a specific solution may be obtained.
Let it be assumed that the reserves are u, risk premiums nP and
that the risk distribution may be assumed to be normal (i.e. n large).
Then we may set up the following approximate equation
u = (nP),
where a is a factor which includes the multiple of the standard
deviation corresponding to the desired ruin probability and t he
factor {m
2/
m
1)
1/ 2
. If A is the loading and r the service rate on the
reserve items we then have ru nP. If the expected loading
(
1
say) is less than A then it will be necessary to write more cases
to provide enough loading to service u. However, the effect of
writing more cases is to increase the premiums and the standard
deviation so that the basic risk equation will no longer be satisfied.
If, however, the amount per case is reduced, then a counteracting
factor can be introduced into the S.D. and a solution found which
balances the equation. Thus if k be the reducing factor, the pre
mium required being nP/
1
the number of cases must be increased
from n to n |K
1.
The S.D. then becomes {n Pk
2
/ k
1
)1/ 2
; since this
has to remain constant at (nP) we find that k =
1
/ ..
This simple and, of course, approximate result applies to a single
type of risk and leads to a rule that the net retention should be
proportional to the relative loading in the premium. To extend the
calculation to the case where a mixed portfolio is concerned, some
conditions must be imposed for a solution to be obtained. We
assume that the proportions of gross business available in the
different loading groups are fixed and have to find a system of
reduction factors for the lower loading groups to provide an overall
average loading factor equal to that required having regard to the
ruin probability and the reserves available. We must then increase
the total net premiums to the level required to provide the amount
of loading required. Finally we must increase the number of cases,
and reduce the sum insured per case to bring the standard deviation
THREE R'S OF I NSURANCE 411
to the required level. To get a unique solution we can then impose
some kind of maximum or minimum condition and a convenient
one is to minimize the total loading with respect to the reduction
factors.
Thus, consider the position of a mixed portfolio in which there
are n
r
P
r
premiums with a loading of A,.. We make the assumption
that enough business is written (from a specified population) t o
provide sufficient loadings to meet the service charge on reserves
of u. We then have for the basic equation u = ( n
r
P
r
) and
ru = n
r
P
r
. The total loadings on net premiums are
and the number of cases must then be increased by a factor
The revised S.D. is then
which has to be equal to ( n
r
, P
r
). We then find
or
Clearly one solution of this is k
r
=
r
/ for all r. By the use of
Lagrange multipliers and the condition that
r
n
r
P
r
k
r
is a maxi
mum the same solution will be found.
One other aspect of the problem needs consideration before the
results can be properly summarized, namely the position of an
expanding business. The foregoing analysis is built up on the
assumption that the premium loadings are just sufficient (on the
average) to provide service for the capital reserves, which are them
selves sufficient to produce a ruin probability of specified amount.
If the business expands there will be an increase in the standard
deviation and thus a rise in the ruin probability and there will also
be a rise in the available loadings. Various theoretical ways of
meeting the position are available: (a) the capital can be increased,
(b) the reinsurance can be adjusted, and (c) the premium can be
L = T,\ n
r
P
r
k
r
412 R. E. BEARD
loaded to provide the necessary increase in reserves, (a) will not
be discussed in the present context as being a method required in
exceptional circumstances and (i) will be discarded as implying no
growth in the retained account; we are then left with (c).
If there is an increase of h% in the premiums the standard
deviation will rise by a factor of approximately (1+h) = I +1/ 2h
and the reserves should therefore be increased by 1/ 2hu to maintain
the ruin probability level. This will require an increase in the service
charge of 1/ 2hur and as the increase in the loading is h nP = hur
there is a balance of 1/ 2hur available towards the required reserve
increase. This leaves a shortage of 1/ 2[hu(1 r)] which has to be made
up by a premium loading. The rate of this loading is thus
We thus see that to finance an increase of h % in premium income
there must be a loading in the premium of 1/ 2h multiplied by a
factor depending on the relative levels of the capital reserves and
the premiums.
In the previous paragraphs, and in fact throughout the paper,
it has been assumed that the loading
r
can be assessed in some
way. Formally, for non life insurance it is a question of knowing
the risk premium rate for a particular class of business and the
properly costed expense for this class. The margin then gives an
estimate of the loading available. To determine the risk premium
rate is not an easy statistical problem when regard is had to the
nature of the variation of the claims, but this is a question in itself.
The expense element should be easier, but in fact there appears
to have been very little in the way of true cost analysis of insurance
business and a whole area of research is available.
The problems of life business are complicated by the long term
nature of the contracts and the fact that other factors are of more
commercial significance than the pure risk factor. However, for
new companies the problem is not without significance and it would
not be without value to develop the ideas for a life portfolio.
It will be noted that in the foregoing analysis no allowance has
been made for interest which may be earned on the 'reserves' and
THREE R'S OF INSURANCE
4I3
which will be available towards the service charges. Theoretically
the reserves and premiums should be held in cash if the conditions
imposed are to be met, so that interest earnings will be small
relative to the other factors involved. If there is any departure
from this then there might be a capital depreciation at the time
when the risk reserve was required to meet an exceptional fluctua-
tion in claims. To meet the originally specified ruin probability it
would then seem necessary to start with a larger reserve, based on
the expected maximum depreciation. This would call for a larger
service charge which could be regarded as made up of the premium
loadings, plus the interest earnings on the invested reserves.
This approach leads to the idea that if reserves are not invested
'dead short' it would be reasonable to use an appropriately
lower figure for u in the risk equation and ignore the interest
earnings.
All the foregoing has been based on the assumption that we are
concerned only with a I-year business and it is desirable to examine
the position where a continuing business is concerned. Having
given the distribution of total net claims, the initial reserve and the
premium income (it being assumed that the loading is appropriate)
the probability of not being ruined by the end of the tth year can
be written down on the basis of a constant premium income. If
this probability was too low, the underwriting policy would need
adjustment (e.g. by reducing the maximum probable loss) until
it was considered that a desirable level had been reached. The net
retentions would then be reviewed in the light of this revised
policy, and its effect on the claim distribution. It will be noted
that basing the underwriting policy on a t-year ruin probability
instead of a I-year affects only the maximum acceptance and not
the relative levels of the various subclasses, which depend on the
relative loadings in the premiums.
Provided some idea of the relative values of the t-year and
I-year ruin probabilities are available there is no need to calculate
the t-year values, it being sufficient to take a sufficiently low value
for the I-year ruin probability. Unfortunately the relationship is
complicated and does not seem to lend itself to elementary methods.
Thus if Q(x) be the probability that claims in a year do not exceed x,
26
ASS I5
414
R. E. BEARD
i.e. p(z)dz where p{z)dz is probability that claims between z and
z + dz arise we have:
Probability not ruined in year I = Q(u+P).
Probability not ruined in year 2
Probability not ruined by end of year t
and so on until
This gives the form of the required probability, but it is difficult to
make any precise statements about the value of w, ... z. Numerical
calculation would be feasible, particularly if an electronic computer
is used.
If p(z) = e
z
an explicit solution can be found for Q
t
i.e.
but other forms of p(z) do not lead to simple expressions.
An approximate value of Q
t
may be found by repeated use of
mean values, the method lending itself very conveniently to a
graphical method of calculation. Thus
where
where
etc.
THREE R'S OF INSURANCE
415
If values of Q(z) and m(z)
wp(w)dw
p{w)dw are calcu-
lated the various Q factors can be obtained successively.
Before concluding, it is proper to add a few words about the
reinsurance angle. Implicit in the principle of reinsuring the
business showing a low loading is the assumption that a reinsurer
may be found who is prepared to accept this business. This can be
justified on the basis that this is possible because the reinsurer has
a greater spread of business and so can operate with a lower loading
factor. I do not propose to analyse this aspect further but merely
make the comment that some nice practical problems can arise.
Discussion of the reinsurance position is a subject in itself, parti-
cularly the question of excess loss and stop loss where the loading
factor is frequently of far greater significance than the net risk
premium.
I have not provided a bibliography but there are two references
which are important background, namely Dubourdieu's Theorie
Mathematiques des Assurances (1952) and Thepaut, Bulletin de I'lnst.
Actuaires Francais(Dec. 1953). A fuller list of reference was provided
in papers read to ASTIN in New York by Depoid and de Finetti,
published in the ASTIN Bulletin, vol. 1, part 11 (December 1959).
CONCLUSION
As I stated at the outset, this paper is largely of an experimental
nature, and is an attempt to use the principles of classical risk
theory in the study of the problem of net retentions. It is original
in the sense that it has not been derived from other published work
on the subject, but having regard to the many continental papers
on risk theory it may well be that the ideas have been recorded
before. It is of course possible to start from Collective Risk
Theory and develop a theory of retention. To my mind, however,
there are certain features about the basic assumptions of Collective
Risk Theory that do not seem to fit in nicely with commercial
realities, and accordingly I have preferred to tread the classical
rather than the elegant path. It will be obvious that I have merely
scratched at the surface of the problem and I hope that others better
equipped than I may find the problems worth solving.
26-2
416 R. E. BEARD
APPENDIX
In this paper I have referred to various types of reinsurance which
are in common usage in the non-life field. This appendix has
therefore been prepared to give a very brief outline of the various
reinsurance arrangements so that the significance of the arguments
of the paper may be appreciated.
Fire insurance is historically the oldest form of non-life business
and it will be convenient first to sketch the development of rein-
surance facilities against the background of fire business. In this
business the amount of a risk retained for the insurers net account
is referred to as a net line. In the early days amounts in excess of
the company's net line were offered to other companies on a
facultative basis, each company specifying its acceptance in terms
of the ceding company's line.
As the business grew the facultative arrangements developed
into obligatory treaties under which the cedant passed his excess
lines (according to a schedule) to a treaty; reinsurers would then
take a share of this treaty, expressed in terms of lines or on a
percentage basis. This automatic arrangement clearly was a great
benefit to direct writers, and apart from convenience was a source
of some economies in operation. Facultative business was still
exchanged for cases in excess of the treaty capacity or for special
risks.
Such treaties, being based on the surplus of each risk over the
company's net retention, go by the general name of surplus
treaties. The normal practice was to list the cessions to the treaty
on bordereaux, copies of which were sent to the reinsurers with the
accounts and enabled the reinsurers to watch closely the risks
being placed on the treaty. It was common to look for accumula-
tions and make special arrangements to avoid excessive exposures
on individual risks.
Terms were fixed on the results of these treaties and the market
soon developed along the lines of looking for substantial profit
margins in these treaties. Between the wars the surplus treaty
flourished and exchanges between offices grew on the basis of
providing a wider spread of risks. However, the competition for
THREE R'S OF INSURANCE
417
better profit margins had its reaction and after the second world
war much greater attention became directed to the question of
'reciprocity'. Premiums ceded under surplus treaties amounted to
perhaps one-third of the direct writings and with the general
increase in competition gross profit margins became thinner and
the effect of such substantial cessions could not be ignored.
Furthermore, even the clerical work involved in treaty bordereaux
became expensive and various devices were developed to try and
economize in operation.
A number or interesting variations on the surplus treaty were
tried in France, these being essentially methods of cutting down
clerical work whilst still retaining the benefits of the surplus treaty.
A more significant development was however the movement to
excess covers of some kind. These covers had come into general
usage between the wars in connexion with catastrophe covers and
for the growing classes of liability insurance. Basically they provide
for the reinsurer to pay the excess of such claims as exceed a certain
figure in return for a premium expressed as a percentage of the
gross premium income.
Clearly such covers afford a considerable economy in operation
since it becomes necessary only to deal with the exceptional claims.
They have, however, given rise to a number of practical difficulties
which have to some extent inhibited their growth. In general terms
they are grouped under the heading of non-proportional reinsurance.
In excess of loss reinsurance the reinsurer agrees to pay the
excess of each and every claim in excess of an agreed amount. This
form is commonly employed in Motor insurance. There is a
statistical justification in that the frequency distribution of claims
by amount from a portfolio of business may be assumed to have
a basic underlying form. When this form is known the proportion
of claims falling in the excess area can be found and a rate deter-
mined for the treaty. The difficulties present are (a) the persistent
inflation of post-war years has meant that the scale of the claim
distribution has been changing so that rates have been behind
experience, (b) even apart from (a) the estimation problem is
difficult enough, (c) the loadings required to cover fluctuations are
substantial and not always appreciated.
418 R. E. BEARD
Various devices have been tried to get over some of these
difficulties. In many treaties an index-linked clause has been used
with some limited success against inflationary tendencies. In one
other treaty an attempt was made to link the excess figure to a
quantile of the claim-distribution curve; unfortunately other
features of this treaty were unsatisfactory. Obviously treaties in
which the excess varies are not completely satisfactory because
neither party can be certain of the position at any point of time.
Nevertheless, such treaties have a great attraction because of the
cost saving and are slowly growing in application in spite of the
handicaps.
A later development has been to develop the so-called 'stop
loss' treaties which are closely linked with' aggregate excess' covers.
If it be assumed that the claim distribution is known then the
probability distribution of the total amount of claims over a total
figure can be found. In the aggregate excess cover the reinsurer
agrees to pay claims over the agreed total. In the stop loss cover the
agreement is to pay in excess of a given claim ratio. Either of these
may be expressed in terms of a fixed amount (i.e. the cover placed
in layers) or in terms of a ratio. The premium estimation problems
will be apparent.
Statistically speaking the estimation of the tail of the claim curve
is similar in the case of individual claims or of the claim ratios
(J.S.S. 13, 139). However, there is a certain amount of 'built in'
protection from inflation in the claim ratio since there will be a
tendency for sums insured to be revised and premiums adjusted
accordingly. In many cases the application of the average clause
will secure the same effect. If average were applied universally,
the protections would be better. The main disadvantage is, of
course, the fact that until the results are fully known the true
relationship of the parties cannot be determined.
To complete this very brief outline it is necessary to mention
those treaties which are based on 'burning cost'. The premium is
based on the actual claim experience on perhaps a 5-year average
with a suitable loading. Clearly if the treaty continued indefinitely
it would amount to no more than a financing arrangement between
the parties. A measure of true reinsurance is however introduced
THREE R'S OF INSURANCE 419
by the fixing of maximum and minimum premium rates. Statisti-
cally speaking the problem of rate fixing is formally determined
when the claim distribution is known.
Many variations are introduced into these basic forms of rein-
surance and different combinations have been devised in attempts
to meet practical demands. Suffice to say here that one of the main
features is the general tendency to treat the treaty result as a whole
and reduce to an absolute minimum any exchange of detailed
information about the finer structure.
One of the problems which lies at the back of all this is, of
course, the incidence of taxation. Briefly, tax is assessed on the
results of a year's working in which the outgoing reserve is expressed
(in the U.K.) as a percentage of the premiums written in the
year. Claims are estimated as closely as can be, subject to a safety
margin. Since, however, the results of the business fluctuate from
both secular trends and random fluctuations, the method of asses-
sing profit tends to encourage insurers to iron out fluctuations to
as great a degree as possible. For example, if a catastrophe cover is
written, and the event happens, the year in which it arises will show
a substantial loss to be met by premiums over many years. These
premiums will fall into profits as they arise and be taxed at the full
rate so that unless other measures exist for spreading the incidence
of the large loss, the company is at a marked disadvantage.
This feature has been recognized in some countries where the
fiscal authorities are now prepared to recognize a calculated risk
equalization reserve. (In the U.K. some limited recognition has
been given in regard to Lloyds underwriters.) The mathematical
problem is clear and is based on developing for non-life insurance
a reserve position parallel to that developed by the actuary for life
business. At present, however, the technical processes involved
are still in course of development and existing actuarial techniques
are unsuited for the particular problems involved.
420 R. E. BEARD
Subsequent to the discussion some members raised points in corre-
spondence with the author, who has replied as follows
In answering a question from Mr W. J. Courcouf it would seem
that the position would be clarified if Table 5 were extended as
follows:
No. of
claims
19
2 0
2 1
2 2
Proby.
00008
00002
00001
00001
Q.S.
- 6 4
- 127
- 191
- 2 5 5
X.S.
-P
-P
-P
-P
Balance
Mr Courcouf also raised some points on the general interpretation
of the first part of the paper and his own comments adequately
cover the position:
'...it might give us a clearer understanding of the position and
the analysis, if we had regard more to the total market than to
the individual companies serving it. I think that the moral
of this part is that if there is 'maximum' reciprocity, the
ruin probability scale and the consequent loadings should be
based on the total market. If there is no reciprocity the smaller
companies cannot afford to write at the same rates as large ones.
If two companies separately decide on a ruin probability of
0012 for each, and the market consists of 2000 cases to be
written, they may hope for 1000 cases each and therefore employ
a loading of 9.4 %. If in fact they get 1500 and 500, then the
loading will have been too much in the one case and too little
in the other, and the one with too little cannot rectify the position
by insurance without reciprocity. If there is reinsurance with
reciprocity, then the smaller office can just rectify the position
it would have 1000 cases at 50 and the larger company would
have 1000 at 100 and 1000 at 50. If reinsurance were effected on
the basis of obtaining an equal share in each case in the market,
then the smaller company would give off 75 % of each it obtained
and receive 25 % of each of the larger company's cases, and this
is what I would define as "maximum reciprocity".'
As a result of Mr S. Benjamin's inquiries on the appropriateness
THREE R'S OF INSURANCE
42 1
of the relation = ru a subtle but interesting point emerged in
that some care is needed in specifying the premium basis if con
sistent mathematical results are to be achieved. The risk premium
is defined as the expected value of claims, i.e.
zp(z)dz. However,
if the business is reinsured in such a way that there is a risk of ruin
by the reinsurer, then to the extent that the cedant cannot obtain
payment from the reinsurer (because the latter is ruined) the
premium is too large. The true risk premium should thus be
P =
zp(z)dz + z

p{z)dz,
where z

is the limit beyond which the reinsurer cannot pay.