Introduction to Actuarial Science:

Elementary Life Contingencies

I. L. MacDonald

1 Introduction
1.1 Payments contingent upon survival
So far, most or all of the payments that you have had to value in this course have been assumed
to be certain; they do not depend on the occurrence of any uncertain event. But many of the
payments made by or to a life assurance company or pension fund are contingent upon (i.e.
depend on) the survival or otherwise of a specified person (or persons), to a specified date, as
a member of some group or category. Death, withdrawal and retirement are all possible causes
of non-survival (as a member of a group), i.e. decrements. For instance, an active member
of a companys pension fund may cease to be an active member of the fund by dying, by
resigning from the company and thereby withdrawing from the fund, or by retiring. And this
last possibility can itself be split into age retirements and ill-health retirements.
Here are three examples of cashflows which involve uncertainty. The sketches depict the
cashflows from the point of view of the annuitant or policyholder. (How would you alter the
sketches to represent the point of view of the life assurance company that pays the annuity or
provides the insurance?)

1. An annuitant receives an annuity payment of R100 on the first of each month provided he
or she is still alive on that date.
+R100 +R100 +R100 +R100 +R100 +R100
-
death 6
1 2 3 4
time (yrs): 0 12 12 12 12

2. A whole-life policyholder pays a premium of RP on 1 January of each year if he or she is

alive, and the sum assured of RS is paid to the estate at the end of the calendar year of
death.
P P P P P P +S
-
time (yrs): 0 1 2 3 4 K 6K + 1
death
 2 I.L. MacDonald: BUS1003H, Elementary life contingencies

3. An endowment assurance policyholder initially aged 40 pays a premium of RP on the first

of each month for 20 years (if alive), and the sum assured of RS is paid immediately on
death, or on survival to age 60, whichever is earlier. One of the following two sketches will
apply:
P P P P P +S
-
time (yrs): 0 1 2 3
6 20
12 12 12
death

P P P P P +S
-
time (yrs): 0 1 2 3
11
19 12 20 6
12 12 12
death

We therefore need to be able to value payments depending on (inter alia) the future lifetime
of a person or persons. That lifetime is of course highly uncertain, and in our calculations we
somehow need to take account of the uncertainty. The tool we use is some probability theory,
which I shall therefore spend a little time on, before returning to the issue of putting a value
on payments which are contingent on such events as survival or death. You may have met some
probability theory already in STA1006S, and you will eventually do a lot more; probability is
the most important mathematical tool used in actuarial science and financial mathematics.

1.2 Curtate and exact future lifetime

In this course we shall frequently refer to the curtate future lifetime, i.e. the number of completed
years of life that a specified person survives from now. We denote it by K, or by Kx if we wish
to emphasize that the person concerned is now aged x. For instance, if someone takes out a life
assurance policy and dies 5 years and 9 months later, we say that the curtate future lifetime,
from the inception of the policy, is K = 5. (The exact, as opposed to curtate, future lifetime
is 5.75 years.) We can describe the curtate future lifetime as the exact future lifetime rounded
down, or as the integer part of the exact future lifetime. The notation [. . . ] is often used for
integer part of, e.g. [5.75] = 5, but another notation you might meet is the floor notation
. . ., which means the same thing.

1.3 Need for expected present values

To motivate the study of probability theory a bit further, let us consider Example 2 above, a
whole-life assurance policy with annual premium RP and sum assured RS. Since the curtate
future lifetime of the policyholder is denoted by K, that means he or she dies between time K
and time K + 1. So how many annual premiums are paid? K + 1. When is the death benefit
paid to the estate? time K + 1. (Refer again to the sketch in Example 2 if you are unsure about
the number of premiums paid or the time at which the death benefit is paid.)
It follows that the present value (at effective rate i) of the premiums paid is:

P aK+1 i

and that the present value of the death benefit (at the same effective rate i) is:

S viK+1 .

But K is of course an uncertain quantity, a random variable, so this does not help us very
much if we want to work out what the premium P should be for S, the specified level of death
Probability 3

cover. We would at least like to know what is the average present value of the premiums that
will be received (the average over a large number of similar lives with the same kind of policy).
That is, we would like to know the average, or expected value, of P aK+1 , the present value of
the premiums. Similarly, we need the average of Sv K+1 , the p.v. of the death benefit.
We will in the next section formally define the expected value (for which we will use the
symbol E). If we can calculate the appropriate expected values we will then be able to find the
right premium P for a given sum assured S, by equating expected values as follows:

P expected value of aK+1 = S expected value of v K+1 ;

that is,
= S E v K+1 ,



P E aK+1
hence 
K+1



P =SE v E aK+1 .

Although equating expected present values of premiums and of benefits is not the only way of
determining premiums for life assurance policies, it is the traditional way, and the only way we
discuss in this part of the course. (I am simplifying somewhat by ignoring expenses, but that is
an aspect that will be considered later.)

2 Elementary probability
We confine ourselves here to discrete random variables, i.e. quantities X which can assume one
of a finite number of values (or possibly denumerable), and do so with probabilities that add
up to 1. (A denumerable set is one that can be placed in 1-to-1 correspondence with the positive
integers. For our purposes a probability is a number between 0 and 1 indicating how likely a
particular value of the random variable is.)

2.1 Examples
1. Suppose you roll a conventional 6-sided die, i.e. a die with 6 faces, marked with 1, 2, 3, . . . ,
6 dots. Define X as the score you get (the number of dots showing when the die comes to
rest). Each of the six possible values of X is assumed to be equally likely, and they have
total probability of 1 (i.e. there are no other possible values of X). Hence each such value
has probability 16 :

1
P(X = x) = for x = 1, 2, . . . , 6
6
= 0 otherwise.

(a) What is the probability that X 3? ( 23 )

(b) Suppose you know that X 3. What now is the probability that X = 6? ( 14 )

2. Suppose you toss a fair coin once, and define Y as the number of times you observe heads
(so Y = 0 or 1). Are these two values equally likely?

3. Suppose you toss a fair coin three times, and define Z as the number of times you observe
heads. What values can Z assume? Are these values equally likely?
In fact the eight possible outcomes HHH, HHT, HTH, HTT, THH, THT, TTH, TTT are
equally likely, so the values that Z can assume (0, 1, 2, 3) are not equally likely. What
are the probabilities associated with these four values of Z?
4 I.L. MacDonald: BUS1003H, Elementary life contingencies

4. To play a game involving dice you have to throw a six (on a fair die) before you can start.
Let K be the number of times you have to throw before you start. What are all the
possible values of K? Note that the set of all possible values of K is in this case not finite,
but it is denumerable.

2.2 Probability (mass) function

The function P(X = x), usually written p(x) or pX (x), is called P the probability function, or
probability mass function, of the random variable X. Notice that x p(x) = 1, where x runs
over all values such that p(x) > 0. We refer to the way in which the total probability of 1 is
divided among the various values X can assume, as the probability distribution of X. Note that
random variables are usually represented by capital letters.
Example A company buys a large number of similar machines at time 0. At times 1, 2, 3,
4, 5 these machines are carefully examined, and 5%/10%/15%/20%/25% of the original number
of machines are scrapped. All the remaining machines are scrapped at time 6. Define X as the
number of years until a particular machine is scrapped (the useful life). Then the probability
mass function of X is as follows:
x 1 2 3 4 5 6
P(X = x) 0.05 0.10 0.15 0.20 0.25 0.25

(a) Find the probability that the useful life of such a machine is at least 3 years (i.e. 3, 4, 5
or 6 years). (0.85)

(b) Find the probability that the useful life is at most 2 years (i.e. 1 or 2). (0.15)

(c) What figure do you think best represents the average useful life? (4.25?)

2.3 Mean of a discrete random variable

We define the mean, or expected value, of a discrete random variable X, as:
X
xpX (x)
x

the summation, as before, being over all x such that pX (x) > 0. Denoted X or E(X). Apply
this to Example 3 on p. 3, i.e. Z = no. of heads inPthree tosses of a fair coin. What is the
expected value of Z? (E(Z) = 3z=0 zpZ (z) = 0 18 + + 3 18 = 1.5.)

2.4 Mean of some function of a discrete random variable

In Example 1 on p. 3 (X = score on rolling a fair die once), we saw that:
1
P(X = x) = (x = 1, 2, . . . , 6)
6
= 0 (otherwise)

(a) P
Suppose some kind soul pays you Rx if you score x. Your expected gain is then E(X) =
x xpX (x) = 3.5, i.e. R3.50.

(b) Suppose some even kinder soul pays you Rx2 if you score x. What then is the probability
mass function of your gain, W = X 2 ?
1
P(W = w) = (w = 1, 4, 9, 16, 25, 36)
6
= 0 (otherwise).
Probability 5

P
The expected gain is E(W ) = Pw w pW (w) = 91/6, i.e. R15.17. This expected gain could
instead be written as E(X 2 ) = x x2 pX (x); you dont actually need to find the probability
mass function of W in order to find the mean of W .

More generally, it can (but wont here) be shown that for any function g(X) of a discrete
random variable X: X
E(g(X)) = g(x)pX (x).
x

This result is used so often that it is sometimes called The Law of the Unconscious Statistician.
(In (b) above, the function g(X) is X 2 .) If you take g(X) to be just aX + b, a and b being some
constants, you should be able to show that E(aX + b) = aE(X) + b.

2.5 Variance of a discrete random variable

We have defined the mean, or expected value, of a discrete r.v. X. We now define a measure of
2 ,
the variability, or dispersion, of X about that mean. The variance of X, written Var(X) or X
is defined as

X
E (X X )2 = (x X )2 pX (x).
x

The square root of the variance, X , is called the standard deviation of X. A convenient formula
for hand calculation of variances is:

Var(X) = E(X 2 ) 2X ;

that is,
X
x2 pX (x) 2X .


Var(X) =
x

(Proof omitted, but not difficult.)

2.6 More examples

1. Suppose you throw a fair die until you get a six for the first time. Define K as the number
of throws needed.

(a) What values can K assume?

(b) Find the probability that K = k.
6 I.L. MacDonald: BUS1003H, Elementary life contingencies

P k1
(c) Find the mean of K. (Hint: prove, and then use, the result that k=1 kx =
(1 x)2 if |x| < 1.)

2. Suppose you throw a fair die until you get a six for the second time. Define K as the total
number of throws needed.

(a) What values can K assume?

(b) Find the probability that K = k.

3 The curtate future lifetime as a discrete random variable

Consider a person now aged x years exactly, often denoted just by the symbol (x). Let K (or
Kx ) denote the curtate future lifetime of that person. This implies that death occurs between
Kx and Kx + 1 years from now (so the age at death lies between x + Kx and x + Kx + 1). Since
there is uncertainty in Kx , and it takes only nonnegative integer values (0, 1, 2, . . . ), we treat
it as a discrete random variable.
The life table is a traditional and convenient way of presenting (inter alia) the probability
distribution of Kx . An example of a life table, which we shall now use for illustrative purposes,
is English Life Table No. 12 (Males). It is based on the 1961 census of the population of England
and Wales (more precisely, the males in that population), and the male deaths which occurred in
Curtate future lifetime 7

those countries in the years 196062.1 The starting value of lx , which can be chosen arbitrarily,
is termed the radix of the table; for ELT 12 (Males) it is l0 = 100 000.
In any life table:
lx is the number of lives who are expected to survive (at least) to age x, out of the
specified starting number;

dx is the number expected to die between age x and age x + 1 (out of the starting
number, or equivalently out of the lx lives expected to survive to age x).

That is, the link between the quantities lx and dx is just:

dx = lx lx+1 .

In ELT 12 (Males), for example, l40 = 93 790 and l41 = 93 570, so d40 = 93 790 93 570 = 220.
Can we find from this table the probability that someone now aged 40 will die within one
year? (i.e. can we find P(K40 = 0)?) Yes,

P(K40 = 0) = d40 /l40 = 220/93 790 = 0.00235.

Similarly
P(K40 = 1) = d41 /l40 = 242/93 790 = 0.00258,
P(K40 = 10) = d50 /l40 = 656/93 790 = 0.00699, etc.
In general, then, the lx and dx columns contain all the information needed to specify the prob-
ability mass function of Kx , which is as follows:

P(Kx = k) = dx+k /lx (k = 0, 1, 2, . . .)

= 0 (otherwise).

The symbol qx is used to denote P(Kx = 0), i.e. qx is the probability that someone now aged
exactly x will die within one year, and can be calculated as dx /lx . Similarly px is defined to be
the probability that such a person survives at least one year: px = 1 qx .
We define also k px = P(Kx k) and k qx = P(Kx < k). What if we want to calculate k px ?
Either do it the hard way, adding up a large number of probabilities:

k px = P(Kx k) = P(Kx = k) + P(Kx = k + 1) + P(Kx = k + 2) + . . . ;

or note that Kx k means survival (at least) to age x + k from age x. Since the number
expected to survive to x + k (out of lx at age x) is lx+k , we therefore have:

k px = P(Kx k) = lx+k /lx (k = 0, 1, 2, . . .).

For example:
p40 = P(K40 1) = l41 /l40 = 93 570/93 790 = 0.99765;
3 p40 = P(K40 3) = l43 /l40 = 93 060/93 790 = 0.99222;
3 p50 = P(K50 3) = l53 /l50 = 87 868/90 085 = 0.97539.
Notice that with the above definitions 1 px and px are just the same thing: similarly 1 qx and qx
are just the same thing.
1
Although several of the life tables we use for illustrative purposes are very old, they have the advantage that
they appear in the old book of Tables for Exam purposes of the Faculty and Institute of Actuaries, which they
have made available free in electronic form to students. The relevant parts of that book are available to you at
the Vula site for this course.
8 I.L. MacDonald: BUS1003H, Elementary life contingencies

Exercise: On the basis of ELT 12 (Males):

1. What proportion (of the original 105 lives) are expected to die in the first year of life?
(q0 = 0.02449)

2. What proportion (of the original 105 ) are expected to die in the first two years of life?
Equivalently, what is (1 the proportion expected to survive the first two years of life)?
(1 2 p0 = 2 q0 = 0.02602)

3. What proportion of those reaching 40 are expected to survive to 42? (2 p40 = 0.99507)

4. What proportion of the original 105 are expected to die between 40 and 42?
(40 p0 2 q40 = 0.00462)

4 Expected present values of (contingent) annuities

4.1 Whole-life annuities
Suppose that (x), a life assurance policyholder initially aged x, pays a premium of RP , annually
in advance, as long as he or she is alive. So Kx + 1 premiums are paid, and the present value
of those premiums is P aKx +1 . But what we really need is the expected present value of those
premiums:

E(P aKx +1 ) = E(g(Kx )) where g(k) = P ak+1

X
= g(k) P(Kx = k)
k=0
= (P a1 ) P(Kx = 0) + (P a2 ) P(Kx = 1) + (P a3 ) P(Kx = 2) + . . .

This would be extremely tedious to calculate by hand, but for many life tables there are available
tables or computer programs or spreadsheets giving the quantity ax defined by:

ax expected p.v. of R1 p.a. in advance while (x) is alive.

So the expected p.v. of premiums that we require here is P ax , with ax looked up in tables.
Example: Suppose that (50) pays R100 p.a. in advance for life, and is subject to the
mortality of the table A196770 (ultimate).2 Suppose also that present values are calculated
here at the interest rate i = 0.04. What is the expected p.v. of these payments?

100a50 = 100 16.003 = R1600.30 approx.

Similarly, but for ages 30 and 90 respectively:

100a30 = R2106.10

100a90 = R352.40.
2
This is the life table based on the mortality experience of assured lives in the United Kingdom during the
years 196770. Ignore for the purposes of this course select mortality, and any symbols like l[x] , l[x]+1 , d[x] , a[x] ,
etc., which have the age x enclosed in square brackets to indicate that they refer to select mortality. We will use
only quantities like l40 , l41 , d40 , a40 , which refer to what is known as ultimate mortality; much simpler!
EPVs of annuities 9

4.2 Temporary annuities

Now suppose instead that policyholder (x) pays RP p.a. in advance while he or she is alive, but
for at most n years. In this case the number of premiums paid is Kx + 1, provided that number
does not exceed n, i.e. min(Kx + 1, n), the smaller of Kx + 1 and n.
Example: Suppose that the policyholder (62) pays premiums of R1 p.a. in advance while
he or she is alive, but for 3 years at most. Assume that (62) is subject to the mortality of the
A196770 (ultimate) table and i = 0.04. What is the expected p.v. of these premiums?
(Remember that at most three premiums are paid, annually in advance. So if the policyholder
survives 2 years or more, i.e. if K62 2, exactly 3 premiums are paid. It follows that in that
case the present value of premiums is R1 a3 .) Therefore (at 4%) the expected p.v. of the
premiums is:

a1 P(K62 = 0) + a2 P(K62 = 1) + a3 P(K62 = 2) + a3 P(K62 = 3) + a3 P(K62 = 4) + . . .

= a1 P(K62 = 0) + a2 P(K62 = 1) + a3 P(K62 2)
= 1 (d62 /l62 ) + 1.96154 (d63 /l62 ) + 2.88609 (l64 /l62 )
1
= (517.08729 + 1.96154 562.41852 + 2.88609 28052.632)
29132.138
= 2.83477

Fortunately this sort of thing is also tabulated! That is, we have tables of:

ax: n expected p.v. of R1 p.a. in advance while (x) is alive, but for at most n years.

For example, a62: 3 = 2.835 in tables based on A196770 (ultimate) mortality and 4% interest,
which confirms our calculation above.

4.3 Another way of calculating temporary annuity values

Consider again the above numerical example, and calculate (at 4%):

v 0 P(K62 0) + v 1 P(K62 1) + v 2 P(K62 2).

You should again get 2.83477. This is not coincidence. Can you prove that this way of calculating
ax: 3 always works? That is, can you prove that the following statement is true in general? (Hint:
start with the right-hand side, and write a1 , a2 and a3 as (respectively) 1, 1 + v and 1 + v + v 2 .)
2
X
v r P(Kx r) = a1 P(Kx = 0) + a2 P(Kx = 1) + a3 P(Kx 2).
r=0
10 I.L. MacDonald: BUS1003H, Elementary life contingencies

You will see in a later course that ax: n and ax can (for all x and n) be written as follows:
n1
X n1
X
X
X
ax: n = v r P(Kx r) = v r r px ; ax = v r P(Kx r) = v r r px .
r=0 r=0 r=0 r=0

vr
P
The sums r r px have a convenient interpretation, because r px is the probability that (x) is
alive to make the payment of R1 at time r, and v r is the present value of such a payment.

5 Expected present values of life assurance benefits, and deter-

mination of net premiums
5.1 Whole-life policies
Consider a whole-life policyholder (x), who takes out a policy now, aged exactly x years, and
whose estate will be paid R1 at the end of the (policy) year of death of (x). (For convenience
we take the sum assured to be R1 here, but in practice it would of course be much larger.) This
means that R1 will be paid at time Kx + 1. The present value of that payment is v Kx +1 , and
we seek the expected present value, that is E(v Kx +1 ). In principle we can calculate it as follows:

E(v Kx +1 ) = vP(Kx = 0) + v 2 P(Kx = 1) + v 3 P(Kx = 2) + . . . ,

but it also has been tabulated for many life tables, and is denoted by Ax . That is,

Ax expected p.v. of R1 payable at the end of the year of death of (x).

For instance, if a life aged 50 takes out a whole-life policy now, with the death benefit of R10 000
payable at the end of the year of death, the expected present value of that payment of R10 000
is 10 000A50 . If the interest rate is 4% and the mortality is that of the A196770 (ultimate)
table, the expected p.v. of the death benefit is 10 000 0.38450 = R3845.
We are now finally in a position to calculate what the net premium should be for such a
whole-life assurance, i.e. what the premium would be if we could ignore expenses completely
and base our calculations on:

expected p.v. of premiums = expected p.v. of the death benefit.

If the sum assured (i.e. the death benefit) under a whole-life policy is RS, and the premium p.a.
(in advance) is denoted by RP , the above equation becomes P ax = SAx , hence P = SAx /ax .
Often Ax /ax is tabulated, and denoted by Px .
Example (of whole-life policy): Consider again the life (50) subject to the mortality of
the A196770 (ultimate) table, who takes out a whole-life policy with sum assured of R10 000
payable at the end of the year of death. Premiums are payable annually in advance while (50)
is alive, and the relevant interest rate is 4%. What is the net premium payable per annum?

10 000A50 /a50 = 10 000 0.38450/16.003 = R240.27.

Using instead the tabulated value of P50 gives us 10 000 0.02403, i.e. R240.30.

5.2 Endowment assurance policies

All of the above relates to whole-life policies, but the quantities needed to do similar calculations
for endowment assurance policies are also tabulated for many life tables. In particular the
following value is usually available:

Ax: n expected p.v. of R1 payable at end of year of death of (x), or at time n if sooner.
Gross premiums, etc 11

Example (of endowment assurance policy): Consider again a life (50) subject to the mortality of
the A196770 (ultimate) table, who takes out an endowment assurance policy with sum assured
of R10 000 payable at the end of the year of death or at age 65, whichever is sooner. Suppose
that premiums are payable annually in advance while (50) is alive (but for at most 15 years).
The relevant interest rate is 4%. What is P , the net premium p.a.?
The expected p.v. of premiums is P a50: 15 , and the expected p.v. of benefits is 10 000A50: 15 .
Equate these two, and you get P = 10 000 0.57711/10.995, so the net annual premium required
is R524.88. (Notice that Px: n , defined as Ax: n /ax: n , is also tabulated, and use it to check this
answer.)
Why is the net premium for the endowment assurance (R524.88 p.a.) so much more than
the premium of R240.27 p.a. for the corresponding whole-life policy? (Look for two reasons.)

5.3 A handy link between Ax and ax : Ax = 1 d ax

For convenience we first repeat some notation. With K denoting the curtate future lifetime of
a life now aged x:


ax = E aK+1 = expected p.v. of R1 p.a. in advance while (x) is alive;
Ax = E v K+1 = expected p.v. of R1 payable at end of year of death of (x).

Since v K+1 = 1 d aK+1 (why?), we can take expected values and get
E v K+1 = E 1 d aK+1 = 1 d E aK+1 .

Hence Ax = 1 d ax , which is a very handy link between Ax and ax . So if you have a table of
values of ax , you dont really need a table of Ax .
Check for yourself that, on A196770 (ultimate) mortality and at 4%, this formula gives
A50 = 0.38450, the figure we used in an example in Section 5.1.
Exercises:
1. Prove the corresponding result for Ax: n and ax: n , i.e. show that Ax: n = 1 d ax: n . (Hint:
in the above proof, replace K + 1 by min(K + 1, n).)
2. Show that Px = 1/ax d and that Px: n = 1/ax: n d. Use these results to check some of
the figures given in the tables for Px and Px: n .
Remember that:
ax: n = expected p.v. of R1 p.a. in advance while (x) is alive, for at most n years; and that
Ax: n = expected p.v. of R1 payable at end of year of death of (x), or at time n if sooner.

6 Gross premiums
6.1 Allowing for expenses
Gross premiums are those which satisfy the equation
EPV of gross premiums = EPV of benefits + EPV of expenses.
Here the abbreviation EPV stands for expected present value. Strictly speaking, expenses
should read estimated expenses or assumed expenses, as expenses are not known in advance.
Compare the above equation with
EPV of net premiums = EPV of benefits.
Which will be bigger: net premiums or gross premiums? An example of the calculation of gross
and net premiums now follows, and a similar example relating to the purchase of an annuity.
12 I.L. MacDonald: BUS1003H, Elementary life contingencies

6.2 Examples
1. Suppose that a life aged 35 and subject to A196770 (ultimate) mortality takes out a
whole-life policy with sum assured R100 000 payable at the end of the year of death.
Premiums are payable annually in advance (for life). The relevant interest rate is 4%.

(a) What is the net premium for such a policy? (R1 136.58)

(b) Suppose that, in determining the gross premium for this policy, the life assurer uses
the following estimate of expenses:
each time a premium is paid, R5 plus 2% of (gross) premium;
additional initial expenses, R100 plus 70% of the (gross) premium; and
claim expenses 1% of sum assured.
Find the gross premium. (R1 225.16)
Other types of life assurance policies 13

2. Suppose a life aged exactly 66, subject to A196770 (ultimate) mortality, buys an annuity,
payable annually in arrear for life. The interest rate used is 4% p.a. and expenses are as
follows:
each time an annuity payment is made, R100 plus 1.25% of the annuity actually paid
to (66);
initial expenses, R500 plus 1.5% of the purchase sum.

(a) Assume that the purchase sum is R2 million. How big is the annuity?
(b) Suppose the amount of the annuity is R150 000 (payable annually in arrear). How
big is the purchase sum?

Solution:
(a) Let the annuity payment be A. Then:
2 106 = A(a66 1) + 500 + 0.015(2 106 ) + (100 + 0.0125A)(a66 1)
= 1.0125A 9.376 + 500 + 30 000 + 937.60.
Hence A = R207 365.52, i.e. approximately R207 366 (better). Or even just approxi-
mately R207 400 (since the tables give you four significant figures only for a66 ).
(b) Purchase sum is X, where:
X = 150 000(a66 1) + 500 + 0.015X + (100 + 1875)(a66 1)
= 1 425 417.6 + 0.015X.
Hence X = R1 447 124.47, i.e. approximately R1 447 124.

7 Other types of life assurance policies

Although this terminology may have been introduced elsewhere in the course, it is probably use-
ful to say here what is meant by temporary assurances, pure endowments and limited-premium
whole-life assurances.
A temporary assurance (also known as a term assurance) pays the sum assured only if
death occurs within a specified term.
A pure endowment pays the sum assured only if the insured life survives to the end of
a specified term. (So an endowment assurance consists of a temporary assurance and a pure
endowment.)
A limited-premium whole-life assurance provides the benefit of a whole-life assurance,
but requires that premiums be paid for a specified term only (or until prior death).
This terminology will be used in some of the tutorial questions.

8 Complete future lifetime, force of mortality

8.1 Curtate and complete future lifetimes
Definitions:
Kx : curtate future lifetime from age (x) a discrete random variable
Tx : complete future lifetime from age (x) a continuous random variable
For instance: if (49) survives for 10 years and 9 months, K49 = 10 and T49 = 10.75.
T0 is the complete future lifetime from birth.
14 I.L. MacDonald: BUS1003H, Elementary life contingencies

8.2 Force of mortality

We defined qx as the probability that someone aged x will die within one year; so

qx (= 1 px ) = P(Kx = 0) = P(0 Tx < 1).

More generally, the probability that (x) will die within t years is

t qx (= 1 t px ) = P(0 Tx < t).

Can we define a measure of mortality at x + t? (i.e. at a single age, not over an interval)
Lets start with the (short) interval from time t to time t + h, hence from age x + t to time
x + t + h. (We shall in due course let h tend to zero.)
The conditional probability of death in that interval, given that (x) survives at least to the
start of the interval, is
P(Tx < t + h | Tx t).
But this depends on h, and decreases to zero as h decreases to zero. So as it stands, it wont do.
But if we divide by h and let h tend to zero, we do get a measure of the intensity of mortality
at age x + t: it is called the force of mortality and denoted by the Greek letter (mu).
Definition:
lim h1 P(Tx < t + h | Tx t) x+t . (1)
h0

In other contexts this is often called the hazard rate or failure rate. One of the convenient
things about the force of mortality is that it can easily be related to survival probabilities n px ;
see below. But first let us note in passing that x+t could also be defined, in terms of T0 , by

lim h1 P(T0 < x + t + h | T0 x + t) x+t ,

h0

and that, subject to a very reasonable assumption, the two definitions can but wont here
be shown to be equivalent.

8.3 Relation of the force of mortality to survival probabilities

The conditional probability appearing in the definition at (1) can be written as follows:

P(Tx < t + h | Tx t) = P(t Tx < t + h)/P(Tx t) = (t+h qx t qx )/t px = (t+h px t px )/t px .

So the force of mortality is just

 
1 t+h px t px d d
x+t = lim h ((t+h px t px )/t px ) = lim /t px = (t px )/t px = (loge t px ).
h0 h0 h dt dt
(2)
Hence Z n
x+t dt = loge n px + loge 0 px = loge n px (why?),
0
and so  Z n   Z x+n 
n px = exp x+t dt = exp t dt . (3)
0 x
Notice that Equation (2) gives you the force of mortality, x+t , in terms of survival probabilities
t px ; Equation (3) enables you to go in the opposite direction.
d
From x+t = dt (t px )/t px see Equation (2) we can extract a useful general expression
for the p.d.f. of Tx . If fx denotes the p.d.f. of Tx , then for all t > 0

fx (t) = t px x+t . (4)

Expectation of life 15

All you need do to establish Equation (4) this is to observe that the p.d.f. is

d d d
P(Tx t) = (1 t px ) = (t px ).
dt dt dt
Example Suppose Tx has constant force of mortality (, say). What are the distribution and
probability density functions of Tx ?
For t > 0, the distribution function is
 Z t 
P(Tx t) = 1 t px = 1 exp x+s ds = 1 exp(t),
0

and the density function is therefore given (for all t > 0) by exp(t); what do you call such
a distribution? 

9 Curtate and complete expectation of life

Definitions:

ex : the mean of Kx is called the curtate expectation of life, and denoted by ex .

ex : the mean of Tx is called the complete expectation of life, and denoted by ex .

Example Suppose Tx has constant force of mortality (equal to ). What is ex ?
Since Tx has an exponential distribution (see previous example), it is easy to find its mean; it is
just Z
tet dt = 1/.
0
(For you, integration by parts is probably the easiest way to do this.) So the complete expectation

of life ex is 1/.
Can we also find the curtate expectation of life, ex ? Yes, but in order to do so we first find the
probability mass function of Kx . For nonnegative integers k,
Z k+1
P(Kx = k) = P(k Tx < k + 1) = et dt = ek (1 e ).
k

Hence the mean is

X
X
kek (1 e ) = (1 ) kk1 = (1 )(1 )2 ,
k=0 k=0

where (for convenience) denotes e . That is, the curtate expectation of life is

ex = /(1 ) = e /(1 e ) = 1/(e 1).

For instance, if e = 0.9, then ex = 9 and ex = 1/ = 9.4912, close to ex + 0.5 but not equal to
it. (Can you think why they might differ by approximately 0.5?)
Try a few more values of ; does the approximation get better or worse for smaller ? 
Comment Constant force of mortality is very unrealistic for human (or animal) lifetimes, but
is useful as a simple example, and might be more realistic for (e.g.) lifetimes of electronic
components.
Example Suppose T40 is uniformly distributed on the interval from 0 to 80 (years). What is the
complete expectation of life at age 40? (Answer: (0+80)/2=40.) What is 40+t , for t [0, 80)?
16 I.L. MacDonald: BUS1003H, Elementary life contingencies

For t [0, 80), the p.d.f. of T40 is 1/80 and the distribution function is t/80. Hence
1
= t p40 40+t = (1 t/80)40+t ,
80
from which we see that
40+t = 1/(80 t), 0 t < 80.
Draw a sketch of 40+t = 1/(80 t), for t [0, 80):

How does 40+t behave as t increases? Is this a realistic model for human lifetimes? 
Example In a certain mortality table:
1
x = (0 < x < 120).
480 4x
Derive, and simplify as far as possible, an expression for t px , valid for 0 t 120 x.

 1/4
Answer: t px = 1 t/(120 x) .
Find the complete expectation of life at age 40. (Answer: 64) Hint: one way of doing it involves
using integration by parts to show that:
Z 80 Z 80
80
e40 = [ t t p40 ]t=0 + t p40 dt = t p40 dt.
0 0

R
In fact you can take it that, for any example that you will get, ex = 0 t px dt. If, as is the case
R
here, there is an upper limit to age (call it , i.e. omega), this reduces to ex = 0 t px dt.
Examples 17

10 Further examples
1. In a certain model for survival, the probabilities of survival from age 60 are given by:

t p60 = k(40 t)(t + 100)2 for 0 t 40.

Find the following quantities:
(a) k
(b) q70
(c) 5 q75
(d) 65 (Hint: first find 60+t for all t [0, 40).)

(e) e60 .
2. (from 2009 exam, BUS1003H) Consider a life aged exactly 62, subject to A196770 (ulti-
mate) mortality.
Find from first principles (i.e. by using tables of lx and/or dx ) the net premiums, payable
annually in advance, required for:
(a) an endowment assurance policy;
(b) a pure endowment; and
(c) a temporary assurance policy.
In all cases the sum assured is R100 000, the term is three years, and the relevant interest
rate is 5% p.a. (effective).
3. In a certain model for survival, the force of mortality is given by:

0.3/(1 + s) for 0 s 2
s =
0.1 for s > 2.
Find the following quantities:
(a) t px , for all x 2 and t 0

(b) e2
(c) t p0 , for all t 0

(d) e0
(e) 3 p1 .

Numerical answers to the above problems:

1. (a) 2.5 106
(b) 0.015678
(c) 0.12892
(d) 0.0095238
(e) 25.867
2. R30 839.28, R28 970.15, R1869.13
3. (b) 10
(d) 8.8460
(e) 0.72496