Actuarial Mathematics

and Life-Table Statistics

Eric V. Slud
Mathematics Department
University of Maryland, College Park

2006
c
Chapter 4

Expected Present Values of

Insurance Contracts

We are now ready to draw together the main strands of the development so
far: (i) expectations of discrete and continuous random variables defined as
functions of a life-table waiting time T until death, and (ii) discounting of
future payment (streams) based on interest-rate assumptions. The approach
is first to define the contractual terms of and discuss relations between the
major sorts of insurance, endowment and life annuity contracts, and next to
use interest theory to define the present value of the contractual payment
stream by the insurer as a nonrandom function of the random individual
lifetime T . In each case, this leads to a formula for the expected present value
of the payout by the insurer, an amount called the net single premium
or net single risk premium of the contract because it is the single cash
payment by the insured at the beginning of the insurance period which would
exactly compensate for the average of the future payments which the insurer
will have to make.
The details of the further mathematical discussion fall into two parts:
first, the specification of formulas in terms of cohort life-table quantities for
net single premiums of insurances and annuities which pay only at whole-year
intervals; and second, the application of the various survival assumptions con-
cerning interpolation between whole years of age, to obtain the corresponding
formulas for insurances and annuities which have m payment times per year.
We close this Chapter with a discussion of instantaneous-payment insurance,

97
98 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

continuous-payment annuity, and mean-residual-life formulas, all of which

involve continuous-time expectation integrals. We also relate these expecta-
tions with their m-payment-per-year discrete analogues, and compare the
corresponding integral and summation formulas.
Similar and parallel discussions can be found in the Life Contingencies
book of Jordan (1967) and the Actuarial Mathematics book of Bowers et
al. (1986). The approach here differs in unifying concepts by discussing
together all of the different contracts, first in the whole-year case, next under
interpolation assumptions in the m-times-per-year case, and finally in the
instantaneous case.

4.1 Expected Present Values of Payments

Throughout the Chapter and from now on, it is helpful to distinguish no-
tationally the expectations relating to present values of life insurances and
annuities for a life aged x. Instead of the notation E(g(T ) | T ≥ x) for
expectations of functions of life-length random variables, we define

Ex ( g(T ) ) = E(g(T ) | T ≥ x)

The expectations formulas can then be written in terms of the residual-

lifetime variable S = T − x (or the change-of-variable s = t − x) as
follows:
Z ∞ Z ∞  
f (t) ∂ S(t)
Ex ( g(T ) ) = g(t) dt = g(t) − dt
x S(x) x ∂t S(x)
Z ∞ Z ∞
∂
= g(s + x) (− s px ) ds = g(s + x) µ(s + x) s px ds
0 ∂s 0

4.1.1 Types of Insurance & Life Annuity Contracts

There are three types of contracts to consider: insurance, life annuities, and
endowments. More complicated kinds of contracts — which we do not discuss
in detail — can be obtained by combining (superposing or subtracting) these
in various ways. A further possibility, which we address in Chapter 10, is
4.1. EXPECTED PAYMENT VALUES 99

to restrict payments to some further contingency (e.g., death-benefits only

under specified cause-of-death).
In what follows, we adopt several uniform notations and assumptions.
Let x denote the initial age of the holder of the insurance, life annuity,
or endowment contract, and assume for convenience that the contract is
initiated on the holder’s birthday. Fix a nonrandom effective (i.e., APR)
interest rate i , and retain the notation v = (1 + i)−1 , together with the
other notations previously discussed for annuities of nonrandom duration.
Next, denote by m the number of payment-periods per year, all times being
measured from the date of policy initiation. Thus, for given m, insurance
will pay off at the end of the fraction 1/m of a year during which death
occurs, and life-annuities pay regularly m times per year until the annuitant
dies. The term or duration n of the contract will always be assumed to
be an integer multiple of 1/m. Note that policy durations are all measured
from policy initiation, and therefore are exactly x smaller than the exact
age of the policyholder at termination.
The random exact age at which the policyholder dies is denoted by T ,
and all of the contracts under discussion have the property that T is the
only random variable upon which either the amount or time of payment can
depend. We assume further that the payment amount depends on the time
T of death only through the attained age Tm measured in multiples of 1/m
year. As before, the survival function of T is denoted S(t), and the density
either f (t). The probabilities of the various possible occurrences under the
policy are therefore calculated using the conditional probability distribution
of T given that T ≥ x, which has density f (t)/S(x) at all times t ≥ x.
Define from the random variable T the related discrete random variable
[T m] 1 th
Tm = = age at beginning of of year of death
m m
which for integer initial age x is equal to x + k/m whenever x + k/m ≤
T < x + (k + 1)/m. Observe that the probability mass function of this
random variable is given by
k  k k + 1 
P (Tm = x + T ≥ x) = P ( ≤ T − x <  T ≥ x)
m m m

 
1 k k+1
= S(x + ) − S(x + ) = k/m px − (k+1)/m px
S(x) m m
100 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

k  k + 1  k
= P (T ≥ x +  T ≥ x) · P (T < x + T ≥ x + ) (4.1)
m m m

= k/m px · 1/m qx+k/m

As has been mentioned previously, a key issue in understanding the special

nature of life insurances and annuities with multiple payment periods is to
understand how to calculate or interpolate these probabilities from the prob-
abilities j py (for integers j, y) which can be deduced or estimated from
life-tables.
An Insurance contract is an agreement to pay a face amount — perhaps
modified by a specified function of the time until death — if the insured, a
life aged x, dies at any time during a specified period, the term of the policy,
with payment to be made at the end of the 1/m year within which the death
occurs. Usually the payment will simply be the face amount F (0), but for
example in decreasing term policies the payment will be F (0) · (1 − k−1 nm
) if
death occurs within the k th successive fraction 1/m year of the policy,
where n is the term. (The insurance is said to be a whole-life policy if
n = ∞, and a term insurance otherwise.) The general form of this contract,
for a specified term n ≤ ∞, payment-amount function F (·), and number
m of possible payment-periods per year, is to

pay F (T − x) at time Tm − x + m1 following policy initiation,

if death occurs at T between x and x + n.

The present value of the insurance company’s payment under the contract is
evidently

F (T − x) v Tm −x+1/m

if x ≤ T < x + n
(4.2)
0 otherwise

The simplest and most common case of this contract and formula arise
when the face-amount F (0) is the constant amount paid whenever a death
within the term occurs. Then the payment is F (0), with present value
F (0) v −x+ ([mT ]+1)/m , if x ≤ T < x + n, and both the payment and present
value are 0 otherwise. In this case, with F (0) ≡ 1, the net single premium
has the standard notation A(m)1x:n⌉ . In the further special case where m = 1,
4.1. EXPECTED PAYMENT VALUES 101

the superscript m is dropped, and the net single premium is denoted A1x:n⌉ .
Similarly, when the insurance is whole-life (n = ∞), the subscript n and
bracket n⌉ are dropped.
A Life Annuity contract is an agreement to pay a scheduled payment to
the policyholder at every interval 1/m of a year while the annuitant is alive,
up to a maximum number of nm payments. Again the payment amounts are
ordinarily constant, but in principle any nonrandom time-dependent schedule
of payments F (k/m) can be used, where F (s) is a fixed function and s
ranges over multiples of 1/m. In this general setting, the life annuity
contract requires the insurer to

pay an amount F (k/m) at each time k/m ≤ T − x, up to a

maximum of nm payments.

To avoid ambiguity, we adopt the convention that in the finite-term life

annuities, either F (0) = 0 or F (n) = 0. As in the case of annuities certain
(i.e., the nonrandom annuities discussed within the theory of interest), we
refer to life annuities with first payment at time 0 as (life) annuities-due
and to those with first payment at time 1/m (and therefore last payment
at time n in the case of a finite term n over which the annuitant survives)
as (life) annuities-immediate. The present value of the insurance company’s
payment under the life annuity contract is
(Tm −x)m
X
F (k/m) v k/m (4.3)
k=0

Here the situation is definitely simpler in the case where the payment
amounts F (k/m) are level or constant, for then the life-annuity-due payment
stream becomes an annuity-due certain (the kind discussed previously under
the Theory of Interest) as soon as the random variable T is fixed. Indeed,
if we replace F (k/m) by 1/m for k = 0, 1, . . . , nm − 1, and by 0 for
(m)
larger indices k, then the present value in equation (4.3) is ämin(Tm +1/m, n)⌉ ,
(m)
and its expected present value (= net single premium) is denoted äx:n⌉ .
In the case of temporary life annuities-immediate, which have payments
commencing at time 1/m and continuing at intervals 1/m either until
death or for a total of nm payments, the expected-present value notation
102 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

(m)
is ax:n⌉ . However, unlike the case of annuities-certain (i.e., nonrandom-
duration annuities), one cannot simply multiply the present value of the life
annuity-due for fixed T by the discount-factor v 1/m in order to obtain the
corresponding present value for the life annuity-immediate with the same
term n. The difference arises because the payment streams (for the life
annuity-due deferred 1/m year and the life-annuity immediate) end at the
same time rather than with the same number of payments when death occurs
before time n. The correct conversion-formula is obtained by treating the
life annuity-immediate of term n as paying, in all circumstances, a present
value of 1/m (equal to the cash payment at policy initiation) less than
the life annuity-due with term n + 1/m. Taking expectations leads to the
formula
(m) (m)
ax:n⌉ = äx:n+1/m⌉ − 1/m (4.4)
In both types of life annuities, the superscripts (m) are dropped from the
net single premium notations when m = 1, and the subscript n is dropped
when n = ∞.
The third major type of insurance contract is the Endowment, which
pays a contractual face amount F (0) at the end of n policy years if the
policyholder initially aged x survives to age x + n. This contract is the
simplest, since neither the amount nor the time of payment is uncertain. The
pure endowment contract commits the insurer to

pay an amount F (0) at time n if T ≥ x + n

The present value of the pure endowment contract payment is

F (0) v n if T ≥ x + n, 0 otherwise (4.5)

The net single premium or expected present value for a pure endowment
contract with face amount F (0) = 1 is denoted Ax:n1⌉ or n Ex and is
evidently equal to
Ax:n1⌉ = n Ex = v n n px (4.6)

The other contract frequently referred to in beginning actuarial texts is

the Endowment Insurance, which for a life aged x and term n is simply
the sum of the pure endowment and the term insurance, both with term n
and the same face amount 1. Here the contract calls for the insurer to
4.1. EXPECTED PAYMENT VALUES 103

1
pay $1 at time Tm + m
if T < n, and at time n if T ≥ n

The present value of this contract has the form v n on the event [T ≥ n]
and the form v Tm −x+1/m on the complementary event [T < n]. Note that
Tm + 1/m ≤ n whenever T < n. Thus, in both cases, the present value is
given by
v min(Tm −x+1/m, n) (4.7)

(m)
The expected present value of the unit endowment insurance is denoted Ax:n⌉ .
Observe (for example in equation (4.10) below) that the notations for the
net single premium of the term insurance and of the pure endowment are
intended to be mnemonic, respectively denoting the parts of the endowment
insurance determined by the expiration of life — and therefore positioning
the superscript 1 above the x — and by the expiration of the fixed term,
with the superscript 1 in the latter case positioned above the n.
Another example of an insurance contract which does not need separate
treatment, because it is built up simply from the contracts already described,
is the n-year deferred insurance. This policy pays a constant face amount at
the end of the time-interval 1/m of death, but only if death occurs after time
n , i.e., after age x + n for a new policyholder aged precisely x. When the
face amount is 1, the contractual payout is precisely the difference between
the unit whole-life insurance and the n-year unit term insurance, and the
formula for the net single premium is

A(m)
x − A(m)1x:n⌉ (4.8)

Since this insurance pays a benefit only if the insured survives at least
n years, it can alternatively be viewed as an endowment with benefit equal
to a whole life insurance to the insured after n years (then aged x + n) if
the insured lives that long. With this interpretation, the n-year deferred
insurance has net single premium = n Ex · Ax+n . This expected present
value must therefore be equal to (4.8), providing the identity:

A(m)
x − A(m)1x:n⌉ = v n n px · Ax+n (4.9)
104 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

4.1.2 Formal Relations among Net Single Premiums

In this subsection, we collect a few useful identities connecting the different

types of contracts, which hold without regard to particular life-table interpo-
lation assumptions. The first, which we have already seen, is the definition of
endowment insurance as the superposition of a constant-face-amount term
insurance with a pure endowment of the same face amount and term. In
terms of net single premiums, this identity is
(m)
Ax:n⌉ = A(m)1x:n⌉ + A(m) x:n1⌉ (4.10)

The other important identity concerns the relation between expected

present values of endowment insurances and life annuities. The great gener-
ality of the identity arises from the fact that, for a fixed value of the random
lifetime T , the present value of the life annuity-due payout coincides with
the annuity-due certain. The unit term-n life annuity-due payout is then
given by
(m) 1 − v min(Tm −x+1/m, n)
ämin(Tm −x+1/m, n)⌉ =
d(m)
The key idea is that the unit life annuity-due has present value which is a
simple linear function of the present value v min(Tm −x+1/m, n) of the unit en-
dowment insurance. Taking expectations (over values of the random variable
T , conditionally given T ≥ x) in the present value formula, and substituting
(m)
Ax:n⌉ as expectation of (4.7), then yields:
 1 − v min(Tm −x+1/m, n)  (m)
(m) 1 − Ax:n⌉
äx:n⌉ = Ex = (4.11)
d(m) d(m)
where recall that Ex ( · ) denotes the conditional expectation E( · | T ≥ x).
A more common and algebraically equivalent form of the identity (4.11) is
(m) (m)
d(m) äx:n⌉ + Ax:n⌉ = 1 (4.12)

To obtain a corresponding identity relating net single premiums for life

annuities-immediate to those of endowment insurances, we appeal to the
conversion-formula (4.4), yielding
(m)
(m) (m) 1 1 − Ax:n+1/m⌉ 1 1 1 (m)
ax:n⌉ = äx:n+1/m⌉ − = (m)
− = (m)
− (m)
Ax:n+1/m⌉ (4.13)
m d m i d
4.1. EXPECTED PAYMENT VALUES 105

and
(m) (m) d(m)
d(m) ax:n⌉ + Ax:n+1/m⌉ = = v 1/m (4.14)
i(m)

In these formulas, we have made use of the definition

m i(m) . i(m)
= (1 + ) ( )
d(m) m m
leading to the simplifications

m m i(m) i(m)
= +1 , = 1 + = v −1/m
d(m) i(m) d(m) m

4.1.3 Formulas for Net Single Premiums

This subsection collects the expectation-formulas for the insurance, annuity,

and endowment contracts defined above. Throughout this Section, the same
conventions as before are in force (integer x and n, fixed m, i, and
conditional survival function t px ) .
First, the expectation of the present value (4.2) of the random term in-
surance payment (with level face value F (0) ≡ 1) is

nm−1
X
1 Tm −x+1/m
v (k+1)/m


Ax:n⌉ = Ex v = k/m px 1/m qx+k/m (4.15)
k=0

The index k in the summation formula given here denotes the multiple of
1/m beginning the interval [k/m, (k + 1)/m) within which the policy age
T − x at death is to lie. The summation itself is simply the weighted sum,
over all indices k such that k/m < n, of the present values v (k+1)/m
to be paid by the insurer in the event that the policy age at death falls in
[k/m, (k + 1)/m) multiplied by the probability, given in formula (4.1), that
this event occurs.
Next, to figure the expected present value of the life annuity-due with
term n, note that payments of 1/m occur at all policy ages k/m, k =
0, . . . , nm − 1, for which T − x ≥ k/m. Therefore, since the present values
106 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

of these payments are (1/m) v k/m and the payment at k/m is made with
probability k/m px ,
nm−1
! nm−1
(m)
X 1 1 X k/m
k/m
äx:n⌉ = Ex v I[T −x≥k/m] = v k/m px (4.16)
k=0
m m k=0

Finally the pure endowment has present value

n
= v n x pn


n Ex = Ex v I[T −x≥n] (4.17)

4.1.4 Expected Present Values for m = 1

It is clear that for the general insurance and life annuity payable at whole-
year intervals ( m = 1 ), with payment amounts determined solely by the
whole-year age [T ] at death, the net single premiums are given by discrete-
random-variable expectation formulas based upon the present values (4.2)
and (4.3). Indeed, since the events {[T ] ≥ x} and {T ≥ x} are identical
for integers x, the discrete random variable [T ] for a life aged x has
conditional probabilities given by
P ([T ] = x + k | T ≥ x) = k px − k+1 px = k px · qx+k
Therefore the expected present value of the term-n insurance paying F (k)
at time k +1 whenever death occurs at age T between x+k and x+k +1
(with k < n) is
   Xn−1
E v [T ]−x+1 F ([T ] − x) I[T ≤x+n]  T ≥ x = F (k) v k+1 k px qx+k

k=0

Here and from now on, for an event B depending on the random lifetime
T , the notation IB denotes the so-called indicator random variable which is
equal to 1 whenever T has a value such that the condition B is satisfied
and is equal to 0 otherwise. The corresponding life annuity which pays
F (k) at each k = 0, . . . , n at which the annuitant is alive has expected
present value
 min(n,
X [T ]−x)  n
X  n
X
k k
Ex v F (k) = Ex v F (k) I[T ≥x+k] = v k F (k) k px
k=0 k=0 k=0
4.1. EXPECTED PAYMENT VALUES 107

In other words, the payment of F (k) at time k is received only if the

annuitant is alive at that time and so contributes expected present value
equal to v k F (k) k px . This makes the annuity equal to the superposition
of pure endowments of terms k = 0, 1, 2, . . . , n and respective face-amounts
F (k).
In the most important special case, where the non-zero face-amounts
F (k) are taken as constant, and for convenience are taken equal to 1 for
k = 0, . . . , n − 1 and equal to 0 otherwise, we obtain the useful formulas
n−1
X
A1x:n⌉ = v k+1 k px qx+k (4.18)
k=0
n−1
X
äx:n⌉ = v k k px (4.19)
k=0
 
1
Ax:n⌉ = Ex v I[T −x≥n] = v n n px
n
(4.20)
∞
X
Ax:n⌉ = v min(n,k+1) k px qx+k
k=0
n−1
X
= v k+1 (k px − k+1 px ) + v n n px (4.21)
k=0

Two further manipulations which will complement this circle of ideas are
left as exercises for the interested reader: (i) first, to verify that formula
(4.19) gives the same answer as the formula Ex (äx:min([T ]−x+1, n)⌉ ) ; and
(ii) second, to sum by parts (collecting terms according to like subscripts k
of k px in formula (4.21)) to obtain the equivalent expression
n−1
X n−1
X
1+ (v k+1 − v k ) k px = 1 − (1 − v) v k k px
k=0 k=0

The reader will observe that this final expression together with formula (4.19)
gives an alternative proof, for the case m = 1, of the identity (4.12).
Let us work out these formulas analytically in the special case where [T ]
has the Geometric(1 − γ) distribution, i.e., where

P ([T ] = k) = P (k ≤ T < k + 1) = γ k (1 − γ) for k = 0, 1, . . .

108 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

with γ a fixed constant parameter between 0 and 1. This would be

true if the force of mortality µ were constant at all ages, i.e., if T were
exponentially distributed with parameter µ, with f (t) = µ e−µt for t ≥ 0.
In that case, P (T ≥ k) = e−µk , and γ = P (T = k|T ≥ k) = 1 − e−µ . Then

k px qx+k = P ([T ] = x + k | T ≥ x) = γ k (1 − γ) , n px = γn

so that
n−1
1 n 1
X 1 − (γv)n
Ax:n⌉ = (γv) , Ax:n⌉ = v k+1 γ k (1 − γ) = v(1 − γ)
k=0
1 − γv

Thus, for the case of interest rate i = 0.05 and γ = 0.97, corresponding
to expected lifetime = γ/(1 − γ) = 32.33 years,

.03 1 − (.97/1.05)20
Ax:20⌉ = (0.97/1.05)20 + · = .503
1.05 (1 − (.97/1.05)
1 .03
which can be compared with Ax ≡ Ax:∞⌉ = .08
= .375.
The formulas (4.18)-(4.21) are benchmarks in the sense that they repre-
sent a complete solution to the problem of determining net single premiums
without the need for interpolation of the life-table survival function between
integer ages. However the insurance, life-annuity, and endowment-insurance
contracts payable only at whole-year intervals are all slightly impractical
as insurance vehicles. In the next chapter, we approach the calculation of
net single premiums for the more realistic context of m-period-per-year in-
surances and life annuities, using only the standard cohort life-table data
collected by integer attained ages.

4.2 Continuous-Time Expectations

So far in this Chapter, all of the expectations considered have been associ-
ated with the discretized random lifetime variables [T ] and Tm = [mT ]/m.
However, Insurance and Annuity contracts can also be defined with re-
spectively instantaneous and continuous payments, as follows. First, an
instantaneous-payment or continuous insurance with face-value F
4.2. CONTINUOUS CONTRACTS & RESIDUAL LIFE 109

is a contract which pays an amount F at the instant of death of the in-

sured. (In practice, this means that when the actual payment is made at
some later time, the amount paid is F together with interest compounded
from the instant of death.) As a function of the random lifetime T for the
insured life initially with exact integer age x, the present value of the amount
paid is F · v T −x for a whole-life insurance and F · v T −x · I[T <x+n] for an
n-year term insurance. The expected present values or net single premiums
on a life aged x are respectively denoted Ax for a whole-life contract and
1
Ax:n⌉ for an n-year temporary insurance. The continuous life annuity is
a contract which provides continuous payments at rate 1 per unit time for
duration equal to the smaller of the remaining lifetime of the annuitant or
the term of n years. Here the present value of the contractual payments,
as a function of the exact age T at death for an annuitant initially of exact
integer age x, is amin(T −x, n)⌉ where n is the (possibly infinite) duration of
the life annuity. Recall that
Z ∞ Z K
t
aK⌉ = v I[t≤K] dt = v t dt = (1 − v K )/δ
0 0

is the present value of a continuous payment stream of 1 per unit time of

duration K units, where v = (1 + i)−1 and δ = ln(1 + i) .
The objective of this section is to develop and interpret formulas for these
continuous-time net single premiums, along with one further quantity which
has been defined as a continuous-time expectation of the lifetime variable T ,
namely the mean residual life (also called complete life expectancy)
◦
ex = Ex (T − x) for a life aged x. The underlying general conditional
expectation formula (1.3) was already derived in Chapter 1, and we reproduce
it here in the form
Z ∞ Z ∞
1
Ex { g(T ) } = g(y) f (y) dy = g(x + t) µ(x + t) t px dt (4.22)
S(x) x 0

We apply this formula directly for the three choices

g(y) = y−x , v y−x , or v y−x · I[y−x<n]

which respectively have the conditional Ex (·) expectations

◦ 1
ex , Ax , Ax:n⌉
110 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

For easy reference, the integral formulas for these three cases are:
Z ∞
◦
ex = Ex (T − x) = t µ(x + t) t px dt (4.23)
0
Z ∞
T −x
Ax = Ex (v ) = v t µ(x + t) t px dt (4.24)
0
Z n
1 T −x
v t µ(x + t) t px dt


Ax:n⌉ = Ex v I[T −x≤n] = (4.25)
0

Next, we obtain two additional formulas, for continuous life annuities-due

ax and ax:n⌉

which correspond to Ex {g(T )} for the two choices

Z ∞ Z n
t
g(t) = v I[t≤y−x] dt or v t I[t≤y−x] dt
0 0

After switching the order of the integrals and the conditional expectations,
and evaluating the conditional expectation of an indicator as a conditional
probability, in the form


Ex I[t≤T −x] = P (T ≥ x + t | T ≥ x) = t px

the resulting two equations become

Z ∞  Z ∞
t
ax = Ex v I[t≤T −x] dt = v t t px dt (4.26)
0 0
Z n  Z n
t
ax:n⌉ = Ex v I[t≤T −x] dt = v t t px dt (4.27)
0 0

As might be expected, the continuous insurance and annuity contracts

have a close relationship to the corresponding contracts with m payment
periods per year for large m. Indeed, it is easy to see that the term insurance
net single premiums

A(m)1x:n⌉ = Ex v Tm −x+1/m

approach the continuous insurance value (4.24) as a limit when m → ∞. A

simple proof can be given because the payments at the end of the fraction
4.2. CONTINUOUS CONTRACTS & RESIDUAL LIFE 111

1/m of year of death are at most 1/m years later than the continuous-
insurance payment at the instant of death, so that the following obvious
inequalities hold:
1 1
Ax:n⌉ ≤ A(m)1x:n⌉ ≤ v 1/m Ax:n⌉ (4.28)
Since the right-hand term in the inequality (4.28) obviously converges for
large m to the leftmost term, the middle term which is sandwiched in
between must converge to the same limit (4.25).
For the continuous annuity, (4.27) can be obtained as a limit of formulas
(4.16) using Riemann sums, as the number m of payments per year goes to
infinity, i.e.,
nm−1
X 1 Z n
(m) k/m
ax:n⌉ = lim äx:n⌉ = lim v k/m px = v t t px ds
m→∞ m→∞
k=0
m 0

The final formula coincides with (4.27), according with the intuition that the
limit as m → ∞ of the payment-stream which pays 1/m at intervals of time
1/m between 0 and Tm − x inclusive is the continuous payment-stream
which pays 1 per unit time throughout the policy-age interval [0, T − x).
Each of the expressions in formulas (4.23), (4.24), and (4.27) can be con-
trasted with a related approximate expectation for a function of the integer-
valued random variable [T ] (taking m = 1). First, alternative general
formulas are developed for the integrals by breaking the formulas down into
sums of integrals over integer-endpoint intervals and substituting the defini-
tion k px /S(x + k) = 1/S(x) :
∞ Z x+k+1
X f (y)
Ex (g(T )) = g(y) dy changing to z = y−x−k
k=0 x+k S(x)
∞ Z 1
X f (x + k + z)
= k px g(x + k + z) dz (4.29)
k=0 0 S(x + k)

Substituting into (4.29) the special function g(y) = y − x, leads to

∞ n S(x + k) − S(x + k + 1) Z 1
X f (x + k + z) o
e̊x = p
k x k + z dz (4.30)
k=0
S(x + k) 0 S(x + k)

Either of two assumptions between integer ages can be applied to simplify

the integrals:
112 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

(a) (Uniform distribution of failures) f (y) = f (x + k) = S(x + k) −

S(x + k + 1) for all y between x + k, x + k + 1 ;
(b) (Constant force of mortality) µ(y) = µ(x + k) for x + k ≤ y <
x + k + 1, in which case 1 − qx+k = exp(−µ(x + k)).
In case (a), the last integral in (4.30) becomes
Z 1 Z 1
f (x + k + z) S(x + k) − S(x + k + 1) 1
z dz = z dz = qx+k
0 S(x + k) 0 S(x + k) 2
and in case (b), we obtain within (4.30)
Z 1 Z 1
f (x + k + z)
z dz = z µ(x + k) e−z µ(x+k) dz
0 S(x + k) 0

which in turn is equal (after integration by parts) to

1 − e−µ(x+k) 1 1
−e−µ(x+k) + ≈ µ(x + k) ≈ qx+k
µ(x + k) 2 2
where the last approximate equalities hold if the death rates are small. It
follows, exactly in the case (a) where failures are uniformly distributed within
integer-age intervals or approximately in case (b) when death rates are small,
that ∞ ∞
◦ X 1 X 1
ex = (k + ) k px qx+k = k k px qx+k + (4.31)
k=0
2 k=0
2
The final summation in (4.31), called the curtate life expectancy
∞
X
ex = k k px qx+k (4.32)
k=0

has an exact interpretation as the expected number of whole years of life

remaining to a life aged x. The behavior of and comparison between com-
plete and curtate life expectancies is explored numerically in subsection 4.2.1
below.
Return now to the general expression for Ex (g(T )), substituting g(y) =
y−x
v but restricting attention to case (a):
n Z x+k+1
1  T −x X f (x + k)
v y−x

Ax:n⌉ = E v I[T <x+n] = dy
k=0 x+k
S(x)
4.2. CONTINUOUS CONTRACTS & RESIDUAL LIFE 113

n Z x+k+1
X S(x + k) − S(x + k + 1)
= v y−x k px dy
k=0 x+k S(x + k)
n 1 n
1 − e−δ
X Z X
k+t
= k px qx+k v dt = k px qx+k v k+1
k=0 0 k=0
vδ

where v = 1/(1+i) = e−δ , and δ is the force of interest. Since 1−e−δ = iv,
we have found in case (a) that
1
Ax:n⌉ = A1x:n⌉ · (i/δ) (4.33)

Finally, return to the formula (4.27) under case (b) to find

n−1 Z
X k+1 n−1 Z
X k+1
t
ax:n⌉ = v t px dt = e−(δ+µ)t dt
k=0 k k=0 k

n−1 −(δ+µ)k n−1

X e − e−(δ+µ)(k+1) X 1 − e−(δ+µ)
= = v k k px ·
k=0
δ+µ k=0
δ+µ

Thus, in case (b) we have shown

1 − e−(δ+µ)n 1 − e−(δ+µ)
ax:n⌉ = = äx:n⌉ · (4.34)
δ+µ δ+µ

In the last two paragraphs, we have obtained formulas (4.33) and (4.34)
respectively under cases (a) and (b) relating net single premiums for con-
tinuous contracts to those of the corresponding single-payment-per-year con-
tracts. More elaborate relations will be given in the next Chapter between
net single premium formulas which do require interpolation-assumptions for
probabilities of survival to times between integer ages to formulas for m = 1,
which do not require such interpolation.

4.2.1 Numerical Calculations of Life Expectancies

Formulas (4.23) or (4.30) and (4.32) above respectively provide the complete
and curtate age-specific life expectancies, in terms respectively of survival
114 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

densities and life-table data. Formula (4.31) provides the actuarial approxi-
mation for complete life expectancy in terms of life-table data, based upon
interpolation-assumption (i) (Uniform mortality within year of age). In this
Section, we illustrate these formulas using the Illustrative simulated and ex-
trapolated life-table data of Table 1.1.
Life expectancy formulas necessarily involve life table data and/or sur-
vival distributions specified out to arbitrarily large ages. While life tables
may be based on large cohorts of insured for ages up to the seventies and even
eighties, beyond that they will be very sparse and very dependent on the par-
ticular small group(s) of aged individuals used in constructing the particular
table(s). On the other hand, the fraction of the cohort at moderate ages who
will survive past 90, say, is extremely small, so a reasonable extrapolation of
a well-established table out to age 80 or so may give sufficiently accurate life-
expectancy values at ages not exceeding 80. Life expectancies are in any case
forecasts based upon an implicit assumption of future mortality following ex-
actly the same pattern as recent past mortality. Life-expectancy calculations
necessarily ignore likely changes in living conditions and medical technology
which many who are currently alive will experience. Thus an assertion of
great accuracy for a particular method of calculation would be misplaced.
All of the numerical life-expectancy calculations produced for the Figure
of this Section are based on the extrapolation (2.9) of the illustrative life table
data from Table 1.1. According to that extrapolation, death-rates qx for
all ages 78 and greater are taken to grow exponentially, with log(qx /q78 ) =
(x − 78) ln(1.0885). This exponential behavior is approximately but not
precisely compatible with a Gompertz-form force-of-mortality function
µ(78 + t) = µ(78) ct
in light of the approximate equality µ(x) ≈ qx , an approximation which
progressively becomes less valid as the force of mortality gets larger. To see
this, note that under a Gompertz survival model,
 
x x c−1
µ(x) = Bc , qx = 1 − exp −Bc
ln c
and with c = 1.0885 in our setting, (c − 1)/ ln c = 1.0436.
Since curtate life expectancy (4.32) relies directly on (extrapolated) life-
table data, its calculation is simplest and most easily interpreted. Figure 4.1
4.3. EXERCISE SET 4 115

presents, as plotted points, the age-specific curtate life expectancies for in-
teger ages x = 0, 1, . . . , 78. Since the complete life expectancy at each age
is larger than the curtate by exactly 1/2 under interpolation assumption
(a), we calculated for comparison the complete life expectancy at all (real-
number) ages, under assumption (b) of piecewise-constant force of mortality
within years of age. Under this assumption, by formula (3.11), mortality
within year of age (0 < t < 1) is t px = (px )t . Using formula (4.31) and
interpolation assumption (b), the exact formula for complete life expectancy
becomes
∞  
X qx+k + px+k ln(px+k )
e̊x − ex = k px
k=0
− ln(px+k )

The complete life expectancies calculated from this formula were found to
exceed the curtate life expectancy by amounts ranging from 0.493 at ages
40 and below, down to 0.485 at age 78 and 0.348 at age 99. Thus there is
essentially no new information in the calculated complete life expectancies,
and they are not plotted.
The aspect of Figure 4.1 which is most startling to the intuition is the
large expected numbers of additional birthdays for individuals of advanced
ages. Moreover, the large life expectancies shown are comparable to actual
US male mortality circa 1959, so would be still larger today.

4.3 Exercise Set 4

(1). For each of the following three lifetime distributions, find (a) the
expected remaining lifetime for an individual aged 20, and (b) 7/12 q40 /q40 .
(i) Weibull(.00634, 1.2), with S(t) = exp(−0.00634 t1.2 ),
(ii) Lognormal(log(50), 0.3252 ), with S(t) = 1−Φ((log(t)−log(50))/0.325),
(iii) Piecewise exponential with force of mortality given the constant value
µt = 0.015 for 20 < t ≤ 50, and µt = 0.03 for t ≥ 50. In these
integrals, you should be prepared to use integrations by parts, gamma function
values, tables of the normal distribution function Φ(x), and/or numerical
integrations via calculators or software.
116 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

Expected number of additional whole years of life, by age

70

••••
••
••
••
60

••
••
••
••
••
50

••
••
Curtate Life Expectancy

••
••
••
••
40

••
••
••
••
••
30

••
••
••
••
••
••
20

••
••
••
••
••
••
••
•••
10

•••
•••
••

0 20 40 60 80
Age in years

Figure 4.1: Curtate life expectancy ex as a function of age, calculated

from the simulated illustrative life table data of Table 1.1, with age-specific
death-rates qx extrapolated as indicated in formula (2.9).
4.3. EXERCISE SET 4 117

(2). (a) Find the expected present value, with respect to the constant
effective interest rate r = 0.07, of an insurance payment of $1000 to be
made at the instant of death of an individual who has just turned 40 and
whose remaining lifetime T − 40 = S is a continuous random variable with
density f (s) = 0.05 e−0.05 s , s > 0.
(b) Find the expected present value of the insurance payment in (a) if
the insurer is allowed to delay the payment to the end of the year in which
the individual dies. Should this answer be larger or smaller than the answer
in (a) ?
(3). If the individual in Problem 2 pays a life insurance premium P at
the beginning of each remaining year of his life (including this one), then
what is the expected total present value of all the premiums he pays before
his death ?
(4). Suppose that an individual has equal probability of dying within each
of the next 40 years, and is certain to die within this time, i.e., his age is x
and
k px − k+1 px = 0.025 for k = 0, 1, . . . , 39
Assume the fixed interest rate r = 0.06.
(a) Find the net single whole-life insurance premium Ax for this indi-
vidual.
(b) Find the net single premium for the term and endowment insurances
1
Ax:20⌉ and Ax:30⌉ .
(5). Show that the expected whole number of years of remaining life for a
life aged x is given by
ω−x−1
X
cx = E([T ] − x | T ≥ x) = k k px qx+k
k=0

and prove that this quantity as a function of integer age x satisfies the
recursion equation
cx = px (1 + cx+1 )

(6). Show that the expected present value bx of an insurance of 1 payable

at the beginning of the year of death (or equivalently, payable at the end of
118 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

the year of death along with interest from the beginning of that same year)
satisfies the recursion relation (4.35) above.
(7). Prove the identity (4.9) algebraically.

For the next two problems, consider a cohort life-table population for
which you know only that l70 = 10, 000, l75 = 7000, l80 = 3000, and l85 =
0, and that the distribution of death-times within 5-year age intervals is
uniform.
(8). Find (a) e̊75 and (b) the probability of an individual aged 70 in
this life-table population dying between ages 72.0 and 78.0.
(9). Find the probability of an individual aged 72 in this life-table popula-
tion dying between ages 75.0 and 83.0, if the assumption of uniform death-
times within 5-year intervals is replaced by:
(a) an assumption of constant force of mortality within 5-year age-
intervals;
(b) the Balducci assumption (of linearity of 1/S(t)) within 5-year age
intervals.
(10). Suppose that a population has survival probabilities governed at all
ages by the force of mortality


 .01 for 0 ≤ t < 1
 .002 for 1 ≤ t < 5


µt = .001 for 5 ≤ t < 20
.004 for 20 ≤ t < 40




.0001 · t for 40 ≤ t


Then (a) find 30 p10 , and (b) find e̊50 .

(11). Suppose that a population has survival probabilities governed at all
ages by the force of mortality

 .01 for 0 ≤ t < 10
µt = .1 for 10 ≤ t < 30
3/t for 30 ≤ t


Then (a) find 30 p20 = the probability that an individual aged 20 survives
for at least 30 more years, and (b) find e̊30 .
4.3. EXERCISE SET 4 119

(12). Assuming the same force of mortality as in the previous problem, find
e̊70 and A60 if i = 0.09.
(13). The force of mortality for impaired lives is three times the standard
force of mortality at all ages. The standard rates qx of mortality at ages 95,
96, and 97 are respectively 0.3, 0.4, and 0.5 . What is the probability that
an impaired life age 95 will live to age 98 ?
(14). You are given a survival function S(x) = (10 − x)2 /100 , 0 ≤ x ≤ 10.
(a) Calculate the average number of future years of life for an individual
who survives to age 1.
(b) Calculate the difference between the force of mortality at age 1, and
the probability that a life aged 1 dies before age 2.
(15). An n-year term life insurance policy to a life aged x provides
that if the insured dies within the n-year period an annuity-certain of yearly
payments of 10 will be paid to the beneficiary, with the first annuity payment
made on the policy-anniversary following death, and the last payment made
on the N th policy anniversary. Here 1 < n ≤ N are fixed integers. If
B(x, n, N ) denotes the net single premium (= expected present value) for
this policy, and if mortality follows the law lx = C(ω − x)/ω for some
terminal integer age ω and constant C, then find a simplified expression
for B(x, n, N ) in terms of interest-rate functions, ω, and the integers
x, n, N . Assume x + n ≤ ω.
(16). The father of a newborn child purchases an endowment and insurance
contract with the following combination of benefits. The child is to receive
$100, 000 for college at her 18th birthday if she lives that long and $500, 000
at her 60th birthday if she lives that long, and the father as beneficiary is
to receive $200, 000 at the end of the year of the child’s death if the child
dies before age 18. Find expressions, both in actuarial notations and in
terms of v = 1/(1 + i) and of the survival probabilities k p0 for the child,
for the net single premium for this contract.
120 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

4.4 Worked Examples

Example 1. Toy Life-Table (assuming uniform failures)
Consider the following life-table with only six equally-spaced ages. (That
is, assume l6 = 0.) Assume that the rate of interest i = .09, so that
v = 1/(1 + i) = 0.9174 and (1 − e−δ )/δ = (1 − v)/δ = 0.9582.

x Age-range lx dx ex Ax
0 0 – 0.99 1000 60 4.2 0.704
1 1 – 1.99 940 80 3.436 0.749
2 2 – 2.99 860 100 2.709 0.795
3 3 – 3.99 760 120 2.0 0.844
4 4 – 4.99 640 140 1.281 0.896
5 5 – 5.99 500 500 0.5 0.958

Using the data in this Table, and interest rate i = .09, we begin by cal-
culating the expected present values for simple contracts for term insurance,
annuity, and endowment. First, for a life aged 0, a term insurance with
payoff amount $1000 to age 3 has present value given by formula (4.18) as
 
1 60 2 80 3 100
1000 A0:3⌉ = 1000 0.917 + (0.917) + (0.917) = 199.60
1000 1000 1000

Second, for a life aged 2, a term annuity-due of $700 per year up to age
5 has present value computed from (4.19) to be
 
760 2 640
700 ä2:3⌉ = 700 1 + 0.917 + (0.917) = 1705.98
860 860

For the same life aged 2, the 3-year Endowment for $700 has present value
500
700 A0:31⌉ = 700 · (0.9174)3 = 314.26
860
Thus we can also calculate (for the life aged 2) the present value of the
3-year annuity-immediate of $700 per year as

700 · ä2:3⌉ − 1 + A0:31⌉ = 1705.98 − 700 + 314.26 = 1320.24

4.4. WORKED EXAMPLES 121

We next apply and interpret the formulas of Section 4.2, together with
the observation that
lx+j dx+j dx+j
j px · qx+j = · =
lx lx+j lx

to show how the last two columns of the Table were computed. In particular,
by (4.31)
100 120 140 500 1 1900
e2 = ·0+ ·1+ ·2+ ·3+ = + 0.5 = 2.709
860 860 860 860 2 860
P5−x
Moreover: observe that cx = k=0 k k px qx+k satisfies the “recursion equa-
tion” cx = px (1 + cx+1 ) (cf. Exercise 5 above), with c5 = 0, from which
the ex column is easily computed by: ex = cx + 0.5.
Now apply the present value formula for conitunous insurance to find
5−x 5−x
X
k1 − e−δ X
k
Ax = k p x qx v = 0.9582 k p x qx v = 0.9582 bx
k=0
δ k=0

where bx is the expected present value of an insurance of 1 payable at the

beginning of the year of death (so that Ax = v bx ) and satisfies b5 = 1
together with the recursion-relation
5−x
X
bx = k p x qx v k = px v bx+1 + qx (4.35)
k=0

(Proof of this recursion is Exercise 6 above.)

Example 2. Find a simplified expression in terms of actuarial exprected

present value notations for the net single premium of an insurance on a life
aged x, which pays F (k) = C än−k⌉ if death occurs at any exact ages
between x + k and x + k + 1, for k = 0, 1, . . . , n − 1, and interpret the
result.
Let us begin with the interpretation: the beneficiary receives at the end
of the year of death a lump-sum equal in present value to a payment stream
of $C annually beginning at the end of the year of death and terminating
at the end of the nth policy year. This payment stream, if superposed upon
122 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

an n-year life annuity-immediate with annual payments $C, would result

in a certain payment of $C at the end of policy years 1, 2, . . . , n. Thus
the expected present value in this example is given by

C an⌉ − C ax:n⌉ (4.36)

Next we re-work this example purely in terms of analytical formulas. By

formula (4.36), the net single premium in the example is equal to
n−1 n−1
X
k+1
X 1 − v n−k
v k px qx+k C än−k+1⌉ = C v k+1 k px qx+k
k=0 k=0
d

( n−1 n−1
)
C X X
= v k+1 k px qx+k − v n+1 (k px − k+1 px )
d k=0 k=0
C  1
Ax:n⌉ − v n+1 (1 − n px )

=
d
C 
Ax:n⌉ − v n n px − v n+1 (1 − n px )

=
d
and finally, by substituting expression (4.14) with m = 1 for Ax:n⌉ , we
have
C 
1 − d äx:n⌉ − (1 − v) v n n px − v n+1

d
C 
1 − d (1 + ax:n⌉ − v n n px ) − d v n n px − v n+1

=
d
1 − vn
 
C  n+1

= v − d ax:n⌉ − v = C − ax:n⌉
d i
= C {an⌉ − ax:n⌉ }

So the analytically derived answer agrees with the one intuitively arrived at
in formula (4.36).
4.5. USEFUL FORMULAS FROM CHAPTER 4 123

4.5 Useful Formulas from Chapter 4

Tm = [T m]/m
p. 99
k
P (Tm = x + | T ≥ x) = k/m px − (k+1)/m px = k/m px · 1/m qx+k/m
m
p. 100

(m) (m)
Term life annuity ax:n⌉ = äx:n+1/m⌉ − 1/m
p. 102

Endowment Ax:n1⌉ = n Ex = v n n px
p. 102

A(m)
x − A(m)1x:n⌉ = v n n px · Ax+n
p. 103

(m)
Ax:n⌉ = A(m)1x:n⌉ + A(m) x:n1⌉ = A(m)1x:n⌉ +n Ex
p. 104

 1 − v min(Tm −x+1/m, n)  (m)

(m) 1 − Ax:n⌉
äx:n⌉ = Ex =
d(m) d(m)
p. 104

(m) (m)
d(m) äx:n⌉ + Ax:n⌉ = 1
p. 104

nm−1
X
1 Tm −x+1/m
v (k+1)/m


Ax:n⌉ = Ex v = k/m px 1/m qx+k/m
k=0
p. 105
124 CHAPTER 4. EXPECTED PRESENT VALUES OF PAYMENTS

n−1
X
1
Ax:n⌉ = v k+1 k px qx+k
k=0

p. 107

n−1
X
äx:n⌉ = v k k px
k=0

p. 107
 
Ax:n1⌉ = Ex v n I[T −x≥n] = v n n px
p. 107

n−1
X
Ax:n⌉ = v k+1 (k px − k+1 px ) + v n n px
k=0

p. 107
Chapter 5

Premium Calculation

This Chapter treats the most important topics related to the calculation
of (risk) premiums for realistic insurance and annuity contracts. We be-
gin by considering at length net single premium formulas for insurance and
annuities, under each of three standard assumptions on interpolation of the
survival function between integer ages, when there are multiple payments per
year. One topic covered more rigorously here than elsewhere is the calculus-
based and numerical comparison between premiums under these slightly dif-
ferent interpolation assumptions, justifying the standard use of the simplest
of the interpolation assumptions, that deaths occur uniformly within whole
years of attained age. Next we introduce the idea of calculating level premi-
ums, setting up equations balancing the stream of level premium payments
coming in to an insurer with the payout under an insurance, endowment, or
annuity contract. Finally, we discuss single and level premium calculation
for insurance contracts where the death benefit is modified by (fractional)
premium amounts, either as refunds or as amounts still due. Here the is-
sue is first of all to write an exact balance equation, then load it appropri-
ately to take account of administrative expenses and the cushion required for
insurance-company profitability, and only then to approximate and obtain
the usual formulas.

125
126 CHAPTER 5. PREMIUM CALCULATION

5.1 m-Payment Net Single Premiums

The objective in this section is to relate the formulas for net single premiums
for life insurance, life annuities, pure endowments and endowment insurances
in the case where there are multiple payment periods per year to the case
where there is just one. Of course, we must now make some interpolation
assumptions about within-year survival in order to do this, and we consider
the three main assumptions previously introduced: piecewise uniform failure
distribution (constant failure density within each year), piecewise exponen-
tial failure distribution (constant force of mortality within each year), and
Balducci assumption. As a practical matter, it usually makes relatively lit-
tle difference which of these is chosen, as we have seen in exercises and will
illustrate further in analytical approximations and numerical tabulations.
However, of the three assumptions, Balducci’s is least important practically,
because of the remark that the force of mortality it induces within years is
actually decreasing (the reciprocal of a linear function with positive slope),
since formula (3.9) gives it under that assumption as
d qx
µ(x + t) = − ln S(x + t) =
dt 1 − (1 − t) qx
Thus the inclusion of the Balducci assumption here is for completeness only,
since it is a recurring topic for examination questions. However, we do not
give separate net single premium formulas for the Balducci case.
In order to display simple formulas, and emphasize closed-form relation-
ships between the net single premiums with and without multiple payments
per year, we adopt a further restriction throughout this Section, namely that
the duration n of the life insurance or annuity is an integer even though
m > 1. There is in principle no reason why all of the formulas cannot be
extended, one by one, to the case where n is assumed only to be an integer
multiple of 1/m, but the formulas are less simple that way.

5.1.1 Dependence Between Integer & Fractional Ages

at Death

One of the clearest ways to distinguish the three interpolation assumptions

is through the probabilistic relationship they impose between the greatest-
5.1. M-PAYMENT NET SINGLE PREMIUMS 127

integer [T ] or attained integer age at death and the fractional age T − [T ]

at death. The first of these is a discrete, nonnegative-integer-valued random
variable, and the second is a continuous random variable with a density on
the time-interval [0, 1). In general, the dependence between these random
variables can be summarized through the calculated joint probability
x+k+t
f (y)
Z
P ([T ] = x + k, T − [T ] < t | T ≥ x) = dy = t qx+k k px (5.1)
x+k S(x)

where k, x are integers and 0 ≤ t < 1. From this we deduce the following
formula (for k ≥ 0) by dividing the formula (5.1) for general t by the
corresponding formula at t = 1 :
t qx+k
P ( T − [T ] ≤ t | [T ] = x + k) = (5.2)
qx+k

where we have used the fact that T − [T ] < 1 with certainty.

In case (i) from Section 3.2, with the density f assumed piecewise
constant, we already know that t qx+k = t qx+k , from which formula (5.2)
immediately implies

P ( T − [T ] ≤ t | [T ] = x + k) = t

In other words, given complete information about the age at death, the
fractional age at death is always uniformly distributed between 0, 1. Since
the conditional probability does not involve the age at death, we say under
the interpolation assumption (i) that the fractional age and whole-year age
at death are independent as random variables.
In case (ii), with piecewise constant force of mortality, we know that

t qx+k = 1 − t px+k = 1 − e−µ(x+k) t

and it is no longer true that fractional and attained ages at death are inde-
pendent except in the very special (completely artificial) case where µ(x+k)
has the same constant value µ for all x, k. In the latter case, where T
is an exponential random variable, it is easy to check from (5.2) that

1 − e−µt
P ( T − [T ] ≤ t | [T ] = x + k) =
1 − e−µ
128 CHAPTER 5. PREMIUM CALCULATION

In that case, T − [T ] is indeed independent of [T ] and has a truncated

exponential distribution on [0, 1), while [T ] has the Geometric(1 − e−µ )
distribution given, according to (5.1), by

P ([T ] = x + k | T ≥ x) = (1 − e−µ )(e−µ )k

In case (iii), under the Balducci assumption, formula (3.8) says that
1−t qx+t = (1 − t) qx , which leads to a special formula for (5.2) but not
a conclusion of conditional independence. The formula comes from the cal-
culation
px+k
(1 − t) qx+k = (1−t) qx+k+t = 1 −
t px+k

leading to
px+k t qx+k
t qx+k = 1 − t px+k = 1 − =
1 − (1 − t) qx+k 1 − (1 − t) qx+k
Thus Balducci implies via (5.2) that
t
P ( T − [T ] ≤ t | [T ] = x + k) =
1 − (1 − t) qx+k

5.1.2 Net Single Premium Formulas — Case (i)

In this setting, the formula (4.15) for insurance net single premium is simpler
than (4.16) for life annuities, because
1
j/m px+k − (j+1)/m px+k = qx+k
m
Here and throughout the rest of this and the following two subsections,
x, k, j are integers and 0 ≤ j < m, and k + mj will index the possi-
ble values for the last multiple Tm − x of 1/m year of the policy age at
death. The formula for net single insurance premium becomes especially
simple when n is an integer, because the double sum over j and k factors
into the product of a sum of terms depending only on j and one depending
only on k :
n−1 m−1
X X 1
A(m)1x:n⌉ = v k+(j+1)/m qx+k k px
k=0 j=0
m
5.1. M-PAYMENT NET SINGLE PREMIUMS 129

n−1
! m−1
X
k+1 v −1+1/m X j/m (m)
= v qx+k k px v = A1x:n⌉ v −1+1/m ä1⌉
k=0
m j=0

1 1−v
−1+1/m i 1
= Ax:n⌉ v (m)
= (m) Ax:n ⌉ (5.3)
d i
The corresponding formula for the case of non-integer n can clearly be
written down in a similar way, but does not bear such a simple relation to
the one-payment-per-year net single premium.
The formulas for life annuities should not be re-derived in this setting but
rather obtained using the general identity connecting endowment insurances
with life annuities. Recall that in the case of integer n the net single premium
for a pure n-year endowment does not depend upon m and is given by
Ax:n1⌉ = n px vn
Thus we continue by displaying the net single premium for an endowment
insurance, related in the m-payment-period-per year case to the formula with
single end-of-year payments:
(m) i
Ax:n⌉ = A(m)1x:n⌉ + Ax:n1⌉ = 1
Ax:n⌉ + n px v
n
(5.4)
i(m)

As a result of (4.11), we obtain the formula for net single premium of a

temporary life-annuity due:
(m)
(m) 1 − Ax:n⌉ 1 h i 1 n
i
äx:n⌉ = = (m) 1 − (m) Ax:n⌉ − n px v
d(m) d i
Re-expressing this formula in terms of annuities on the right-hand side, using
äx:n⌉ = d−1 (1 − v n n px − A1x:n⌉ ), immediately yields
1 − v n n px
 
(m) di i
äx:n⌉ = (m) (m) äx:n⌉ + 1 − (m) (5.5)
d i i d(m)
The last formula has the form that the life-annuity due with m payments
per year is a weighted linear combination of the life-annuity due with a single
payment per year, the n-year pure endowment, and a constant, where the
weights and constant depend only on interest rates and m but not on
survival probabilities:
(m)
äx:n⌉ = α(m) äx:n⌉ − β(m) (1 − n px v n )
= α(m) äx:n⌉ − β(m) + β(m) Ax:n1⌉ (5.6)
130 CHAPTER 5. PREMIUM CALCULATION

Table 5.1: Values of α(m), β(m) for Selected m, i

i m= 2 3 4 6 12

0.03
α(m) 1.0001 1.0001 1.0001 1.0001 1.0001
β(m) 0.2537 0.3377 0.3796 0.4215 0.4633
0.05
α(m) 1.0002 1.0002 1.0002 1.0002 1.0002
β(m) 0.2562 0.3406 0.3827 0.4247 0.4665
0.07
α(m) 1.0003 1.0003 1.0004 1.0004 1.0004
β(m) 0.2586 0.3435 0.3858 0.4278 0.4697
0.08
α(m) 1.0004 1.0004 1.0005 1.0005 1.0005
β(m) 0.2598 0.3450 0.3873 0.4294 0.4713
0.10
α(m) 1.0006 1.0007 1.0007 1.0007 1.0008
β(m) 0.2622 0.3478 0.3902 0.4325 0.4745

Here the interest-rate related constants α(m), β(m) are given by

di i − i(m)
α(m) = (m) (m) , β(m) = (m) (m)
d i d i
Their values for some practically interesting values of m, i are given in
(m)
Table 5.1. Note that α(1) = 1, β(1) = 0, reflecting that äx:n⌉ coincides
with äx:n⌉ by definition when m = 1. The limiting case for i = 0 is given
in Exercises 6 and 7:
m−1
for i = 0 , m ≥ 1 , α(m) = 1 , β(m) =
2m

Equations (5.3), (5.5), and (5.6) are useful because they summarize con-
cisely the modification needed for one-payment-per-year formulas (which
used only life-table and interest-rate-related quantities) to accommodate mul-
tiple payment-periods per year. Let us specialize them to cases where either
5.1. M-PAYMENT NET SINGLE PREMIUMS 131

the duration n, the number of payment-periods m, or both approach ∞.

Recall that failures continue to be assumed uniformly distributed within years
of age.
Consider first the case where the insurances and life-annuities are whole-
life, with n = ∞. The net single premium formulas for insurance and life
annuity due reduce to
i
A(m)
x = Ax , ä(m)
x = α(m) äx − β(m)
i(m)
Next consider the case where n is again allowed to be finite, but where
m is taken to go to ∞, or in other words, the payments are taken to be
instantaneous. Recall that both i(m) and d(m) tend in the limit to the
force-of-interest δ, so that the limits of the constants α(m), β(m) are
respectively
di i−δ
α(∞) = 2 , β(∞) =
δ δ2
Recall also that the instantaneous-payment notations replace the superscripts
(m)
by an overbar. The single-premium formulas for instantaneous-payment
insurance and life-annuities due become:
1 i di i−δ
Ax:n⌉ = A1x:n⌉ , ax:n⌉ = 2 äx:n⌉ − (1 − v n n px )
δ δ δ2

5.1.3 Net Single Premium Formulas — Case (ii)

In this setting, where the force of mortality is constant within single years of
age, the formula for life-annuity net single premium is simpler than the one
for insurance, because for integers j, k ≥ 0,
k+j/m px = k px e−jµx+k /m
Again restrict attention to the case where n is a positive integer, and
calculate from first principles (as in 4.16)
n−1 m−1
X 1
(m)
X
äx:n⌉ = v k+j/m j/m px+k k px (5.7)
k=0 j=0
m
n−1 m−1 n−1
X X 1 X 1 − vpx+k
= v k k px (ve−µx+k )j/m = v k k px
k=0 j=0
m k=0
m(1 − (vpx+k )1/m )
132 CHAPTER 5. PREMIUM CALCULATION

where we have used the fact that when force of mortality is constant within
years, px+k = e−µx+k . In order to compare this formula with equation (5.5)
established under the assumption of uniform distribution of deaths within
years of policy age, we apply the first-order Taylor series approximation
about 0 for formula (5.7) with respect to the death-rates qx+k inside
the denominator-expression 1 − (vpx+k )1/m = 1 − (v − vqx+k )1/m . (These
annual death-rates qx+k are actually small over a large range of ages for U.S.
life tables.) The final expression in (5.7) will be Taylor-approximated in a
slightly modified form: the numerator and denominator are both multiplied
by the factor 1 − v 1/m , and the term

(1 − v 1/m )/(1 − (vpx+k )1/m )

will be analyzed first. The first-order Taylor-series approximation about

z = 1 for the function (1 − v 1/m )/(1 − (vz)1/m ) is

1 − v 1/m h v 1/m (1 − v 1/m ) z −1+1/m i

≈ 1 − (1 − z)
1 − (vz)1/m m (1 − (vz)1/m )2 z=1
1/m
v 1−z
= 1 − (1 − z) 1/m
= 1 − (m)
m (1 − v ) i

Evaluating this Taylor-series approximation at z = px+k = 1 − qx+k then

yields
1 − v 1/m qx+k
1/m
≈ 1 − (m)
1 − (vpx+k ) i

Substituting this final approximate expression into equation (5.7), with

numerator and denominator both multiplied by 1 − v 1/m , we find for
piecewise-constant force of mortality which is assumed small
n−1
(m)
X 1 − vpx+k
äx:n⌉ ≈ v k k px (1 − qx+k /i(m) )
k=0
m(1 − v 1/m )
n−1
X
k 1 n 1−v o
≈ v k px 1 − vpx+k − (m) qx+k (5.8)
k=0
d(m) i

where in the last line we have applied the identity m(1 − v 1/m ) = d(m) and
discarded a quadratic term in qx+k within the large curly bracket.
5.1. M-PAYMENT NET SINGLE PREMIUMS 133

We are now close to our final objective: proving that the formulas (5.5)
and (5.6) of the previous subsection are in the present setting still valid as
approximate formulas. Indeed, we now prove that the final expression (5.8)
is precisely equal to the right-hand side of formula (5.6). The interest of
this result is that (5.6) applied to piecewise-uniform mortality (Case (i)),
while we are presently operating under the assumption of piecewise-constant
hazards (Case ii). The proof of our assertion requires us to apply simple
identities in several steps. First, observe that (5.8) is equal by definition to

1 h 1−v 1 i
äx:n⌉ − ax:n⌉ − v −1 A (5.9)
d(m) i(m) x:n⌉
Second, apply the general formula for äx:n⌉ as a sum to check the identity

n−1
X
äx:n⌉ = v k k px = 1 − v n n px + ax:n⌉ (5.10)
k=0

and third, recall the identity

1 
äx:n⌉ = 1 − A1x:n⌉ − v n n px (5.11)
d
Substitute the identities (5.10) and (5.11) into expression (5.9) to re-express
the latter as
1 h n i n
i
1 − v p
n x − (1 − v p
n x − d äx:n⌉ )
d(m) i(m)
di 1 i
= (m) (m) äx:n⌉ + (m) (1 − v n n px ) (1 − (m) ) (5.12)
d i d i
The proof is completed by remarking that (5.12) coincides with expression
(5.6) in the previous subsection.
Since formulas for the insurance and life annuity net single premiums can
each be used to obtain the other when there are m payments per year, and
since in the case of integer n, the pure endowment single premium Ax:n1⌉
does not depend upon m, it follows from the result of this section that all
of the formulas derived in the previous section for case (i) can be used as
approximate formulas (to first order in the death-rates qx+k ) also in case
(ii).
134 CHAPTER 5. PREMIUM CALCULATION

5.2 Approximate Formulas via Case(i)

The previous Section developed a Taylor-series justification for using the very
convenient net-single-premium formulas derived in case (i) (of uniform distri-
bution of deaths within whole years of age) to approximate the corresponding
formulas in case (ii) (constant force of mortality within whole years of age.
The approximation was derived as a first-order Taylor series, up to linear
terms in qx+k . However, some care is needed in interpreting the result,
because for this step of the approximation to be accurate, the year-by-year
death-rates qx+k must be small compared to the nominal rate of interest
i(m) . While this may be roughly valid at ages 15 to 50, at least in developed
countries, this is definitely not the case, even roughly, at ages larger than
around 55.
Accordingly, it is interesting to compare numerically, under several as-
sumed death- and interest- rates, the individual terms A(m)1x:k+1⌉ − A(m)1x:k⌉
which arise as summands under the different interpolation assumptions. (Here
and throughout this Section, k is an integer.) We first recall the formulas for
cases (i) and (ii), and for completeness supply also the formula for case (iii)
(the Balducci interpolation assumption). Recall that Balducci’s assumption
was previously faulted both for complexity of premium formulas and lack
of realism, because of its consequence that the force of mortality decreases
within whole years of age. The following three formulas are exactly valid
under the interpolation assumptions of cases (i), (ii), and (iii) respectively.

i
A(m)1x:k+1⌉ − A(m)1x:k⌉ = v k+1 k px · qx+k (5.13)
i(m)

1/m i + qx+k
A(m)1x:k+1⌉ − A(m)1x:k⌉ = v k+1 k px (1 − px+k ) 1/m
(5.14)
1 + (i(m) /m) − px+k
m−1
(m)1 (m)1 k+1
X px+k v −j/m
A x:k+1⌉ − A = v k px qx+k
m (1 − j+1 j
x:k⌉
j=0 m
qx+k ) (1 − m
qx+k )
(5.15)
Formula (5.13) is an immediate consequence of the formula A(m)1x:n⌉ =
1 (m)
i Ax:n⌉ /i derived in the previous section. To prove (5.14), assume (ii) and
calculate from first principles and the identities v −1/m = 1 + i(m) /m and
5.2. APPROXIMATE FORMULAS VIA CASE(I) 135

px+k = exp(−µx+k ) that

m−1
X
v k+(j+1)/m k px ( j/m px+k − (j+1)/m px+k )
j=0
m−1
X
k+1 −1+1/m −µx+k /m
= v k px v (1 − e ) (v e−µx+k )j/m
j=0

1 − v px+k v −1
= v k+1 k px (1 − e−µx+k /m ) ·
1 − (vpx+k )1/m v −1/m
i + qx+k
= v k+1 k px (1 − e−µx+k /m ) 1/m
1 + i(m) /m − px+k

Finally, for the Balducci case, (5.15) is established by calculating first

px+k px+k
j/m px+k = =
1 − 1−j/m qx+k+j/m 1 − m−j
m
qx+k

Then the left-hand side of (5.15) is equal to

m−1
X
v k+(j+1)/m k px ( j/m px+k − (j+1)/m px+k )
j=0
m−1
k+1 −1+1/m
X px+k v j/m
= v k px qx+k v
j=0
m (1 − m−j q ) (1 −
m x+k
m−j−1
m
qx+k )

which is seen to be equal to the right-hand side of (5.15) after the change of
summation-index j ′ = m − j − 1.
Formulas (5.13), (5.14), and (5.15) are progressively more complicated,
and it would be very desirable to stop with the first one if the choice of
interpolation assumption actually made no difference. In preparing the fol-
lowing Table, the ratios both of formulas (5.14)/(5.13) and of (5.15)/(5.13)
were calculated for a range of possible death-rates q = qx+k , interest-rates
i, and payment-periods-per-year m. We do not tabulate the results for
the ratios (5.14)/(5.13) because these ratios were equal to 1 to three decimal
places except in the following cases: the ratio was 1.001 when i ranged
from 0.05 to 0.12 and q = 0.15 or when i was .12 or .15 and q was .12,
achieving a value of 1.002 only in the cases where q = i = 0.15, m ≥ 4.
136 CHAPTER 5. PREMIUM CALCULATION

Such remarkable correspondence between the net single premium formulas

in cases (i), (ii) was by no means guaranteed by the previous Taylor series
calculation, and is made only somewhat less surprising by the remark that
the ratio of formulas (5.14)/(5.13) is smooth in both parameters qx+k , i and
exactly equal to 1 when either of these parameters is 0.
The Table shows a bit more variety in the ratios of (5.15)/(5.13), showing
in part why the Balducci assumption is not much used in practice, but also
showing that for a large range of ages and interest rates it also gives correct
answers within 1 or 2 %. Here also there are many cases where the Balducci
formula (5.15) agrees extremely closely with the usual actuarial (case (i))
formula (5.13). This also can be partially justified through the observation
(a small exercise for the reader) that the ratio of the right-hand sides of
formulas (5.15) divided by (5.13) are identical in either of the two limiting
cases where i = 0 or where qx+k = 0. The Table shows that the deviations
from 1 of the ratio (5.15) divided by (5.13) are controlled by the parameter
m and the interest rate, with the death-rate much less important within the
broad range of values commonly encountered.

5.3 Net Level (Risk) Premiums

The general principle previously enunciated regarding equivalence of two dif-

ferent (certain) payment-streams if their present values are equal, has the
following extension to the case of uncertain (time-of-death-dependent) pay-
ment streams: two such payment streams are equivalent (in the sense of
having equal ‘risk premiums’) if their expected present values are equal. This
definition makes sense if each such equivalence is regarded as the matching
of random income and payout for the insurer with respect to each of a large
number of independent (and identical) policies. Then the Law of Large
Numbers has the interpretation that the actual random net payout minus
income for the aggregate of the policies per policy is with very high prob-
ability very close (percentagewise) to the mathematical expectation of the
difference between the single-policy payout and income. That is why, from
a pure-risk perspective, before allowing for administrative expenses and the
‘loading’ or cushion which an insurer needs to maintain a very tiny proba-
bility of going bankrupt after starting with a large but fixed fund of reserve
5.3. NET LEVEL PREMIUMS 137

Table 5.2: Ratios of Values (5.15)/(5.13)

qx+k i m= 2 m= 4 m= 12
.002 .03 1.015 1.007 1.002
.006 .03 1.015 1.007 1.002
.02 .03 1.015 1.008 1.003
.06 .03 1.015 1.008 1.003
.15 .03 1.015 1.008 1.003

.002 .05 1.025 1.012 1.004

.006 .05 1.025 1.012 1.004
.02 .05 1.025 1.012 1.004
.06 .05 1.025 1.013 1.005
.15 .05 1.026 1.014 1.005

.002 .07 1.034 1.017 1.006

.006 .07 1.034 1.017 1.006
.02 .07 1.035 1.017 1.006
.06 .07 1.035 1.018 1.006
.15 .07 1.036 1.019 1.007

.002 .10 1.049 1.024 1.008

.006 .10 1.049 1.024 1.008
.02 .10 1.049 1.024 1.008
.06 .10 1.050 1.025 1.009
.15 .10 1.051 1.027 1.011

.002 .12 1.058 1.029 1.010

.006 .12 1.058 1.029 1.010
.02 .12 1.059 1.029 1.010
.06 .12 1.059 1.030 1.011
.15 .12 1.061 1.032 1.013

.002 .15 1.072 1.036 1.012

.006 .15 1.072 1.036 1.012
.02 .15 1.073 1.036 1.012
.06 .15 1.074 1.037 1.013
.15 .15 1.075 1.039 1.016
138 CHAPTER 5. PREMIUM CALCULATION

capital, this expected difference should be set equal to 0 in figuring premi-

ums. The resulting rule for calculation of the premium amount P which
must multiply the unit amount in a specified payment pattern is as follows:

P = Expected present value of life insurance, annuity, or endow-

ment contract proceeds divided by the expected present value of
a unit amount paid regularly, according to the specified payment
pattern, until death or expiration of term.

5.4 Benefits Involving Fractional Premiums

The general principle for calculating risk premiums sets up a balance be-
tween expected payout by an insurer and expected payment stream received
as premiums. In the simplest case of level payment streams, the insurer re-
ceives a life-annuity due with level premium P , and pays out according
to the terms of the insurance product purchased, say a term insurance. If
the insurance purchased pays only at the end of the year of death, but the
premium payments are made m times per year, then the balance equation
becomes
(m)
A1x:n⌉ = P · m äx:n⌉
for which the solution P is called the level risk premium for a term insurance.
The reader should distinguish this premium from the level premium payable
m times yearly for an insurance which pays at the end of the (1/m)th year
of death. In the latter case, where the number of payment periods per year
for the premium agrees with that for the insurance, the balance equation is
(m)
A(m)1x:n⌉ = P · m äx:n⌉

In standard actuarial notations for premiums, not given here, level premiums
are annualized (which would result in the removal of a factor m from the
right-hand sides of the last two equations).

Two other applications of the balancing-equation principle can be made

in calculating level premiums for insurances which either (a) deduct the
additional premium payments for the remainder of the year of death from
the insurance proceeds, or (b) refund a pro-rata share of the premium for
5.4. BENEFITS INVOLVING FRACTIONAL PREMIUMS 139

the portion of the 1/m year of death from the instant of death to the
end of the 1/m year of death. Insurance contracts with provision (a) are
called insurances with installment premiums: the meaning of this term is
that the insurer views the full year’s premium as due at the beginning of the
year, but that for the convenience of the insured, payments are allowed to be
made in installments at m regularly spaced times in the year. Insurances
with provision (b) are said to have apportionable refund of premium, with
the implication that premiums are understood to cover only the period of
the year during which the insured is alive . First in case (a), the expected
amount paid out by the insurer, if each level premium payment is P and
the face amount of the policy is F (0), is equal to
n−1 m−1
X X
(m)1
F (0) A x:n⌉ − v k+(j+1)/m k+j/m px · 1/m qx+k+j/m (m − 1 − j) P
k=0 j=0

and the exact balance equation is obtained by setting this equal to the ex-
(m)
pected amount paid in, which is again P m äx:n⌉ . Under the interpolation
assumption of case (i), using the same reasoning which previously led to the
simplified formulas in that case, this balance equation becomes
m−1
(m)1 1 P X −(m−j−1)/m (m)
F (0) A x:n⌉ − Ax:n⌉ v (m − j − 1) = P m äx:n⌉ (5.16)
m j=0

Although one could base an exact calculation of P on this equation, a further

standard approximation leads to a simpler formula. If the term P (m − j − 1)
−1 m−1
is replaced in the final sum by its average over j, or by m j=0 (m −
j − 1) = m−1 (m − 1)m/2 = (m − 1)/2, we obtain the installment premium
formula
F (0) A(m)1x:n⌉
P = (m)
mäx:n⌉ + m−1
2
A(m)1x:n⌉
and this formula could be related using previous formulas derived in Section
5.1 to the insurance and annuity net single premiums with only one payment
period per year.
In the case of the apportionable return of premium, the only assumption
usually considered is that of case (i), that the fraction of a single premium
payment which will be returned is on average 1/2 regardless of which of
140 CHAPTER 5. PREMIUM CALCULATION

the 1/m fractions of the year contains the instant of death. The balance
equation is then very simple:
1 (m)
A(m)1x:n⌉ (F (0) + P ) = P m äx:n⌉ (5.17)
2
and this equation has the straightforward solution
F (0) A(m)1x:n⌉
P = (m) 1
m äx:n⌉ − 2
A(m)1x:n⌉

It remains only to remark what is the effect of loading for administrative

expenses and profit on insurance premium calculation. If all amounts paid
out by the insurer were equally loaded (i.e., multiplied) by the factor 1 +
L, then formula (5.17) would involve the loading in the second term of
the denominator, but this is apparently not the usual practice. In both
the apportionable refund and installment premium contracts, as well as the
insurance contracts which do not modify proceeds by premium fractions, it is
apparently the practice to load the level premiums P directly by the factor
1 + L, which can easily be seen to be equivalent to inflating the face-amount
F (0) in the balance-formulas by this factor.

5.5 Exercise Set 5

(1). Show from first principles that for all integers x, n, and all fixed
interest-rates and life-distributions

ax:n⌉ = äx:n⌉ − 1 + v n n px

(2). Show from first principles that for all integers x, and all fixed interest-
rates and life-distributions

Ax = v äx − ax

Show further that this relation is obtained by taking the expectation on both
sides of an identity in terms of present values of payment-streams, an identity
whoch holds for each value of (the greatest integer [T ] less than or equal
to) the exact-age-at-death random variable T .
5.5. EXERCISE SET 5 141

(3). Using the same idea as in problem (2), show that (for all x, n, interest
rates, and life-distributions)
1
Ax:n⌉ = v äx:n⌉ − ax:n⌉

(4). Suppose that a life aged x (precisely, where x is an integer) has

the survival probabilities px+k = 0.98 for k = 0, 1, , . . . , 9. Suppose that
he wants to purchase a term insurance which will pay $30,000 at the end of
the quarter-year of death if he dies within the first five years, and will pay
$10,000 (also at the end of the quarter-year of death) if he dies between exact
ages 5, 10. In both parts (a), (b) of the problem, assume that the interest
rate is fixed at 5%, and assume wherever necessary that the individual’s
distribution of death-time is uniform within each whole year of age.
(a) Find the net single premium of the insurance contract described.
(b) Suppose that the individual purchasing the insurance described
wants to pay level premiums semi-annually, beginning immediately. Find
the amount of each semi-annual payment.
(5). Re-do problem (4) assuming in place of the uniform distribution of age
at death that the insured individual has constant force of mortality within
each whole year of age. Give your numerical answers to at least 6 significant
figures so that you can compare the exact numerical answers in these two
problems.
(6). Using the exact expression for the interest-rate functions i(m) , d(m)
respectively as functions of i and d, expand these functions in Taylor
series about 0 up to quadratic terms. Use the resulting expressions to
approximate the coefficients α(m), β(m) which were derived in the Chapter.
Hence justify the so-called traditional approximation
m−1
ä(m)
x ≈ äx −
2m

(7). Justify the ‘traditional approximation’ (the displayed formula in Exer-

cise 6) as an exact formula in the case (i) in the limit i → 0, by filling in
the details of the following argument.
No matter which policy-year is the year of death of the annuitant, the
policy with m = 1 (and expected present value äx ) pays 1 at the beginning
142 CHAPTER 5. PREMIUM CALCULATION

of that year while the policy with m > 1 pays amounts 1/m at the beginning
of each 1/m’th year in which the annuitant is alive. Thus, the annuity with
one payment per years pays more than the annuity with m > 1 by an absolute
amount 1 − (Tm − [T ] + 1/m). Under assumption (i), Tm − [T ] is a discrete
random variable taking on the possible values 0, 1, . . . , (m − 1)/m each with
probability 1/m. Disregard the interest and present-value discounting on the
excess amount 1 − (Tm − [T ])/m paid by the m-payment-per year annuity,
and show that it is exactly (m − 1)/2m.
(8). Give an exact formula for the error of the ‘traditional approximation’
given in the previous problem, in terms of m, the constant interest rate i (or
v = (1 + i)−1 ), and the constant force µ of mortality, when the lifetime T
is assumed to be distributed precisely as an Exponential(µ) random variable.
(9). Show that the ratio of formulas (5.14)/(5.13) is 1 whenever either
qx+k or i is set equal to 0.
(10). Show that the ratio of formulas (5.15)/(5.13) is 1 whenever either
qx+k or i is set equal to 0.
(11). For a temporary life annuity on a life aged 57, with benefits deferred
for three years, you are given that µx = 0.04 is constant, δ = .06, that
premiums are paid continuously (with m = ∞) only for the first two years,
at rate P per year, and that the annuity benefits are payable at beginnings
of years according to the following schedule:

Year 0 1 2 3 4 5 6 7 8+
Benefit 0 0 0 10 8 6 4 2 0

(a) In terms of P , calculate the expected present value of the premiums

paid.
(b) Using the equivalence principle, calculate P numerically.
(12). You are given that (i) q60 = 0.3, q61 = 0.4, (ii) f denotes the
probability that a life aged 60 will die between ages 60.5 and 61.5 under the
assumption of uniform distribution of deaths within whole years of age, and
(iii) g denotes the probability that a life aged 60 will die between ages
60.5 and 61.5 under the Balducci assumption. Calculate 10, 000 · (g − f )
numerically, with accuracy to the nearest whole number.
5.5. EXERCISE SET 5 143

(13). You are given that S(40) = 0.500, S(41) = 0.475, i = 0.06, A41 =
0.54, and that deaths are uniformly distributed over each year of age. Find
A40 exactly.
(14). If a mortality table follows Gompertz’ law (with exponent c), prove
that .
µx = Ax a′x

where Ax is calculated at interest rate i while a′x is calculated at a rate

of interest i′ = 1+i
c
− 1.
(15). You are given that i = 0.10, qx = 0.05, and qx+1 = 0.08, and that
1
deaths are uniformly distributed over each year of age. Calculate Ax:2⌉ .
(16). A special life insurance policy to a life aged x provides that if death
occurs at any time within 15 years, then the only benefit is the return of
premiums with interest compounded to the end of the year of death. If death
occurs after 15 years, the benefit is $10, 000. In either case, the benefit
is paid at the end of the year of death. If the premiums for this policy are
to be paid yearly for only the first 5 years (starting at the time of issuance
of the policy), then find a simplified expression for the level annual pure-risk
premium for the policy, in terms of standard actuarial and interest functions.
(17). Prove that for every m, n, x, k, the net single premium for an
n-year term insurance for a life aged x, with benefit deferred for k years,
and payable at the end of the 1/m year of death is given by either side of
the identity
An+k
x⌉
m
− Akx⌉ m = k Ex Ax+k
nm
⌉

First prove the identity algebraically; then give an alternative, intuitive ex-
planation of why the right-hand side represents the expected present value
of the same contingent payment stream as the left-hand side.
144 CHAPTER 5. PREMIUM CALCULATION

5.6 Worked Examples

Overview of Premium Calculation for Single-Life Insurance & Annuities
Here is a schematic overview of the calculation of net single and level pre-
miums for life insurances and life annuities, based on life-table or theoretical
survival probabilities and constant interest or discount rate. We describe the
general situation and follow a specific case study/example throughout.
(I) First you will be given information about the constant assumed in-
terest rate in any of the equivalent forms i(m) , d(m) , or δ, and you should
immediately convert to find the effective annual interest rate (APR) i and
one-year discount factor v = 1/(1 + i). In our case-study, assume that

the force of interest δ is constant = − ln(0.94)

so that i = exp(δ) − 1 = (1/0.94) − 1 = 6/94, and v = 0.94. In terms

of this quantity, one immediately answers a question such as “what is the
present value of $1 at the end of 7 12 years ?” by: v 7.5 .
(II) Next you must be given either a theoretical survival function for
the random age at death of a life aged x, in any of the equivalent forms
S(x + t), t px , f (x + t), or µ(x + t), or a cohort-form life-table, e.g.,

l25 = 10,000 0 p25 = 1.0

l26 = 9,726 1 p25 = 0.9726
l27 = 9,443 2 p25 = 0.9443
l28 = 9,137 3 p25 = 0.9137
l29 = 8,818 4 p25 = 0.8818
l30 = 8,504 5 p25 = 0.8504

From such data, one calculates immediately that (for example) the probabil-
ity of dying at an odd attained-age between 25 and 30 inclusive is

(1 − 0.9726) + (0.9443 − 0.9137) + (0.8818 − 0.8504) = 0.0894

The generally useful additional column to compute is:

q25 = 1 − 1 p25 = 0.0274, 1 p25 − 2 p25 = 0.0283, 2 p25 − 3 p25 = 0.0306

5.6. WORKED EXAMPLES 145

3 p25 − 4 p25 = 0.0319, 4 p25 − 5 p25 = 0.0314

(III) In any problem, the terms of the life insurance or annuity to be

purchased will be specified, and you should re-express its present value in
terms of standard functions such as ax:n⌉ or A1x:n⌉ . For example, suppose a
life aged x purchases an endowment/annuity according to which he receives
$10,000 once a year starting at age x+1 until either death occurs or n years
have elapsed, and if he is alive at the end of n years he receives $15,000.
This contract is evidently a superposition of a n-year pure endowment with
face value $15,000 and a n-year temporary life annuity-immediate with
yearly payments $10,000. Thus, the expected present value (= net single
premium) is
10, 000 ax:n⌉ + 15, 000 n px v n
In our case-study example, this expected present value is
= 10000 0.94(0.9726) + 0.942 (0.9443) + 0.943 (0.9137) +
+ 0.944 (0.8818) + 0.945 (0.8504) + 15000(0.945 · 0.8504)

The annuity part of this net single premium is $38,201.09 , and the pure-
endowment part is $9,361.68 , for a total net single premium of $47,562.77
(IV) The final part of the premium computation problem is to specify
the type of payment stream with which the insured life intends to pay for
the contract whose expected present value has been figured in step (III). If
the payment is to be made at time 0 in one lump sum, then the net single
premium has already been figured and we are done. If the payments are to
be constant in amount (level premiums), once a year, to start immediately,
and to terminate at death or a maximum of n payments, then we divide
the net single premium by the expected present value of a unit life annuity
äx:n⌉ . In general, to find the premium we divide the net single premium of
(III) by the expected present value of a unit amount paid according to the
desired premium-payment stream.
In the case-study example, consider two cases. The first is that the pur-
chaser aged x wishes to pay in two equal installments, one at time 0 and
one after 3 years (with the second payment to be made only if he is alive at
that time). The expected present value of a unit amount paid in this fashion
is
1 + v 3 3 px = 1 + (0.94)3 0.9137 = 1.7589
146 CHAPTER 5. PREMIUM CALCULATION

Thus the premium amount to be paid at each payment time is

$ 47, 563 / 1.7589 = $27, 041
Alternatively, as a second example, suppose that the purchaser is in effect
taking out his annuity/endowment in the form of a loan, and agrees to (have
his estate) repay the loan unconditionally (i.e. without regard to the event
of his death) over a period of 29 years, with 25 equal payments to be made
every year beginning at the end of 5 years. In this case, no probabilities
are involved in valuing the payment stream, and the present value of such a
payment stream of unit amounts is
v 4 a25⌉ = (0.94)4 (.94/.06) (1 − (0.94)25 ) = 9.627
In this setting, the amount of each of the equal payments must be
$ 47, 563 / 9.627 = $ 4941

(V) To complete the circle of ideas given here, let us re-do the case-
study calculation of paragraphs (III) to cover the case where the insurance
has quarterly instead of annual payments. Throughout, assume that deaths
within years of attained age are uniformly distributed (case(i)).
First, the expected present value to find becomes
(4)

(4) 1 
10, 000 ax:n⌉ + 15, 000 Ax:n1⌉ = 10000 äx:n⌉ − (1 − v n n px ) + 15000 v n n px
4
which by virtue of (5.6) is equal to
= 10000 α(4) äx:n⌉ − (1 − v n n px ) (10000β(4) + 2500) + 15000 v n n px
In the particular case with v = 0.94, x = 25, n = 5, and cohort life-table
given in (II), the net single premium for the endowment part of the contract
has exactly the same value $9361.68 as before, while the annuity part now
has the value
10000 (1.0002991) (1 + 0.94(0.9726) + 0.942 (0.9443) + 0.943 (0.9137) +
+ 0.944 (0.8818)) − (6348.19) (1 − 0.945 (0.8504)) = 39586.31
Thus the combined present value is 48947.99: the increase of 1385 in value
arises mostly from the earlier annuity payments: consider that the interest
on the annuity value for one-half year is 38201(0.94−0.5 − 1) = 1200 .
5.7. USEFUL FORMULAS FROM CHAPTER 5 147

5.7 Useful Formulas from Chapter 5

x+k+t
f (y)
Z
P ([T ] = x + k, T − [T ] < t | T ≥ x) = dy = t qx+k k px
x+k S(x)
p. 127

t qx+k
P ( T − [T ] ≤ t | [T ] = x + k) =
qx+k
p. 127

i
A(m)1x:n⌉ = 1
Ax:n⌉ under (i)
i(m)
p. 129

(m)
(m) 1 − Ax:n⌉ 1 h i 1 n
i
äx:n⌉ = = 1 − A − p
n x v
d(m) d(m) i(m) x:n⌉
p. 129

1 − v n n px
 
(m) di i
äx:n⌉ = (m) (m) äx:n⌉ + 1− under (i)
d i i(m) d(m)

p. 129

(m)
äx:n⌉ = α(m) äx:n⌉ − β(m) (1 − n px v n ) under (i)
p. 129

di i − i(m)
α(m) = , β(m) =
d(m) i(m) d(m) i(m)
p. 130

m−1
α(m) = 1 , β(m) = when i = 0
2m
p. 130
148 CHAPTER 5. PREMIUM CALCULATION

n−1
(m)
X 1 − vpx+k
äx:n⌉ = v k k px under (ii)
k=0
m(1 − (vpx+k )1/m )
p. 131

i
A(m)1x:k+1⌉ − A(m)1x:k⌉ = v k+1 k px · qx+k under (i)
i(m)
p. 134

1/m i + qx+k
A(m)1x:k+1⌉ − A(m)1x:k⌉ = v k+1 k px (1 − px+k ) 1/m
under (ii)
1 + (i(m) /m) − px+k

p. 134

m−1
(m)1 (m)1 k+1
X px+k v −j/m
A x:k+1⌉ − A = v k px qx+k
m (1 − j+1 j
x:k⌉
j=0 m
qx+k ) (1 − m
qx+k )

under (iii) p. 134

F (0) A(m)1x:n⌉
Level Installment Risk Premium = (m) m−1
mäx:n⌉ + 2
A(m)1x:n⌉

p. 139

F (0) A(m)1x:n⌉
Apportionable Refund Risk Premium = (m) 1
m äx:n⌉ − 2
A(m)1x:n⌉

p. 140
176 CHAPTER 5. PREMIUM CALCULATION
Bibliography

[1] Bowers, N., Gerber, H., Hickman, J., Jones, D. and Nesbitt, C. Actu-
arial Mathematics Society of Actuaries, Itasca, Ill. 1986

[2] Cox, D. R. and Oakes, D. Analysis of Survival Data, Chapman and

Hall, London 1984

[3] Feller, W. An Introduction to Probability Theory and its Ap-

plications, vol. I, 2nd ed. Wiley, New York, 1957

[4] Feller, W. An Introduction to Probability Theory and its Ap-

plications, vol. II, 2nd ed. Wiley, New York, 1971

[5] Gerber, H. Life Insurance Mathematics, 3rd ed. Springer-Verlag,

New York, 1997

[6] Hogg, R. V. and Tanis, E. Probability and Statistical Inference,

5th ed. Prentice-Hall Simon & Schuster, New York, 1997

[7] Jordan, C. W. Life Contingencies, 2nd ed. Society of Actuaries,

Chicago, 1967

[8] Kalbfleisch, J. and Prentice, R. The Statistical Analysis of Failure

Time Data, Wiley, New York 1980

[9] Kellison, S. The Theory of Interest. Irwin, Homewood, Ill. 1970

[10] Larsen, R. and Marx, M. An Introduction to Probability and its

Applications. Prentice-Hall, Englewood Cliffs, NJ 1985

[11] Larson, H. Introduction to Probability Theory and Statistical

Inference, 3rd ed. Wiley, New York, 1982

177
178 BIBLIOGRAPHY

[12] Lee, E. T. Statistical Models for Survival Data Analysis, Lifetime

Learning, Belmont Calif. 1980

[13] The R Development Core Team, R: a Language and Environment.

[14] Splus, version 7.0. MathSoft Inc., 1988, 1996, 2005

[15] Spiegelman, M. Introduction to Demography, Revised ed. Univ. of

Chicago, Chicago, 1968

[16] Venables, W. and Ripley, B. Modern Applied Statistics with S, 4th

ed. Springer-Verlag, New York, 2002